;;- Machine description for Renesas / SuperH SH.
;; Copyright (C) 1993-2014 Free Software Foundation, Inc.
;; Contributed by Steve Chamberlain (sac@cygnus.com).
;; Improved by Jim Wilson (wilson@cygnus.com).
;; This file is part of GCC.
;; GCC is free software; you can redistribute it and/or modify
;; it under the terms of the GNU General Public License as published by
;; the Free Software Foundation; either version 3, or (at your option)
;; any later version.
;; GCC is distributed in the hope that it will be useful,
;; but WITHOUT ANY WARRANTY; without even the implied warranty of
;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
;; GNU General Public License for more details.
;; You should have received a copy of the GNU General Public License
;; along with GCC; see the file COPYING3. If not see
;; .
;; ??? Should prepend a * to all pattern names which are not used.
;; This will make the compiler smaller, and rebuilds after changes faster.
;; ??? Should be enhanced to include support for many more GNU superoptimizer
;; sequences. Especially the sequences for arithmetic right shifts.
;; ??? Should check all DImode patterns for consistency and usefulness.
;; ??? The MAC.W and MAC.L instructions are not supported. There is no
;; way to generate them.
;; BSR is not generated by the compiler proper, but when relaxing, it
;; generates .uses pseudo-ops that allow linker relaxation to create
;; BSR. This is actually implemented in bfd/{coff,elf32}-sh.c
;; Special constraints for SH machine description:
;;
;; t -- T
;; x -- mac
;; l -- pr
;; z -- r0
;;
;; Special formats used for outputting SH instructions:
;;
;; %. -- print a .s if insn needs delay slot
;; %@ -- print rte/rts if is/isn't an interrupt function
;; %# -- output a nop if there is nothing to put in the delay slot
;; %O -- print a constant without the #
;; %R -- print the lsw reg of a double
;; %S -- print the msw reg of a double
;; %T -- print next word of a double REG or MEM
;;
;; Special predicates:
;;
;; arith_operand -- operand is valid source for arithmetic op
;; arith_reg_operand -- operand is valid register for arithmetic op
;; general_movdst_operand -- operand is valid move destination
;; general_movsrc_operand -- operand is valid move source
;; logical_operand -- operand is valid source for logical op
;; -------------------------------------------------------------------------
;; Constants
;; -------------------------------------------------------------------------
(define_constants [
(AP_REG 145)
(PR_REG 146)
(T_REG 147)
(GBR_REG 144)
(MACH_REG 148)
(MACL_REG 149)
(FPUL_REG 150)
(RAP_REG 152)
(FPSCR_REG 151)
;; Virtual FPSCR - bits that are used by FP ops.
(FPSCR_MODES_REG 154)
;; Virtual FPSCR - bits that are updated by FP ops.
(FPSCR_STAT_REG 155)
(PIC_REG 12)
(FP_REG 14)
(SP_REG 15)
(PR_MEDIA_REG 18)
(T_MEDIA_REG 19)
(R0_REG 0)
(R1_REG 1)
(R2_REG 2)
(R3_REG 3)
(R4_REG 4)
(R5_REG 5)
(R6_REG 6)
(R7_REG 7)
(R8_REG 8)
(R9_REG 9)
(R10_REG 10)
(R20_REG 20)
(R21_REG 21)
(R22_REG 22)
(R23_REG 23)
(DR0_REG 64)
(DR2_REG 66)
(DR4_REG 68)
(FR23_REG 87)
(TR0_REG 128)
(TR1_REG 129)
(TR2_REG 130)
(XD0_REG 136)
(FPSCR_PR 524288) ;; 1 << 19
(FPSCR_SZ 1048576) ;; 1 << 20
(FPSCR_FR 2097152) ;; 1 << 21
])
(define_c_enum "unspec" [
;; These are used with unspec.
UNSPEC_COMPACT_ARGS
UNSPEC_MOVA
UNSPEC_CASESI
UNSPEC_DATALABEL
UNSPEC_BBR
UNSPEC_SFUNC
UNSPEC_PIC
UNSPEC_GOT
UNSPEC_GOTOFF
UNSPEC_PLT
UNSPEC_CALLER
UNSPEC_GOTPLT
UNSPEC_ICACHE
UNSPEC_INIT_TRAMP
UNSPEC_FCOSA
UNSPEC_FSRRA
UNSPEC_FSINA
UNSPEC_NSB
UNSPEC_ALLOCO
UNSPEC_TLSGD
UNSPEC_TLSLDM
UNSPEC_TLSIE
UNSPEC_DTPOFF
UNSPEC_GOTTPOFF
UNSPEC_TPOFF
UNSPEC_RA
UNSPEC_DIV_INV_M0
UNSPEC_DIV_INV_M1
UNSPEC_DIV_INV_M2
UNSPEC_DIV_INV_M3
UNSPEC_DIV_INV20
UNSPEC_DIV_INV_TABLE
UNSPEC_ASHIFTRT
UNSPEC_THUNK
UNSPEC_CHKADD
UNSPEC_SP_SET
UNSPEC_SP_TEST
UNSPEC_MOVUA
;; (unspec [VAL SHIFT] UNSPEC_EXTRACT_S16) computes (short) (VAL >> SHIFT).
;; UNSPEC_EXTRACT_U16 is the unsigned equivalent.
UNSPEC_EXTRACT_S16
UNSPEC_EXTRACT_U16
;; (unspec [TARGET ANCHOR] UNSPEC_SYMOFF) == TARGET - ANCHOR.
UNSPEC_SYMOFF
;; (unspec [OFFSET ANCHOR] UNSPEC_PCREL_SYMOFF) == OFFSET - (ANCHOR - .).
UNSPEC_PCREL_SYMOFF
;; Misc builtins
UNSPEC_BUILTIN_STRLEN
])
(define_c_enum "unspecv" [
;; These are used with unspec_volatile.
UNSPECV_BLOCKAGE
UNSPECV_ALIGN
UNSPECV_CONST2
UNSPECV_CONST4
UNSPECV_CONST8
UNSPECV_WINDOW_END
UNSPECV_CONST_END
UNSPECV_EH_RETURN
UNSPECV_GBR
UNSPECV_SP_SWITCH_B
UNSPECV_SP_SWITCH_E
UNSPECV_FPSCR_MODES
UNSPECV_FPSCR_STAT
])
;; -------------------------------------------------------------------------
;; Attributes
;; -------------------------------------------------------------------------
;; Target CPU.
(define_attr "cpu"
"sh1,sh2,sh2e,sh2a,sh3,sh3e,sh4,sh4a,sh5"
(const (symbol_ref "sh_cpu_attr")))
(define_attr "endian" "big,little"
(const (if_then_else (symbol_ref "TARGET_LITTLE_ENDIAN")
(const_string "little") (const_string "big"))))
;; Indicate if the default fpu mode is single precision.
(define_attr "fpu_single" "yes,no"
(const (if_then_else (symbol_ref "TARGET_FPU_SINGLE")
(const_string "yes") (const_string "no"))))
(define_attr "fmovd" "yes,no"
(const (if_then_else (symbol_ref "TARGET_FMOVD")
(const_string "yes") (const_string "no"))))
;; pipeline model
(define_attr "pipe_model" "sh1,sh4,sh5media"
(const
(cond [(symbol_ref "TARGET_SHMEDIA") (const_string "sh5media")
(symbol_ref "TARGET_SUPERSCALAR") (const_string "sh4")]
(const_string "sh1"))))
;; cbranch conditional branch instructions
;; jump unconditional jumps
;; arith ordinary arithmetic
;; arith3 a compound insn that behaves similarly to a sequence of
;; three insns of type arith
;; arith3b like above, but might end with a redirected branch
;; load from memory
;; load_si Likewise, SImode variant for general register.
;; fload Likewise, but load to fp register.
;; store to memory
;; fstore floating point register to memory
;; move general purpose register to register
;; movi8 8-bit immediate to general purpose register
;; mt_group other sh4 mt instructions
;; fmove register to register, floating point
;; smpy word precision integer multiply
;; dmpy longword or doublelongword precision integer multiply
;; return rts
;; pload load of pr reg, which can't be put into delay slot of rts
;; prset copy register to pr reg, ditto
;; pstore store of pr reg, which can't be put into delay slot of jsr
;; prget copy pr to register, ditto
;; pcload pc relative load of constant value
;; pcfload Likewise, but load to fp register.
;; pcload_si Likewise, SImode variant for general register.
;; rte return from exception
;; sfunc special function call with known used registers
;; call function call
;; fp floating point
;; fpscr_toggle toggle a bit in the fpscr
;; fdiv floating point divide (or square root)
;; gp_fpul move from general purpose register to fpul
;; fpul_gp move from fpul to general purpose register
;; mac_gp move from mac[lh] to general purpose register
;; gp_mac move from general purpose register to mac[lh]
;; mac_mem move from mac[lh] to memory
;; mem_mac move from memory to mac[lh]
;; dfp_arith,dfp_mul, fp_cmp,dfp_cmp,dfp_conv
;; ftrc_s fix_truncsfsi2_i4
;; dfdiv double precision floating point divide (or square root)
;; cwb ic_invalidate_line_i
;; movua SH4a unaligned load
;; fsrra square root reciprocal approximate
;; fsca sine and cosine approximate
;; tls_load load TLS related address
;; arith_media SHmedia arithmetic, logical, and shift instructions
;; cbranch_media SHmedia conditional branch instructions
;; cmp_media SHmedia compare instructions
;; dfdiv_media SHmedia double precision divide and square root
;; dfmul_media SHmedia double precision multiply instruction
;; dfparith_media SHmedia double precision floating point arithmetic
;; dfpconv_media SHmedia double precision floating point conversions
;; dmpy_media SHmedia longword multiply
;; fcmp_media SHmedia floating point compare instructions
;; fdiv_media SHmedia single precision divide and square root
;; fload_media SHmedia floating point register load instructions
;; fmove_media SHmedia floating point register moves (inc. fabs and fneg)
;; fparith_media SHmedia single precision floating point arithmetic
;; fpconv_media SHmedia single precision floating point conversions
;; fstore_media SHmedia floating point register store instructions
;; gettr_media SHmedia gettr instruction
;; invalidate_line_media SHmedia invalidate_line sequence
;; jump_media SHmedia unconditional branch instructions
;; load_media SHmedia general register load instructions
;; pt_media SHmedia pt instruction (expanded by assembler)
;; ptabs_media SHmedia ptabs instruction
;; store_media SHmedia general register store instructions
;; mcmp_media SHmedia multimedia compare, absolute, saturating ops
;; mac_media SHmedia mac-style fixed point operations
;; d2mpy_media SHmedia: two 32-bit integer multiplies
;; atrans_media SHmedia approximate transcendental functions
;; ustore_media SHmedia unaligned stores
;; nil no-op move, will be deleted.
(define_attr "type"
"mt_group,cbranch,jump,jump_ind,arith,arith3,arith3b,dyn_shift,load,load_si,
fload,store,fstore,move,movi8,fmove,smpy,dmpy,return,pload,prset,pstore,
prget,pcload,pcload_si,pcfload,rte,sfunc,call,fp,fpscr_toggle,fdiv,ftrc_s,
dfp_arith,dfp_mul,fp_cmp,dfp_cmp,dfp_conv,dfdiv,gp_fpul,fpul_gp,mac_gp,
gp_mac,mac_mem,mem_mac,mem_fpscr,gp_fpscr,cwb,movua,fsrra,fsca,tls_load,
arith_media,cbranch_media,cmp_media,dfdiv_media,dfmul_media,dfparith_media,
dfpconv_media,dmpy_media,fcmp_media,fdiv_media,fload_media,fmove_media,
fparith_media,fpconv_media,fstore_media,gettr_media,invalidate_line_media,
jump_media,load_media,pt_media,ptabs_media,store_media,mcmp_media,mac_media,
d2mpy_media,atrans_media,ustore_media,nil,other"
(const_string "other"))
;; We define a new attribute namely "insn_class".We use
;; this for the DFA based pipeline description.
;;
;; mt_group SH4 "mt" group instructions.
;;
;; ex_group SH4 "ex" group instructions.
;;
;; ls_group SH4 "ls" group instructions.
;;
(define_attr "insn_class"
"mt_group,ex_group,ls_group,br_group,fe_group,co_group,none"
(cond [(eq_attr "type" "move,mt_group") (const_string "mt_group")
(eq_attr "type" "movi8,arith,dyn_shift") (const_string "ex_group")
(eq_attr "type" "fmove,load,pcload,load_si,pcload_si,fload,pcfload,
store,fstore,gp_fpul,fpul_gp") (const_string "ls_group")
(eq_attr "type" "cbranch,jump") (const_string "br_group")
(eq_attr "type" "fp,fp_cmp,fdiv,ftrc_s,dfp_arith,dfp_mul,dfp_conv,dfdiv")
(const_string "fe_group")
(eq_attr "type" "jump_ind,smpy,dmpy,mac_gp,return,pload,prset,pstore,
prget,rte,sfunc,call,dfp_cmp,mem_fpscr,gp_fpscr,cwb,
gp_mac,mac_mem,mem_mac") (const_string "co_group")]
(const_string "none")))
;; nil are zero instructions, and arith3 / arith3b are multiple instructions,
;; so these do not belong in an insn group, although they are modeled
;; with their own define_insn_reservations.
;; Indicate what precision must be selected in fpscr for this insn, if any.
(define_attr "fp_mode" "single,double,none" (const_string "none"))
;; Indicate if the fpu mode is set by this instruction
;; "unknown" must have the value as "none" in fp_mode, and means
;; that the instruction/abi has left the processor in an unknown
;; state.
;; "none" means that nothing has changed and no mode is set.
;; This attribute is only used for the Renesas ABI.
(define_attr "fp_set" "single,double,unknown,none" (const_string "none"))
; If a conditional branch destination is within -252..258 bytes away
; from the instruction it can be 2 bytes long. Something in the
; range -4090..4100 bytes can be 6 bytes long. All other conditional
; branches are initially assumed to be 16 bytes long.
; In machine_dependent_reorg, we split all branches that are longer than
; 2 bytes.
;; The maximum range used for SImode constant pool entries is 1018. A final
;; instruction can add 8 bytes while only being 4 bytes in size, thus we
;; can have a total of 1022 bytes in the pool. Add 4 bytes for a branch
;; instruction around the pool table, 2 bytes of alignment before the table,
;; and 30 bytes of alignment after the table. That gives a maximum total
;; pool size of 1058 bytes.
;; Worst case code/pool content size ratio is 1:2 (using asms).
;; Thus, in the worst case, there is one instruction in front of a maximum
;; sized pool, and then there are 1052 bytes of pool for every 508 bytes of
;; code. For the last n bytes of code, there are 2n + 36 bytes of pool.
;; If we have a forward branch, the initial table will be put after the
;; unconditional branch.
;;
;; ??? We could do much better by keeping track of the actual pcloads within
;; the branch range and in the pcload range in front of the branch range.
;; ??? This looks ugly because genattrtab won't allow if_then_else or cond
;; inside an le.
(define_attr "short_cbranch_p" "no,yes"
(cond [(match_test "mdep_reorg_phase <= SH_FIXUP_PCLOAD")
(const_string "no")
(leu (plus (minus (match_dup 0) (pc)) (const_int 252)) (const_int 506))
(const_string "yes")
(match_test "NEXT_INSN (PREV_INSN (insn)) != insn")
(const_string "no")
(leu (plus (minus (match_dup 0) (pc)) (const_int 252)) (const_int 508))
(const_string "yes")
] (const_string "no")))
(define_attr "med_branch_p" "no,yes"
(cond [(leu (plus (minus (match_dup 0) (pc)) (const_int 990))
(const_int 1988))
(const_string "yes")
(match_test "mdep_reorg_phase <= SH_FIXUP_PCLOAD")
(const_string "no")
(leu (plus (minus (match_dup 0) (pc)) (const_int 4092))
(const_int 8186))
(const_string "yes")
] (const_string "no")))
(define_attr "med_cbranch_p" "no,yes"
(cond [(leu (plus (minus (match_dup 0) (pc)) (const_int 988))
(const_int 1986))
(const_string "yes")
(match_test "mdep_reorg_phase <= SH_FIXUP_PCLOAD")
(const_string "no")
(leu (plus (minus (match_dup 0) (pc)) (const_int 4090))
(const_int 8184))
(const_string "yes")
] (const_string "no")))
(define_attr "braf_branch_p" "no,yes"
(cond [(match_test "! TARGET_SH2")
(const_string "no")
(leu (plus (minus (match_dup 0) (pc)) (const_int 10330))
(const_int 20660))
(const_string "yes")
(match_test "mdep_reorg_phase <= SH_FIXUP_PCLOAD")
(const_string "no")
(leu (plus (minus (match_dup 0) (pc)) (const_int 32764))
(const_int 65530))
(const_string "yes")
] (const_string "no")))
(define_attr "braf_cbranch_p" "no,yes"
(cond [(match_test "! TARGET_SH2")
(const_string "no")
(leu (plus (minus (match_dup 0) (pc)) (const_int 10328))
(const_int 20658))
(const_string "yes")
(match_test "mdep_reorg_phase <= SH_FIXUP_PCLOAD")
(const_string "no")
(leu (plus (minus (match_dup 0) (pc)) (const_int 32762))
(const_int 65528))
(const_string "yes")
] (const_string "no")))
;; An unconditional jump in the range -4092..4098 can be 2 bytes long.
;; For wider ranges, we need a combination of a code and a data part.
;; If we can get a scratch register for a long range jump, the code
;; part can be 4 bytes long; otherwise, it must be 8 bytes long.
;; If the jump is in the range -32764..32770, the data part can be 2 bytes
;; long; otherwise, it must be 6 bytes long.
;; All other instructions are two bytes long by default.
;; ??? This should use something like *branch_p (minus (match_dup 0) (pc)),
;; but getattrtab doesn't understand this.
(define_attr "length" ""
(cond [(eq_attr "type" "cbranch")
(cond [(eq_attr "short_cbranch_p" "yes")
(const_int 2)
(eq_attr "med_cbranch_p" "yes")
(const_int 6)
(eq_attr "braf_cbranch_p" "yes")
(const_int 12)
;; ??? using pc is not computed transitively.
(ne (match_dup 0) (match_dup 0))
(const_int 14)
(match_test "flag_pic")
(const_int 24)
] (const_int 16))
(eq_attr "type" "jump")
(cond [(eq_attr "med_branch_p" "yes")
(const_int 2)
(and (match_test "prev_nonnote_insn (insn)")
(and (eq (symbol_ref "GET_CODE (prev_nonnote_insn (insn))")
(symbol_ref "INSN"))
(eq (symbol_ref "INSN_CODE (prev_nonnote_insn (insn))")
(symbol_ref "code_for_indirect_jump_scratch"))))
(cond [(eq_attr "braf_branch_p" "yes")
(const_int 6)
(not (match_test "flag_pic"))
(const_int 10)
(match_test "TARGET_SH2")
(const_int 10)] (const_int 18))
(eq_attr "braf_branch_p" "yes")
(const_int 10)
;; ??? using pc is not computed transitively.
(ne (match_dup 0) (match_dup 0))
(const_int 12)
(match_test "flag_pic")
(const_int 22)
] (const_int 14))
(eq_attr "type" "pt_media")
(if_then_else (match_test "TARGET_SHMEDIA64")
(const_int 20) (const_int 12))
(and (eq_attr "type" "jump_media")
(match_test "TARGET_SH5_CUT2_WORKAROUND"))
(const_int 8)
] (if_then_else (match_test "TARGET_SHMEDIA")
(const_int 4)
(const_int 2))))
;; DFA descriptions for the pipelines
(include "sh1.md")
(include "shmedia.md")
(include "sh4.md")
(include "iterators.md")
(include "predicates.md")
(include "constraints.md")
;; Definitions for filling delay slots
(define_attr "needs_delay_slot" "yes,no" (const_string "no"))
(define_attr "banked" "yes,no"
(cond [(match_test "sh_loads_bankedreg_p (insn)")
(const_string "yes")]
(const_string "no")))
;; ??? This should be (nil) instead of (const_int 0)
(define_attr "hit_stack" "yes,no"
(cond [(not (match_test "find_regno_note (insn, REG_INC, SP_REG)"))
(const_string "no")]
(const_string "yes")))
(define_attr "interrupt_function" "no,yes"
(const (symbol_ref "current_function_interrupt")))
(define_attr "in_delay_slot" "yes,no"
(cond [(eq_attr "type" "cbranch") (const_string "no")
(eq_attr "type" "pcload,pcload_si") (const_string "no")
(eq_attr "type" "fpscr_toggle") (const_string "no")
(eq_attr "needs_delay_slot" "yes") (const_string "no")
(eq_attr "length" "2") (const_string "yes")
] (const_string "no")))
(define_attr "cond_delay_slot" "yes,no"
(cond [(eq_attr "in_delay_slot" "yes") (const_string "yes")
] (const_string "no")))
(define_attr "is_sfunc" ""
(if_then_else (eq_attr "type" "sfunc") (const_int 1) (const_int 0)))
(define_attr "is_mac_media" ""
(if_then_else (eq_attr "type" "mac_media") (const_int 1) (const_int 0)))
(define_attr "branch_zero" "yes,no"
(cond [(eq_attr "type" "!cbranch") (const_string "no")
(ne (symbol_ref "(next_active_insn (insn)\
== (prev_active_insn\
(XEXP (SET_SRC (PATTERN (insn)), 1))))\
&& get_attr_length (next_active_insn (insn)) == 2")
(const_int 0))
(const_string "yes")]
(const_string "no")))
;; SH4 Double-precision computation with double-precision result -
;; the two halves are ready at different times.
(define_attr "dfp_comp" "yes,no"
(cond [(eq_attr "type" "dfp_arith,dfp_mul,dfp_conv,dfdiv") (const_string "yes")]
(const_string "no")))
;; Insns for which the latency of a preceding fp insn is decreased by one.
(define_attr "late_fp_use" "yes,no" (const_string "no"))
;; And feeding insns for which this relevant.
(define_attr "any_fp_comp" "yes,no"
(cond [(eq_attr "type" "fp,fdiv,ftrc_s,dfp_arith,dfp_mul,dfp_conv,dfdiv")
(const_string "yes")]
(const_string "no")))
(define_attr "any_int_load" "yes,no"
(cond [(eq_attr "type" "load,load_si,pcload,pcload_si")
(const_string "yes")]
(const_string "no")))
(define_attr "highpart" "user, ignore, extend, depend, must_split"
(const_string "user"))
(define_delay
(eq_attr "needs_delay_slot" "yes")
[(eq_attr "in_delay_slot" "yes") (nil) (nil)])
;; Since a normal return (rts) implicitly uses the PR register,
;; we can't allow PR register loads in an rts delay slot.
;; On the SH1* and SH2*, the rte instruction reads the return pc from the
;; stack, and thus we can't put a pop instruction in its delay slot.
;; On the SH3* and SH4*, the rte instruction does not use the stack, so a
;; pop instruction can go in the delay slot, unless it references a banked
;; register (the register bank is switched by rte).
(define_delay
(eq_attr "type" "return")
[(and (eq_attr "in_delay_slot" "yes")
(ior (and (eq_attr "interrupt_function" "no")
(eq_attr "type" "!pload,prset"))
(and (eq_attr "interrupt_function" "yes")
(ior (match_test "TARGET_SH3") (eq_attr "hit_stack" "no"))
(eq_attr "banked" "no"))))
(nil) (nil)])
;; Since a call implicitly uses the PR register, we can't allow
;; a PR register store in a jsr delay slot.
(define_delay
(ior (eq_attr "type" "call") (eq_attr "type" "sfunc"))
[(and (eq_attr "in_delay_slot" "yes")
(eq_attr "type" "!pstore,prget")) (nil) (nil)])
;; Say that we have annulled true branches, since this gives smaller and
;; faster code when branches are predicted as not taken.
;; ??? The non-annulled condition should really be "in_delay_slot",
;; but insns that can be filled in non-annulled get priority over insns
;; that can only be filled in anulled.
(define_delay
(and (eq_attr "type" "cbranch")
(match_test "TARGET_SH2"))
;; SH2e has a hardware bug that pretty much prohibits the use of
;; annulled delay slots.
[(eq_attr "cond_delay_slot" "yes") (and (eq_attr "cond_delay_slot" "yes")
(not (eq_attr "cpu" "sh2e"))) (nil)])
;; -------------------------------------------------------------------------
;; SImode signed integer comparisons
;; -------------------------------------------------------------------------
;; Various patterns to generate the TST #imm, R0 instruction.
;; Although this adds some pressure on the R0 register, it can potentially
;; result in faster code, even if the operand has to be moved to R0 first.
;; This is because on SH4 TST #imm, R0 and MOV Rm, Rn are both MT group
;; instructions and thus will be executed in parallel. On SH4A TST #imm, R0
;; is an EX group instruction but still can be executed in parallel with the
;; MT group MOV Rm, Rn instruction.
;; Usual TST #imm, R0 patterns for SI, HI and QI
;; This is usually used for bit patterns other than contiguous bits
;; and single bits.
(define_insn "tstsi_t"
[(set (reg:SI T_REG)
(eq:SI (and:SI (match_operand:SI 0 "logical_operand" "%z,r")
(match_operand:SI 1 "logical_operand" "K08,r"))
(const_int 0)))]
"TARGET_SH1"
"tst %1,%0"
[(set_attr "type" "mt_group")])
(define_insn "tsthi_t"
[(set (reg:SI T_REG)
(eq:SI (subreg:SI (and:HI (match_operand:HI 0 "logical_operand" "%z")
(match_operand 1 "const_int_operand")) 0)
(const_int 0)))]
"TARGET_SH1
&& CONST_OK_FOR_K08 (INTVAL (operands[1]))"
"tst %1,%0"
[(set_attr "type" "mt_group")])
(define_insn "tstqi_t"
[(set (reg:SI T_REG)
(eq:SI (subreg:SI (and:QI (match_operand:QI 0 "logical_operand" "%z")
(match_operand 1 "const_int_operand")) 0)
(const_int 0)))]
"TARGET_SH1
&& (CONST_OK_FOR_K08 (INTVAL (operands[1]))
|| CONST_OK_FOR_I08 (INTVAL (operands[1])))"
{
operands[1] = GEN_INT (INTVAL (operands[1]) & 255);
return "tst %1,%0";
}
[(set_attr "type" "mt_group")])
;; Test low QI subreg against zero.
;; This avoids unnecessary zero extension before the test.
(define_insn "*tstqi_t_zero"
[(set (reg:SI T_REG)
(eq:SI (match_operand:QI 0 "logical_operand" "z") (const_int 0)))]
"TARGET_SH1"
"tst #255,%0"
[(set_attr "type" "mt_group")])
;; This pattern might be risky because it also tests the upper bits and not
;; only the subreg. However, it seems that combine will get to this only
;; when testing sign/zero extended values. In this case the extended upper
;; bits do not matter.
(define_insn "*tst_t_zero"
[(set (reg:SI T_REG)
(eq:SI
(subreg:QIHI
(and:SI (match_operand:SI 0 "arith_reg_operand" "%r")
(match_operand:SI 1 "arith_reg_operand" "r")) )
(const_int 0)))]
"TARGET_SH1 && TARGET_LITTLE_ENDIAN"
"tst %0,%1"
[(set_attr "type" "mt_group")])
(define_insn "*tst_t_zero"
[(set (reg:SI T_REG)
(eq:SI
(subreg:QIHI
(and:SI (match_operand:SI 0 "arith_reg_operand" "%r")
(match_operand:SI 1 "arith_reg_operand" "r")) )
(const_int 0)))]
"TARGET_SH1 && TARGET_BIG_ENDIAN"
"tst %0,%1"
[(set_attr "type" "mt_group")])
;; Extract LSB, negate and store in T bit.
(define_insn "tstsi_t_and_not"
[(set (reg:SI T_REG)
(and:SI (not:SI (match_operand:SI 0 "logical_operand" "z"))
(const_int 1)))]
"TARGET_SH1"
"tst #1,%0"
[(set_attr "type" "mt_group")])
;; Extract contiguous bits and compare them against zero.
(define_insn "tst_t_zero_extract_eq"
[(set (reg:SI T_REG)
(eq:SI (zero_extract:SI (match_operand:QIHISIDI 0 "logical_operand" "z")
(match_operand:SI 1 "const_int_operand")
(match_operand:SI 2 "const_int_operand"))
(const_int 0)))]
"TARGET_SH1
&& CONST_OK_FOR_K08 (ZERO_EXTRACT_ANDMASK (operands[1], operands[2]))"
{
operands[1] = GEN_INT (ZERO_EXTRACT_ANDMASK (operands[1], operands[2]));
return "tst %1,%0";
}
[(set_attr "type" "mt_group")])
;; This split is required when testing bits in a QI subreg.
(define_split
[(set (reg:SI T_REG)
(eq:SI
(if_then_else:SI
(zero_extract:SI (match_operand 0 "logical_operand")
(match_operand 1 "const_int_operand")
(match_operand 2 "const_int_operand"))
(match_operand 3 "const_int_operand")
(const_int 0))
(const_int 0)))]
"TARGET_SH1
&& ZERO_EXTRACT_ANDMASK (operands[1], operands[2]) == INTVAL (operands[3])
&& CONST_OK_FOR_K08 (INTVAL (operands[3]))"
[(set (reg:SI T_REG) (eq:SI (and:SI (match_dup 0) (match_dup 3))
(const_int 0)))]
{
if (GET_MODE (operands[0]) == QImode)
operands[0] = simplify_gen_subreg (SImode, operands[0], QImode, 0);
})
;; Extract single bit, negate and store it in the T bit.
;; Not used for SH4A.
(define_insn "tstsi_t_zero_extract_xor"
[(set (reg:SI T_REG)
(zero_extract:SI (xor:SI (match_operand:SI 0 "logical_operand" "z")
(match_operand:SI 3 "const_int_operand"))
(match_operand:SI 1 "const_int_operand")
(match_operand:SI 2 "const_int_operand")))]
"TARGET_SH1
&& ZERO_EXTRACT_ANDMASK (operands[1], operands[2]) == INTVAL (operands[3])
&& CONST_OK_FOR_K08 (INTVAL (operands[3]))"
"tst %3,%0"
[(set_attr "type" "mt_group")])
;; Extract single bit, negate and store it in the T bit.
;; Used for SH4A little endian.
(define_insn "tstsi_t_zero_extract_subreg_xor_little"
[(set (reg:SI T_REG)
(zero_extract:SI
(subreg:QI (xor:SI (match_operand:SI 0 "logical_operand" "z")
(match_operand:SI 3 "const_int_operand")) 0)
(match_operand:SI 1 "const_int_operand")
(match_operand:SI 2 "const_int_operand")))]
"TARGET_SH1 && TARGET_LITTLE_ENDIAN
&& ZERO_EXTRACT_ANDMASK (operands[1], operands[2])
== (INTVAL (operands[3]) & 255)
&& CONST_OK_FOR_K08 (INTVAL (operands[3]) & 255)"
{
operands[3] = GEN_INT (INTVAL (operands[3]) & 255);
return "tst %3,%0";
}
[(set_attr "type" "mt_group")])
;; Extract single bit, negate and store it in the T bit.
;; Used for SH4A big endian.
(define_insn "tstsi_t_zero_extract_subreg_xor_big"
[(set (reg:SI T_REG)
(zero_extract:SI
(subreg:QI (xor:SI (match_operand:SI 0 "logical_operand" "z")
(match_operand:SI 3 "const_int_operand")) 3)
(match_operand:SI 1 "const_int_operand")
(match_operand:SI 2 "const_int_operand")))]
"TARGET_SH1 && TARGET_BIG_ENDIAN
&& ZERO_EXTRACT_ANDMASK (operands[1], operands[2])
== (INTVAL (operands[3]) & 255)
&& CONST_OK_FOR_K08 (INTVAL (operands[3]) & 255)"
{
operands[3] = GEN_INT (INTVAL (operands[3]) & 255);
return "tst %3,%0";
}
[(set_attr "type" "mt_group")])
(define_insn "cmpeqsi_t"
[(set (reg:SI T_REG)
(eq:SI (match_operand:SI 0 "arith_reg_operand" "r,z,r")
(match_operand:SI 1 "arith_operand" "N,rI08,r")))]
"TARGET_SH1"
"@
tst %0,%0
cmp/eq %1,%0
cmp/eq %1,%0"
[(set_attr "type" "mt_group")])
;; FIXME: For some reason, on SH4A and SH2A combine fails to simplify this
;; pattern by itself. What this actually does is:
;; x == 0: (1 >> 0-0) & 1 = 1
;; x != 0: (1 >> 0-x) & 1 = 0
;; Without this the test pr51244-8.c fails on SH2A and SH4A.
(define_insn_and_split "*cmpeqsi_t"
[(set (reg:SI T_REG)
(and:SI (lshiftrt:SI
(const_int 1)
(neg:SI (match_operand:SI 0 "arith_reg_operand" "r")))
(const_int 1)))]
"TARGET_SH1"
"#"
"&& 1"
[(set (reg:SI T_REG) (eq:SI (match_dup 0) (const_int 0)))])
(define_insn "cmpgtsi_t"
[(set (reg:SI T_REG)
(gt:SI (match_operand:SI 0 "arith_reg_operand" "r,r")
(match_operand:SI 1 "arith_reg_or_0_operand" "N,r")))]
"TARGET_SH1"
"@
cmp/pl %0
cmp/gt %1,%0"
[(set_attr "type" "mt_group")])
(define_insn "cmpgesi_t"
[(set (reg:SI T_REG)
(ge:SI (match_operand:SI 0 "arith_reg_operand" "r,r")
(match_operand:SI 1 "arith_reg_or_0_operand" "N,r")))]
"TARGET_SH1"
"@
cmp/pz %0
cmp/ge %1,%0"
[(set_attr "type" "mt_group")])
;; FIXME: This is actually wrong. There is no way to literally move a
;; general reg to t reg. Luckily, it seems that this pattern will be only
;; used when the general reg is known be either '0' or '1' during combine.
;; What we actually need is reg != 0 -> T, but we have only reg == 0 -> T.
;; Due to interactions with other patterns, combine fails to pick the latter
;; and invert the dependent logic.
(define_insn "*negtstsi"
[(set (reg:SI T_REG) (match_operand:SI 0 "arith_reg_operand" "r"))]
"TARGET_SH1"
"cmp/pl %0"
[(set_attr "type" "mt_group")])
;; Some integer sign comparison patterns can be realized with the div0s insn.
;; div0s Rm,Rn T = (Rm >> 31) ^ (Rn >> 31)
(define_insn "cmp_div0s_0"
[(set (reg:SI T_REG)
(lshiftrt:SI (xor:SI (match_operand:SI 0 "arith_reg_operand" "%r")
(match_operand:SI 1 "arith_reg_operand" "r"))
(const_int 31)))]
"TARGET_SH1"
"div0s %0,%1"
[(set_attr "type" "arith")])
(define_insn "cmp_div0s_1"
[(set (reg:SI T_REG)
(lt:SI (xor:SI (match_operand:SI 0 "arith_reg_operand" "%r")
(match_operand:SI 1 "arith_reg_operand" "r"))
(const_int 0)))]
"TARGET_SH1"
"div0s %0,%1"
[(set_attr "type" "arith")])
(define_insn_and_split "*cmp_div0s_0"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(lshiftrt:SI (xor:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "arith_reg_operand" ""))
(const_int 31)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(set (reg:SI T_REG)
(lshiftrt:SI (xor:SI (match_dup 1) (match_dup 2)) (const_int 31)))
(set (match_dup 0) (reg:SI T_REG))])
(define_insn "*cmp_div0s_0"
[(set (reg:SI T_REG)
(eq:SI (lshiftrt:SI (match_operand:SI 0 "arith_reg_operand" "%r")
(const_int 31))
(ge:SI (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 0))))]
"TARGET_SH1"
"div0s %0,%1"
[(set_attr "type" "arith")])
(define_insn_and_split "*cmp_div0s_1"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(ge:SI (xor:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "arith_reg_operand" ""))
(const_int 0)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
;; We have to go through the movnegt expander here which will handle the
;; SH2A vs non-SH2A cases.
{
emit_insn (gen_cmp_div0s_1 (operands[1], operands[2]));
emit_insn (gen_movnegt (operands[0], get_t_reg_rtx ()));
DONE;
})
(define_insn_and_split "*cmp_div0s_1"
[(set (reg:SI T_REG)
(ge:SI (xor:SI (match_operand:SI 0 "arith_reg_operand" "")
(match_operand:SI 1 "arith_reg_operand" ""))
(const_int 0)))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(set (reg:SI T_REG) (lt:SI (xor:SI (match_dup 0) (match_dup 1))
(const_int 0)))
(set (reg:SI T_REG) (xor:SI (reg:SI T_REG) (const_int 1)))])
(define_insn_and_split "*cmp_div0s_1"
[(set (reg:SI T_REG)
(eq:SI (lshiftrt:SI (match_operand:SI 0 "arith_reg_operand")
(const_int 31))
(lshiftrt:SI (match_operand:SI 1 "arith_reg_operand")
(const_int 31))))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(set (reg:SI T_REG) (lt:SI (xor:SI (match_dup 0) (match_dup 1))
(const_int 0)))
(set (reg:SI T_REG) (xor:SI (reg:SI T_REG) (const_int 1)))])
;; -------------------------------------------------------------------------
;; SImode compare and branch
;; -------------------------------------------------------------------------
(define_expand "cbranchsi4"
[(set (pc)
(if_then_else (match_operator 0 "comparison_operator"
[(match_operand:SI 1 "arith_operand" "")
(match_operand:SI 2 "arith_operand" "")])
(label_ref (match_operand 3 "" ""))
(pc)))
(clobber (reg:SI T_REG))]
""
{
if (TARGET_SHMEDIA)
emit_jump_insn (gen_cbranchint4_media (operands[0], operands[1],
operands[2], operands[3]));
else
expand_cbranchsi4 (operands, LAST_AND_UNUSED_RTX_CODE, -1);
DONE;
})
;; Combine patterns to invert compare and branch operations for which we
;; don't have actual comparison insns. These patterns are used in cases
;; which appear after the initial cbranchsi expansion, which also does
;; some condition inversion.
(define_split
[(set (pc)
(if_then_else (ne (match_operand:SI 0 "arith_reg_operand" "")
(match_operand:SI 1 "arith_reg_or_0_operand" ""))
(label_ref (match_operand 2))
(pc)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
[(set (reg:SI T_REG) (eq:SI (match_dup 0) (match_dup 1)))
(set (pc) (if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 2))
(pc)))])
;; FIXME: Similar to the *cmpeqsi_t pattern above, for some reason, on SH4A
;; and SH2A combine fails to simplify this pattern by itself.
;; What this actually does is:
;; x == 0: (1 >> 0-0) & 1 = 1
;; x != 0: (1 >> 0-x) & 1 = 0
;; Without this the test pr51244-8.c fails on SH2A and SH4A.
(define_split
[(set (pc)
(if_then_else
(eq (and:SI (lshiftrt:SI
(const_int 1)
(neg:SI (match_operand:SI 0 "arith_reg_operand" "")))
(const_int 1))
(const_int 0))
(label_ref (match_operand 2))
(pc)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
[(set (reg:SI T_REG) (eq:SI (match_dup 0) (const_int 0)))
(set (pc) (if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 2))
(pc)))])
;; FIXME: These could probably use code iterators for the compare op.
(define_split
[(set (pc)
(if_then_else (le (match_operand:SI 0 "arith_reg_operand" "")
(match_operand:SI 1 "arith_reg_or_0_operand" ""))
(label_ref (match_operand 2))
(pc)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
[(set (reg:SI T_REG) (gt:SI (match_dup 0) (match_dup 1)))
(set (pc) (if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 2))
(pc)))])
(define_split
[(set (pc)
(if_then_else (lt (match_operand:SI 0 "arith_reg_operand" "")
(match_operand:SI 1 "arith_reg_or_0_operand" ""))
(label_ref (match_operand 2))
(pc)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
[(set (reg:SI T_REG) (ge:SI (match_dup 0) (match_dup 1)))
(set (pc) (if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 2))
(pc)))])
(define_split
[(set (pc)
(if_then_else (leu (match_operand:SI 0 "arith_reg_operand" "")
(match_operand:SI 1 "arith_reg_operand" ""))
(label_ref (match_operand 2))
(pc)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
[(set (reg:SI T_REG) (gtu:SI (match_dup 0) (match_dup 1)))
(set (pc) (if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 2))
(pc)))])
(define_split
[(set (pc)
(if_then_else (ltu (match_operand:SI 0 "arith_reg_operand" "")
(match_operand:SI 1 "arith_reg_operand" ""))
(label_ref (match_operand 2))
(pc)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
[(set (reg:SI T_REG) (geu:SI (match_dup 0) (match_dup 1)))
(set (pc) (if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 2))
(pc)))])
;; Compare and branch combine patterns for div0s comparisons.
(define_insn_and_split "*cbranch_div0s"
[(set (pc)
(if_then_else (lt (xor:SI (match_operand:SI 0 "arith_reg_operand" "")
(match_operand:SI 1 "arith_reg_operand" ""))
(const_int 0))
(label_ref (match_operand 2))
(pc)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(set (reg:SI T_REG)
(lt:SI (xor:SI (match_dup 0) (match_dup 1)) (const_int 0)))
(set (pc)
(if_then_else (ne (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 2))
(pc)))])
(define_insn_and_split "*cbranch_div0s"
[(set (pc)
(if_then_else (ge (xor:SI (match_operand:SI 0 "arith_reg_operand" "")
(match_operand:SI 1 "arith_reg_operand" ""))
(const_int 0))
(label_ref (match_operand 2))
(pc)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(set (reg:SI T_REG)
(lt:SI (xor:SI (match_dup 0) (match_dup 1)) (const_int 0)))
(set (pc)
(if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 2))
(pc)))])
;; Conditional move combine pattern for div0s comparisons.
;; This is used when TARGET_PRETEND_CMOVE is in effect.
(define_insn_and_split "*movsicc_div0s"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(if_then_else:SI (ge (xor:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "arith_reg_operand" ""))
(const_int 0))
(match_operand:SI 3 "arith_reg_operand" "")
(match_operand:SI 4 "general_movsrc_operand" "")))
(clobber (reg:SI T_REG))]
"TARGET_PRETEND_CMOVE"
"#"
"&& 1"
[(set (reg:SI T_REG) (lt:SI (xor:SI (match_dup 1) (match_dup 2))
(const_int 0)))
(set (match_dup 0)
(if_then_else (ne (reg:SI T_REG) (const_int 0))
(match_dup 4)
(match_dup 3)))])
(define_insn_and_split "*movsicc_div0s"
[(set (match_operand:SI 0 "arith_reg_dest")
(if_then_else:SI (eq (lshiftrt:SI
(match_operand:SI 1 "arith_reg_operand")
(const_int 31))
(lshiftrt:SI
(match_operand:SI 2 "arith_reg_operand")
(const_int 31)))
(match_operand:SI 3 "arith_reg_operand")
(match_operand:SI 4 "general_movsrc_operand")))
(clobber (reg:SI T_REG))]
"TARGET_PRETEND_CMOVE"
"#"
"&& 1"
[(set (reg:SI T_REG) (lt:SI (xor:SI (match_dup 1) (match_dup 2))
(const_int 0)))
(set (match_dup 0)
(if_then_else (ne (reg:SI T_REG) (const_int 0))
(match_dup 4)
(match_dup 3)))])
;; -------------------------------------------------------------------------
;; SImode unsigned integer comparisons
;; -------------------------------------------------------------------------
;; Usually comparisons of 'unsigned int >= 0' are optimized away completely.
;; However, especially when optimizations are off (e.g. -O0) such comparisons
;; might remain and we have to handle them. If the '>= 0' case wasn't
;; handled here, something else would just load a '0' into the second operand
;; and do the comparison. We can do slightly better by just setting the
;; T bit to '1'.
(define_insn_and_split "cmpgeusi_t"
[(set (reg:SI T_REG)
(geu:SI (match_operand:SI 0 "arith_reg_operand" "r")
(match_operand:SI 1 "arith_reg_or_0_operand" "r")))]
"TARGET_SH1"
"cmp/hs %1,%0"
"&& satisfies_constraint_Z (operands[1])"
[(set (reg:SI T_REG) (const_int 1))]
""
[(set_attr "type" "mt_group")])
(define_insn "cmpgtusi_t"
[(set (reg:SI T_REG)
(gtu:SI (match_operand:SI 0 "arith_reg_operand" "r")
(match_operand:SI 1 "arith_reg_operand" "r")))]
"TARGET_SH1"
"cmp/hi %1,%0"
[(set_attr "type" "mt_group")])
;; -------------------------------------------------------------------------
;; DImode compare and branch
;; -------------------------------------------------------------------------
;; arith3 patterns don't work well with the sh4-300 branch prediction mechanism.
;; Therefore, we aim to have a set of three branches that go straight to the
;; destination, i.e. only one of them is taken at any one time.
;; This mechanism should also be slightly better for the sh4-200.
(define_expand "cbranchdi4"
[(set (pc)
(if_then_else (match_operator 0 "comparison_operator"
[(match_operand:DI 1 "arith_operand" "")
(match_operand:DI 2 "arith_operand" "")])
(label_ref (match_operand 3 "" ""))
(pc)))
(clobber (match_dup 4))
(clobber (reg:SI T_REG))]
"TARGET_CBRANCHDI4 || TARGET_SH2 || TARGET_SHMEDIA"
{
enum rtx_code comparison;
if (TARGET_SHMEDIA)
{
emit_jump_insn (gen_cbranchint4_media (operands[0], operands[1],
operands[2], operands[3]));
DONE;
}
else if (!TARGET_CBRANCHDI4)
{
sh_emit_compare_and_branch (operands, DImode);
DONE;
}
else
{
if (expand_cbranchdi4 (operands, LAST_AND_UNUSED_RTX_CODE))
DONE;
comparison = prepare_cbranch_operands (operands, DImode,
LAST_AND_UNUSED_RTX_CODE);
if (comparison != GET_CODE (operands[0]))
operands[0]
= gen_rtx_fmt_ee (comparison, VOIDmode, operands[1], operands[2]);
operands[4] = gen_rtx_SCRATCH (SImode);
}
})
(define_insn_and_split "cbranchdi4_i"
[(set (pc)
(if_then_else (match_operator 0 "comparison_operator"
[(match_operand:DI 1 "arith_operand" "r,r")
(match_operand:DI 2 "arith_operand" "rN,I08")])
(label_ref (match_operand 3 "" ""))
(pc)))
(clobber (match_scratch:SI 4 "=X,&r"))
(clobber (reg:SI T_REG))]
"TARGET_CBRANCHDI4"
"#"
"&& reload_completed"
[(pc)]
{
if (!expand_cbranchdi4 (operands, GET_CODE (operands[0])))
FAIL;
DONE;
})
;; -------------------------------------------------------------------------
;; DImode signed integer comparisons
;; -------------------------------------------------------------------------
(define_insn ""
[(set (reg:SI T_REG)
(eq:SI (and:DI (match_operand:DI 0 "arith_reg_operand" "r")
(match_operand:DI 1 "arith_operand" "r"))
(const_int 0)))]
"TARGET_SH1"
{
return output_branchy_insn (EQ, "tst\t%S1,%S0;bf\t%l9;tst\t%R1,%R0",
insn, operands);
}
[(set_attr "length" "6")
(set_attr "type" "arith3b")])
(define_insn "cmpeqdi_t"
[(set (reg:SI T_REG)
(eq:SI (match_operand:DI 0 "arith_reg_operand" "r,r")
(match_operand:DI 1 "arith_reg_or_0_operand" "N,r")))]
"TARGET_SH1"
{
static const char* alt[] =
{
"tst %S0,%S0" "\n"
" bf 0f" "\n"
" tst %R0,%R0" "\n"
"0:",
"cmp/eq %S1,%S0" "\n"
" bf 0f" "\n"
" cmp/eq %R1,%R0" "\n"
"0:"
};
return alt[which_alternative];
}
[(set_attr "length" "6")
(set_attr "type" "arith3b")])
(define_split
[(set (reg:SI T_REG)
(eq:SI (match_operand:DI 0 "arith_reg_operand" "")
(match_operand:DI 1 "arith_reg_or_0_operand" "")))]
;; If we applied this split when not optimizing, it would only be
;; applied during the machine-dependent reorg, when no new basic blocks
;; may be created.
"TARGET_SH1 && reload_completed && optimize"
[(set (reg:SI T_REG) (eq:SI (match_dup 2) (match_dup 3)))
(set (pc) (if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 6))
(pc)))
(set (reg:SI T_REG) (eq:SI (match_dup 4) (match_dup 5)))
(match_dup 6)]
{
operands[2] = gen_highpart (SImode, operands[0]);
operands[3] = operands[1] == const0_rtx
? const0_rtx
: gen_highpart (SImode, operands[1]);
operands[4] = gen_lowpart (SImode, operands[0]);
operands[5] = gen_lowpart (SImode, operands[1]);
operands[6] = gen_label_rtx ();
})
(define_insn "cmpgtdi_t"
[(set (reg:SI T_REG)
(gt:SI (match_operand:DI 0 "arith_reg_operand" "r,r")
(match_operand:DI 1 "arith_reg_or_0_operand" "r,N")))]
"TARGET_SH2"
{
static const char* alt[] =
{
"cmp/eq %S1,%S0" "\n"
" bf{.|/}s 0f" "\n"
" cmp/gt %S1,%S0" "\n"
" cmp/hi %R1,%R0" "\n"
"0:",
"tst %S0,%S0" "\n"
" bf{.|/}s 0f" "\n"
" cmp/pl %S0" "\n"
" cmp/hi %S0,%R0" "\n"
"0:"
};
return alt[which_alternative];
}
[(set_attr "length" "8")
(set_attr "type" "arith3")])
(define_insn "cmpgedi_t"
[(set (reg:SI T_REG)
(ge:SI (match_operand:DI 0 "arith_reg_operand" "r,r")
(match_operand:DI 1 "arith_reg_or_0_operand" "r,N")))]
"TARGET_SH2"
{
static const char* alt[] =
{
"cmp/eq %S1,%S0" "\n"
" bf{.|/}s 0f" "\n"
" cmp/ge %S1,%S0" "\n"
" cmp/hs %R1,%R0" "\n"
"0:",
"cmp/pz %S0"
};
return alt[which_alternative];
}
[(set_attr "length" "8,2")
(set_attr "type" "arith3,mt_group")])
;; -------------------------------------------------------------------------
;; DImode unsigned integer comparisons
;; -------------------------------------------------------------------------
(define_insn "cmpgeudi_t"
[(set (reg:SI T_REG)
(geu:SI (match_operand:DI 0 "arith_reg_operand" "r")
(match_operand:DI 1 "arith_reg_operand" "r")))]
"TARGET_SH2"
{
return "cmp/eq %S1,%S0" "\n"
" bf{.|/}s 0f" "\n"
" cmp/hs %S1,%S0" "\n"
" cmp/hs %R1,%R0" "\n"
"0:";
}
[(set_attr "length" "8")
(set_attr "type" "arith3")])
(define_insn "cmpgtudi_t"
[(set (reg:SI T_REG)
(gtu:SI (match_operand:DI 0 "arith_reg_operand" "r")
(match_operand:DI 1 "arith_reg_operand" "r")))]
"TARGET_SH2"
{
return "cmp/eq %S1,%S0" "\n"
" bf{.|/}s 0f" "\n"
" cmp/hi %S1,%S0" "\n"
" cmp/hi %R1,%R0" "\n"
"0:";
}
[(set_attr "length" "8")
(set_attr "type" "arith3")])
(define_insn "cmpeqsi_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(eq:SI (match_operand:SI 1 "logical_operand" "%r")
(match_operand:SI 2 "cmp_operand" "Nr")))]
"TARGET_SHMEDIA"
"cmpeq %1, %N2, %0"
[(set_attr "type" "cmp_media")])
(define_insn "cmpeqdi_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(eq:SI (match_operand:DI 1 "register_operand" "%r")
(match_operand:DI 2 "cmp_operand" "Nr")))]
"TARGET_SHMEDIA"
"cmpeq %1, %N2, %0"
[(set_attr "type" "cmp_media")])
(define_insn "cmpgtsi_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(gt:SI (match_operand:SI 1 "cmp_operand" "Nr")
(match_operand:SI 2 "cmp_operand" "rN")))]
"TARGET_SHMEDIA"
"cmpgt %N1, %N2, %0"
[(set_attr "type" "cmp_media")])
(define_insn "cmpgtdi_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(gt:SI (match_operand:DI 1 "arith_reg_or_0_operand" "Nr")
(match_operand:DI 2 "arith_reg_or_0_operand" "rN")))]
"TARGET_SHMEDIA"
"cmpgt %N1, %N2, %0"
[(set_attr "type" "cmp_media")])
(define_insn "cmpgtusi_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(gtu:SI (match_operand:SI 1 "cmp_operand" "Nr")
(match_operand:SI 2 "cmp_operand" "rN")))]
"TARGET_SHMEDIA"
"cmpgtu %N1, %N2, %0"
[(set_attr "type" "cmp_media")])
(define_insn "cmpgtudi_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(gtu:SI (match_operand:DI 1 "arith_reg_or_0_operand" "Nr")
(match_operand:DI 2 "arith_reg_or_0_operand" "rN")))]
"TARGET_SHMEDIA"
"cmpgtu %N1, %N2, %0"
[(set_attr "type" "cmp_media")])
; This pattern is for combine.
(define_insn "*cmpne0sisi_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(ne:SI (match_operand:SI 1 "arith_reg_operand" "r") (const_int 0)))]
"TARGET_SHMEDIA"
"cmpgtu %1,r63,%0"
[(set_attr "type" "cmp_media")])
;; -------------------------------------------------------------------------
;; Conditional move instructions
;; -------------------------------------------------------------------------
;; The insn names may seem reversed, but note that cmveq performs the move
;; if op1 == 0, and cmvne does it if op1 != 0.
(define_insn "movdicc_false"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(if_then_else:DI (eq (match_operand:DI 1 "arith_reg_operand" "r")
(const_int 0))
(match_operand:DI 2 "arith_reg_or_0_operand" "rN")
(match_operand:DI 3 "arith_reg_operand" "0")))]
"TARGET_SHMEDIA"
"cmveq %1, %N2, %0"
[(set_attr "type" "arith_media")])
(define_insn "movdicc_true"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(if_then_else:DI (ne (match_operand:DI 1 "arith_reg_operand" "r")
(const_int 0))
(match_operand:DI 2 "arith_reg_or_0_operand" "rN")
(match_operand:DI 3 "arith_reg_operand" "0")))]
"TARGET_SHMEDIA"
"cmvne %1, %N2, %0"
[(set_attr "type" "arith_media")])
(define_peephole2
[(set (match_operand:DI 0 "arith_reg_dest" "")
(if_then_else:DI (match_operator 3 "equality_comparison_operator"
[(match_operand:DI 1 "arith_reg_operand" "")
(const_int 0)])
(match_operand:DI 2 "arith_reg_dest" "")
(match_dup 0)))
(set (match_dup 2) (match_dup 0))]
"TARGET_SHMEDIA && peep2_reg_dead_p (2, operands[0])"
[(set (match_dup 2)
(if_then_else:DI (match_dup 3) (match_dup 0) (match_dup 2)))]
{
operands[3] = gen_rtx_fmt_ee (reverse_condition (GET_CODE (operands[3])),
VOIDmode, operands[1], CONST0_RTX (DImode));
})
(define_peephole2
[(set (match_operand:DI 0 "general_movdst_operand" "")
(match_operand:DI 1 "arith_reg_or_0_operand" ""))
(set (match_operand:DI 2 "arith_reg_dest" "")
(if_then_else:DI (match_operator 4 "equality_comparison_operator"
[(match_operand:DI 3 "arith_reg_operand" "")
(const_int 0)])
(match_dup 0)
(match_dup 2)))]
"TARGET_SHMEDIA && peep2_reg_dead_p (2, operands[0])"
[(set (match_dup 2)
(if_then_else:DI (match_dup 4) (match_dup 1) (match_dup 2)))]
"")
(define_expand "movdicc"
[(set (match_operand:DI 0 "register_operand" "")
(if_then_else:DI (match_operand 1 "comparison_operator" "")
(match_operand:DI 2 "register_operand" "")
(match_operand:DI 3 "register_operand" "")))]
"TARGET_SHMEDIA"
{
if ((GET_CODE (operands[1]) == EQ || GET_CODE (operands[1]) == NE)
&& GET_MODE (XEXP (operands[1], 0)) == DImode
&& XEXP (operands[1], 1) == const0_rtx)
;
else
{
if (!can_create_pseudo_p ())
FAIL;
operands[1] = sh_emit_cheap_store_flag (GET_MODE (operands[0]),
GET_CODE (operands[1]),
XEXP (operands[1], 0),
XEXP (operands[1], 1));
if (!operands[1])
FAIL;
}
})
;; Add SImode variants for cmveq / cmvne to compensate for not promoting
;; SImode to DImode.
(define_insn "movsicc_false"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(if_then_else:SI (eq (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 0))
(match_operand:SI 2 "arith_reg_or_0_operand" "rN")
(match_operand:SI 3 "arith_reg_operand" "0")))]
"TARGET_SHMEDIA"
"cmveq %1, %N2, %0"
[(set_attr "type" "arith_media")])
(define_insn "movsicc_true"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(if_then_else:SI (ne (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 0))
(match_operand:SI 2 "arith_reg_or_0_operand" "rN")
(match_operand:SI 3 "arith_reg_operand" "0")))]
"TARGET_SHMEDIA"
"cmvne %1, %N2, %0"
[(set_attr "type" "arith_media")])
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(if_then_else:SI (match_operator 3 "equality_comparison_operator"
[(match_operand:SI 1 "arith_reg_operand" "")
(const_int 0)])
(match_operand:SI 2 "arith_reg_dest" "")
(match_dup 0)))
(set (match_dup 2) (match_dup 0))]
"TARGET_SHMEDIA && peep2_reg_dead_p (2, operands[0])"
[(set (match_dup 2)
(if_then_else:SI (match_dup 3) (match_dup 0) (match_dup 2)))]
{
operands[3] = gen_rtx_fmt_ee (reverse_condition (GET_CODE (operands[3])),
VOIDmode, operands[1], CONST0_RTX (SImode));
})
(define_peephole2
[(set (match_operand:SI 0 "general_movdst_operand" "")
(match_operand:SI 1 "arith_reg_or_0_operand" ""))
(set (match_operand:SI 2 "arith_reg_dest" "")
(if_then_else:SI (match_operator 4 "equality_comparison_operator"
[(match_operand:SI 3 "arith_reg_operand" "")
(const_int 0)])
(match_dup 0)
(match_dup 2)))]
"TARGET_SHMEDIA && peep2_reg_dead_p (2, operands[0])
&& (!REG_P (operands[1]) || GENERAL_REGISTER_P (REGNO (operands[1])))"
[(set (match_dup 2)
(if_then_else:SI (match_dup 4) (match_dup 1) (match_dup 2)))]
{
replace_rtx (operands[4], operands[0], operands[1]);
})
(define_peephole2
[(set (match_operand 0 "any_register_operand" "")
(match_operand 1 "any_register_operand" ""))
(set (match_operand 2 "any_register_operand" "") (match_operand 3 "" ""))
(set (match_operand 4 "" "") (match_operand 5 "" ""))]
"(HARD_REGNO_NREGS (REGNO (operands[0]), GET_MODE (operands[2]))
<= HARD_REGNO_NREGS (REGNO (operands[0]), GET_MODE (operands[0])))
&& peep2_reg_dead_p (3, operands[0]) && peep2_reg_dead_p (3, operands[2])
&& ! FIND_REG_INC_NOTE (peep2_next_insn (2), operands[0])
&& ! FIND_REG_INC_NOTE (peep2_next_insn (2), operands[2])
&& ! reg_overlap_mentioned_p (operands[0], operands[3])
&& ! reg_overlap_mentioned_p (operands[2], operands[0])
&& ! reg_overlap_mentioned_p (operands[0], operands[1])
&& (REGNO_REG_CLASS (REGNO (operands[0]))
== REGNO_REG_CLASS (REGNO (operands[2])))
&& (REGNO_REG_CLASS (REGNO (operands[1]))
== REGNO_REG_CLASS (REGNO (operands[0])))"
[(set (match_dup 0) (match_dup 3))
(set (match_dup 4) (match_dup 5))]
{
rtx set1, set2;
rtx_insn *insn1, *insn2;
rtx replacements[4];
/* We want to replace occurrences of operands[0] with operands[1] and
operands[2] with operands[0] in operands[4]/operands[5].
Doing just two replace_rtx calls naively would result in the second
replacement undoing all that the first did if operands[1] and operands[2]
are identical, so we must do this simultaneously. */
replacements[0] = operands[0];
replacements[1] = operands[1];
replacements[2] = operands[2];
replacements[3] = operands[0];
if (!replace_n_hard_rtx (operands[5], replacements, 2, 0)
|| !replace_n_hard_rtx (operands[4], replacements, 2, 0)
|| !replace_n_hard_rtx (operands[2], replacements, 2, 0))
FAIL;
operands[5] = replace_n_hard_rtx (operands[5], replacements, 2, 1);
replace_n_hard_rtx (operands[4], replacements, 2, 1);
operands[2] = replace_n_hard_rtx (operands[2], replacements, 2, 1);
/* The operands array is aliased to recog_data.operand, which gets
clobbered by extract_insn, so finish with it now. */
set1 = gen_rtx_SET (VOIDmode, operands[2], operands[3]);
set2 = gen_rtx_SET (VOIDmode, operands[4], operands[5]);
/* ??? The last insn might be a jump insn, but the generic peephole2 code
always uses emit_insn. */
/* Check that we don't violate matching constraints or earlyclobbers. */
basic_block bb = BLOCK_FOR_INSN (peep2_next_insn (2));
insn1 = emit_insn (set1);
extract_insn (insn1);
if (! constrain_operands (1, get_preferred_alternatives (insn1, bb)))
goto failure;
insn2 = emit (set2);
if (GET_CODE (insn2) == BARRIER)
goto failure;
extract_insn (insn2);
if (! constrain_operands (1, get_preferred_alternatives (insn2, bb)))
{
rtx tmp;
failure:
tmp = replacements[0];
replacements[0] = replacements[1];
replacements[1] = tmp;
tmp = replacements[2];
replacements[2] = replacements[3];
replacements[3] = tmp;
replace_n_hard_rtx (SET_DEST (set1), replacements, 2, 1);
replace_n_hard_rtx (SET_DEST (set2), replacements, 2, 1);
replace_n_hard_rtx (SET_SRC (set2), replacements, 2, 1);
FAIL;
}
DONE;
})
;; The register allocator is rather clumsy in handling multi-way conditional
;; moves, so allow the combiner to make them, and we split them up after
;; reload. */
(define_insn_and_split "*movsicc_umin"
[(set (match_operand:SI 0 "arith_reg_dest" "=&r")
(umin:SI (if_then_else:SI
(eq (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 0))
(match_operand:SI 2 "arith_reg_or_0_operand" "rN")
(match_operand:SI 3 "register_operand" "0"))
(match_operand:SI 4 "arith_reg_or_0_operand" "r")))
(clobber (match_scratch:SI 5 "=&r"))]
"TARGET_SHMEDIA && !can_create_pseudo_p ()"
"#"
"TARGET_SHMEDIA && reload_completed"
[(pc)]
{
emit_insn (gen_movsicc_false (operands[0], operands[1], operands[2],
operands[3]));
emit_insn (gen_cmpgtusi_media (operands[5], operands[4], operands[0]));
emit_insn (gen_movsicc_false (operands[0], operands[5], operands[4],
operands[0]));
DONE;
})
(define_insn "*movsicc_t_false"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(if_then_else (eq (reg:SI T_REG) (const_int 0))
(match_operand:SI 1 "general_movsrc_operand" "r,I08")
(match_operand:SI 2 "arith_reg_operand" "0,0")))]
"TARGET_PRETEND_CMOVE
&& (arith_reg_operand (operands[1], SImode)
|| (immediate_operand (operands[1], SImode)
&& satisfies_constraint_I08 (operands[1])))"
{
return "bt 0f" "\n"
" mov %1,%0" "\n"
"0:";
}
[(set_attr "type" "mt_group,arith") ;; poor approximation
(set_attr "length" "4")])
(define_insn "*movsicc_t_true"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(if_then_else (ne (reg:SI T_REG) (const_int 0))
(match_operand:SI 1 "general_movsrc_operand" "r,I08")
(match_operand:SI 2 "arith_reg_operand" "0,0")))]
"TARGET_PRETEND_CMOVE
&& (arith_reg_operand (operands[1], SImode)
|| (immediate_operand (operands[1], SImode)
&& satisfies_constraint_I08 (operands[1])))"
{
return "bf 0f" "\n"
" mov %1,%0" "\n"
"0:";
}
[(set_attr "type" "mt_group,arith") ;; poor approximation
(set_attr "length" "4")])
(define_expand "movsicc"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(if_then_else:SI (match_operand 1 "comparison_operator" "")
(match_operand:SI 2 "arith_reg_or_0_operand" "")
(match_operand:SI 3 "arith_reg_operand" "")))]
"TARGET_SHMEDIA || TARGET_PRETEND_CMOVE"
{
if ((GET_CODE (operands[1]) == EQ || GET_CODE (operands[1]) == NE)
&& GET_MODE (XEXP (operands[1], 0)) == SImode
&& (TARGET_SHMEDIA
|| (REG_P (XEXP (operands[1], 0))
&& REGNO (XEXP (operands[1], 0)) == T_REG))
&& XEXP (operands[1], 1) == const0_rtx)
;
else if (TARGET_PRETEND_CMOVE)
{
enum rtx_code code = GET_CODE (operands[1]);
enum rtx_code new_code = code;
rtx op0 = XEXP (operands[1], 0);
rtx op1 = XEXP (operands[1], 1);
if (! currently_expanding_to_rtl)
FAIL;
switch (code)
{
case LT: case LE: case LEU: case LTU:
if (GET_MODE_CLASS (GET_MODE (op0)) != MODE_INT)
break;
case NE:
new_code = reverse_condition (code);
break;
case EQ: case GT: case GE: case GEU: case GTU:
break;
default:
FAIL;
}
sh_emit_scc_to_t (new_code, op0, op1);
operands[1] = gen_rtx_fmt_ee (new_code == code ? NE : EQ, VOIDmode,
gen_rtx_REG (SImode, T_REG), const0_rtx);
}
else
{
if (!can_create_pseudo_p ())
FAIL;
operands[1] = sh_emit_cheap_store_flag (GET_MODE (operands[0]),
GET_CODE (operands[1]),
XEXP (operands[1], 0),
XEXP (operands[1], 1));
if (!operands[1])
FAIL;
}
})
(define_expand "movqicc"
[(set (match_operand:QI 0 "register_operand" "")
(if_then_else:QI (match_operand 1 "comparison_operator" "")
(match_operand:QI 2 "register_operand" "")
(match_operand:QI 3 "register_operand" "")))]
"TARGET_SHMEDIA"
{
operands[0] = simplify_gen_subreg (SImode, operands[0], QImode, 0);
operands[2] = simplify_gen_subreg (SImode, operands[2], QImode, 0);
operands[3] = simplify_gen_subreg (SImode, operands[3], QImode, 0);
emit (gen_movsicc (operands[0], operands[1], operands[2], operands[3]));
DONE;
})
;; -------------------------------------------------------------------------
;; Addition instructions
;; -------------------------------------------------------------------------
(define_expand "adddi3"
[(set (match_operand:DI 0 "arith_reg_operand")
(plus:DI (match_operand:DI 1 "arith_reg_operand")
(match_operand:DI 2 "arith_operand")))]
""
{
if (TARGET_SH1)
{
operands[2] = force_reg (DImode, operands[2]);
emit_insn (gen_adddi3_compact (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*adddi3_media"
[(set (match_operand:DI 0 "arith_reg_dest" "=r,r")
(plus:DI (match_operand:DI 1 "arith_reg_operand" "%r,r")
(match_operand:DI 2 "arith_operand" "r,I10")))]
"TARGET_SHMEDIA"
"@
add %1, %2, %0
addi %1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn "*adddisi3_media"
[(set (subreg:DI (match_operand:SI 0 "arith_reg_operand" "=r,r") 0)
(plus:DI (match_operand:DI 1 "arith_reg_operand" "%r,r")
(match_operand:DI 2 "arith_operand" "r,I10")))]
"TARGET_SHMEDIA"
"@
add.l %1, %2, %0
addi.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn "adddi3z_media"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extend:DI
(plus:SI (match_operand:SI 1 "extend_reg_operand" "r")
(match_operand:SI 2 "extend_reg_or_0_operand" "rN"))))]
"TARGET_SHMEDIA"
"addz.l %1, %N2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn_and_split "adddi3_compact"
[(set (match_operand:DI 0 "arith_reg_dest")
(plus:DI (match_operand:DI 1 "arith_reg_operand")
(match_operand:DI 2 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
emit_insn (gen_clrt ());
emit_insn (gen_addc (gen_lowpart (SImode, operands[0]),
gen_lowpart (SImode, operands[1]),
gen_lowpart (SImode, operands[2])));
emit_insn (gen_addc (gen_highpart (SImode, operands[0]),
gen_highpart (SImode, operands[1]),
gen_highpart (SImode, operands[2])));
DONE;
})
(define_insn "addc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(plus:SI (plus:SI (match_operand:SI 1 "arith_reg_operand" "%0")
(match_operand:SI 2 "arith_reg_operand" "r"))
(reg:SI T_REG)))
(set (reg:SI T_REG)
(ltu:SI (plus:SI (match_dup 1) (match_dup 2)) (match_dup 1)))]
"TARGET_SH1"
"addc %2,%0"
[(set_attr "type" "arith")])
;; A simplified version of the addc insn, where the exact value of the
;; T bit doesn't matter. This is easier for combine to pick up.
;; We allow a reg or 0 for one of the operands in order to be able to
;; do 'reg + T' sequences. Reload will load the constant 0 into the reg
;; as needed.
;; FIXME: The load of constant 0 should be split out before reload, or else
;; it will be difficult to hoist or combine the constant load.
(define_insn "*addc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(plus:SI (plus:SI (match_operand:SI 1 "arith_reg_operand" "%0")
(match_operand:SI 2 "arith_reg_or_0_operand" "r"))
(match_operand:SI 3 "t_reg_operand" "")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"addc %2,%0"
[(set_attr "type" "arith")])
;; Split 'reg + reg + 1' into a sett addc sequence, as it can be scheduled
;; better, if the sett insn can be done early.
(define_insn_and_split "*addc_r_r_1"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(plus:SI (plus:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "arith_reg_operand" ""))
(const_int 1)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(set (reg:SI T_REG) (const_int 1))
(parallel [(set (match_dup 0) (plus:SI (plus:SI (match_dup 1) (match_dup 2))
(reg:SI T_REG)))
(clobber (reg:SI T_REG))])])
;; Left shifts by one are usually done with an add insn to avoid T_REG
;; clobbers. Thus addc can also be used to do something like '(x << 1) + 1'.
(define_insn_and_split "*addc_2r_1"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (mult:SI (match_operand:SI 1 "arith_reg_operand")
(const_int 2))
(const_int 1)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(set (reg:SI T_REG) (const_int 1))
(parallel [(set (match_dup 0) (plus:SI (plus:SI (match_dup 1) (match_dup 1))
(reg:SI T_REG)))
(clobber (reg:SI T_REG))])])
;; Sometimes combine will try to do 'reg + (0-reg) + 1' if the *addc pattern
;; matched. Split this up into a simple sub add sequence, as this will save
;; us one sett insn.
(define_insn_and_split "*minus_plus_one"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(plus:SI (minus:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "arith_reg_operand" ""))
(const_int 1)))]
"TARGET_SH1"
"#"
"&& 1"
[(set (match_dup 0) (minus:SI (match_dup 1) (match_dup 2)))
(set (match_dup 0) (plus:SI (match_dup 0) (const_int 1)))])
;; Split 'reg + T' into 'reg + 0 + T' to utilize the addc insn.
;; If the 0 constant can be CSE-ed, this becomes a one instruction
;; operation, as opposed to sequences such as
;; movt r2
;; add r2,r3
;;
;; Even if the constant is not CSE-ed, a sequence such as
;; mov #0,r2
;; addc r2,r3
;; can be scheduled much better since the load of the constant can be
;; done earlier, before any comparison insns that store the result in
;; the T bit.
(define_insn_and_split "*addc_t_r"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (match_operand:SI 1 "t_reg_operand")
(match_operand:SI 2 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(parallel [(set (match_dup 0)
(plus:SI (plus:SI (match_dup 2) (const_int 0))
(match_dup 1)))
(clobber (reg:SI T_REG))])])
(define_insn_and_split "*addc_r_t"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "t_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(parallel [(set (match_dup 0)
(plus:SI (plus:SI (match_dup 1) (const_int 0))
(match_dup 2)))
(clobber (reg:SI T_REG))])])
;; Use shlr-addc to do 'reg + (reg & 1)'.
(define_insn_and_split "*addc_r_lsb"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (and:SI (match_operand:SI 1 "arith_reg_operand")
(const_int 1))
(match_operand:SI 2 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0) (plus:SI (reg:SI T_REG) (match_dup 2)))
(clobber (reg:SI T_REG))])]
{
emit_insn (gen_shlr (gen_reg_rtx (SImode), operands[1]));
})
;; Use shlr-addc to do 'reg + reg + (reg & 1)'.
(define_insn_and_split "*addc_r_r_lsb"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (plus:SI (and:SI (match_operand:SI 1 "arith_reg_operand")
(const_int 1))
(match_operand:SI 2 "arith_reg_operand"))
(match_operand:SI 3 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0) (plus:SI (plus:SI (match_dup 2) (match_dup 3))
(reg:SI T_REG)))
(clobber (reg:SI T_REG))])]
{
emit_insn (gen_shlr (gen_reg_rtx (SImode), operands[1]));
})
;; Canonicalize 'reg + (reg & 1) + reg' into 'reg + reg + (reg & 1)'.
(define_insn_and_split "*addc_r_lsb_r"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (and:SI (match_operand:SI 1 "arith_reg_operand")
(const_int 1))
(plus:SI (match_operand:SI 2 "arith_reg_operand")
(match_operand:SI 3 "arith_reg_operand"))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(plus:SI (plus:SI (and:SI (match_dup 1) (const_int 1))
(match_dup 2))
(match_dup 3)))
(clobber (reg:SI T_REG))])])
;; Canonicalize '2 * reg + (reg & 1)' into 'reg + reg + (reg & 1)'.
(define_insn_and_split "*addc_2r_lsb"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (and:SI (match_operand:SI 1 "arith_reg_operand")
(const_int 1))
(mult:SI (match_operand:SI 2 "arith_reg_operand")
(const_int 2))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(plus:SI (plus:SI (and:SI (match_dup 1) (const_int 1))
(match_dup 2))
(match_dup 2)))
(clobber (reg:SI T_REG))])])
;; Use shll-addc to do 'reg + ((unsigned int)reg >> 31)'.
(define_insn_and_split "*addc_r_msb"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (lshiftrt:SI (match_operand:SI 1 "arith_reg_operand")
(const_int 31))
(match_operand:SI 2 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0) (plus:SI (reg:SI T_REG) (match_dup 2)))
(clobber (reg:SI T_REG))])]
{
emit_insn (gen_shll (gen_reg_rtx (SImode), operands[1]));
})
;; Use shll-addc to do 'reg + reg + ((unsigned int)reg >> 31)'.
(define_insn_and_split "*addc_r_r_msb"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (plus:SI (lshiftrt:SI (match_operand:SI 1 "arith_reg_operand")
(const_int 31))
(match_operand:SI 2 "arith_reg_operand"))
(match_operand:SI 3 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0) (plus:SI (plus:SI (match_dup 2) (match_dup 3))
(reg:SI T_REG)))
(clobber (reg:SI T_REG))])]
{
emit_insn (gen_shll (gen_reg_rtx (SImode), operands[1]));
})
;; Canonicalize '2 * reg + ((unsigned int)reg >> 31)'
;; into 'reg + reg + (reg & 1)'.
(define_insn_and_split "*addc_2r_msb"
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (mult:SI (match_operand:SI 1 "arith_reg_operand")
(const_int 2))
(lshiftrt:SI (match_operand:SI 2 "arith_reg_operand")
(const_int 31))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(plus:SI (plus:SI (lshiftrt:SI (match_dup 2) (const_int 31))
(match_dup 1))
(match_dup 1)))
(clobber (reg:SI T_REG))])])
(define_expand "addsi3"
[(set (match_operand:SI 0 "arith_reg_operand" "")
(plus:SI (match_operand:SI 1 "arith_operand" "")
(match_operand:SI 2 "arith_operand" "")))]
""
{
if (TARGET_SHMEDIA)
operands[1] = force_reg (SImode, operands[1]);
})
(define_insn "addsi3_media"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(plus:SI (match_operand:SI 1 "extend_reg_operand" "%r,r")
(match_operand:SI 2 "arith_operand" "r,I10")))]
"TARGET_SHMEDIA"
"@
add.l %1, %2, %0
addi.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn "addsidi3_media"
[(set (match_operand:DI 0 "arith_reg_dest" "=r,r")
(sign_extend:DI (plus:SI (match_operand:SI 1 "extend_reg_operand"
"%r,r")
(match_operand:SI 2 "arith_operand"
"r,I10"))))]
"TARGET_SHMEDIA"
"@
add.l %1, %2, %0
addi.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn "*addsi3_compact"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(plus:SI (match_operand:SI 1 "arith_operand" "%0")
(match_operand:SI 2 "arith_operand" "rI08")))]
"TARGET_SH1"
"add %2,%0"
[(set_attr "type" "arith")])
;; -------------------------------------------------------------------------
;; Subtraction instructions
;; -------------------------------------------------------------------------
(define_expand "subdi3"
[(set (match_operand:DI 0 "arith_reg_operand" "")
(minus:DI (match_operand:DI 1 "arith_reg_or_0_operand" "")
(match_operand:DI 2 "arith_reg_operand" "")))]
""
{
if (TARGET_SH1)
{
operands[1] = force_reg (DImode, operands[1]);
emit_insn (gen_subdi3_compact (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*subdi3_media"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(minus:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rN")
(match_operand:DI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"sub %N1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn "subdisi3_media"
[(set (subreg:DI (match_operand:SI 0 "arith_reg_operand" "=r") 0)
(minus:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rN")
(match_operand:DI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"sub.l %N1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn_and_split "subdi3_compact"
[(set (match_operand:DI 0 "arith_reg_dest")
(minus:DI (match_operand:DI 1 "arith_reg_operand")
(match_operand:DI 2 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
emit_insn (gen_clrt ());
emit_insn (gen_subc (gen_lowpart (SImode, operands[0]),
gen_lowpart (SImode, operands[1]),
gen_lowpart (SImode, operands[2])));
emit_insn (gen_subc (gen_highpart (SImode, operands[0]),
gen_highpart (SImode, operands[1]),
gen_highpart (SImode, operands[2])));
DONE;
})
(define_insn "subc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(minus:SI (minus:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "arith_reg_operand" "r"))
(reg:SI T_REG)))
(set (reg:SI T_REG)
(gtu:SI (minus:SI (minus:SI (match_dup 1) (match_dup 2))
(reg:SI T_REG))
(match_dup 1)))]
"TARGET_SH1"
"subc %2,%0"
[(set_attr "type" "arith")])
;; A simplified version of the subc insn, where the exact value of the
;; T bit doesn't matter. This is easier for combine to pick up.
;; We allow a reg or 0 for one of the operands in order to be able to
;; do 'reg - T' sequences. Reload will load the constant 0 into the reg
;; as needed.
(define_insn "*subc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(minus:SI (minus:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "arith_reg_or_0_operand" "r"))
(match_operand:SI 3 "t_reg_operand" "")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"subc %2,%0"
[(set_attr "type" "arith")])
;; Split reg - reg - 1 into a sett subc sequence, as it can be scheduled
;; better, if the sett insn can be done early.
;; Notice that combine turns 'a - b - 1' into 'a + (~b)'.
(define_insn_and_split "*subc"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(plus:SI (not:SI (match_operand:SI 1 "arith_reg_operand" ""))
(match_operand:SI 2 "arith_reg_operand" "")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(set (reg:SI T_REG) (const_int 1))
(parallel [(set (match_dup 0)
(minus:SI (minus:SI (match_dup 2) (match_dup 1))
(reg:SI T_REG)))
(clobber (reg:SI T_REG))])])
;; Split 'reg - T' into 'reg - 0 - T' to utilize the subc insn.
;; If the 0 constant can be CSE-ed, this becomes a one instruction
;; operation, as opposed to sequences such as
;; movt r2
;; sub r2,r3
;;
;; Even if the constant is not CSE-ed, a sequence such as
;; mov #0,r2
;; subc r2,r3
;; can be scheduled much better since the load of the constant can be
;; done earlier, before any comparison insns that store the result in
;; the T bit.
(define_insn_and_split "*subc"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(minus:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "t_reg_operand" "")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(parallel [(set (match_dup 0)
(minus:SI (minus:SI (match_dup 1) (const_int 0))
(match_dup 2)))
(clobber (reg:SI T_REG))])])
(define_insn "*subsi3_internal"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(minus:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "arith_reg_operand" "r")))]
"TARGET_SH1"
"sub %2,%0"
[(set_attr "type" "arith")])
(define_insn_and_split "*subsi3_media"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(minus:SI (match_operand:SI 1 "minuend_operand" "rN")
(match_operand:SI 2 "extend_reg_operand" "r")))]
"TARGET_SHMEDIA
&& (operands[1] != constm1_rtx
|| (GET_CODE (operands[2]) != TRUNCATE
&& GET_CODE (operands[2]) != SUBREG))"
"sub.l %N1, %2, %0"
"operands[1] == constm1_rtx"
[(set (match_dup 0) (xor:SI (match_dup 2) (match_dup 1)))]
""
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_split
[(set (match_operand:SI 0 "arith_reg_dest" "")
(zero_extend:SI (subreg:QI (not:SI (subreg:SI (match_operand:QI 1
"general_extend_operand"
"") 0)) 0)))]
"TARGET_SHMEDIA && TARGET_LITTLE_ENDIAN"
[(set (match_dup 0) (zero_extend:SI (match_dup 1)))
(set (match_dup 0) (xor:SI (match_dup 0) (const_int 255)))]
"")
(define_split
[(set (match_operand:SI 0 "arith_reg_dest" "")
(zero_extend:SI (subreg:QI (not:SI (subreg:SI (match_operand:QI 1
"general_extend_operand"
"") 0)) 3)))]
"TARGET_SHMEDIA && TARGET_BIG_ENDIAN"
[(set (match_dup 0) (zero_extend:SI (match_dup 1)))
(set (match_dup 0) (xor:SI (match_dup 0) (const_int 255)))]
"")
;; Convert
;; constant - reg
;; to
;; neg reg
;; add reg, #const
;; since this will sometimes save one instruction.
;; Otherwise we might get a sequence like
;; mov #const, rY
;; sub rY, rX
;; mov rX, rY
;; if the source and dest regs are the same.
(define_expand "subsi3"
[(set (match_operand:SI 0 "arith_reg_operand" "")
(minus:SI (match_operand:SI 1 "arith_operand" "")
(match_operand:SI 2 "arith_reg_operand" "")))]
""
{
if (TARGET_SH1 && CONST_INT_P (operands[1]))
{
emit_insn (gen_negsi2 (operands[0], operands[2]));
emit_insn (gen_addsi3 (operands[0], operands[0], operands[1]));
DONE;
}
if (TARGET_SHMEDIA)
{
if (!can_create_pseudo_p ()
&& ! arith_reg_or_0_operand (operands[1], SImode))
FAIL;
if (operands[1] != const0_rtx && GET_CODE (operands[1]) != SUBREG)
operands[1] = force_reg (SImode, operands[1]);
}
})
;; -------------------------------------------------------------------------
;; Division instructions
;; -------------------------------------------------------------------------
;; We take advantage of the library routines which don't clobber as many
;; registers as a normal function call would.
;; The INSN_REFERENCES_ARE_DELAYED in sh.h is problematic because it
;; also has an effect on the register that holds the address of the sfunc.
;; To make this work, we have an extra dummy insn that shows the use
;; of this register for reorg.
(define_insn "use_sfunc_addr"
[(set (reg:SI PR_REG)
(unspec:SI [(match_operand:SI 0 "register_operand" "r")] UNSPEC_SFUNC))]
"TARGET_SH1 && check_use_sfunc_addr (insn, operands[0])"
""
[(set_attr "length" "0")])
(define_insn "udivsi3_sh2a"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(udiv:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "arith_reg_operand" "z")))]
"TARGET_SH2A"
"divu %2,%1"
[(set_attr "type" "arith")
(set_attr "in_delay_slot" "no")])
;; We must use a pseudo-reg forced to reg 0 in the SET_DEST rather than
;; hard register 0. If we used hard register 0, then the next instruction
;; would be a move from hard register 0 to a pseudo-reg. If the pseudo-reg
;; gets allocated to a stack slot that needs its address reloaded, then
;; there is nothing to prevent reload from using r0 to reload the address.
;; This reload would clobber the value in r0 we are trying to store.
;; If we let reload allocate r0, then this problem can never happen.
(define_insn "udivsi3_i1"
[(set (match_operand:SI 0 "register_operand" "=z")
(udiv:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R4_REG))
(use (match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_SH1 && TARGET_DIVIDE_CALL_DIV1"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
; Since shmedia-nofpu code could be linked against shcompact code, and
; the udivsi3 libcall has the same name, we must consider all registers
; clobbered that are in the union of the registers clobbered by the
; shmedia and the shcompact implementation. Note, if the shcompact
; implementation actually used shcompact code, we'd need to clobber
; also r23 and fr23.
(define_insn "udivsi3_i1_media"
[(set (match_operand:SI 0 "register_operand" "=z")
(udiv:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI T_MEDIA_REG))
(clobber (reg:SI PR_MEDIA_REG))
(clobber (reg:SI R20_REG))
(clobber (reg:SI R21_REG))
(clobber (reg:SI R22_REG))
(clobber (reg:DI TR0_REG))
(clobber (reg:DI TR1_REG))
(clobber (reg:DI TR2_REG))
(use (match_operand 1 "target_reg_operand" "b"))]
"TARGET_SHMEDIA && (! TARGET_SHMEDIA_FPU || ! TARGET_DIVIDE_FP)"
"blink %1, r18"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_expand "udivsi3_i4_media"
[(set (match_dup 3)
(zero_extend:DI (match_operand:SI 1 "register_operand" "")))
(set (match_dup 4)
(zero_extend:DI (match_operand:SI 2 "register_operand" "")))
(set (match_dup 5) (float:DF (match_dup 3)))
(set (match_dup 6) (float:DF (match_dup 4)))
(set (match_dup 7) (div:DF (match_dup 5) (match_dup 6)))
(set (match_dup 8) (fix:DI (match_dup 7)))
(set (match_operand:SI 0 "register_operand" "")
(truncate:SI (match_dup 8)))]
"TARGET_SHMEDIA_FPU"
{
operands[3] = gen_reg_rtx (DImode);
operands[4] = gen_reg_rtx (DImode);
operands[5] = gen_reg_rtx (DFmode);
operands[6] = gen_reg_rtx (DFmode);
operands[7] = gen_reg_rtx (DFmode);
operands[8] = gen_reg_rtx (DImode);
})
(define_insn "udivsi3_i4"
[(set (match_operand:SI 0 "register_operand" "=y")
(udiv:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:DF DR0_REG))
(clobber (reg:DF DR2_REG))
(clobber (reg:DF DR4_REG))
(clobber (reg:SI R0_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R4_REG))
(clobber (reg:SI R5_REG))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))
(use (match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_FPU_DOUBLE && ! TARGET_FPU_SINGLE"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "fp_mode" "double")
(set_attr "needs_delay_slot" "yes")])
(define_insn "udivsi3_i4_single"
[(set (match_operand:SI 0 "register_operand" "=y")
(udiv:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:DF DR0_REG))
(clobber (reg:DF DR2_REG))
(clobber (reg:DF DR4_REG))
(clobber (reg:SI R0_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R4_REG))
(clobber (reg:SI R5_REG))
(use (match_operand:SI 1 "arith_reg_operand" "r"))]
"(TARGET_FPU_SINGLE_ONLY || TARGET_FPU_DOUBLE || TARGET_SHCOMPACT)
&& TARGET_FPU_SINGLE"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_insn "udivsi3_i4_int"
[(set (match_operand:SI 0 "register_operand" "=z")
(udiv:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI T_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI MACH_REG))
(clobber (reg:SI MACL_REG))
(use (match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_SH1"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_expand "udivsi3"
[(set (match_dup 3) (symbol_ref:SI "__udivsi3"))
(set (reg:SI R4_REG) (match_operand:SI 1 "general_operand" ""))
(set (reg:SI R5_REG) (match_operand:SI 2 "general_operand" ""))
(parallel [(set (match_operand:SI 0 "register_operand" "")
(udiv:SI (reg:SI R4_REG)
(reg:SI R5_REG)))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R4_REG))
(use (match_dup 3))])]
""
{
rtx last;
operands[3] = gen_reg_rtx (Pmode);
/* Emit the move of the address to a pseudo outside of the libcall. */
if (TARGET_DIVIDE_CALL_TABLE)
{
/* libgcc2:__udivmoddi4 is not supposed to use an actual division, since
that causes problems when the divide code is supposed to come from a
separate library. Division by zero is undefined, so dividing 1 can be
implemented by comparing with the divisor. */
if (operands[1] == const1_rtx && currently_expanding_to_rtl)
{
rtx test = gen_rtx_GEU (VOIDmode, operands[1], operands[2]);
emit_insn (gen_cstoresi4 (operands[0], test,
operands[1], operands[2]));
DONE;
}
else if (operands[2] == const0_rtx)
{
emit_move_insn (operands[0], operands[2]);
DONE;
}
function_symbol (operands[3], "__udivsi3_i4i", SFUNC_GOT);
last = gen_udivsi3_i4_int (operands[0], operands[3]);
}
else if (TARGET_DIVIDE_CALL_FP)
{
function_symbol (operands[3], "__udivsi3_i4", SFUNC_STATIC);
if (TARGET_FPU_SINGLE)
last = gen_udivsi3_i4_single (operands[0], operands[3]);
else
last = gen_udivsi3_i4 (operands[0], operands[3]);
}
else if (TARGET_SHMEDIA_FPU)
{
operands[1] = force_reg (SImode, operands[1]);
operands[2] = force_reg (SImode, operands[2]);
emit_insn (gen_udivsi3_i4_media (operands[0], operands[1], operands[2]));
DONE;
}
else if (TARGET_SH2A)
{
operands[1] = force_reg (SImode, operands[1]);
operands[2] = force_reg (SImode, operands[2]);
emit_insn (gen_udivsi3_sh2a (operands[0], operands[1], operands[2]));
DONE;
}
else if (TARGET_SH5)
{
function_symbol (operands[3],
TARGET_FPU_ANY ? "__udivsi3_i4" : "__udivsi3",
SFUNC_STATIC);
if (TARGET_SHMEDIA)
last = gen_udivsi3_i1_media (operands[0], operands[3]);
else if (TARGET_FPU_ANY)
last = gen_udivsi3_i4_single (operands[0], operands[3]);
else
last = gen_udivsi3_i1 (operands[0], operands[3]);
}
else
{
function_symbol (operands[3], "__udivsi3", SFUNC_STATIC);
last = gen_udivsi3_i1 (operands[0], operands[3]);
}
emit_move_insn (gen_rtx_REG (SImode, 4), operands[1]);
emit_move_insn (gen_rtx_REG (SImode, 5), operands[2]);
emit_insn (last);
DONE;
})
(define_insn "divsi3_sh2a"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(div:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "arith_reg_operand" "z")))]
"TARGET_SH2A"
"divs %2,%1"
[(set_attr "type" "arith")
(set_attr "in_delay_slot" "no")])
(define_insn "divsi3_i1"
[(set (match_operand:SI 0 "register_operand" "=z")
(div:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R2_REG))
(clobber (reg:SI R3_REG))
(use (match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_SH1 && TARGET_DIVIDE_CALL_DIV1"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_insn "divsi3_i1_media"
[(set (match_operand:SI 0 "register_operand" "=z")
(div:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI T_MEDIA_REG))
(clobber (reg:SI PR_MEDIA_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R20_REG))
(clobber (reg:SI R21_REG))
(clobber (reg:SI TR0_REG))
(use (match_operand 1 "target_reg_operand" "b"))]
"TARGET_SHMEDIA && (! TARGET_SHMEDIA_FPU || ! TARGET_DIVIDE_FP)"
"blink %1, r18"
[(set_attr "type" "sfunc")])
(define_insn "divsi3_media_2"
[(set (match_operand:SI 0 "register_operand" "=z")
(div:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI T_MEDIA_REG))
(clobber (reg:SI PR_MEDIA_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R21_REG))
(clobber (reg:SI TR0_REG))
(use (reg:SI R20_REG))
(use (match_operand 1 "target_reg_operand" "b"))]
"TARGET_SHMEDIA && (! TARGET_SHMEDIA_FPU || ! TARGET_DIVIDE_FP)"
"blink %1, r18"
[(set_attr "type" "sfunc")])
;; This pattern acts as a placeholder for -mdiv=inv:call to carry
;; hard reg clobbers and data dependencies that we need when we want
;; to rematerialize the division into a call.
(define_insn_and_split "divsi_inv_call"
[(set (match_operand:SI 0 "register_operand" "=r")
(div:SI (match_operand:SI 1 "register_operand" "r")
(match_operand:SI 2 "register_operand" "r")))
(clobber (reg:SI R4_REG))
(clobber (reg:SI R5_REG))
(clobber (reg:SI T_MEDIA_REG))
(clobber (reg:SI PR_MEDIA_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R21_REG))
(clobber (reg:SI TR0_REG))
(clobber (reg:SI R20_REG))
(use (match_operand:SI 3 "register_operand" "r"))]
"TARGET_SHMEDIA"
"#"
"&& (reload_in_progress || reload_completed)"
[(set (match_dup 0) (match_dup 3))]
""
[(set_attr "highpart" "must_split")])
;; This is the combiner pattern for -mdiv=inv:call .
(define_insn_and_split "*divsi_inv_call_combine"
[(set (match_operand:SI 0 "register_operand" "=z")
(div:SI (match_operand:SI 1 "register_operand" "r")
(match_operand:SI 2 "register_operand" "r")))
(clobber (reg:SI R4_REG))
(clobber (reg:SI R5_REG))
(clobber (reg:SI T_MEDIA_REG))
(clobber (reg:SI PR_MEDIA_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R21_REG))
(clobber (reg:SI TR0_REG))
(clobber (reg:SI R20_REG))
(use (unspec:SI [(match_dup 1)
(match_operand:SI 3 "" "")
(unspec:SI [(match_operand:SI 4 "" "")
(match_dup 3)
(match_operand:DI 5 "" "")]
UNSPEC_DIV_INV_M2)
(match_operand:DI 6 "" "")
(const_int 0)
(const_int 0)]
UNSPEC_DIV_INV_M3))]
"TARGET_SHMEDIA"
"#"
"&& (reload_in_progress || reload_completed)"
[(pc)]
{
const char *name = sh_divsi3_libfunc;
enum sh_function_kind kind = SFUNC_GOT;
rtx sym;
emit_move_insn (gen_rtx_REG (SImode, R4_REG), operands[1]);
emit_move_insn (gen_rtx_REG (SImode, R5_REG), operands[2]);
while (TARGET_DIVIDE_INV_CALL2)
{
rtx x = operands[3];
if (GET_CODE (x) != UNSPEC || XINT (x, 1) != UNSPEC_DIV_INV_M1)
break;
x = XVECEXP (x, 0, 0);
name = "__sdivsi3_2";
kind = SFUNC_STATIC;
emit_move_insn (gen_rtx_REG (DImode, R20_REG), x);
break;
}
sym = function_symbol (NULL, name, kind);
emit_insn (gen_divsi3_media_2 (operands[0], sym));
DONE;
}
[(set_attr "highpart" "must_split")])
(define_expand "divsi3_i4_media"
[(set (match_dup 3) (float:DF (match_operand:SI 1 "register_operand" "r")))
(set (match_dup 4) (float:DF (match_operand:SI 2 "register_operand" "r")))
(set (match_dup 5) (div:DF (match_dup 3) (match_dup 4)))
(set (match_operand:SI 0 "register_operand" "=r")
(fix:SI (match_dup 5)))]
"TARGET_SHMEDIA_FPU"
{
operands[3] = gen_reg_rtx (DFmode);
operands[4] = gen_reg_rtx (DFmode);
operands[5] = gen_reg_rtx (DFmode);
})
(define_insn "divsi3_i4"
[(set (match_operand:SI 0 "register_operand" "=y")
(div:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI PR_REG))
(clobber (reg:DF DR0_REG))
(clobber (reg:DF DR2_REG))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))
(use (match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_FPU_DOUBLE && ! TARGET_FPU_SINGLE"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "fp_mode" "double")
(set_attr "needs_delay_slot" "yes")])
(define_insn "divsi3_i4_single"
[(set (match_operand:SI 0 "register_operand" "=y")
(div:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI PR_REG))
(clobber (reg:DF DR0_REG))
(clobber (reg:DF DR2_REG))
(clobber (reg:SI R2_REG))
(use (match_operand:SI 1 "arith_reg_operand" "r"))]
"(TARGET_FPU_SINGLE_ONLY || TARGET_FPU_DOUBLE || TARGET_SHCOMPACT)
&& TARGET_FPU_SINGLE"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_insn "divsi3_i4_int"
[(set (match_operand:SI 0 "register_operand" "=z")
(div:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI MACH_REG))
(clobber (reg:SI MACL_REG))
(use (match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_SH1"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_expand "divsi3"
[(set (match_dup 3) (symbol_ref:SI "__sdivsi3"))
(set (reg:SI R4_REG) (match_operand:SI 1 "general_operand" ""))
(set (reg:SI R5_REG) (match_operand:SI 2 "general_operand" ""))
(parallel [(set (match_operand:SI 0 "register_operand" "")
(div:SI (reg:SI R4_REG)
(reg:SI R5_REG)))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R2_REG))
(clobber (reg:SI R3_REG))
(use (match_dup 3))])]
""
{
rtx last;
operands[3] = gen_reg_rtx (Pmode);
/* Emit the move of the address to a pseudo outside of the libcall. */
if (TARGET_DIVIDE_CALL_TABLE)
{
function_symbol (operands[3], sh_divsi3_libfunc, SFUNC_GOT);
last = gen_divsi3_i4_int (operands[0], operands[3]);
}
else if (TARGET_DIVIDE_CALL_FP)
{
function_symbol (operands[3], sh_divsi3_libfunc, SFUNC_STATIC);
if (TARGET_FPU_SINGLE)
last = gen_divsi3_i4_single (operands[0], operands[3]);
else
last = gen_divsi3_i4 (operands[0], operands[3]);
}
else if (TARGET_SH2A)
{
operands[1] = force_reg (SImode, operands[1]);
operands[2] = force_reg (SImode, operands[2]);
emit_insn (gen_divsi3_sh2a (operands[0], operands[1], operands[2]));
DONE;
}
else if (TARGET_DIVIDE_INV)
{
rtx dividend = operands[1];
rtx divisor = operands[2];
rtx tab_base;
rtx nsb_res = gen_reg_rtx (DImode);
rtx norm64 = gen_reg_rtx (DImode);
rtx tab_ix = gen_reg_rtx (DImode);
rtx norm32 = gen_reg_rtx (SImode);
rtx i92 = force_reg (DImode, GEN_INT (92));
rtx scratch0a = gen_reg_rtx (DImode);
rtx scratch0b = gen_reg_rtx (DImode);
rtx inv0 = gen_reg_rtx (SImode);
rtx scratch1a = gen_reg_rtx (DImode);
rtx scratch1b = gen_reg_rtx (DImode);
rtx shift = gen_reg_rtx (DImode);
rtx i2p27, i43;
rtx inv1 = gen_reg_rtx (SImode);
rtx scratch2a = gen_reg_rtx (DImode);
rtx scratch2b = gen_reg_rtx (SImode);
rtx inv2 = gen_reg_rtx (SImode);
rtx scratch3a = gen_reg_rtx (DImode);
rtx scratch3b = gen_reg_rtx (DImode);
rtx scratch3c = gen_reg_rtx (DImode);
rtx scratch3d = gen_reg_rtx (SImode);
rtx scratch3e = gen_reg_rtx (DImode);
rtx result = gen_reg_rtx (SImode);
if (! arith_reg_or_0_operand (dividend, SImode))
dividend = force_reg (SImode, dividend);
if (! arith_reg_operand (divisor, SImode))
divisor = force_reg (SImode, divisor);
if (flag_pic && Pmode != DImode)
{
tab_base = gen_rtx_SYMBOL_REF (Pmode, "__div_table");
tab_base = gen_datalabel_ref (tab_base);
tab_base = force_reg (DImode, gen_rtx_SIGN_EXTEND (DImode, tab_base));
}
else
{
tab_base = gen_rtx_SYMBOL_REF (DImode, "__div_table");
tab_base = gen_datalabel_ref (tab_base);
tab_base = force_reg (DImode, tab_base);
}
if (TARGET_DIVIDE_INV20U)
i2p27 = force_reg (DImode, GEN_INT (-2 << 27));
else
i2p27 = GEN_INT (0);
if (TARGET_DIVIDE_INV20U || TARGET_DIVIDE_INV20L)
i43 = force_reg (DImode, GEN_INT (43));
else
i43 = GEN_INT (0);
emit_insn (gen_nsbdi (nsb_res,
simplify_gen_subreg (DImode, divisor, SImode, 0)));
emit_insn (gen_ashldi3_media (norm64,
gen_rtx_SUBREG (DImode, divisor, 0),
nsb_res));
emit_insn (gen_ashrdi3_media (tab_ix, norm64, GEN_INT (58)));
emit_insn (gen_ashrdisi3_media_high (norm32, norm64, GEN_INT (32)));
emit_insn (gen_divsi_inv_m1 (inv1, tab_base, tab_ix, norm32,
inv0, scratch0a, scratch0b,
scratch1a, scratch1b));
emit_insn (gen_subdi3 (shift, i92, nsb_res));
emit_insn (gen_divsi_inv_m2 (inv2, norm32, inv1, i92,
scratch2a));
emit_insn (gen_divsi_inv_m3 (result, dividend, inv1, inv2, shift,
i2p27, i43,
scratch3a, scratch3b, scratch3c,
scratch2a, scratch2b, scratch3d, scratch3e));
if (TARGET_DIVIDE_INV_CALL || TARGET_DIVIDE_INV_CALL2)
emit_insn (gen_divsi_inv_call (operands[0], dividend, divisor, result));
else if (TARGET_DIVIDE_INV_FP)
emit_insn (gen_divsi_inv_fp (operands[0], dividend, divisor, result,
gen_reg_rtx (SImode), gen_reg_rtx (SImode),
gen_reg_rtx (DFmode), gen_reg_rtx (DFmode),
gen_reg_rtx (DFmode)));
else
emit_move_insn (operands[0], result);
DONE;
}
else if (TARGET_SHMEDIA_FPU && TARGET_DIVIDE_FP)
{
operands[1] = force_reg (SImode, operands[1]);
operands[2] = force_reg (SImode, operands[2]);
emit_insn (gen_divsi3_i4_media (operands[0], operands[1], operands[2]));
DONE;
}
else if (TARGET_SH5)
{
if (TARGET_DIVIDE_CALL2)
{
rtx tab_base = gen_rtx_SYMBOL_REF (Pmode, "__div_table");
tab_base = gen_datalabel_ref (tab_base);
emit_move_insn (gen_rtx_REG (Pmode, R20_REG), tab_base);
}
if (TARGET_FPU_ANY && TARGET_SH1)
function_symbol (operands[3], sh_divsi3_libfunc, SFUNC_STATIC);
else if (TARGET_DIVIDE_CALL2)
function_symbol (operands[3], "__sdivsi3_2", SFUNC_STATIC);
else
function_symbol (operands[3], sh_divsi3_libfunc, SFUNC_GOT);
if (TARGET_SHMEDIA)
last = ((TARGET_DIVIDE_CALL2 ? gen_divsi3_media_2 : gen_divsi3_i1_media)
(operands[0], operands[3]));
else if (TARGET_FPU_ANY)
last = gen_divsi3_i4_single (operands[0], operands[3]);
else
last = gen_divsi3_i1 (operands[0], operands[3]);
}
else
{
function_symbol (operands[3], sh_divsi3_libfunc, SFUNC_GOT);
last = gen_divsi3_i1 (operands[0], operands[3]);
}
emit_move_insn (gen_rtx_REG (SImode, 4), operands[1]);
emit_move_insn (gen_rtx_REG (SImode, 5), operands[2]);
emit_insn (last);
DONE;
})
;; operands: scratch, tab_base, tab_ix
;; These are unspecs because we could generate an indexed addressing mode
;; even if -m5-32media, where INDEX_REG_CLASS == NO_REGS, and this would
;; confuse reload. See PR27117.
(define_insn "divsi_inv_qitable"
[(set (match_operand:DI 0 "register_operand" "=r")
(zero_extend:DI (unspec:QI [(match_operand:DI 1 "register_operand" "r")
(match_operand:DI 2 "register_operand" "r")]
UNSPEC_DIV_INV_TABLE)))]
"TARGET_SHMEDIA"
"ldx.ub %1, %2, %0"
[(set_attr "type" "load_media")
(set_attr "highpart" "user")])
;; operands: scratch, tab_base, tab_ix
(define_insn "divsi_inv_hitable"
[(set (match_operand:DI 0 "register_operand" "=r")
(sign_extend:DI (unspec:HI [(match_operand:DI 1 "register_operand" "r")
(match_operand:DI 2 "register_operand" "r")]
UNSPEC_DIV_INV_TABLE)))]
"TARGET_SHMEDIA"
"ldx.w %1, %2, %0"
[(set_attr "type" "load_media")
(set_attr "highpart" "user")])
;; operands: inv0, tab_base, tab_ix, norm32
;; scratch equiv in sdivsi3_2: r19, r21
(define_expand "divsi_inv_m0"
[(set (match_operand:SI 0 "register_operand" "=r")
(unspec:SI [(match_operand:DI 1 "register_operand" "r")
(match_operand:DI 2 "register_operand" "r")
(match_operand:SI 3 "register_operand" "r")]
UNSPEC_DIV_INV_M0))
(clobber (match_operand:DI 4 "register_operand" "=r"))
(clobber (match_operand:DI 5 "register_operand" "=r"))]
"TARGET_SHMEDIA"
{
/*
tab_base: r20
tab_ix: r21
norm32: r25
ldx.ub r20, r21, r19 // u0.8
shlli r21, 1, r21
muls.l r25, r19, r19 // s2.38
ldx.w r20, r21, r21 // s2.14
shari r19, 24, r19 // truncate to s2.14
sub r21, r19, r19 // some 11 bit inverse in s1.14
*/
rtx inv0 = operands[0];
rtx tab_base = operands[1];
rtx tab_ix = operands[2];
rtx norm32 = operands[3];
rtx scratch0 = operands[4];
rtx scratch0_si = gen_lowpart (SImode, scratch0);
rtx scratch1 = operands[5];
emit_insn (gen_divsi_inv_qitable (scratch0, tab_base, tab_ix));
emit_insn (gen_ashldi3_media (scratch1, tab_ix, GEN_INT (1)));
emit_insn (gen_mulsidi3_media (scratch0, norm32, scratch0_si));
emit_insn (gen_divsi_inv_hitable (scratch1, tab_base, scratch1));
emit_insn (gen_ashrdi3_media (scratch0, scratch0, GEN_INT (24)));
emit_insn (gen_subdisi3_media (inv0, scratch1, scratch0));
DONE;
})
;; operands: inv1, tab_base, tab_ix, norm32
(define_insn_and_split "divsi_inv_m1"
[(set (match_operand:SI 0 "register_operand" "=r")
(unspec:SI [(match_operand:DI 1 "register_operand" "r")
(match_operand:DI 2 "register_operand" "r")
(match_operand:SI 3 "register_operand" "r")]
UNSPEC_DIV_INV_M1))
(clobber (match_operand:SI 4 "register_operand" "=r"))
(clobber (match_operand:DI 5 "register_operand" "=r"))
(clobber (match_operand:DI 6 "register_operand" "=r"))
(clobber (match_operand:DI 7 "register_operand" "=r"))
(clobber (match_operand:DI 8 "register_operand" "=r"))]
"TARGET_SHMEDIA"
"#"
"&& !can_create_pseudo_p ()"
[(pc)]
{
/* inv0: r19
muls.l r19, r19, r18 // u0.28
muls.l r25, r18, r18 // s2.58
shlli r19, 45, r0 // multiply by two and convert to s2.58
sub r0, r18, r18
shari r18, 28, r18 // some 18 bit inverse in s1.30
*/
rtx inv1 = operands[0];
rtx tab_base = operands[1];
rtx tab_ix = operands[2];
rtx norm32 = operands[3];
rtx inv0 = operands[4];
rtx inv0_di = simplify_gen_subreg (DImode, inv0, SImode, 0);
rtx scratch0a = operands[5];
rtx scratch0b = operands[6];
rtx scratch0 = operands[7];
rtx scratch1 = operands[8];
rtx scratch1_si = gen_lowpart (SImode, scratch1);
emit_insn (gen_divsi_inv_m0 (inv0, tab_base, tab_ix, norm32,
scratch0a, scratch0b));
emit_insn (gen_mulsidi3_media (scratch1, inv0, inv0));
emit_insn (gen_mulsidi3_media (scratch1, norm32, scratch1_si));
emit_insn (gen_ashldi3_media (scratch0, inv0_di, GEN_INT (45)));
emit_insn (gen_subdi3 (scratch1, scratch0, scratch1));
emit_insn (gen_ashrdisi3_media_opaque (inv1, scratch1, GEN_INT (28)));
DONE;
})
;; operands: inv2, norm32, inv1, i92
(define_insn_and_split "divsi_inv_m2"
[(set (match_operand:SI 0 "register_operand" "=r")
(unspec:SI [(match_operand:SI 1 "register_operand" "r")
(match_operand:SI 2 "register_operand" "r")
(match_operand:DI 3 "register_operand" "r")]
UNSPEC_DIV_INV_M2))
(clobber (match_operand:DI 4 "register_operand" "=r"))]
"TARGET_SHMEDIA"
"#"
"&& !can_create_pseudo_p ()"
[(pc)]
{
/*
muls.l r18, r25, r0 // s2.60
shari r0, 16, r0 // s-16.44
sub
muls.l r0, r18, r19 // s-16.74
shari r19, 30, r19 // s-16.44
*/
rtx inv2 = operands[0];
rtx norm32 = operands[1];
rtx inv1 = operands[2];
rtx i92 = operands[3];
rtx scratch0 = operands[4];
rtx scratch0_si = gen_lowpart (SImode, scratch0);
emit_insn (gen_mulsidi3_media (scratch0, inv1, norm32));
emit_insn (gen_ashrdi3_media (scratch0, scratch0, GEN_INT (16)));
emit_insn (gen_subdi3 (scratch0, i92, scratch0));
emit_insn (gen_mulsidi3_media (scratch0, scratch0_si, inv1));
emit_insn (gen_ashrdisi3_media_opaque (inv2, scratch0, GEN_INT (30)));
DONE;
})
(define_insn_and_split "divsi_inv_m3"
[(set (match_operand:SI 0 "register_operand" "=r")
(unspec:SI [(match_operand:SI 1 "arith_reg_or_0_operand" "rN")
(match_operand:SI 2 "register_operand" "r")
(match_operand:SI 3 "register_operand" "r")
(match_operand:DI 4 "register_operand" "r")
(match_operand:DI 5 "arith_reg_or_0_operand" "rN")
(match_operand:DI 6 "arith_reg_or_0_operand" "rN")]
UNSPEC_DIV_INV_M3))
(clobber (match_operand:DI 7 "register_operand" "=r"))
(clobber (match_operand:DI 8 "register_operand" "=r"))
(clobber (match_operand:DI 9 "register_operand" "=r"))
(clobber (match_operand:DI 10 "register_operand" "=r"))
(clobber (match_operand:SI 11 "register_operand" "=r"))
(clobber (match_operand:SI 12 "register_operand" "=r"))
(clobber (match_operand:DI 13 "register_operand" "=r"))]
"TARGET_SHMEDIA"
"#"
"&& !can_create_pseudo_p ()"
[(pc)]
{
/*
r0: result r1: shift r4: dividend r18: inv1 r19: inv2
r0: scratch0 r19: scratch1 r21: scratch2
muls.l r18, r4, r25 // s32.30
muls.l r19, r4, r19 // s15.30
shari r25, 63, r21
shari r19, 14, r19 // s18.-14
sub r25, r19, r0
shard r0, r1, r0
sub r0, r21, r0
*/
rtx result = operands[0];
rtx dividend = operands[1];
rtx inv1 = operands[2];
rtx inv2 = operands[3];
rtx shift = operands[4];
rtx scratch0 = operands[7];
rtx scratch1 = operands[8];
rtx scratch2 = operands[9];
if (satisfies_constraint_N (dividend))
{
emit_move_insn (result, dividend);
DONE;
}
emit_insn (gen_mulsidi3_media (scratch0, inv1, dividend));
emit_insn (gen_mulsidi3_media (scratch1, inv2, dividend));
emit_insn (gen_ashrdi3_media (scratch2, scratch0, GEN_INT (63)));
emit_insn (gen_ashrdi3_media (scratch1, scratch1, GEN_INT (14)));
emit_insn (gen_adddi3 (scratch0, scratch0, scratch1));
emit_insn (gen_ashrdi3_media (scratch0, scratch0, shift));
emit_insn (gen_subdisi3_media (result, scratch0, scratch2));
DONE;
})
;; operands: quotient, dividend, inv1, inv2, shift, i2p27, i43
;; inv1: tab_base, tab_ix, norm32
;; inv2: norm32, inv1, i92
(define_insn_and_split "divsi_inv_m1_3"
[(set (match_operand:SI 0 "register_operand" "=r")
(unspec:SI [(match_operand:SI 1 "arith_reg_or_0_operand" "rN")
(unspec:SI [(match_operand:DI 2 "register_operand" "r")
(match_operand:DI 3 "register_operand" "r")
(match_operand:SI 4 "register_operand" "r")]
UNSPEC_DIV_INV_M1)
(unspec:SI [(match_dup 4)
(unspec:SI [(match_dup 2)
(match_dup 3)
(match_dup 4)] UNSPEC_DIV_INV_M1)
(match_operand:SI 5 "" "")]
UNSPEC_DIV_INV_M2)
(match_operand:DI 6 "register_operand" "r")
(match_operand:DI 7 "arith_reg_or_0_operand" "rN")
(match_operand:DI 8 "arith_reg_or_0_operand" "rN")]
UNSPEC_DIV_INV_M3))
(clobber (match_operand:DI 9 "register_operand" "=r"))
(clobber (match_operand:DI 10 "register_operand" "=r"))
(clobber (match_operand:DI 11 "register_operand" "=r"))
(clobber (match_operand:DI 12 "register_operand" "=r"))
(clobber (match_operand:SI 13 "register_operand" "=r"))
(clobber (match_operand:SI 14 "register_operand" "=r"))
(clobber (match_operand:DI 15 "register_operand" "=r"))]
"TARGET_SHMEDIA
&& (TARGET_DIVIDE_INV_MINLAT
|| TARGET_DIVIDE_INV20U || TARGET_DIVIDE_INV20L)"
"#"
"&& !can_create_pseudo_p ()"
[(pc)]
{
rtx result = operands[0];
rtx dividend = operands[1];
rtx tab_base = operands[2];
rtx tab_ix = operands[3];
rtx norm32 = operands[4];
/* rtx i92 = operands[5]; */
rtx shift = operands[6];
rtx i2p27 = operands[7];
rtx i43 = operands[8];
rtx scratch0 = operands[9];
rtx scratch0_si = gen_lowpart (SImode, scratch0);
rtx scratch1 = operands[10];
rtx scratch1_si = gen_lowpart (SImode, scratch1);
rtx scratch2 = operands[11];
rtx scratch3 = operands[12];
rtx scratch4 = operands[13];
rtx scratch4_di = simplify_gen_subreg (DImode, scratch4, SImode, 0);
rtx scratch5 = operands[14];
rtx scratch5_di = simplify_gen_subreg (DImode, scratch5, SImode, 0);
rtx scratch6 = operands[15];
emit_insn (gen_divsi_inv_m0 (scratch4, tab_base, tab_ix, norm32,
scratch0, scratch1));
/* inv0 == scratch4 */
if (! TARGET_DIVIDE_INV20U)
{
emit_insn (gen_mulsidi3_media (scratch0, scratch4, scratch4));
i2p27 = scratch0;
emit_insn (gen_mulsidi3_media (scratch1, norm32, scratch0_si));
}
else
{
emit_insn (gen_mulsidi3_media (scratch1, scratch4, scratch4));
emit_insn (gen_mulsidi3_media (scratch1, norm32, scratch1_si));
}
emit_insn (gen_ashldi3_media (scratch2, scratch4_di, GEN_INT (45)));
emit_insn (gen_subdi3 (scratch1, scratch2, scratch1));
emit_insn (gen_ashrdisi3_media_opaque (scratch4, scratch1, GEN_INT (28)));
/* inv1 == scratch4 */
if (TARGET_DIVIDE_INV_MINLAT)
{
emit_insn (gen_mulsidi3_media (scratch1, scratch4, norm32));
emit_insn (gen_mulsidi3_media (scratch2, dividend, scratch4));
emit_insn (gen_ashrdi3_media (scratch1, scratch1, GEN_INT (16)));
emit_insn (gen_mulsidi3_media (scratch1, scratch1_si, scratch4));
emit_insn (gen_ashrdi3_media (scratch3, scratch2, GEN_INT (63)));
emit_insn (gen_ashrsi3_media (scratch5, dividend, GEN_INT (14)));
emit_insn (gen_ashrdi3_media (scratch1, scratch1, GEN_INT (30)));
emit_insn (gen_mulsidi3_media (scratch1, scratch1_si, scratch5));
emit_insn (gen_xordi3 (scratch0, scratch3, i2p27));
emit_insn (gen_adddi3 (scratch2, scratch2, scratch0));
emit_insn (gen_subdi3 (scratch2, scratch2, scratch1));
}
else
{
rtx label = gen_rtx_LABEL_REF (Pmode, gen_label_rtx ());
/* Use separate scratch regs for nsb and sign to allow scheduling. */
emit_insn (gen_nsbdi (scratch6,
simplify_gen_subreg (DImode, dividend, SImode, 0)));
emit_insn (gen_xorsi3 (scratch5, dividend, norm32));
emit_insn (gen_ashrdi3_media (scratch3, scratch5_di, GEN_INT (63)));
emit_insn (gen_divsi_inv20 (scratch2,
norm32, scratch4, dividend,
scratch6, scratch3, i43,
/* scratch0 may be shared with i2p27. */
scratch0, scratch1, scratch5,
label, label, i2p27));
}
emit_insn (gen_ashrdi3_media (scratch2, scratch2, shift));
emit_insn (gen_subdisi3_media (result, scratch2, scratch3));
DONE;
})
(define_insn "divsi_inv20"
[(set (match_operand:DI 0 "register_operand" "=&r")
(unspec:DI [(match_operand:SI 1 "register_operand" "r")
(match_operand:SI 2 "register_operand" "r")
(match_operand:SI 3 "register_operand" "r")
(match_operand:DI 4 "register_operand" "r")
(match_operand:DI 5 "register_operand" "r")
(match_operand:DI 6 "register_operand" "r")
(match_operand:DI 12 "register_operand" "r")
(match_operand 10 "target_operand" "b")
(match_operand 11 "immediate_operand" "i")]
UNSPEC_DIV_INV20))
(clobber (match_operand:DI 7 "register_operand" "=&r"))
(clobber (match_operand:DI 8 "register_operand" "=&r"))
(clobber (match_operand:SI 9 "register_operand" "=r"))]
"TARGET_SHMEDIA
&& (TARGET_DIVIDE_INV20U || TARGET_DIVIDE_INV20L)"
{
/* operands: %0 div_result, %1 norm32, %2 inv1, %3 dividend,
%4 dividend_nsb, %5 result_sign, %6 i43, %12 i2p27,
%7 round_scratch, %8 scratch0 (di), %9 scratch1 (si)
%10 label (tr), %11 label (imm)
muls.l inv1, norm32, scratch0 // s2.60
muls.l inv1, dividend, result // s32.30
xor i2p27, result_sign, round_scratch
bge/u dividend_nsb, i43, tr.. (label)
shari scratch0, 16, scratch0 // s-16.44
muls.l sratch0_si, inv1, scratch0 // s-16.74
sub result, round_scratch, result
shari dividend, 14, scratch1 // s19.-14
shari scratch0, 30, scratch0 // s-16.44
muls.l scratch0, scratch1, round_scratch // s15.30
label:
sub result, round_scratch, result */
const bool likely = TARGET_DIVIDE_INV20L;
if (likely)
return
"muls.l %2, %3, %0" "\n"
" xor %12, %5, %7" "\n"
" bge/l %4, %6, %10" "\n"
" muls.l %2, %1, %8" "\n"
" shari %8, 16, %8" "\n"
" muls.l %8, %2, %8" "\n"
" shari %3, 14, %9" "\n"
" shari %8, 30, %8" "\n"
" muls.l %8, %9, %8" "\n"
" sub %0, %8, %0" "\n"
"%11: add %0, %7, %0";
else
return
"muls.l %2, %1, %8" "\n"
" muls.l %2, %3, %0" "\n"
" xor %12, %5, %7" "\n"
" bge/u %4, %6, %10" "\n"
" shari %8, 16, %8" "\n"
" muls.l %8, %2, %8" "\n"
" sub %0, %7, %0" "\n"
" shari %3, 14, %9" "\n"
" shari %8, 30, %8" "\n"
" muls.l %8, %9, %7" "\n"
"%11: sub %0, %7, %0";
})
(define_insn_and_split "divsi_inv_fp"
[(set (match_operand:SI 0 "general_movdst_operand" "=rf")
(div:SI (match_operand:SI 1 "general_movsrc_operand" "rf")
(match_operand:SI 2 "register_operand" "rf")))
(use (match_operand:SI 3 "general_movsrc_operand" "r"))
(clobber (match_operand:SI 4 "register_operand" "=r"))
(clobber (match_operand:SI 5 "register_operand" "=r"))
(clobber (match_operand:DF 6 "register_operand" "=r"))
(clobber (match_operand:DF 7 "register_operand" "=r"))
(clobber (match_operand:DF 8 "register_operand" "=r"))]
"TARGET_SHMEDIA_FPU"
"#"
"&& (reload_in_progress || reload_completed)"
[(set (match_dup 0) (match_dup 3))]
""
[(set_attr "highpart" "must_split")])
;; If a matching group of divide-by-inverse instructions is in the same
;; basic block after gcse & loop optimizations, we want to transform them
;; to a straight division using floating point for TARGET_DIVIDE_INV_FP.
(define_insn_and_split "*divsi_inv_fp_combine"
[(set (match_operand:SI 0 "register_operand" "=f")
(div:SI (match_operand:SI 1 "register_operand" "f")
(match_operand:SI 2 "register_operand" "f")))
(use (unspec:SI [(match_dup 1)
(match_operand:SI 3 "" "")
(unspec:SI [(match_operand:SI 4 "" "")
(match_dup 3)
(match_operand:DI 5 "" "")] UNSPEC_DIV_INV_M2)
(match_operand:DI 6 "" "")
(const_int 0)
(const_int 0)] UNSPEC_DIV_INV_M3))
(clobber (match_operand:SI 7 "fp_arith_reg_operand" ""))
(clobber (match_operand:SI 8 "fp_arith_reg_operand" ""))
(clobber (match_operand:DF 9 "fp_arith_reg_operand" ""))
(clobber (match_operand:DF 10 "fp_arith_reg_operand" ""))
(clobber (match_operand:DF 11 "fp_arith_reg_operand" ""))]
"TARGET_SHMEDIA_FPU && TARGET_DIVIDE_INV_FP && !can_create_pseudo_p ()"
"#"
"&& 1"
[(set (match_dup 9) (float:DF (match_dup 1)))
(set (match_dup 10) (float:DF (match_dup 2)))
(set (match_dup 11) (div:DF (match_dup 9) (match_dup 10)))
(set (match_dup 8)
(fix:SI (match_dup 11)))
(set (match_dup 0) (match_dup 8))]
{
if (! fp_arith_reg_operand (operands[1], SImode))
{
emit_move_insn (operands[7], operands[1]);
operands[1] = operands[7];
}
if (! fp_arith_reg_operand (operands[2], SImode))
{
emit_move_insn (operands[8], operands[2]);
operands[2] = operands[8];
}
}
[(set_attr "highpart" "must_split")])
;; -------------------------------------------------------------------------
;; Multiplication instructions
;; -------------------------------------------------------------------------
(define_insn "umulhisi3_i"
[(set (reg:SI MACL_REG)
(mult:SI (zero_extend:SI
(match_operand:HI 0 "arith_reg_operand" "r"))
(zero_extend:SI
(match_operand:HI 1 "arith_reg_operand" "r"))))]
"TARGET_SH1"
"mulu.w %1,%0"
[(set_attr "type" "smpy")])
(define_insn "mulhisi3_i"
[(set (reg:SI MACL_REG)
(mult:SI (sign_extend:SI
(match_operand:HI 0 "arith_reg_operand" "r"))
(sign_extend:SI
(match_operand:HI 1 "arith_reg_operand" "r"))))]
"TARGET_SH1"
"muls.w %1,%0"
[(set_attr "type" "smpy")])
(define_expand "mulhisi3"
[(set (reg:SI MACL_REG)
(mult:SI (sign_extend:SI
(match_operand:HI 1 "arith_reg_operand" ""))
(sign_extend:SI
(match_operand:HI 2 "arith_reg_operand" ""))))
(set (match_operand:SI 0 "arith_reg_operand" "")
(reg:SI MACL_REG))]
"TARGET_SH1"
{
rtx_insn *insn;
rtx macl;
macl = gen_rtx_REG (SImode, MACL_REG);
start_sequence ();
emit_insn (gen_mulhisi3_i (operands[1], operands[2]));
insn = get_insns ();
end_sequence ();
/* expand_binop can't find a suitable code in umul_widen_optab to
make a REG_EQUAL note from, so make one here.
See also smulsi3_highpart.
??? Alternatively, we could put this at the calling site of expand_binop,
i.e. expand_expr. */
/* Use emit_libcall_block for loop invariant code motion and to make
a REG_EQUAL note. */
emit_libcall_block (insn, operands[0], macl, SET_SRC (single_set (insn)));
DONE;
})
(define_expand "umulhisi3"
[(set (reg:SI MACL_REG)
(mult:SI (zero_extend:SI
(match_operand:HI 1 "arith_reg_operand" ""))
(zero_extend:SI
(match_operand:HI 2 "arith_reg_operand" ""))))
(set (match_operand:SI 0 "arith_reg_operand" "")
(reg:SI MACL_REG))]
"TARGET_SH1"
{
rtx_insn *insn;
rtx macl;
macl = gen_rtx_REG (SImode, MACL_REG);
start_sequence ();
emit_insn (gen_umulhisi3_i (operands[1], operands[2]));
insn = get_insns ();
end_sequence ();
/* expand_binop can't find a suitable code in umul_widen_optab to
make a REG_EQUAL note from, so make one here.
See also smulsi3_highpart.
??? Alternatively, we could put this at the calling site of expand_binop,
i.e. expand_expr. */
/* Use emit_libcall_block for loop invariant code motion and to make
a REG_EQUAL note. */
emit_libcall_block (insn, operands[0], macl, SET_SRC (single_set (insn)));
DONE;
})
;; mulsi3 on the SH2 can be done in one instruction, on the SH1 we generate
;; a call to a routine which clobbers known registers.
(define_insn ""
[(set (match_operand:SI 1 "register_operand" "=z")
(mult:SI (reg:SI R4_REG) (reg:SI R5_REG)))
(clobber (reg:SI MACL_REG))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R3_REG))
(clobber (reg:SI R2_REG))
(clobber (reg:SI R1_REG))
(use (match_operand:SI 0 "arith_reg_operand" "r"))]
"TARGET_SH1"
"jsr @%0%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_expand "mulsi3_call"
[(set (reg:SI R4_REG) (match_operand:SI 1 "general_operand" ""))
(set (reg:SI R5_REG) (match_operand:SI 2 "general_operand" ""))
(parallel[(set (match_operand:SI 0 "register_operand" "")
(mult:SI (reg:SI R4_REG)
(reg:SI R5_REG)))
(clobber (reg:SI MACL_REG))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R3_REG))
(clobber (reg:SI R2_REG))
(clobber (reg:SI R1_REG))
(use (match_operand:SI 3 "register_operand" ""))])]
"TARGET_SH1"
"")
(define_insn "mul_r"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(mult:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "arith_reg_operand" "z")))]
"TARGET_SH2A"
"mulr %2,%0"
[(set_attr "type" "dmpy")])
(define_insn "mul_l"
[(set (reg:SI MACL_REG)
(mult:SI (match_operand:SI 0 "arith_reg_operand" "r")
(match_operand:SI 1 "arith_reg_operand" "r")))]
"TARGET_SH2"
"mul.l %1,%0"
[(set_attr "type" "dmpy")])
(define_expand "mulsi3"
[(set (reg:SI MACL_REG)
(mult:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "arith_reg_operand" "")))
(set (match_operand:SI 0 "arith_reg_operand" "")
(reg:SI MACL_REG))]
"TARGET_SH1"
{
if (!TARGET_SH2)
{
/* The address must be set outside the libcall,
since it goes into a pseudo. */
rtx sym = function_symbol (NULL, "__mulsi3", SFUNC_STATIC);
rtx addr = force_reg (SImode, sym);
rtx insns = gen_mulsi3_call (operands[0], operands[1],
operands[2], addr);
emit_insn (insns);
}
else
{
rtx macl = gen_rtx_REG (SImode, MACL_REG);
emit_insn (gen_mul_l (operands[1], operands[2]));
/* consec_sets_giv can only recognize the first insn that sets a
giv as the giv insn. So we must tag this also with a REG_EQUAL
note. */
emit_insn (gen_movsi_i ((operands[0]), macl));
}
DONE;
})
(define_insn "mulsidi3_i"
[(set (reg:SI MACH_REG)
(truncate:SI
(lshiftrt:DI
(mult:DI
(sign_extend:DI (match_operand:SI 0 "arith_reg_operand" "r"))
(sign_extend:DI (match_operand:SI 1 "arith_reg_operand" "r")))
(const_int 32))))
(set (reg:SI MACL_REG)
(mult:SI (match_dup 0)
(match_dup 1)))]
"TARGET_SH2"
"dmuls.l %1,%0"
[(set_attr "type" "dmpy")])
(define_expand "mulsidi3"
[(set (match_operand:DI 0 "arith_reg_dest" "")
(mult:DI (sign_extend:DI (match_operand:SI 1 "arith_reg_operand" ""))
(sign_extend:DI (match_operand:SI 2 "arith_reg_operand" ""))))]
"TARGET_SH2 || TARGET_SHMEDIA"
{
if (TARGET_SH2)
{
emit_insn (gen_mulsidi3_compact (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "mulsidi3_media"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(mult:DI (sign_extend:DI (match_operand:SI 1 "extend_reg_operand" "%r"))
(sign_extend:DI (match_operand:SI 2 "extend_reg_operand" "r"))))]
"TARGET_SHMEDIA"
"muls.l %1, %2, %0"
[(set_attr "type" "dmpy_media")
(set_attr "highpart" "ignore")])
(define_insn_and_split "mulsidi3_compact"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(mult:DI
(sign_extend:DI (match_operand:SI 1 "arith_reg_operand" "r"))
(sign_extend:DI (match_operand:SI 2 "arith_reg_operand" "r"))))
(clobber (reg:SI MACH_REG))
(clobber (reg:SI MACL_REG))]
"TARGET_SH2"
"#"
"&& 1"
[(const_int 0)]
{
rtx low_dst = gen_lowpart (SImode, operands[0]);
rtx high_dst = gen_highpart (SImode, operands[0]);
emit_insn (gen_mulsidi3_i (operands[1], operands[2]));
emit_move_insn (low_dst, gen_rtx_REG (SImode, MACL_REG));
emit_move_insn (high_dst, gen_rtx_REG (SImode, MACH_REG));
/* We need something to tag the possible REG_EQUAL notes on to. */
emit_move_insn (operands[0], operands[0]);
DONE;
})
(define_insn "umulsidi3_i"
[(set (reg:SI MACH_REG)
(truncate:SI
(lshiftrt:DI
(mult:DI
(zero_extend:DI (match_operand:SI 0 "arith_reg_operand" "r"))
(zero_extend:DI (match_operand:SI 1 "arith_reg_operand" "r")))
(const_int 32))))
(set (reg:SI MACL_REG)
(mult:SI (match_dup 0)
(match_dup 1)))]
"TARGET_SH2"
"dmulu.l %1,%0"
[(set_attr "type" "dmpy")])
(define_expand "umulsidi3"
[(set (match_operand:DI 0 "arith_reg_dest" "")
(mult:DI (zero_extend:DI (match_operand:SI 1 "arith_reg_operand" ""))
(zero_extend:DI (match_operand:SI 2 "arith_reg_operand" ""))))]
"TARGET_SH2 || TARGET_SHMEDIA"
{
if (TARGET_SH2)
{
emit_insn (gen_umulsidi3_compact (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "umulsidi3_media"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(mult:DI (zero_extend:DI (match_operand:SI 1 "extend_reg_operand" "%r"))
(zero_extend:DI (match_operand:SI 2 "extend_reg_operand" "r"))))]
"TARGET_SHMEDIA"
"mulu.l %1, %2, %0"
[(set_attr "type" "dmpy_media")
(set_attr "highpart" "ignore")])
(define_insn_and_split "umulsidi3_compact"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(mult:DI
(zero_extend:DI (match_operand:SI 1 "arith_reg_operand" "r"))
(zero_extend:DI (match_operand:SI 2 "arith_reg_operand" "r"))))
(clobber (reg:SI MACH_REG))
(clobber (reg:SI MACL_REG))]
"TARGET_SH2"
"#"
"&& 1"
[(const_int 0)]
{
rtx low_dst = gen_lowpart (SImode, operands[0]);
rtx high_dst = gen_highpart (SImode, operands[0]);
emit_insn (gen_umulsidi3_i (operands[1], operands[2]));
emit_move_insn (low_dst, gen_rtx_REG (SImode, MACL_REG));
emit_move_insn (high_dst, gen_rtx_REG (SImode, MACH_REG));
/* We need something to tag the possible REG_EQUAL notes on to. */
emit_move_insn (operands[0], operands[0]);
DONE;
})
(define_insn "smulsi3_highpart_i"
[(set (reg:SI MACH_REG)
(truncate:SI
(lshiftrt:DI
(mult:DI
(sign_extend:DI (match_operand:SI 0 "arith_reg_operand" "r"))
(sign_extend:DI (match_operand:SI 1 "arith_reg_operand" "r")))
(const_int 32))))
(clobber (reg:SI MACL_REG))]
"TARGET_SH2"
"dmuls.l %1,%0"
[(set_attr "type" "dmpy")])
(define_expand "smulsi3_highpart"
[(parallel
[(set (reg:SI MACH_REG)
(truncate:SI
(lshiftrt:DI
(mult:DI
(sign_extend:DI (match_operand:SI 1 "arith_reg_operand" ""))
(sign_extend:DI (match_operand:SI 2 "arith_reg_operand" "")))
(const_int 32))))
(clobber (reg:SI MACL_REG))])
(set (match_operand:SI 0 "arith_reg_operand" "")
(reg:SI MACH_REG))]
"TARGET_SH2"
{
rtx_insn *insn;
rtx mach;
mach = gen_rtx_REG (SImode, MACH_REG);
start_sequence ();
emit_insn (gen_smulsi3_highpart_i (operands[1], operands[2]));
insn = get_insns ();
end_sequence ();
/* expand_binop can't find a suitable code in mul_highpart_optab to
make a REG_EQUAL note from, so make one here.
See also {,u}mulhisi.
??? Alternatively, we could put this at the calling site of expand_binop,
i.e. expand_mult_highpart. */
/* Use emit_libcall_block for loop invariant code motion and to make
a REG_EQUAL note. */
emit_libcall_block (insn, operands[0], mach, SET_SRC (single_set (insn)));
DONE;
})
(define_insn "umulsi3_highpart_i"
[(set (reg:SI MACH_REG)
(truncate:SI
(lshiftrt:DI
(mult:DI
(zero_extend:DI (match_operand:SI 0 "arith_reg_operand" "r"))
(zero_extend:DI (match_operand:SI 1 "arith_reg_operand" "r")))
(const_int 32))))
(clobber (reg:SI MACL_REG))]
"TARGET_SH2"
"dmulu.l %1,%0"
[(set_attr "type" "dmpy")])
(define_expand "umulsi3_highpart"
[(parallel
[(set (reg:SI MACH_REG)
(truncate:SI
(lshiftrt:DI
(mult:DI
(zero_extend:DI (match_operand:SI 1 "arith_reg_operand" ""))
(zero_extend:DI (match_operand:SI 2 "arith_reg_operand" "")))
(const_int 32))))
(clobber (reg:SI MACL_REG))])
(set (match_operand:SI 0 "arith_reg_operand" "")
(reg:SI MACH_REG))]
"TARGET_SH2"
{
rtx_insn *insn;
rtx mach;
mach = gen_rtx_REG (SImode, MACH_REG);
start_sequence ();
emit_insn (gen_umulsi3_highpart_i (operands[1], operands[2]));
insn = get_insns ();
end_sequence ();
/* Use emit_libcall_block for loop invariant code motion and to make
a REG_EQUAL note. */
emit_libcall_block (insn, operands[0], mach, SET_SRC (single_set (insn)));
DONE;
})
(define_insn_and_split "muldi3"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(mult:DI (match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "arith_reg_operand" "r")))
(clobber (match_scratch:DI 3 "=&r"))
(clobber (match_scratch:DI 4 "=r"))]
"TARGET_SHMEDIA"
"#"
"reload_completed"
[(const_int 0)]
{
rtx op3_v2si, op2_v2si;
op3_v2si = operands[3];
if (GET_CODE (op3_v2si) == SIGN_EXTEND)
{
op3_v2si = XEXP (op3_v2si, 0);
op3_v2si = simplify_gen_subreg (DImode, op3_v2si, GET_MODE (op3_v2si), 0);
}
op3_v2si = simplify_gen_subreg (V2SImode, op3_v2si, DImode, 0);
op2_v2si = operands[2];
if (GET_CODE (op2_v2si) == SIGN_EXTEND)
{
op2_v2si = XEXP (op2_v2si, 0);
op2_v2si = simplify_gen_subreg (DImode, op2_v2si, GET_MODE (op2_v2si), 0);
}
op2_v2si = simplify_gen_subreg (V2SImode, op2_v2si, DImode, 0);
emit_insn (gen_rotldi3 (operands[3], operands[1], GEN_INT (32)));
emit_insn (gen_mulv2si3 (op3_v2si, op3_v2si, op2_v2si));
emit_insn (gen_umulsidi3_media (operands[4],
sh_gen_truncate (SImode, operands[1], 0),
sh_gen_truncate (SImode, operands[2], 0)));
emit_insn (gen_anddi3 (operands[0], operands[3], GEN_INT (0xffffffff00000000LL)));
emit_insn (gen_ashldi3_media (operands[3], operands[3], GEN_INT (32)));
emit_insn (gen_adddi3 (operands[0], operands[3], operands[0]));
emit_insn (gen_adddi3 (operands[0], operands[4], operands[0]));
DONE;
})
;; -------------------------------------------------------------------------
;; Logical operations
;; -------------------------------------------------------------------------
(define_expand "andsi3"
[(set (match_operand:SI 0 "arith_reg_operand" "")
(and:SI (match_operand:SI 1 "logical_reg_operand" "")
(match_operand:SI 2 "logical_and_operand" "")))]
""
{
/* If it is possible to turn the and insn into a zero extension
already, redundant zero extensions will be folded, which results
in better code.
Ideally the splitter of *andsi_compact would be enough, if redundant
zero extensions were detected after the combine pass, which does not
happen at the moment. */
if (TARGET_SH1)
{
if (satisfies_constraint_Jmb (operands[2]))
{
emit_insn (gen_zero_extendqisi2 (operands[0],
gen_lowpart (QImode, operands[1])));
DONE;
}
else if (satisfies_constraint_Jmw (operands[2]))
{
emit_insn (gen_zero_extendhisi2 (operands[0],
gen_lowpart (HImode, operands[1])));
DONE;
}
}
})
(define_insn_and_split "*andsi_compact"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r,z,r")
(and:SI (match_operand:SI 1 "arith_reg_operand" "%r,r,0,0")
(match_operand:SI 2 "logical_and_operand" "Jmb,Jmw,K08,r")))]
"TARGET_SH1"
"@
extu.b %1,%0
extu.w %1,%0
and %2,%0
and %2,%0"
"&& 1"
[(set (match_dup 0) (zero_extend:SI (match_dup 1)))]
{
if (satisfies_constraint_Jmb (operands[2]))
operands[1] = gen_lowpart (QImode, operands[1]);
else if (satisfies_constraint_Jmw (operands[2]))
operands[1] = gen_lowpart (HImode, operands[1]);
else
FAIL;
}
[(set_attr "type" "arith")])
(define_insn "*andsi3_media"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(and:SI (match_operand:SI 1 "logical_reg_operand" "%r,r")
(match_operand:SI 2 "logical_operand" "r,I10")))]
"TARGET_SHMEDIA"
"@
and %1, %2, %0
andi %1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn "*andsi3_bclr"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(and:SI (match_operand:SI 1 "arith_reg_operand" "%0")
(match_operand:SI 2 "const_int_operand" "Psz")))]
"TARGET_SH2A && satisfies_constraint_Psz (operands[2])"
"bclr %W2,%0"
[(set_attr "type" "arith")])
(define_insn_and_split "anddi3"
[(set (match_operand:DI 0 "arith_reg_dest" "=r,r,r")
(and:DI (match_operand:DI 1 "arith_reg_operand" "%r,r,r")
(match_operand:DI 2 "and_operand" "r,I10,J16")))]
"TARGET_SHMEDIA"
"@
and %1, %2, %0
andi %1, %2, %0
#"
"reload_completed
&& ! logical_operand (operands[2], DImode)"
[(const_int 0)]
{
if ((unsigned)INTVAL (operands[2]) == (unsigned) 0xffffffff)
emit_insn (gen_mshflo_l_di (operands[0], operands[1], CONST0_RTX (DImode)));
else
emit_insn (gen_mshfhi_l_di (operands[0], CONST0_RTX (DImode), operands[1]));
DONE;
}
[(set_attr "type" "arith_media")])
(define_insn "andcsi3"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(and:SI (match_operand:SI 1 "arith_reg_operand" "r")
(not:SI (match_operand:SI 2 "arith_reg_operand" "r"))))]
"TARGET_SHMEDIA"
"andc %1,%2,%0"
[(set_attr "type" "arith_media")])
(define_insn "andcdi3"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(and:DI (match_operand:DI 1 "arith_reg_operand" "r")
(not:DI (match_operand:DI 2 "arith_reg_operand" "r"))))]
"TARGET_SHMEDIA"
"andc %1,%2,%0"
[(set_attr "type" "arith_media")])
(define_expand "iorsi3"
[(set (match_operand:SI 0 "arith_reg_operand" "")
(ior:SI (match_operand:SI 1 "logical_reg_operand" "")
(match_operand:SI 2 "logical_operand" "")))]
""
"")
(define_insn "*iorsi3_compact"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,z")
(ior:SI (match_operand:SI 1 "arith_reg_operand" "%0,0")
(match_operand:SI 2 "logical_operand" "r,K08")))]
"TARGET_SH1
&& !(TARGET_SH2A && satisfies_constraint_Pso (operands[2]))"
"or %2,%0"
[(set_attr "type" "arith")])
(define_insn "*iorsi3_media"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(ior:SI (match_operand:SI 1 "logical_reg_operand" "%r,r")
(match_operand:SI 2 "logical_operand" "r,I10")))]
"TARGET_SHMEDIA"
"@
or %1, %2, %0
ori %1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn "*iorsi3_bset"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ior:SI (match_operand:SI 1 "arith_reg_operand" "%0")
(match_operand:SI 2 "const_int_operand" "Pso")))]
"TARGET_SH2A && satisfies_constraint_Pso (operands[2])"
"bset %V2,%0"
[(set_attr "type" "arith")])
(define_insn "iordi3"
[(set (match_operand:DI 0 "arith_reg_dest" "=r,r")
(ior:DI (match_operand:DI 1 "arith_reg_operand" "%r,r")
(match_operand:DI 2 "logical_operand" "r,I10")))]
"TARGET_SHMEDIA"
"@
or %1, %2, %0
ori %1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn_and_split "*logical_sidi3"
[(set (match_operand:DI 0 "arith_reg_dest" "=r,r")
(sign_extend:DI (match_operator:SI 3 "logical_operator"
[(match_operand:SI 1 "arith_reg_operand" "%r,r")
(match_operand:SI 2 "logical_operand" "r,I10")])))]
"TARGET_SHMEDIA"
"#"
"&& reload_completed"
[(set (match_dup 0) (match_dup 3))]
{
operands[3]
= gen_rtx_fmt_ee (GET_CODE (operands[3]), DImode,
simplify_gen_subreg (DImode, operands[1], SImode, 0),
simplify_gen_subreg (DImode, operands[2], SImode, 0));
})
(define_insn_and_split "*logical_sidisi3"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(truncate:SI (sign_extend:DI
(match_operator:SI 3 "logical_operator"
[(match_operand:SI 1 "arith_reg_operand" "%r,r")
(match_operand:SI 2 "logical_operand" "r,I10")]))))]
"TARGET_SHMEDIA"
"#"
"&& 1"
[(set (match_dup 0) (match_dup 3))])
(define_insn_and_split "*logical_sidi3_2"
[(set (match_operand:DI 0 "arith_reg_dest" "=r,r")
(sign_extend:DI (truncate:SI (sign_extend:DI
(match_operator:SI 3 "logical_operator"
[(match_operand:SI 1 "arith_reg_operand" "%r,r")
(match_operand:SI 2 "logical_operand" "r,I10")])))))]
"TARGET_SHMEDIA"
"#"
"&& 1"
[(set (match_dup 0) (sign_extend:DI (match_dup 3)))])
(define_expand "xorsi3"
[(set (match_operand:SI 0 "arith_reg_operand" "")
(xor:SI (match_operand:SI 1 "logical_reg_operand" "")
(match_operand:SI 2 "xor_operand" "")))]
""
"")
(define_insn "*xorsi3_compact"
[(set (match_operand:SI 0 "arith_reg_dest" "=z,r")
(xor:SI (match_operand:SI 1 "arith_reg_operand" "%0,0")
(match_operand:SI 2 "logical_operand" "K08,r")))]
"TARGET_SH1"
"xor %2,%0"
[(set_attr "type" "arith")])
;; The *logical_op_t pattern helps combine eliminating sign/zero extensions
;; of results where one of the inputs is a T bit store. Notice that this
;; pattern must not match during reload. If reload picks this pattern it
;; will be impossible to split it afterwards.
(define_insn_and_split "*logical_op_t"
[(set (match_operand:SI 0 "arith_reg_dest")
(match_operator:SI 3 "logical_operator"
[(match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "t_reg_operand")]))]
"TARGET_SH1 && can_create_pseudo_p ()"
"#"
"&& 1"
[(set (match_dup 4) (reg:SI T_REG))
(set (match_dup 0) (match_dup 3))]
{
operands[4] = gen_reg_rtx (SImode);
operands[3] = gen_rtx_fmt_ee (GET_CODE (operands[3]), SImode,
operands[1], operands[4]);
})
(define_insn "*xorsi3_media"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(xor:SI (match_operand:SI 1 "logical_reg_operand" "%r,r")
(match_operand:SI 2 "xor_operand" "r,I06")))]
"TARGET_SHMEDIA"
"@
xor %1, %2, %0
xori %1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn "xordi3"
[(set (match_operand:DI 0 "arith_reg_dest" "=r,r")
(xor:DI (match_operand:DI 1 "arith_reg_operand" "%r,r")
(match_operand:DI 2 "xor_operand" "r,I06")))]
"TARGET_SHMEDIA"
"@
xor %1, %2, %0
xori %1, %2, %0"
[(set_attr "type" "arith_media")])
;; Combiner bridge pattern for 2 * sign extend -> logical op -> truncate.
;; converts 2 * sign extend -> logical op into logical op -> sign extend
(define_split
[(set (match_operand:DI 0 "arith_reg_dest" "")
(sign_extend:DI (match_operator 4 "binary_logical_operator"
[(match_operand 1 "any_register_operand" "")
(match_operand 2 "any_register_operand" "")])))]
"TARGET_SHMEDIA"
[(set (match_dup 5) (match_dup 4))
(set (match_dup 0) (sign_extend:DI (match_dup 5)))]
{
enum machine_mode inmode = GET_MODE (operands[1]);
int offset = 0;
if (GET_CODE (operands[0]) == SUBREG)
{
offset = SUBREG_BYTE (operands[0]);
operands[0] = SUBREG_REG (operands[0]);
}
gcc_assert (REG_P (operands[0]));
if (TARGET_BIG_ENDIAN)
offset += 8 - GET_MODE_SIZE (inmode);
operands[5] = gen_rtx_SUBREG (inmode, operands[0], offset);
})
;; -------------------------------------------------------------------------
;; Shifts and rotates
;; -------------------------------------------------------------------------
(define_expand "rotldi3"
[(set (match_operand:DI 0 "arith_reg_dest" "")
(rotate:DI (match_operand:DI 1 "arith_reg_operand" "")
(match_operand:HI 2 "mextr_bit_offset" "")))]
"TARGET_SHMEDIA"
{
if (! mextr_bit_offset (operands[2], HImode))
FAIL;
})
(define_insn "rotldi3_mextr"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(rotate:DI (match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:HI 2 "mextr_bit_offset" "i")))]
"TARGET_SHMEDIA"
{
static char templ[16];
sprintf (templ, "mextr%d %%1,%%1,%%0",
8 - (int) (INTVAL (operands[2]) >> 3));
return templ;
}
[(set_attr "type" "arith_media")])
(define_expand "rotrdi3"
[(set (match_operand:DI 0 "arith_reg_dest" "")
(rotatert:DI (match_operand:DI 1 "arith_reg_operand" "")
(match_operand:HI 2 "mextr_bit_offset" "")))]
"TARGET_SHMEDIA"
{
if (! mextr_bit_offset (operands[2], HImode))
FAIL;
})
(define_insn "rotrdi3_mextr"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(rotatert:DI (match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:HI 2 "mextr_bit_offset" "i")))]
"TARGET_SHMEDIA"
{
static char templ[16];
sprintf (templ, "mextr%d %%1,%%1,%%0", (int) INTVAL (operands[2]) >> 3);
return templ;
}
[(set_attr "type" "arith_media")])
(define_split
[(set (match_operand:DI 0 "arith_reg_dest" "")
(ior:DI (zero_extend:DI (mem:QI (match_operand 1
"ua_address_operand" "")))
(ashift:DI (match_operand:DI 2 "arith_reg_operand" "")
(const_int 8))))
(clobber (match_operand:DI 3 "register_operand" ""))]
"TARGET_SHMEDIA"
[(match_dup 4) (match_dup 5)]
{
operands[4] = ((TARGET_LITTLE_ENDIAN ? gen_ldhi_q : gen_ldlo_q)
(operands[3], operands[1]));
operands[5] = gen_mextr_rl (operands[0], operands[3], operands[2],
GEN_INT (56), GEN_INT (8));
})
(define_expand "rotrsi3"
[(set (match_operand:SI 0 "arith_reg_dest")
(rotatert:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "const_int_operand")))]
"TARGET_SH1"
{
HOST_WIDE_INT ival = INTVAL (operands[2]);
if (ival == 1)
{
emit_insn (gen_rotrsi3_1 (operands[0], operands[1]));
DONE;
}
FAIL;
})
(define_insn "rotrsi3_1"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(rotatert:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 1)))
(set (reg:SI T_REG)
(and:SI (match_dup 1) (const_int 1)))]
"TARGET_SH1"
"rotr %0"
[(set_attr "type" "arith")])
;; A slimplified version of rotr for combine.
(define_insn "*rotrsi3_1"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(rotatert:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 1)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"rotr %0"
[(set_attr "type" "arith")])
(define_insn "rotlsi3_1"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(rotate:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 1)))
(set (reg:SI T_REG)
(lshiftrt:SI (match_dup 1) (const_int 31)))]
"TARGET_SH1"
"rotl %0"
[(set_attr "type" "arith")])
;; A simplified version of rotl for combine.
(define_insn "*rotlsi3_1"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(rotate:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 1)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"rotl %0"
[(set_attr "type" "arith")])
(define_insn "rotlsi3_31"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(rotate:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 31)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"rotr %0"
[(set_attr "type" "arith")])
(define_insn "rotlsi3_16"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(rotate:SI (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 16)))]
"TARGET_SH1"
"swap.w %1,%0"
[(set_attr "type" "arith")])
(define_expand "rotlsi3"
[(set (match_operand:SI 0 "arith_reg_dest")
(rotate:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "const_int_operand")))]
"TARGET_SH1"
{
static const char rot_tab[] = {
000, 000, 000, 000, 000, 000, 010, 001,
001, 001, 011, 013, 003, 003, 003, 003,
003, 003, 003, 003, 003, 013, 012, 002,
002, 002, 010, 000, 000, 000, 000, 000,
};
int count = INTVAL (operands[2]);
int choice = rot_tab[count];
if (choice & 010 && SH_DYNAMIC_SHIFT_COST <= 1)
FAIL;
choice &= 7;
switch (choice)
{
case 0:
emit_move_insn (operands[0], operands[1]);
count -= (count & 16) * 2;
break;
case 3:
emit_insn (gen_rotlsi3_16 (operands[0], operands[1]));
count -= 16;
break;
case 1:
case 2:
{
rtx parts[2];
parts[0] = gen_reg_rtx (SImode);
parts[1] = gen_reg_rtx (SImode);
emit_insn (gen_rotlsi3_16 (parts[2-choice], operands[1]));
emit_move_insn (parts[choice-1], operands[1]);
emit_insn (gen_ashlsi3 (parts[0], parts[0], GEN_INT (8)));
emit_insn (gen_lshrsi3 (parts[1], parts[1], GEN_INT (8)));
emit_insn (gen_iorsi3 (operands[0], parts[0], parts[1]));
count = (count & ~16) - 8;
}
}
for (; count > 0; count--)
emit_insn (gen_rotlsi3_1 (operands[0], operands[0]));
for (; count < 0; count++)
emit_insn (gen_rotlsi3_31 (operands[0], operands[0]));
DONE;
})
(define_insn "rotlhi3_8"
[(set (match_operand:HI 0 "arith_reg_dest" "=r")
(rotate:HI (match_operand:HI 1 "arith_reg_operand" "r")
(const_int 8)))]
"TARGET_SH1"
"swap.b %1,%0"
[(set_attr "type" "arith")])
(define_expand "rotlhi3"
[(set (match_operand:HI 0 "arith_reg_operand")
(rotate:HI (match_operand:HI 1 "arith_reg_operand")
(match_operand:HI 2 "const_int_operand")))]
"TARGET_SH1"
{
if (INTVAL (operands[2]) != 8)
FAIL;
})
;; The rotcr and rotcl insns are used primarily in DImode shifts by one.
;; They can also be used to implement things like
;; bool t = a == b;
;; int x0 = (y >> 1) | (t << 31); // rotcr
;; int x1 = (y << 1) | t; // rotcl
(define_insn "rotcr"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ior:SI (lshiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 1))
(ashift:SI (match_operand:SI 2 "t_reg_operand")
(const_int 31))))
(set (reg:SI T_REG)
(and:SI (match_dup 1) (const_int 1)))]
"TARGET_SH1"
"rotcr %0"
[(set_attr "type" "arith")])
(define_insn "rotcl"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ior:SI (ashift:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 1))
(match_operand:SI 2 "t_reg_operand")))
(set (reg:SI T_REG)
(lshiftrt:SI (match_dup 1) (const_int 31)))]
"TARGET_SH1"
"rotcl %0"
[(set_attr "type" "arith")])
;; Simplified rotcr version for combine, which allows arbitrary shift
;; amounts for the reg. If the shift amount is '1' rotcr can be used
;; directly. Otherwise we have to insert a shift in between.
(define_insn_and_split "*rotcr"
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (lshiftrt:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "const_int_operand"))
(ashift:SI (match_operand:SI 3 "arith_reg_or_t_reg_operand")
(const_int 31))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
if (INTVAL (operands[2]) > 1)
{
const rtx shift_count = GEN_INT (INTVAL (operands[2]) - 1);
rtx prev_set_t_insn = NULL_RTX;
rtx tmp_t_reg = NULL_RTX;
/* If we're going to emit a shift sequence that clobbers the T_REG,
try to find the previous insn that sets the T_REG and emit the
shift insn before that insn, to remove the T_REG dependency.
If the insn that sets the T_REG cannot be found, store the T_REG
in a temporary reg and restore it after the shift. */
if (sh_lshrsi_clobbers_t_reg_p (shift_count)
&& ! sh_dynamicalize_shift_p (shift_count))
{
prev_set_t_insn = prev_nonnote_insn_bb (curr_insn);
/* Skip the nott insn, which was probably inserted by the splitter
of *rotcr_neg_t. Don't use one of the recog functions
here during insn splitting, since that causes problems in later
passes. */
if (prev_set_t_insn != NULL_RTX)
{
rtx pat = PATTERN (prev_set_t_insn);
if (GET_CODE (pat) == SET
&& t_reg_operand (XEXP (pat, 0), SImode)
&& negt_reg_operand (XEXP (pat, 1), SImode))
prev_set_t_insn = prev_nonnote_insn_bb (prev_set_t_insn);
}
if (! (prev_set_t_insn != NULL_RTX
&& reg_set_p (get_t_reg_rtx (), prev_set_t_insn)
&& ! reg_referenced_p (get_t_reg_rtx (),
PATTERN (prev_set_t_insn))))
{
prev_set_t_insn = NULL_RTX;
tmp_t_reg = gen_reg_rtx (SImode);
emit_insn (gen_move_insn (tmp_t_reg, get_t_reg_rtx ()));
}
}
rtx shift_result = gen_reg_rtx (SImode);
rtx shift_insn = gen_lshrsi3 (shift_result, operands[1], shift_count);
operands[1] = shift_result;
/* Emit the shift insn before the insn that sets T_REG, if possible. */
if (prev_set_t_insn != NULL_RTX)
emit_insn_before (shift_insn, prev_set_t_insn);
else
emit_insn (shift_insn);
/* Restore T_REG if it has been saved before. */
if (tmp_t_reg != NULL_RTX)
emit_insn (gen_cmpgtsi_t (tmp_t_reg, const0_rtx));
}
/* For the rotcr insn to work, operands[3] must be in T_REG.
If it is not we can get it there by shifting it right one bit.
In this case T_REG is not an input for this insn, thus we don't have to
pay attention as of where to insert the shlr insn. */
if (! t_reg_operand (operands[3], SImode))
{
/* We don't care about the shifted result here, only the T_REG. */
emit_insn (gen_shlr (gen_reg_rtx (SImode), operands[3]));
operands[3] = get_t_reg_rtx ();
}
emit_insn (gen_rotcr (operands[0], operands[1], operands[3]));
DONE;
})
;; If combine tries the same as above but with swapped operands, split
;; it so that it will try the pattern above.
(define_split
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (ashift:SI (match_operand:SI 1 "arith_reg_or_t_reg_operand")
(const_int 31))
(lshiftrt:SI (match_operand:SI 2 "arith_reg_operand")
(match_operand:SI 3 "const_int_operand"))))]
"TARGET_SH1 && can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(ior:SI (lshiftrt:SI (match_dup 2) (match_dup 3))
(ashift:SI (match_dup 1) (const_int 31))))
(clobber (reg:SI T_REG))])])
;; Basically the same as the rotcr pattern above, but for rotcl.
;; FIXME: Fold copy pasted split code for rotcr and rotcl.
(define_insn_and_split "*rotcl"
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (ashift:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "const_int_operand"))
(and:SI (match_operand:SI 3 "arith_reg_or_t_reg_operand")
(const_int 1))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
gcc_assert (INTVAL (operands[2]) > 0);
if (INTVAL (operands[2]) > 1)
{
const rtx shift_count = GEN_INT (INTVAL (operands[2]) - 1);
rtx prev_set_t_insn = NULL_RTX;
rtx tmp_t_reg = NULL_RTX;
/* If we're going to emit a shift sequence that clobbers the T_REG,
try to find the previous insn that sets the T_REG and emit the
shift insn before that insn, to remove the T_REG dependency.
If the insn that sets the T_REG cannot be found, store the T_REG
in a temporary reg and restore it after the shift. */
if (sh_ashlsi_clobbers_t_reg_p (shift_count)
&& ! sh_dynamicalize_shift_p (shift_count))
{
prev_set_t_insn = prev_nonnote_insn_bb (curr_insn);
/* Skip the nott insn, which was probably inserted by the splitter
of *rotcl_neg_t. Don't use one of the recog functions
here during insn splitting, since that causes problems in later
passes. */
if (prev_set_t_insn != NULL_RTX)
{
rtx pat = PATTERN (prev_set_t_insn);
if (GET_CODE (pat) == SET
&& t_reg_operand (XEXP (pat, 0), SImode)
&& negt_reg_operand (XEXP (pat, 1), SImode))
prev_set_t_insn = prev_nonnote_insn_bb (prev_set_t_insn);
}
if (! (prev_set_t_insn != NULL_RTX
&& reg_set_p (get_t_reg_rtx (), prev_set_t_insn)
&& ! reg_referenced_p (get_t_reg_rtx (),
PATTERN (prev_set_t_insn))))
{
prev_set_t_insn = NULL_RTX;
tmp_t_reg = gen_reg_rtx (SImode);
emit_insn (gen_move_insn (tmp_t_reg, get_t_reg_rtx ()));
}
}
rtx shift_result = gen_reg_rtx (SImode);
rtx shift_insn = gen_ashlsi3 (shift_result, operands[1], shift_count);
operands[1] = shift_result;
/* Emit the shift insn before the insn that sets T_REG, if possible. */
if (prev_set_t_insn != NULL_RTX)
emit_insn_before (shift_insn, prev_set_t_insn);
else
emit_insn (shift_insn);
/* Restore T_REG if it has been saved before. */
if (tmp_t_reg != NULL_RTX)
emit_insn (gen_cmpgtsi_t (tmp_t_reg, const0_rtx));
}
/* For the rotcl insn to work, operands[3] must be in T_REG.
If it is not we can get it there by shifting it right one bit.
In this case T_REG is not an input for this insn, thus we don't have to
pay attention as of where to insert the shlr insn. */
if (! t_reg_operand (operands[3], SImode))
{
/* We don't care about the shifted result here, only the T_REG. */
emit_insn (gen_shlr (gen_reg_rtx (SImode), operands[3]));
operands[3] = get_t_reg_rtx ();
}
emit_insn (gen_rotcl (operands[0], operands[1], operands[3]));
DONE;
})
;; rotcl combine pattern variations
(define_insn_and_split "*rotcl"
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (ashift:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "const_int_operand"))
(match_operand:SI 3 "t_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(ior:SI (ashift:SI (match_dup 1) (match_dup 2))
(and:SI (match_dup 3) (const_int 1))))
(clobber (reg:SI T_REG))])])
(define_insn_and_split "*rotcl"
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (and:SI (match_operand:SI 1 "arith_reg_or_t_reg_operand")
(const_int 1))
(ashift:SI (match_operand:SI 2 "arith_reg_operand")
(match_operand:SI 3 "const_int_operand"))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(ior:SI (ashift:SI (match_dup 2) (match_dup 3))
(and:SI (match_dup 1) (const_int 1))))
(clobber (reg:SI T_REG))])])
(define_insn_and_split "*rotcl"
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (ashift:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "const_int_operand"))
(lshiftrt:SI (match_operand:SI 3 "arith_reg_operand")
(const_int 31))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(ior:SI (ashift:SI (match_dup 1) (match_dup 2))
(and:SI (reg:SI T_REG) (const_int 1))))
(clobber (reg:SI T_REG))])]
{
/* We don't care about the result of the left shift, only the T_REG. */
emit_insn (gen_shll (gen_reg_rtx (SImode), operands[3]));
})
(define_insn_and_split "*rotcl"
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (lshiftrt:SI (match_operand:SI 3 "arith_reg_operand")
(const_int 31))
(ashift:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "const_int_operand"))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(ior:SI (ashift:SI (match_dup 1) (match_dup 2))
(and:SI (reg:SI T_REG) (const_int 1))))
(clobber (reg:SI T_REG))])]
{
/* We don't care about the result of the left shift, only the T_REG. */
emit_insn (gen_shll (gen_reg_rtx (SImode), operands[3]));
})
;; rotcr combine bridge pattern which will make combine try out more
;; complex patterns.
(define_insn_and_split "*rotcr"
[(set (match_operand:SI 0 "arith_reg_dest")
(ashift:SI (match_operand:SI 1 "t_reg_operand") (const_int 31)))]
"TARGET_SH1"
"#"
"&& 1"
[(set (match_dup 0) (match_dup 1))
(parallel [(set (match_dup 0)
(ior:SI (lshiftrt:SI (match_dup 0) (const_int 1))
(ashift:SI (match_dup 1) (const_int 31))))
(set (reg:SI T_REG)
(and:SI (match_dup 0) (const_int 1)))])])
(define_insn_and_split "*rotcr"
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (and:SI (match_operand:SI 1 "arith_reg_operand")
(const_int -2147483648)) ;; 0xffffffff80000000
(lshiftrt:SI (match_operand:SI 2 "arith_reg_operand")
(const_int 1))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
rtx tmp = gen_reg_rtx (SImode);
emit_insn (gen_shll (tmp, operands[1]));
emit_insn (gen_rotcr (operands[0], operands[2], get_t_reg_rtx ()));
DONE;
})
;; rotcr combine patterns for rotating in the negated T_REG value.
(define_insn_and_split "*rotcr_neg_t"
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (match_operand:SI 1 "negt_reg_shl31_operand")
(lshiftrt:SI (match_operand:SI 2 "arith_reg_operand")
(match_operand:SI 3 "const_int_operand"))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(ior:SI (lshiftrt:SI (match_dup 2) (match_dup 3))
(ashift:SI (reg:SI T_REG) (const_int 31))))
(clobber (reg:SI T_REG))])]
{
emit_insn (gen_nott (get_t_reg_rtx ()));
})
(define_insn_and_split "*rotcr_neg_t"
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (lshiftrt:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "const_int_operand"))
(match_operand:SI 3 "negt_reg_shl31_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(ior:SI (lshiftrt:SI (match_dup 1) (match_dup 2))
(ashift:SI (reg:SI T_REG) (const_int 31))))
(clobber (reg:SI T_REG))])]
{
emit_insn (gen_nott (get_t_reg_rtx ()));
})
;; rotcl combine patterns for rotating in the negated T_REG value.
;; For some strange reason these have to be specified as splits which combine
;; will pick up. If they are specified as insn_and_split like the
;; *rotcr_neg_t patterns above, combine would recognize them successfully
;; but not emit them on non-SH2A targets.
(define_split
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (match_operand:SI 1 "negt_reg_operand")
(ashift:SI (match_operand:SI 2 "arith_reg_operand")
(match_operand:SI 3 "const_int_operand"))))]
"TARGET_SH1"
[(set (reg:SI T_REG) (xor:SI (reg:SI T_REG) (const_int 1)))
(parallel [(set (match_dup 0)
(ior:SI (ashift:SI (match_dup 2) (match_dup 3))
(and:SI (reg:SI T_REG) (const_int 1))))
(clobber (reg:SI T_REG))])])
(define_split
[(set (match_operand:SI 0 "arith_reg_dest")
(ior:SI (ashift:SI (match_operand:SI 2 "arith_reg_operand")
(match_operand:SI 3 "const_int_operand"))
(match_operand:SI 1 "negt_reg_operand")))]
"TARGET_SH1"
[(set (reg:SI T_REG) (xor:SI (reg:SI T_REG) (const_int 1)))
(parallel [(set (match_dup 0)
(ior:SI (ashift:SI (match_dup 2) (match_dup 3))
(and:SI (reg:SI T_REG) (const_int 1))))
(clobber (reg:SI T_REG))])])
;; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
;; SImode shift left
(define_expand "ashlsi3"
[(set (match_operand:SI 0 "arith_reg_operand" "")
(ashift:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "shift_count_operand" "")))]
""
{
if (TARGET_SHMEDIA)
{
if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) < 0)
{
operands[2] = GEN_INT (-INTVAL (operands[2]));
emit_insn (gen_ashrsi3_media (operands[0], operands[1], operands[2]));
DONE;
}
emit_insn (gen_ashlsi3_media (operands[0], operands[1], operands[2]));
DONE;
}
if (TARGET_DYNSHIFT
&& CONST_INT_P (operands[2]) && sh_dynamicalize_shift_p (operands[2]))
operands[2] = force_reg (SImode, operands[2]);
/* If the ashlsi3_* insn is going to clobber the T_REG it must be
expanded here. */
if (CONST_INT_P (operands[2])
&& sh_ashlsi_clobbers_t_reg_p (operands[2])
&& ! sh_dynamicalize_shift_p (operands[2]))
{
emit_insn (gen_ashlsi3_n_clobbers_t (operands[0], operands[1],
operands[2]));
DONE;
}
/* Expand a library call for the dynamic shift. */
if (!CONST_INT_P (operands[2]) && !TARGET_DYNSHIFT)
{
emit_move_insn (gen_rtx_REG (SImode, R4_REG), operands[1]);
rtx funcaddr = gen_reg_rtx (Pmode);
function_symbol (funcaddr, "__ashlsi3_r0", SFUNC_STATIC);
emit_insn (gen_ashlsi3_d_call (operands[0], operands[2], funcaddr));
DONE;
}
})
(define_insn "ashlsi3_k"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(ashift:SI (match_operand:SI 1 "arith_reg_operand" "0,0")
(match_operand:SI 2 "p27_shift_count_operand" "M,P27")))]
"TARGET_SH1"
"@
add %0,%0
shll%O2 %0"
[(set_attr "type" "arith")])
(define_insn_and_split "ashlsi3_d"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ashift:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "shift_count_operand" "r")))]
"TARGET_DYNSHIFT"
"shld %2,%0"
"&& CONST_INT_P (operands[2]) && ! sh_dynamicalize_shift_p (operands[2])
&& ! sh_ashlsi_clobbers_t_reg_p (operands[2])"
[(const_int 0)]
{
if (satisfies_constraint_P27 (operands[2]))
{
emit_insn (gen_ashlsi3_k (operands[0], operands[1], operands[2]));
DONE;
}
else if (! satisfies_constraint_P27 (operands[2]))
{
/* This must happen before reload, otherwise the constant will be moved
into a register due to the "r" constraint, after which this split
cannot be done anymore.
Unfortunately the move insn will not always be eliminated.
Also, here we must not create a shift sequence that clobbers the
T_REG. */
emit_move_insn (operands[0], operands[1]);
gen_shifty_op (ASHIFT, operands);
DONE;
}
FAIL;
}
[(set_attr "type" "dyn_shift")])
;; If dynamic shifts are not available use a library function.
;; By specifying the pattern we reduce the number of call clobbered regs.
;; In order to make combine understand the truncation of the shift amount
;; operand we have to allow it to use pseudo regs for the shift operands.
(define_insn "ashlsi3_d_call"
[(set (match_operand:SI 0 "arith_reg_dest" "=z")
(ashift:SI (reg:SI R4_REG)
(and:SI (match_operand:SI 1 "arith_reg_operand" "z")
(const_int 31))))
(use (match_operand:SI 2 "arith_reg_operand" "r"))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SH1 && !TARGET_DYNSHIFT"
"jsr @%2%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_insn_and_split "ashlsi3_n"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ashift:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "not_p27_shift_count_operand" "")))]
"TARGET_SH1 && ! sh_ashlsi_clobbers_t_reg_p (operands[2])"
"#"
"&& (reload_completed
|| (sh_dynamicalize_shift_p (operands[2]) && can_create_pseudo_p ()))"
[(const_int 0)]
{
if (sh_dynamicalize_shift_p (operands[2]) && can_create_pseudo_p ())
{
/* If this pattern was picked and dynamic shifts are supported, switch
to dynamic shift pattern before reload. */
operands[2] = force_reg (SImode, operands[2]);
emit_insn (gen_ashlsi3_d (operands[0], operands[1], operands[2]));
}
else
gen_shifty_op (ASHIFT, operands);
DONE;
})
(define_insn_and_split "ashlsi3_n_clobbers_t"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ashift:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "not_p27_shift_count_operand" "")))
(clobber (reg:SI T_REG))]
"TARGET_SH1 && sh_ashlsi_clobbers_t_reg_p (operands[2])"
"#"
"&& (reload_completed || INTVAL (operands[2]) == 31
|| (sh_dynamicalize_shift_p (operands[2]) && can_create_pseudo_p ()))"
[(const_int 0)]
{
if (INTVAL (operands[2]) == 31)
{
/* If the shift amount is 31 we split into a different sequence before
reload so that it gets a chance to allocate R0 for the sequence.
If it fails to do so (due to pressure on R0), it will take one insn
more for the and. */
emit_insn (gen_andsi3 (operands[0], operands[1], const1_rtx));
emit_insn (gen_rotlsi3_31 (operands[0], operands[0]));
}
else if (sh_dynamicalize_shift_p (operands[2]) && can_create_pseudo_p ())
{
/* If this pattern was picked and dynamic shifts are supported, switch
to dynamic shift pattern before reload. */
operands[2] = force_reg (SImode, operands[2]);
emit_insn (gen_ashlsi3_d (operands[0], operands[1], operands[2]));
}
else
gen_shifty_op (ASHIFT, operands);
DONE;
})
(define_insn "shll"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ashift:SI (match_operand:SI 1 "arith_reg_operand" "0") (const_int 1)))
(set (reg:SI T_REG)
(lt:SI (match_dup 1) (const_int 0)))]
"TARGET_SH1"
"shll %0"
[(set_attr "type" "arith")])
(define_insn "*ashlsi_c_void"
[(set (reg:SI T_REG)
(lt:SI (match_operand:SI 0 "arith_reg_operand" "r") (const_int 0)))
(clobber (match_scratch:SI 1 "=0"))]
"TARGET_SH1 && cse_not_expected"
"shll %0"
[(set_attr "type" "arith")])
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "") (const_int 0))
(set (reg:SI T_REG)
(gt:SI (match_dup 0) (match_operand:SI 1 "arith_reg_operand" "")))]
"TARGET_SH1
&& peep2_reg_dead_p (2, operands[0])
&& peep2_reg_dead_p (2, operands[1])"
[(const_int 0)]
{
emit_insn (gen_shll (operands[1], operands[1]));
DONE;
})
(define_insn "ashlsi3_media"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(ashift:SI (match_operand:SI 1 "extend_reg_operand" "r,r")
(match_operand:SI 2 "shift_count_operand" "r,n")))]
"TARGET_SHMEDIA"
"@
shlld.l %1, %2, %0
shlli.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
;; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
;; HImode shift left
(define_expand "ashlhi3"
[(parallel [(set (match_operand:HI 0 "arith_reg_operand" "")
(ashift:HI (match_operand:HI 1 "arith_reg_operand" "")
(match_operand:SI 2 "nonmemory_operand" "")))
(clobber (reg:SI T_REG))])]
"TARGET_SH1"
{
if (!CONST_INT_P (operands[2]))
FAIL;
/* It may be possible to call gen_ashlhi3 directly with more generic
operands. Make sure operands[1] is a HImode register here. */
if (!arith_reg_operand (operands[1], HImode))
operands[1] = copy_to_mode_reg (HImode, operands[1]);
})
(define_insn "ashlhi3_k"
[(set (match_operand:HI 0 "arith_reg_dest" "=r,r")
(ashift:HI (match_operand:HI 1 "arith_reg_operand" "0,0")
(match_operand:HI 2 "const_int_operand" "M,P27")))]
"TARGET_SH1 && satisfies_constraint_P27 (operands[2])"
"@
add %0,%0
shll%O2 %0"
[(set_attr "type" "arith")])
(define_insn_and_split "*ashlhi3_n"
[(set (match_operand:HI 0 "arith_reg_dest" "=r")
(ashift:HI (match_operand:HI 1 "arith_reg_operand" "0")
(match_operand:HI 2 "const_int_operand" "n")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& reload_completed"
[(use (reg:SI R0_REG))]
{
gen_shifty_hi_op (ASHIFT, operands);
DONE;
})
;; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
;; DImode shift left
(define_expand "ashldi3"
[(parallel [(set (match_operand:DI 0 "arith_reg_operand" "")
(ashift:DI (match_operand:DI 1 "arith_reg_operand" "")
(match_operand:DI 2 "immediate_operand" "")))
(clobber (reg:SI T_REG))])]
""
{
if (TARGET_SHMEDIA)
{
if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) < 0)
{
operands[2] = GEN_INT (-INTVAL (operands[2]));
emit_insn (gen_ashrdi3_media (operands[0], operands[1], operands[2]));
DONE;
}
emit_insn (gen_ashldi3_media (operands[0], operands[1], operands[2]));
DONE;
}
if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) == 1)
{
emit_insn (gen_ashldi3_k (operands[0], operands[1]));
DONE;
}
else if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) < 32)
{
emit_insn (gen_ashldi3_std (operands[0], operands[1], operands[2]));
DONE;
}
else
FAIL;
})
;; Expander for DImode shift left with SImode operations.
(define_expand "ashldi3_std"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ashift:DI (match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "const_int_operand" "n")))]
"TARGET_SH1 && INTVAL (operands[2]) < 32"
{
rtx low_src = gen_lowpart (SImode, operands[1]);
rtx high_src = gen_highpart (SImode, operands[1]);
rtx dst = gen_reg_rtx (DImode);
rtx low_dst = gen_lowpart (SImode, dst);
rtx high_dst = gen_highpart (SImode, dst);
rtx tmp0 = gen_reg_rtx (SImode);
rtx tmp1 = gen_reg_rtx (SImode);
emit_insn (gen_lshrsi3 (tmp0, low_src, GEN_INT (32 - INTVAL (operands[2]))));
emit_insn (gen_ashlsi3 (low_dst, low_src, operands[2]));
emit_insn (gen_ashlsi3 (tmp1, high_src, operands[2]));
emit_insn (gen_iorsi3 (high_dst, tmp0, tmp1));
emit_move_insn (operands[0], dst);
DONE;
})
(define_insn_and_split "ashldi3_k"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ashift:DI (match_operand:DI 1 "arith_reg_operand" "0")
(const_int 1)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& reload_completed"
[(const_int 0)]
{
rtx high = gen_highpart (SImode, operands[0]);
rtx low = gen_lowpart (SImode, operands[0]);
emit_insn (gen_shll (low, low));
emit_insn (gen_rotcl (high, high, get_t_reg_rtx ()));
DONE;
})
(define_insn "ashldi3_media"
[(set (match_operand:DI 0 "arith_reg_dest" "=r,r")
(ashift:DI (match_operand:DI 1 "arith_reg_operand" "r,r")
(match_operand:DI 2 "shift_count_operand" "r,n")))]
"TARGET_SHMEDIA"
"@
shlld %1, %2, %0
shlli %1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn "*ashldisi3_media"
[(set (subreg:DI (match_operand:SI 0 "arith_reg_operand" "=r") 0)
(ashift:DI (match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "const_int_operand" "n")))]
"TARGET_SHMEDIA && INTVAL (operands[2]) < 32"
"shlli.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
;; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
;; SImode arithmetic shift right
;;
;; We can't do HImode right shifts correctly unless we start out with an
;; explicit zero / sign extension; doing that would result in worse overall
;; code, so just let the machine independent code widen the mode.
;; That's why we don't have ashrhi3_k / lshrhi3_k / lshrhi3_m / lshrhi3 .
(define_expand "ashrsi3"
[(parallel [(set (match_operand:SI 0 "arith_reg_dest" "")
(ashiftrt:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "nonmemory_operand" "")))
(clobber (reg:SI T_REG))])]
""
{
if (TARGET_SHMEDIA)
{
if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) < 0)
{
operands[2] = GEN_INT (-INTVAL (operands[2]));
emit_insn (gen_ashlsi3_media (operands[0], operands[1], operands[2]));
DONE;
}
emit_insn (gen_ashrsi3_media (operands[0], operands[1], operands[2]));
DONE;
}
if (expand_ashiftrt (operands))
DONE;
else
FAIL;
})
(define_insn "shar"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ashiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 1)))
(set (reg:SI T_REG)
(and:SI (match_dup 1) (const_int 1)))]
"TARGET_SH1"
"shar %0"
[(set_attr "type" "arith")])
(define_insn "ashrsi3_k"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ashiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "const_int_operand" "M")))
(clobber (reg:SI T_REG))]
"TARGET_SH1 && INTVAL (operands[2]) == 1"
"shar %0"
[(set_attr "type" "arith")])
(define_insn_and_split "ashrsi2_16"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ashiftrt:SI (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 16)))]
"TARGET_SH1"
"#"
"&& 1"
[(set (match_dup 0) (rotate:SI (match_dup 1) (const_int 16)))
(set (match_dup 0) (sign_extend:SI (match_dup 2)))]
{
operands[2] = gen_lowpart (HImode, operands[0]);
})
(define_insn_and_split "ashrsi2_31"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ashiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 31)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(const_int 0)]
{
emit_insn (gen_shll (operands[0], operands[1]));
emit_insn (gen_mov_neg_si_t (operands[0], get_t_reg_rtx ()));
DONE;
})
(define_insn "ashrsi3_d"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ashiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(neg:SI (match_operand:SI 2 "arith_reg_operand" "r"))))]
"TARGET_DYNSHIFT"
"shad %2,%0"
[(set_attr "type" "dyn_shift")])
(define_insn "ashrsi3_n"
[(set (reg:SI R4_REG)
(ashiftrt:SI (reg:SI R4_REG)
(match_operand:SI 0 "const_int_operand" "i")))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))
(use (match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_SH1"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_insn "ashrsi3_media"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(ashiftrt:SI (match_operand:SI 1 "extend_reg_operand" "r,r")
(match_operand:SI 2 "shift_count_operand" "r,n")))]
"TARGET_SHMEDIA"
"@
shard.l %1, %2, %0
shari.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
;; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
;; DImode arithmetic shift right
(define_expand "ashrdi3"
[(parallel [(set (match_operand:DI 0 "arith_reg_operand" "")
(ashiftrt:DI (match_operand:DI 1 "arith_reg_operand" "")
(match_operand:DI 2 "immediate_operand" "")))
(clobber (reg:SI T_REG))])]
""
{
if (TARGET_SHMEDIA)
{
if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) < 0)
{
operands[2] = GEN_INT (-INTVAL (operands[2]));
emit_insn (gen_ashldi3_media (operands[0], operands[1], operands[2]));
DONE;
}
emit_insn (gen_ashrdi3_media (operands[0], operands[1], operands[2]));
DONE;
}
if (!CONST_INT_P (operands[2]) || INTVAL (operands[2]) != 1)
FAIL;
})
(define_insn_and_split "ashrdi3_k"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ashiftrt:DI (match_operand:DI 1 "arith_reg_operand" "0")
(const_int 1)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& reload_completed"
[(const_int 0)]
{
rtx high = gen_highpart (SImode, operands[0]);
rtx low = gen_lowpart (SImode, operands[0]);
emit_insn (gen_shar (high, high));
emit_insn (gen_rotcr (low, low, get_t_reg_rtx ()));
DONE;
})
(define_insn "ashrdi3_media"
[(set (match_operand:DI 0 "ext_dest_operand" "=r,r")
(ashiftrt:DI (match_operand:DI 1 "arith_reg_operand" "r,r")
(match_operand:DI 2 "shift_count_operand" "r,n")))]
"TARGET_SHMEDIA
&& (arith_reg_dest (operands[0], DImode)
|| (CONST_INT_P (operands[2]) && INTVAL (operands[2]) >= 32))"
"@
shard %1, %2, %0
shari %1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn "*ashrdisi3_media"
[(set (subreg:DI (match_operand:SI 0 "arith_reg_operand" "=r") 0)
(ashiftrt:DI (match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "const_int_operand" "n")))]
"TARGET_SHMEDIA && INTVAL (operands[2]) < 32"
"shari.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn "ashrdisi3_media_high"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(truncate:SI
(ashiftrt:DI (match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "const_int_operand" "n"))))]
"TARGET_SHMEDIA && INTVAL (operands[2]) >= 32"
"shari %1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn "ashrdisi3_media_opaque"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(unspec:SI [(match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "const_int_operand" "n")]
UNSPEC_ASHIFTRT))]
"TARGET_SHMEDIA"
"shari %1, %2, %0"
[(set_attr "type" "arith_media")])
;; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
;; SImode logical shift right
(define_expand "lshrsi3"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(lshiftrt:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "shift_count_operand" "")))]
""
{
if (TARGET_SHMEDIA)
{
if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) < 0)
{
operands[2] = GEN_INT (-INTVAL (operands[2]));
emit_insn (gen_ashlsi3_media (operands[0], operands[1], operands[2]));
DONE;
}
emit_insn (gen_lshrsi3_media (operands[0], operands[1], operands[2]));
DONE;
}
/* If a dynamic shift is supposed to be used, expand the lshrsi3_d insn
here, otherwise the pattern will never match due to the shift amount reg
negation. */
if (TARGET_DYNSHIFT
&& CONST_INT_P (operands[2]) && sh_dynamicalize_shift_p (operands[2]))
{
rtx neg_count = force_reg (SImode,
gen_int_mode (- INTVAL (operands[2]), SImode));
emit_insn (gen_lshrsi3_d (operands[0], operands[1], neg_count));
DONE;
}
if (TARGET_DYNSHIFT && ! CONST_INT_P (operands[2]))
{
rtx neg_count = gen_reg_rtx (SImode);
emit_insn (gen_negsi2 (neg_count, operands[2]));
emit_insn (gen_lshrsi3_d (operands[0], operands[1], neg_count));
DONE;
}
/* If the lshrsi3_* insn is going to clobber the T_REG it must be
expanded here. */
if (CONST_INT_P (operands[2])
&& sh_lshrsi_clobbers_t_reg_p (operands[2])
&& ! sh_dynamicalize_shift_p (operands[2]))
{
emit_insn (gen_lshrsi3_n_clobbers_t (operands[0], operands[1],
operands[2]));
DONE;
}
/* Expand a library call for the dynamic shift. */
if (!CONST_INT_P (operands[2]) && !TARGET_DYNSHIFT)
{
emit_move_insn (gen_rtx_REG (SImode, R4_REG), operands[1]);
rtx funcaddr = gen_reg_rtx (Pmode);
function_symbol (funcaddr, "__lshrsi3_r0", SFUNC_STATIC);
emit_insn (gen_lshrsi3_d_call (operands[0], operands[2], funcaddr));
DONE;
}
})
(define_insn "lshrsi3_k"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(lshiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "p27_rshift_count_operand" "P27")))]
"TARGET_SH1"
"shlr%O2 %0"
[(set_attr "type" "arith")])
(define_insn_and_split "lshrsi3_d"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(lshiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(neg:SI (match_operand:SI 2 "shift_count_operand" "r"))))]
"TARGET_DYNSHIFT"
"shld %2,%0"
"&& CONST_INT_P (operands[2]) && ! sh_dynamicalize_shift_p (operands[2])
&& ! sh_lshrsi_clobbers_t_reg_p (operands[2])"
[(const_int 0)]
{
if (satisfies_constraint_P27 (operands[2]))
{
/* This will not be done for a shift amount of 1, because it would
clobber the T_REG. */
emit_insn (gen_lshrsi3_k (operands[0], operands[1], operands[2]));
DONE;
}
else if (! satisfies_constraint_P27 (operands[2]))
{
/* This must happen before reload, otherwise the constant will be moved
into a register due to the "r" constraint, after which this split
cannot be done anymore.
Unfortunately the move insn will not always be eliminated.
Also, here we must not create a shift sequence that clobbers the
T_REG. */
emit_move_insn (operands[0], operands[1]);
gen_shifty_op (LSHIFTRT, operands);
DONE;
}
FAIL;
}
[(set_attr "type" "dyn_shift")])
;; If dynamic shifts are not available use a library function.
;; By specifying the pattern we reduce the number of call clobbered regs.
;; In order to make combine understand the truncation of the shift amount
;; operand we have to allow it to use pseudo regs for the shift operands.
(define_insn "lshrsi3_d_call"
[(set (match_operand:SI 0 "arith_reg_dest" "=z")
(lshiftrt:SI (reg:SI R4_REG)
(and:SI (match_operand:SI 1 "arith_reg_operand" "z")
(const_int 31))))
(use (match_operand:SI 2 "arith_reg_operand" "r"))
(clobber (reg:SI T_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SH1 && !TARGET_DYNSHIFT"
"jsr @%2%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_insn_and_split "lshrsi3_n"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(lshiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "not_p27_rshift_count_operand")))]
"TARGET_SH1 && ! sh_lshrsi_clobbers_t_reg_p (operands[2])"
"#"
"&& (reload_completed
|| (sh_dynamicalize_shift_p (operands[2]) && can_create_pseudo_p ()))"
[(const_int 0)]
{
if (sh_dynamicalize_shift_p (operands[2]) && can_create_pseudo_p ())
{
/* If this pattern was picked and dynamic shifts are supported, switch
to dynamic shift pattern before reload. */
operands[2] = force_reg (SImode,
gen_int_mode (- INTVAL (operands[2]), SImode));
emit_insn (gen_lshrsi3_d (operands[0], operands[1], operands[2]));
}
else
gen_shifty_op (LSHIFTRT, operands);
DONE;
})
;; The lshrsi3_n_clobbers_t pattern also works as a simplified version of
;; the shlr pattern.
(define_insn_and_split "lshrsi3_n_clobbers_t"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(lshiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand:SI 2 "not_p27_rshift_count_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1 && sh_lshrsi_clobbers_t_reg_p (operands[2])"
"#"
"&& (reload_completed || INTVAL (operands[2]) == 31
|| (sh_dynamicalize_shift_p (operands[2]) && can_create_pseudo_p ()))"
[(const_int 0)]
{
if (INTVAL (operands[2]) == 31)
{
emit_insn (gen_shll (operands[0], operands[1]));
emit_insn (gen_movt (operands[0], get_t_reg_rtx ()));
}
else if (sh_dynamicalize_shift_p (operands[2]) && can_create_pseudo_p ())
{
/* If this pattern was picked and dynamic shifts are supported, switch
to dynamic shift pattern before reload. */
operands[2] = force_reg (SImode,
gen_int_mode (- INTVAL (operands[2]), SImode));
emit_insn (gen_lshrsi3_d (operands[0], operands[1], operands[2]));
}
else
gen_shifty_op (LSHIFTRT, operands);
DONE;
})
(define_insn "shlr"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(lshiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 1)))
(set (reg:SI T_REG)
(and:SI (match_dup 1) (const_int 1)))]
"TARGET_SH1"
"shlr %0"
[(set_attr "type" "arith")])
(define_insn "lshrsi3_media"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(lshiftrt:SI (match_operand:SI 1 "extend_reg_operand" "r,r")
(match_operand:SI 2 "shift_count_operand" "r,n")))]
"TARGET_SHMEDIA"
"@
shlrd.l %1, %2, %0
shlri.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
;; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
;; DImode logical shift right
(define_expand "lshrdi3"
[(parallel [(set (match_operand:DI 0 "arith_reg_operand" "")
(lshiftrt:DI (match_operand:DI 1 "arith_reg_operand" "")
(match_operand:DI 2 "immediate_operand" "")))
(clobber (reg:SI T_REG))])]
""
{
if (TARGET_SHMEDIA)
{
if (CONST_INT_P (operands[2]) && INTVAL (operands[2]) < 0)
{
operands[2] = GEN_INT (-INTVAL (operands[2]));
emit_insn (gen_ashldi3_media (operands[0], operands[1], operands[2]));
DONE;
}
emit_insn (gen_lshrdi3_media (operands[0], operands[1], operands[2]));
DONE;
}
if (!CONST_INT_P (operands[2]) || INTVAL (operands[2]) != 1)
FAIL;
})
(define_insn_and_split "lshrdi3_k"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(lshiftrt:DI (match_operand:DI 1 "arith_reg_operand" "0")
(const_int 1)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& reload_completed"
[(const_int 0)]
{
rtx high = gen_highpart (SImode, operands[0]);
rtx low = gen_lowpart (SImode, operands[0]);
emit_insn (gen_shlr (high, high));
emit_insn (gen_rotcr (low, low, get_t_reg_rtx ()));
DONE;
})
(define_insn "lshrdi3_media"
[(set (match_operand:DI 0 "ext_dest_operand" "=r,r")
(lshiftrt:DI (match_operand:DI 1 "arith_reg_operand" "r,r")
(match_operand:DI 2 "shift_count_operand" "r,n")))]
"TARGET_SHMEDIA
&& (arith_reg_dest (operands[0], DImode)
|| (CONST_INT_P (operands[2]) && INTVAL (operands[2]) > 32))"
"@
shlrd %1, %2, %0
shlri %1, %2, %0"
[(set_attr "type" "arith_media")])
(define_insn "*lshrdisi3_media"
[(set (subreg:DI (match_operand:SI 0 "arith_reg_operand" "=r") 0)
(lshiftrt:DI (match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "const_int_operand" "n")))]
"TARGET_SHMEDIA && INTVAL (operands[2]) < 32"
"shlri.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
;; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
;; Combined left/right shifts
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(and:SI (ashift:SI (match_operand:SI 1 "register_operand" "")
(match_operand:SI 2 "const_int_operand" ""))
(match_operand:SI 3 "const_int_operand" "")))]
"TARGET_SH1 && reload_completed && (unsigned)INTVAL (operands[2]) < 32"
[(use (reg:SI R0_REG))]
{
if (gen_shl_and (operands[0], operands[2], operands[3], operands[1]))
FAIL;
DONE;
})
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(and:SI (ashift:SI (match_operand:SI 1 "register_operand" "")
(match_operand:SI 2 "const_int_operand" ""))
(match_operand:SI 3 "const_int_operand" "")))
(clobber (reg:SI T_REG))]
"TARGET_SH1 && reload_completed && (unsigned)INTVAL (operands[2]) < 32"
[(use (reg:SI R0_REG))]
{
if (gen_shl_and (operands[0], operands[2], operands[3], operands[1]))
FAIL;
DONE;
})
(define_insn ""
[(set (match_operand:SI 0 "register_operand" "=r")
(and:SI (ashift:SI (match_operand:SI 1 "register_operand" "0")
(match_operand:SI 2 "const_int_operand" "n"))
(match_operand:SI 3 "const_int_operand" "n")))
(clobber (reg:SI T_REG))]
"TARGET_SH1 && shl_and_kind (operands[2], operands[3], 0) == 1"
"#"
[(set (attr "length")
(cond [(eq (symbol_ref "shl_and_length (insn)") (const_int 2))
(const_string "4")
(eq (symbol_ref "shl_and_length (insn)") (const_int 3))
(const_string "6")
(eq (symbol_ref "shl_and_length (insn)") (const_int 4))
(const_string "8")
(eq (symbol_ref "shl_and_length (insn)") (const_int 5))
(const_string "10")
(eq (symbol_ref "shl_and_length (insn)") (const_int 6))
(const_string "12")
(eq (symbol_ref "shl_and_length (insn)") (const_int 7))
(const_string "14")
(eq (symbol_ref "shl_and_length (insn)") (const_int 8))
(const_string "16")]
(const_string "18")))
(set_attr "type" "arith")])
(define_insn ""
[(set (match_operand:SI 0 "register_operand" "=z")
(and:SI (ashift:SI (match_operand:SI 1 "register_operand" "0")
(match_operand:SI 2 "const_int_operand" "n"))
(match_operand:SI 3 "const_int_operand" "n")))
(clobber (reg:SI T_REG))]
"TARGET_SH1 && shl_and_kind (operands[2], operands[3], 0) == 2"
"#"
[(set (attr "length")
(cond [(eq (symbol_ref "shl_and_length (insn)") (const_int 2))
(const_string "4")
(eq (symbol_ref "shl_and_length (insn)") (const_int 3))
(const_string "6")
(eq (symbol_ref "shl_and_length (insn)") (const_int 4))
(const_string "8")]
(const_string "10")))
(set_attr "type" "arith")])
;; shift left / and combination with a scratch register: The combine pass
;; does not accept the individual instructions, even though they are
;; cheap. But it needs a precise description so that it is usable after
;; reload.
(define_insn "and_shl_scratch"
[(set (match_operand:SI 0 "register_operand" "=r,&r")
(lshiftrt:SI
(ashift:SI
(and:SI
(lshiftrt:SI (match_operand:SI 1 "register_operand" "r,0")
(match_operand:SI 2 "const_int_operand" "N,n"))
(match_operand:SI 3 "" "0,r"))
(match_operand:SI 4 "const_int_operand" "n,n"))
(match_operand:SI 5 "const_int_operand" "n,n")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
[(set (attr "length")
(cond [(eq (symbol_ref "shl_and_scr_length (insn)") (const_int 2))
(const_string "4")
(eq (symbol_ref "shl_and_scr_length (insn)") (const_int 3))
(const_string "6")
(eq (symbol_ref "shl_and_scr_length (insn)") (const_int 4))
(const_string "8")
(eq (symbol_ref "shl_and_scr_length (insn)") (const_int 5))
(const_string "10")]
(const_string "12")))
(set_attr "type" "arith")])
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(lshiftrt:SI
(ashift:SI
(and:SI
(lshiftrt:SI (match_operand:SI 1 "register_operand" "")
(match_operand:SI 2 "const_int_operand" ""))
(match_operand:SI 3 "register_operand" ""))
(match_operand:SI 4 "const_int_operand" ""))
(match_operand:SI 5 "const_int_operand" "")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
[(use (reg:SI R0_REG))]
{
rtx and_source = operands[rtx_equal_p (operands[0], operands[1]) ? 3 : 1];
if (INTVAL (operands[2]))
{
gen_shifty_op (LSHIFTRT, operands);
}
emit_insn (gen_andsi3 (operands[0], operands[0], and_source));
operands[2] = operands[4];
gen_shifty_op (ASHIFT, operands);
if (INTVAL (operands[5]))
{
operands[2] = operands[5];
gen_shifty_op (LSHIFTRT, operands);
}
DONE;
})
;; signed left/right shift combination.
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(sign_extract:SI
(ashift:SI (match_operand:SI 1 "register_operand" "")
(match_operand:SI 2 "const_int_operand" ""))
(match_operand:SI 3 "const_int_operand" "")
(const_int 0)))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
[(use (reg:SI R0_REG))]
{
if (gen_shl_sext (operands[0], operands[2], operands[3], operands[1]))
FAIL;
DONE;
})
(define_insn "shl_sext_ext"
[(set (match_operand:SI 0 "register_operand" "=r")
(sign_extract:SI
(ashift:SI (match_operand:SI 1 "register_operand" "0")
(match_operand:SI 2 "const_int_operand" "n"))
(match_operand:SI 3 "const_int_operand" "n")
(const_int 0)))
(clobber (reg:SI T_REG))]
"TARGET_SH1 && (unsigned)shl_sext_kind (operands[2], operands[3], 0) - 1 < 5"
"#"
[(set (attr "length")
(cond [(match_test "shl_sext_length (insn)")
(const_string "2")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 2))
(const_string "4")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 3))
(const_string "6")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 4))
(const_string "8")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 5))
(const_string "10")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 6))
(const_string "12")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 7))
(const_string "14")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 8))
(const_string "16")]
(const_string "18")))
(set_attr "type" "arith")])
(define_insn "shl_sext_sub"
[(set (match_operand:SI 0 "register_operand" "=z")
(sign_extract:SI
(ashift:SI (match_operand:SI 1 "register_operand" "0")
(match_operand:SI 2 "const_int_operand" "n"))
(match_operand:SI 3 "const_int_operand" "n")
(const_int 0)))
(clobber (reg:SI T_REG))]
"TARGET_SH1 && (shl_sext_kind (operands[2], operands[3], 0) & ~1) == 6"
"#"
[(set (attr "length")
(cond [(eq (symbol_ref "shl_sext_length (insn)") (const_int 3))
(const_string "6")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 4))
(const_string "8")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 5))
(const_string "10")
(eq (symbol_ref "shl_sext_length (insn)") (const_int 6))
(const_string "12")]
(const_string "14")))
(set_attr "type" "arith")])
;; The xtrct_left and xtrct_right patterns are used in expansions of DImode
;; shifts by 16, and allow the xtrct instruction to be generated from C
;; source.
(define_insn "xtrct_left"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ior:SI (ashift:SI (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 16))
(lshiftrt:SI (match_operand:SI 2 "arith_reg_operand" "0")
(const_int 16))))]
"TARGET_SH1"
"xtrct %1,%0"
[(set_attr "type" "arith")])
(define_insn "xtrct_right"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ior:SI (lshiftrt:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 16))
(ashift:SI (match_operand:SI 2 "arith_reg_operand" "r")
(const_int 16))))]
"TARGET_SH1"
"xtrct %2,%0"
[(set_attr "type" "arith")])
;; -------------------------------------------------------------------------
;; Unary arithmetic
;; -------------------------------------------------------------------------
(define_insn "negc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(neg:SI (plus:SI (reg:SI T_REG)
(match_operand:SI 1 "arith_reg_operand" "r"))))
(set (reg:SI T_REG)
(ne:SI (ior:SI (reg:SI T_REG) (match_dup 1))
(const_int 0)))]
"TARGET_SH1"
"negc %1,%0"
[(set_attr "type" "arith")])
;; A simplified version of the negc insn, where the exact value of the
;; T bit doesn't matter. This is easier for combine to pick up.
;; Notice that '0 - x - 1' is the same as '~x', thus we don't specify
;; extra patterns for this case.
(define_insn "*negc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(minus:SI (neg:SI (match_operand:SI 1 "arith_reg_operand" "r"))
(match_operand:SI 2 "t_reg_operand" "")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"negc %1,%0"
[(set_attr "type" "arith")])
(define_insn "*negdi_media"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(neg:DI (match_operand:DI 1 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"sub r63, %1, %0"
[(set_attr "type" "arith_media")])
;; Don't split into individual negc insns immediately so that neg:DI (abs:DI)
;; can be combined.
(define_expand "negdi2"
[(parallel [(set (match_operand:DI 0 "arith_reg_dest")
(neg:DI (match_operand:DI 1 "arith_reg_operand")))
(clobber (reg:SI T_REG))])]
"TARGET_SH1")
(define_insn_and_split "*negdi2"
[(set (match_operand:DI 0 "arith_reg_dest")
(neg:DI (match_operand:DI 1 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
emit_insn (gen_clrt ());
emit_insn (gen_negc (gen_lowpart (SImode, operands[0]),
gen_lowpart (SImode, operands[1])));
emit_insn (gen_negc (gen_highpart (SImode, operands[0]),
gen_highpart (SImode, operands[1])));
DONE;
})
(define_insn "negsi2"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(neg:SI (match_operand:SI 1 "arith_reg_operand" "r")))]
"TARGET_SH1"
"neg %1,%0"
[(set_attr "type" "arith")])
(define_insn_and_split "one_cmplsi2"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(not:SI (match_operand:SI 1 "arith_reg_operand" "r")))]
"TARGET_SH1"
"not %1,%0"
"&& can_create_pseudo_p ()"
[(set (reg:SI T_REG) (ge:SI (match_dup 1) (const_int 0)))
(set (match_dup 0) (reg:SI T_REG))]
{
/* PR 54685
If the result of 'unsigned int <= 0x7FFFFFFF' ends up as the following
sequence:
(set (reg0) (not:SI (reg0) (reg1)))
(parallel [(set (reg2) (lshiftrt:SI (reg0) (const_int 31)))
(clobber (reg:SI T_REG))])
... match and combine the sequence manually in the split pass after the
combine pass. Notice that combine does try the target pattern of this
split, but if the pattern is added it interferes with other patterns, in
particular with the div0s comparisons.
This could also be done with a peephole but doing it here before register
allocation can save one temporary.
When we're here, the not:SI pattern obviously has been matched already
and we only have to see whether the following insn is the left shift. */
rtx i = next_nonnote_insn_bb (curr_insn);
if (i == NULL_RTX || !NONJUMP_INSN_P (i))
FAIL;
rtx p = PATTERN (i);
if (GET_CODE (p) != PARALLEL || XVECLEN (p, 0) != 2)
FAIL;
rtx p0 = XVECEXP (p, 0, 0);
rtx p1 = XVECEXP (p, 0, 1);
if (/* (set (reg2) (lshiftrt:SI (reg0) (const_int 31))) */
GET_CODE (p0) == SET
&& GET_CODE (XEXP (p0, 1)) == LSHIFTRT
&& REG_P (XEXP (XEXP (p0, 1), 0))
&& REGNO (XEXP (XEXP (p0, 1), 0)) == REGNO (operands[0])
&& CONST_INT_P (XEXP (XEXP (p0, 1), 1))
&& INTVAL (XEXP (XEXP (p0, 1), 1)) == 31
/* (clobber (reg:SI T_REG)) */
&& GET_CODE (p1) == CLOBBER && REG_P (XEXP (p1, 0))
&& REGNO (XEXP (p1, 0)) == T_REG)
{
operands[0] = XEXP (p0, 0);
set_insn_deleted (i);
}
else
FAIL;
}
[(set_attr "type" "arith")])
(define_expand "one_cmpldi2"
[(set (match_operand:DI 0 "arith_reg_dest" "")
(xor:DI (match_operand:DI 1 "arith_reg_operand" "")
(const_int -1)))]
"TARGET_SHMEDIA" "")
(define_expand "abs2"
[(parallel [(set (match_operand:SIDI 0 "arith_reg_dest")
(abs:SIDI (match_operand:SIDI 1 "arith_reg_operand")))
(clobber (reg:SI T_REG))])]
"TARGET_SH1")
(define_insn_and_split "*abs2"
[(set (match_operand:SIDI 0 "arith_reg_dest")
(abs:SIDI (match_operand:SIDI 1 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
if (mode == SImode)
emit_insn (gen_cmpgesi_t (operands[1], const0_rtx));
else
{
rtx high_src = gen_highpart (SImode, operands[1]);
emit_insn (gen_cmpgesi_t (high_src, const0_rtx));
}
emit_insn (gen_neg_cond (operands[0], operands[1], operands[1],
const1_rtx));
DONE;
})
(define_insn_and_split "*negabs2"
[(set (match_operand:SIDI 0 "arith_reg_dest")
(neg:SIDI (abs:SIDI (match_operand:SIDI 1 "arith_reg_operand"))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
if (mode == SImode)
emit_insn (gen_cmpgesi_t (operands[1], const0_rtx));
else
{
rtx high_src = gen_highpart (SImode, operands[1]);
emit_insn (gen_cmpgesi_t (high_src, const0_rtx));
}
emit_insn (gen_neg_cond (operands[0], operands[1], operands[1],
const0_rtx));
DONE;
})
;; The SH4 202 can do zero-offset branches without pipeline stalls.
;; This can be used as some kind of conditional execution, which is useful
;; for abs.
;; Actually the instruction scheduling should decide whether to use a
;; zero-offset branch or not for any generic case involving a single
;; instruction on SH4 202.
(define_insn_and_split "negsi_cond"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(if_then_else
(eq:SI (reg:SI T_REG) (match_operand:SI 3 "const_int_operand" "M,N"))
(match_operand:SI 1 "arith_reg_operand" "0,0")
(neg:SI (match_operand:SI 2 "arith_reg_operand" "r,r"))))]
"TARGET_SH1 && TARGET_ZDCBRANCH"
{
static const char* alt[] =
{
"bt 0f" "\n"
" neg %2,%0" "\n"
"0:",
"bf 0f" "\n"
" neg %2,%0" "\n"
"0:"
};
return alt[which_alternative];
}
"TARGET_SH1 && ! TARGET_ZDCBRANCH"
[(const_int 0)]
{
rtx skip_neg_label = gen_label_rtx ();
emit_move_insn (operands[0], operands[1]);
emit_jump_insn (INTVAL (operands[3])
? gen_branch_true (skip_neg_label)
: gen_branch_false (skip_neg_label));
emit_label_after (skip_neg_label,
emit_insn (gen_negsi2 (operands[0], operands[1])));
DONE;
}
[(set_attr "type" "arith") ;; poor approximation
(set_attr "length" "4")])
(define_insn_and_split "negdi_cond"
[(set (match_operand:DI 0 "arith_reg_dest")
(if_then_else
(eq:SI (reg:SI T_REG) (match_operand:SI 3 "const_int_operand"))
(match_operand:DI 1 "arith_reg_operand")
(neg:DI (match_operand:DI 2 "arith_reg_operand"))))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
rtx skip_neg_label = gen_label_rtx ();
emit_move_insn (operands[0], operands[1]);
emit_jump_insn (INTVAL (operands[3])
? gen_branch_true (skip_neg_label)
: gen_branch_false (skip_neg_label));
if (!INTVAL (operands[3]))
emit_insn (gen_clrt ());
emit_insn (gen_negc (gen_lowpart (SImode, operands[0]),
gen_lowpart (SImode, operands[1])));
emit_label_after (skip_neg_label,
emit_insn (gen_negc (gen_highpart (SImode, operands[0]),
gen_highpart (SImode, operands[1]))));
DONE;
})
(define_expand "bswapsi2"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(bswap:SI (match_operand:SI 1 "arith_reg_operand" "")))]
"TARGET_SH1"
{
if (! can_create_pseudo_p ())
FAIL;
else
{
rtx tmp0 = gen_reg_rtx (SImode);
rtx tmp1 = gen_reg_rtx (SImode);
emit_insn (gen_swapbsi2 (tmp0, operands[1]));
emit_insn (gen_rotlsi3_16 (tmp1, tmp0));
emit_insn (gen_swapbsi2 (operands[0], tmp1));
DONE;
}
})
(define_insn "swapbsi2"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ior:SI (and:SI (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 4294901760))
(ior:SI (and:SI (ashift:SI (match_dup 1) (const_int 8))
(const_int 65280))
(and:SI (ashiftrt:SI (match_dup 1) (const_int 8))
(const_int 255)))))]
"TARGET_SH1"
"swap.b %1,%0"
[(set_attr "type" "arith")])
;; The *swapbisi2_and_shl8 pattern helps the combine pass simplifying
;; partial byte swap expressions such as...
;; ((x & 0xFF) << 8) | ((x >> 8) & 0xFF).
;; ...which are currently not handled by the tree optimizers.
;; The combine pass will not initially try to combine the full expression,
;; but only some sub-expressions. In such a case the *swapbisi2_and_shl8
;; pattern acts as an intermediate pattern that will eventually lead combine
;; to the swapbsi2 pattern above.
;; As a side effect this also improves code that does (x & 0xFF) << 8
;; or (x << 8) & 0xFF00.
(define_insn_and_split "*swapbisi2_and_shl8"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ior:SI (and:SI (ashift:SI (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 8))
(const_int 65280))
(match_operand:SI 2 "arith_reg_operand" "r")))]
"TARGET_SH1 && ! reload_in_progress && ! reload_completed"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
rtx tmp0 = gen_reg_rtx (SImode);
rtx tmp1 = gen_reg_rtx (SImode);
emit_insn (gen_zero_extendqisi2 (tmp0, gen_lowpart (QImode, operands[1])));
emit_insn (gen_swapbsi2 (tmp1, tmp0));
emit_insn (gen_iorsi3 (operands[0], tmp1, operands[2]));
DONE;
})
;; The *swapbhisi2 pattern is, like the *swapbisi2_and_shl8 pattern, another
;; intermediate pattern that will help the combine pass arriving at swapbsi2.
(define_insn_and_split "*swapbhisi2"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ior:SI (and:SI (ashift:SI (match_operand:SI 1 "arith_reg_operand" "r")
(const_int 8))
(const_int 65280))
(zero_extract:SI (match_dup 1) (const_int 8) (const_int 8))))]
"TARGET_SH1 && ! reload_in_progress && ! reload_completed"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
rtx tmp = gen_reg_rtx (SImode);
emit_insn (gen_zero_extendhisi2 (tmp, gen_lowpart (HImode, operands[1])));
emit_insn (gen_swapbsi2 (operands[0], tmp));
DONE;
})
;; In some cases the swapbsi2 pattern might leave a sequence such as...
;; swap.b r4,r4
;; mov r4,r0
;;
;; which can be simplified to...
;; swap.b r4,r0
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(ior:SI (and:SI (match_operand:SI 1 "arith_reg_operand" "")
(const_int 4294901760))
(ior:SI (and:SI (ashift:SI (match_dup 1) (const_int 8))
(const_int 65280))
(and:SI (ashiftrt:SI (match_dup 1) (const_int 8))
(const_int 255)))))
(set (match_operand:SI 2 "arith_reg_dest" "")
(match_dup 0))]
"TARGET_SH1 && peep2_reg_dead_p (2, operands[0])"
[(set (match_dup 2)
(ior:SI (and:SI (match_operand:SI 1 "arith_reg_operand" "")
(const_int 4294901760))
(ior:SI (and:SI (ashift:SI (match_dup 1) (const_int 8))
(const_int 65280))
(and:SI (ashiftrt:SI (match_dup 1) (const_int 8))
(const_int 255)))))])
;; -------------------------------------------------------------------------
;; Zero extension instructions
;; -------------------------------------------------------------------------
(define_insn "zero_extendsidi2"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extend:DI (match_operand:SI 1 "extend_reg_operand" "r")))]
"TARGET_SHMEDIA"
"addz.l %1, r63, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "extend")])
(define_insn "zero_extendhidi2"
[(set (match_operand:DI 0 "register_operand" "=r,r")
(zero_extend:DI (match_operand:HI 1 "general_extend_operand" "r,m")))]
"TARGET_SHMEDIA"
"@
#
ld%M1.uw %m1, %0"
[(set_attr "type" "*,load_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_split
[(set (match_operand:DI 0 "register_operand" "")
(zero_extend:DI (match_operand:HI 1 "extend_reg_operand" "")))]
"TARGET_SHMEDIA && reload_completed"
[(set (match_dup 0) (ashift:DI (subreg:DI (match_dup 1) 0) (const_int 48)))
(set (match_dup 0) (lshiftrt:DI (match_dup 0) (const_int 48)))]
{
if (GET_CODE (operands[1]) == TRUNCATE)
operands[1] = XEXP (operands[1], 0);
})
;; ??? when a truncated input to a zero_extend is reloaded, reload will
;; reload the entire truncate expression.
(define_insn_and_split "*loaddi_trunc"
[(set (match_operand 0 "any_register_operand" "=r")
(truncate (match_operand:DI 1 "memory_operand" "m")))]
"TARGET_SHMEDIA && reload_completed"
"#"
"TARGET_SHMEDIA && reload_completed"
[(set (match_dup 0) (match_dup 1))]
{
operands[0] = gen_rtx_REG (DImode, true_regnum (operands[0]));
})
(define_insn "zero_extendqidi2"
[(set (match_operand:DI 0 "register_operand" "=r,r")
(zero_extend:DI (match_operand:QI 1 "general_extend_operand" "r,m")))]
"TARGET_SHMEDIA"
"@
andi %1, 255, %0
ld%M1.ub %m1, %0"
[(set_attr "type" "arith_media,load_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_expand "zero_extendsi2"
[(set (match_operand:SI 0 "arith_reg_dest")
(zero_extend:SI (match_operand:QIHI 1 "zero_extend_operand")))])
(define_insn_and_split "*zero_extendsi2_compact"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(zero_extend:SI (match_operand:QIHI 1 "arith_reg_operand" "r")))]
"TARGET_SH1"
"extu. %1,%0"
"&& can_create_pseudo_p ()"
[(set (match_dup 0) (match_dup 2))]
{
/* Sometimes combine fails to combine a T bit or negated T bit store to a
reg with a following zero extension. In the split pass after combine,
try to figure out how the extended reg was set. If it originated from
the T bit we can replace the zero extension with a reg move, which will
be eliminated. Notice that this also helps the *cbranch_t splitter when
it tries to post-combine tests and conditional branches, as it does not
check for zero extensions. */
operands[2] = sh_try_omit_signzero_extend (operands[1], curr_insn);
if (operands[2] == NULL_RTX)
FAIL;
}
[(set_attr "type" "arith")])
(define_insn "*zero_extendhisi2_media"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(zero_extend:SI (match_operand:HI 1 "general_extend_operand" "r,m")))]
"TARGET_SHMEDIA"
"@
#
ld%M1.uw %m1, %0"
[(set_attr "type" "arith_media,load_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(zero_extend:SI (match_operand:HI 1 "extend_reg_operand" "")))]
"TARGET_SHMEDIA && reload_completed"
[(set (match_dup 0) (ashift:SI (match_dup 2) (const_int 16)))
(set (match_dup 0) (lshiftrt:SI (match_dup 0) (const_int 16)))]
{
rtx op1 = operands[1];
if (GET_CODE (op1) == TRUNCATE)
op1 = XEXP (op1, 0);
operands[2]
= simplify_gen_subreg (SImode, op1, GET_MODE (op1),
subreg_lowpart_offset (SImode, GET_MODE (op1)));
})
(define_insn "*zero_extendqisi2_media"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(zero_extend:SI (match_operand:QI 1 "general_extend_operand" "r,m")))]
"TARGET_SHMEDIA"
"@
andi %1, 255, %0
ld%M1.ub %m1, %0"
[(set_attr "type" "arith_media,load_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_insn "zero_extendqihi2"
[(set (match_operand:HI 0 "arith_reg_dest" "=r")
(zero_extend:HI (match_operand:QI 1 "arith_reg_operand" "r")))]
"TARGET_SH1"
"extu.b %1,%0"
[(set_attr "type" "arith")])
;; SH2A supports two zero extending load instructions: movu.b and movu.w.
;; They could also be used for simple memory addresses like @Rn by setting
;; the displacement value to zero. However, doing so too early results in
;; missed opportunities for other optimizations such as post-inc or index
;; addressing loads.
;; Although the 'zero_extend_movu_operand' predicate does not allow simple
;; register addresses (an address without a displacement, index, post-inc),
;; zero-displacement addresses might be generated during reload, wich are
;; simplified to simple register addresses in turn. Thus, we have to
;; provide the Sdd and Sra alternatives in the patterns.
(define_insn "*zero_extendsi2_disp_mem"
[(set (match_operand:SI 0 "arith_reg_dest" "=r,r")
(zero_extend:SI
(match_operand:QIHI 1 "zero_extend_movu_operand" "Sdd,Sra")))]
"TARGET_SH2A"
"@
movu. %1,%0
movu. @(0,%t1),%0"
[(set_attr "type" "load")
(set_attr "length" "4")])
;; Convert the zero extending loads in sequences such as:
;; movu.b @(1,r5),r0 movu.w @(2,r5),r0
;; mov.b r0,@(1,r4) mov.b r0,@(1,r4)
;;
;; back to sign extending loads like:
;; mov.b @(1,r5),r0 mov.w @(2,r5),r0
;; mov.b r0,@(1,r4) mov.b r0,@(1,r4)
;;
;; if the extension type is irrelevant. The sign extending mov.{b|w} insn
;; is only 2 bytes in size if the displacement is {K04|K05}.
;; If the displacement is greater it doesn't matter, so we convert anyways.
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(zero_extend:SI (match_operand 1 "displacement_mem_operand" "")))
(set (match_operand 2 "nonimmediate_operand" "")
(match_operand 3 "arith_reg_operand" ""))]
"TARGET_SH2A
&& REGNO (operands[0]) == REGNO (operands[3])
&& peep2_reg_dead_p (2, operands[0])
&& GET_MODE_SIZE (GET_MODE (operands[2]))
<= GET_MODE_SIZE (GET_MODE (operands[1]))"
[(set (match_dup 0) (sign_extend:SI (match_dup 1)))
(set (match_dup 2) (match_dup 3))])
;; Fold sequences such as
;; mov.b @r3,r7
;; extu.b r7,r7
;; into
;; movu.b @(0,r3),r7
;; This does not reduce the code size but the number of instructions is
;; halved, which results in faster code.
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(sign_extend:SI (match_operand 1 "simple_mem_operand" "")))
(set (match_operand:SI 2 "arith_reg_dest" "")
(zero_extend:SI (match_operand 3 "arith_reg_operand" "")))]
"TARGET_SH2A
&& GET_MODE (operands[1]) == GET_MODE (operands[3])
&& (GET_MODE (operands[1]) == QImode || GET_MODE (operands[1]) == HImode)
&& REGNO (operands[0]) == REGNO (operands[3])
&& (REGNO (operands[2]) == REGNO (operands[0])
|| peep2_reg_dead_p (2, operands[0]))"
[(set (match_dup 2) (zero_extend:SI (match_dup 4)))]
{
operands[4]
= replace_equiv_address (operands[1],
gen_rtx_PLUS (SImode, XEXP (operands[1], 0),
const0_rtx));
})
;; -------------------------------------------------------------------------
;; Sign extension instructions
;; -------------------------------------------------------------------------
;; ??? This should be a define expand.
;; ??? Or perhaps it should be dropped?
;; convert_move generates good code for SH[1-4].
(define_insn "extendsidi2"
[(set (match_operand:DI 0 "register_operand" "=r,r,r")
(sign_extend:DI (match_operand:SI 1 "nonimmediate_operand" "r,m,?f")))]
"TARGET_SHMEDIA"
"@
add.l %1, r63, %0
ld%M1.l %m1, %0
fmov.sl %1, %0"
[(set_attr "type" "arith_media,load_media,fpconv_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "extend")))])
(define_insn "extendhidi2"
[(set (match_operand:DI 0 "register_operand" "=r,r")
(sign_extend:DI (match_operand:HI 1 "general_extend_operand" "r,m")))]
"TARGET_SHMEDIA"
"@
#
ld%M1.w %m1, %0"
[(set_attr "type" "*,load_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_split
[(set (match_operand:DI 0 "register_operand" "")
(sign_extend:DI (match_operand:HI 1 "extend_reg_operand" "")))]
"TARGET_SHMEDIA && reload_completed"
[(set (match_dup 0) (ashift:DI (subreg:DI (match_dup 1) 0) (const_int 48)))
(set (match_dup 0) (ashiftrt:DI (match_dup 0) (const_int 48)))]
{
if (GET_CODE (operands[1]) == TRUNCATE)
operands[1] = XEXP (operands[1], 0);
})
(define_insn "extendqidi2"
[(set (match_operand:DI 0 "register_operand" "=r,r")
(sign_extend:DI (match_operand:QI 1 "general_extend_operand" "r,m")))]
"TARGET_SHMEDIA"
"@
#
ld%M1.b %m1, %0"
[(set_attr "type" "*,load_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_split
[(set (match_operand:DI 0 "register_operand" "")
(sign_extend:DI (match_operand:QI 1 "extend_reg_operand" "")))]
"TARGET_SHMEDIA && reload_completed"
[(set (match_dup 0) (ashift:DI (subreg:DI (match_dup 1) 0) (const_int 56)))
(set (match_dup 0) (ashiftrt:DI (match_dup 0) (const_int 56)))]
{
if (GET_CODE (operands[1]) == TRUNCATE)
operands[1] = XEXP (operands[1], 0);
})
(define_expand "extendsi2"
[(set (match_operand:SI 0 "arith_reg_dest")
(sign_extend:SI (match_operand:QIHI 1 "general_extend_operand")))])
(define_insn "*extendhisi2_media"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(sign_extend:SI (match_operand:HI 1 "general_extend_operand" "r,m")))]
"TARGET_SHMEDIA"
"@
#
ld%M1.w %m1, %0"
[(set_attr "type" "arith_media,load_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(sign_extend:SI (match_operand:HI 1 "extend_reg_operand" "")))]
"TARGET_SHMEDIA && reload_completed"
[(set (match_dup 0) (ashift:SI (match_dup 2) (const_int 16)))
(set (match_dup 0) (ashiftrt:SI (match_dup 0) (const_int 16)))]
{
rtx op1 = operands[1];
if (GET_CODE (op1) == TRUNCATE)
op1 = XEXP (op1, 0);
operands[2]
= simplify_gen_subreg (SImode, op1, GET_MODE (op1),
subreg_lowpart_offset (SImode, GET_MODE (op1)));
})
(define_insn_and_split "*extendsi2_compact_reg"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(sign_extend:SI (match_operand:QIHI 1 "arith_reg_operand" "r")))]
"TARGET_SH1"
"exts. %1,%0"
"&& can_create_pseudo_p ()"
[(set (match_dup 0) (match_dup 2))]
{
/* Sometimes combine fails to combine a T bit or negated T bit store to a
reg with a following sign extension. In the split pass after combine,
try to figure the extended reg was set. If it originated from the T
bit we can replace the sign extension with a reg move, which will be
eliminated. */
operands[2] = sh_try_omit_signzero_extend (operands[1], curr_insn);
if (operands[2] == NULL_RTX)
FAIL;
}
[(set_attr "type" "arith")])
;; FIXME: Fold non-SH2A and SH2A alternatives with "enabled" attribute.
;; See movqi insns.
(define_insn "*extendsi2_compact_mem_disp"
[(set (match_operand:SI 0 "arith_reg_dest" "=z,r")
(sign_extend:SI
(mem:QIHI
(plus:SI
(match_operand:SI 1 "arith_reg_operand" "%r,r")
(match_operand:SI 2 "const_int_operand" ",N")))))]
"TARGET_SH1 && ! TARGET_SH2A
&& sh_legitimate_index_p (mode, operands[2], false, true)"
"@
mov. @(%O2,%1),%0
mov. @%1,%0"
[(set_attr "type" "load")])
(define_insn "*extendsi2_compact_mem_disp"
[(set (match_operand:SI 0 "arith_reg_dest" "=z,r,r")
(sign_extend:SI
(mem:QIHI
(plus:SI
(match_operand:SI 1 "arith_reg_operand" "%r,r,r")
(match_operand:SI 2 "const_int_operand" ",N,")))))]
"TARGET_SH2A && sh_legitimate_index_p (mode, operands[2], true, true)"
"@
mov. @(%O2,%1),%0
mov. @%1,%0
mov. @(%O2,%1),%0"
[(set_attr "type" "load")
(set_attr "length" "2,2,4")])
;; The *_snd patterns will take care of other QImode/HImode addressing
;; modes than displacement addressing. They must be defined _after_ the
;; displacement addressing patterns. Otherwise the displacement addressing
;; patterns will not be picked.
(define_insn "*extendsi2_compact_snd"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(sign_extend:SI
(match_operand:QIHI 1 "movsrc_no_disp_mem_operand" "Snd")))]
"TARGET_SH1"
"mov. %1,%0"
[(set_attr "type" "load")])
(define_insn "*extendqisi2_media"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(sign_extend:SI (match_operand:QI 1 "general_extend_operand" "r,m")))]
"TARGET_SHMEDIA"
"@
#
ld%M1.b %m1, %0"
[(set_attr "type" "arith_media,load_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(sign_extend:SI (match_operand:QI 1 "extend_reg_operand" "")))]
"TARGET_SHMEDIA && reload_completed"
[(set (match_dup 0) (ashift:SI (match_dup 2) (const_int 24)))
(set (match_dup 0) (ashiftrt:SI (match_dup 0) (const_int 24)))]
{
rtx op1 = operands[1];
if (GET_CODE (op1) == TRUNCATE)
op1 = XEXP (op1, 0);
operands[2]
= simplify_gen_subreg (SImode, op1, GET_MODE (op1),
subreg_lowpart_offset (SImode, GET_MODE (op1)));
})
(define_expand "extendqihi2"
[(set (match_operand:HI 0 "arith_reg_dest" "")
(sign_extend:HI (match_operand:QI 1 "arith_reg_operand" "")))]
""
"")
(define_insn "*extendqihi2_compact_reg"
[(set (match_operand:HI 0 "arith_reg_dest" "=r")
(sign_extend:HI (match_operand:QI 1 "arith_reg_operand" "r")))]
"TARGET_SH1"
"exts.b %1,%0"
[(set_attr "type" "arith")])
;; It would seem useful to combine the truncXi patterns into the movXi
;; patterns, but unary operators are ignored when matching constraints,
;; so we need separate patterns.
(define_insn "truncdisi2"
[(set (match_operand:SI 0 "general_movdst_operand" "=r,m,m,f,r,f")
(truncate:SI (match_operand:DI 1 "register_operand" "r,r,f,r,f,f")))]
"TARGET_SHMEDIA"
"@
add.l %1, r63, %0
st%M0.l %m0, %1
fst%M0.s %m0, %T1
fmov.ls %1, %0
fmov.sl %T1, %0
fmov.s %T1, %0"
[(set_attr "type" "arith_media,store_media,fstore_media,fload_media,
fpconv_media,fmove_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "extend")))])
(define_insn "truncdihi2"
[(set (match_operand:HI 0 "general_movdst_operand" "=?r,m")
(truncate:HI (match_operand:DI 1 "register_operand" "r,r")))]
"TARGET_SHMEDIA"
{
static const char* alt[] =
{
"shlli %1,48,%0" "\n"
" shlri %0,48,%0",
"st%M0.w %m0, %1"
};
return alt[which_alternative];
}
[(set_attr "type" "arith_media,store_media")
(set_attr "length" "8,4")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "extend")))])
; N.B. This should agree with LOAD_EXTEND_OP and movqi.
; Because we use zero extension, we can't provide signed QImode compares
; using a simple compare or conditional branch insn.
(define_insn "truncdiqi2"
[(set (match_operand:QI 0 "general_movdst_operand" "=r,m")
(truncate:QI (match_operand:DI 1 "register_operand" "r,r")))]
"TARGET_SHMEDIA"
"@
andi %1, 255, %0
st%M0.b %m0, %1"
[(set_attr "type" "arith_media,store")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "extend")))])
;; -------------------------------------------------------------------------
;; Move instructions
;; -------------------------------------------------------------------------
;; define push and pop so it is easy for sh.c
;; We can't use push and pop on SHcompact because the stack must always
;; be 8-byte aligned.
(define_expand "push"
[(set (mem:SI (pre_dec:SI (reg:SI SP_REG)))
(match_operand:SI 0 "register_operand" "r,l,x"))]
"TARGET_SH1 && ! TARGET_SH5"
"")
(define_expand "pop"
[(set (match_operand:SI 0 "register_operand" "=r,l,x")
(mem:SI (post_inc:SI (reg:SI SP_REG))))]
"TARGET_SH1 && ! TARGET_SH5"
"")
(define_expand "push_e"
[(parallel [(set (mem:SF (pre_dec:SI (reg:SI SP_REG)))
(match_operand:SF 0 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (scratch:SI))])]
"TARGET_SH1 && ! TARGET_SH5"
"")
(define_insn "push_fpul"
[(set (mem:SF (pre_dec:SI (reg:SI SP_REG))) (reg:SF FPUL_REG))]
"TARGET_SH2E && ! TARGET_SH5"
"sts.l fpul,@-r15"
[(set_attr "type" "fstore")
(set_attr "late_fp_use" "yes")
(set_attr "hit_stack" "yes")])
;; DFmode pushes for sh4 require a lot of what is defined for movdf_i4,
;; so use that.
(define_expand "push_4"
[(parallel [(set (mem:DF (pre_dec:SI (reg:SI SP_REG)))
(match_operand:DF 0 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (scratch:SI))])]
"TARGET_SH1 && ! TARGET_SH5"
"")
(define_expand "pop_e"
[(parallel [(set (match_operand:SF 0 "" "")
(mem:SF (post_inc:SI (reg:SI SP_REG))))
(use (reg:SI FPSCR_MODES_REG))
(clobber (scratch:SI))])]
"TARGET_SH1 && ! TARGET_SH5"
"")
(define_insn "pop_fpul"
[(set (reg:SF FPUL_REG) (mem:SF (post_inc:SI (reg:SI SP_REG))))]
"TARGET_SH2E && ! TARGET_SH5"
"lds.l @r15+,fpul"
[(set_attr "type" "load")
(set_attr "hit_stack" "yes")])
(define_expand "pop_4"
[(parallel [(set (match_operand:DF 0 "" "")
(mem:DF (post_inc:SI (reg:SI SP_REG))))
(use (reg:SI FPSCR_MODES_REG))
(clobber (scratch:SI))])]
"TARGET_SH1 && ! TARGET_SH5"
"")
(define_expand "push_fpscr"
[(const_int 0)]
"TARGET_SH2E"
{
add_reg_note (
emit_insn (
gen_sts_fpscr (
gen_frame_mem (SImode, gen_rtx_PRE_DEC (Pmode, stack_pointer_rtx)))),
REG_INC, stack_pointer_rtx);
DONE;
})
(define_expand "pop_fpscr"
[(const_int 0)]
"TARGET_SH2E"
{
add_reg_note (
emit_insn (
gen_lds_fpscr (
gen_frame_mem (SImode, gen_rtx_POST_INC (Pmode, stack_pointer_rtx)))),
REG_INC, stack_pointer_rtx);
DONE;
})
;; The clrt and sett patterns can happen as the result of optimization and
;; insn expansion.
;; Comparisons might get simplified to a move of zero or 1 into the T reg.
;; In this case they might not disappear completely, because the T reg is
;; a fixed hard reg.
;; When DImode operations that use the T reg as carry/borrow are split into
;; individual SImode operations, the T reg is usually cleared before the
;; first SImode insn.
(define_insn "clrt"
[(set (reg:SI T_REG) (const_int 0))]
"TARGET_SH1"
"clrt"
[(set_attr "type" "mt_group")])
(define_insn "sett"
[(set (reg:SI T_REG) (const_int 1))]
"TARGET_SH1"
"sett"
[(set_attr "type" "mt_group")])
;; Use the combine pass to transform sequences such as
;; mov r5,r0
;; add #1,r0
;; shll2 r0
;; mov.l @(r0,r4),r0
;; into
;; shll2 r5
;; add r4,r5
;; mov.l @(4,r5),r0
;;
;; See also PR 39423.
;; Notice that these patterns have a T_REG clobber, because the shift
;; sequence that will be split out might clobber the T_REG. Ideally, the
;; clobber would be added conditionally, depending on the result of
;; sh_ashlsi_clobbers_t_reg_p. When splitting out the shifts we must go
;; through the ashlsi3 expander in order to get the right shift insn --
;; a T_REG clobbering or non-clobbering shift sequence or dynamic shift.
;; FIXME: Combine never tries this kind of patterns for DImode.
(define_insn_and_split "*movsi_index_disp_load"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(match_operand:SI 1 "mem_index_disp_operand" "m"))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(set (match_dup 6) (plus:SI (match_dup 5) (match_dup 3)))
(set (match_dup 0) (match_dup 7))]
{
rtx mem = operands[1];
rtx plus0_rtx = XEXP (mem, 0);
rtx plus1_rtx = XEXP (plus0_rtx, 0);
rtx mult_rtx = XEXP (plus1_rtx, 0);
operands[1] = XEXP (mult_rtx, 0);
operands[2] = GEN_INT (exact_log2 (INTVAL (XEXP (mult_rtx, 1))));
operands[3] = XEXP (plus1_rtx, 1);
operands[4] = XEXP (plus0_rtx, 1);
operands[5] = gen_reg_rtx (SImode);
operands[6] = gen_reg_rtx (SImode);
operands[7] =
replace_equiv_address (mem,
gen_rtx_PLUS (SImode, operands[6], operands[4]));
emit_insn (gen_ashlsi3 (operands[5], operands[1], operands[2]));
})
(define_insn_and_split "*movhi_index_disp_load"
[(set (match_operand:SI 0 "arith_reg_dest")
(SZ_EXTEND:SI (match_operand:HI 1 "mem_index_disp_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
rtx mem = operands[1];
rtx plus0_rtx = XEXP (mem, 0);
rtx plus1_rtx = XEXP (plus0_rtx, 0);
rtx mult_rtx = XEXP (plus1_rtx, 0);
rtx op_1 = XEXP (mult_rtx, 0);
rtx op_2 = GEN_INT (exact_log2 (INTVAL (XEXP (mult_rtx, 1))));
rtx op_3 = XEXP (plus1_rtx, 1);
rtx op_4 = XEXP (plus0_rtx, 1);
rtx op_5 = gen_reg_rtx (SImode);
rtx op_6 = gen_reg_rtx (SImode);
rtx op_7 = replace_equiv_address (mem, gen_rtx_PLUS (SImode, op_6, op_4));
emit_insn (gen_ashlsi3 (op_5, op_1, op_2));
emit_insn (gen_addsi3 (op_6, op_5, op_3));
if ( == SIGN_EXTEND)
{
emit_insn (gen_extendhisi2 (operands[0], op_7));
DONE;
}
else if ( == ZERO_EXTEND)
{
/* On SH2A the movu.w insn can be used for zero extending loads. */
if (TARGET_SH2A)
emit_insn (gen_zero_extendhisi2 (operands[0], op_7));
else
{
emit_insn (gen_extendhisi2 (operands[0], op_7));
emit_insn (gen_zero_extendhisi2 (operands[0],
gen_lowpart (HImode, operands[0])));
}
DONE;
}
else
FAIL;
})
(define_insn_and_split "*mov_index_disp_store"
[(set (match_operand:HISI 0 "mem_index_disp_operand" "=m")
(match_operand:HISI 1 "arith_reg_operand" "r"))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(set (match_dup 6) (plus:SI (match_dup 5) (match_dup 3)))
(set (match_dup 7) (match_dup 1))]
{
rtx mem = operands[0];
rtx plus0_rtx = XEXP (mem, 0);
rtx plus1_rtx = XEXP (plus0_rtx, 0);
rtx mult_rtx = XEXP (plus1_rtx, 0);
operands[0] = XEXP (mult_rtx, 0);
operands[2] = GEN_INT (exact_log2 (INTVAL (XEXP (mult_rtx, 1))));
operands[3] = XEXP (plus1_rtx, 1);
operands[4] = XEXP (plus0_rtx, 1);
operands[5] = gen_reg_rtx (SImode);
operands[6] = gen_reg_rtx (SImode);
operands[7] =
replace_equiv_address (mem,
gen_rtx_PLUS (SImode, operands[6], operands[4]));
emit_insn (gen_ashlsi3 (operands[5], operands[0], operands[2]));
})
;; t/r must come after r/r, lest reload will try to reload stuff like
;; (set (subreg:SI (mem:QI (plus:SI (reg:SI SP_REG) (const_int 12)) 0) 0)
;; (made from (set (subreg:SI (reg:QI ###) 0) ) into T.
(define_insn "movsi_i"
[(set (match_operand:SI 0 "general_movdst_operand"
"=r,r,r,r,r,r,m,<,<,x,l,x,l,r")
(match_operand:SI 1 "general_movsrc_operand"
"Q,r,I08,mr,x,l,r,x,l,r,r,>,>,i"))]
"TARGET_SH1
&& ! TARGET_SH2E
&& ! TARGET_SH2A
&& (register_operand (operands[0], SImode)
|| register_operand (operands[1], SImode))"
"@
mov.l %1,%0
mov %1,%0
mov %1,%0
mov.l %1,%0
sts %1,%0
sts %1,%0
mov.l %1,%0
sts.l %1,%0
sts.l %1,%0
lds %1,%0
lds %1,%0
lds.l %1,%0
lds.l %1,%0
fake %1,%0"
[(set_attr "type" "pcload_si,move,movi8,load_si,mac_gp,prget,store,mac_mem,
pstore,gp_mac,prset,mem_mac,pload,pcload_si")
(set_attr "length" "*,*,*,*,*,*,*,*,*,*,*,*,*,*")])
;; t/r must come after r/r, lest reload will try to reload stuff like
;; (subreg:SI (reg:SF FR14_REG) 0) into T (compiling stdlib/strtod.c -m3e -O2)
;; ??? This allows moves from macl to fpul to be recognized, but these moves
;; will require a reload.
;; ??? We can't include f/f because we need the proper FPSCR setting when
;; TARGET_FMOVD is in effect, and mode switching is done before reload.
(define_insn "movsi_ie"
[(set (match_operand:SI 0 "general_movdst_operand"
"=r,r,r,r,r,r,r,r,m,<,<,x,l,x,l,y,<,r,y,r,*f,y,*f,y")
(match_operand:SI 1 "general_movsrc_operand"
"Q,r,I08,I20,I28,mr,x,l,r,x,l,r,r,>,>,>,y,i,r,y,y,*f,*f,y"))]
"(TARGET_SH2E || TARGET_SH2A)
&& ((register_operand (operands[0], SImode)
&& !fpscr_operand (operands[0], SImode))
|| (register_operand (operands[1], SImode)
&& !fpscr_operand (operands[1], SImode)))"
"@
mov.l %1,%0
mov %1,%0
mov %1,%0
movi20 %1,%0
movi20s %1,%0
mov.l %1,%0
sts %1,%0
sts %1,%0
mov.l %1,%0
sts.l %1,%0
sts.l %1,%0
lds %1,%0
lds %1,%0
lds.l %1,%0
lds.l %1,%0
lds.l %1,%0
sts.l %1,%0
fake %1,%0
lds %1,%0
sts %1,%0
fsts fpul,%0
flds %1,fpul
fmov %1,%0
! move optimized away"
[(set_attr "type" "pcload_si,move,movi8,move,move,load_si,mac_gp,prget,store,
mac_mem,pstore,gp_mac,prset,mem_mac,pload,load,fstore,
pcload_si,gp_fpul,fpul_gp,fmove,fmove,fmove,nil")
(set_attr "late_fp_use" "*,*,*,*,*,*,*,*,*,*,*,*,*,*,*,*,yes,*,*,yes,*,*,*,*")
(set_attr_alternative "length"
[(const_int 2)
(const_int 2)
(const_int 2)
(const_int 4)
(const_int 4)
(if_then_else
(match_test "TARGET_SH2A")
(const_int 4) (const_int 2))
(const_int 2)
(const_int 2)
(if_then_else
(match_test "TARGET_SH2A")
(const_int 4) (const_int 2))
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 0)])])
(define_insn "movsi_i_lowpart"
[(set (strict_low_part
(match_operand:SI 0 "general_movdst_operand" "+r,r,r,r,r,r,m,r"))
(match_operand:SI 1 "general_movsrc_operand" "Q,r,I08,mr,x,l,r,i"))]
"TARGET_SH1
&& (register_operand (operands[0], SImode)
|| register_operand (operands[1], SImode))"
"@
mov.l %1,%0
mov %1,%0
mov %1,%0
mov.l %1,%0
sts %1,%0
sts %1,%0
mov.l %1,%0
fake %1,%0"
[(set_attr "type" "pcload,move,arith,load,mac_gp,prget,store,pcload")])
(define_insn_and_split "load_ra"
[(set (match_operand:SI 0 "general_movdst_operand" "")
(unspec:SI [(match_operand:SI 1 "register_operand" "")] UNSPEC_RA))]
"TARGET_SH1"
"#"
"&& ! currently_expanding_to_rtl"
[(set (match_dup 0) (match_dup 1))]
{
if (TARGET_SHCOMPACT && crtl->saves_all_registers)
operands[1] = gen_frame_mem (SImode, return_address_pointer_rtx);
})
;; The '?'s in the following constraints may not reflect the time taken
;; to perform the move. They are there to discourage the use of floating-
;; point registers for storing integer values.
(define_insn "*movsi_media"
[(set (match_operand:SI 0 "general_movdst_operand"
"=r,r,r,r,m,f?,m,f?,r,f?,*b,r,b")
(match_operand:SI 1 "general_movsrc_operand"
"r,I16Css,nCpg,m,rZ,m,f?,rZ,f?,f?,r,*b,Csy"))]
"TARGET_SHMEDIA_FPU
&& (register_operand (operands[0], SImode)
|| sh_register_operand (operands[1], SImode)
|| GET_CODE (operands[1]) == TRUNCATE)"
"@
add.l %1, r63, %0
movi %1, %0
#
ld%M1.l %m1, %0
st%M0.l %m0, %N1
fld%M1.s %m1, %0
fst%M0.s %m0, %1
fmov.ls %N1, %0
fmov.sl %1, %0
fmov.s %1, %0
ptabs %1, %0
gettr %1, %0
pt %1, %0"
[(set_attr "type" "arith_media,arith_media,*,load_media,store_media,
fload_media,fstore_media,fload_media,fpconv_media,
fmove_media,ptabs_media,gettr_media,pt_media")
(set_attr "length" "4,4,8,4,4,4,4,4,4,4,4,4,12")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_insn "*movsi_media_nofpu"
[(set (match_operand:SI 0 "general_movdst_operand"
"=r,r,r,r,m,*b,r,*b")
(match_operand:SI 1 "general_movsrc_operand"
"r,I16Css,nCpg,m,rZ,r,*b,Csy"))]
"TARGET_SHMEDIA
&& (register_operand (operands[0], SImode)
|| sh_register_operand (operands[1], SImode)
|| GET_CODE (operands[1]) == TRUNCATE)"
"@
add.l %1, r63, %0
movi %1, %0
#
ld%M1.l %m1, %0
st%M0.l %m0, %N1
ptabs %1, %0
gettr %1, %0
pt %1, %0"
[(set_attr "type" "arith_media,arith_media,*,load_media,store_media,
ptabs_media,gettr_media,pt_media")
(set_attr "length" "4,4,8,4,4,4,4,12")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_expand "movsi_const"
[(set (match_operand:SI 0 "arith_reg_operand" "=r")
(const:SI (unspec:SI [(match_operand:DI 1 "immediate_operand" "s")
(const_int 16)] UNSPEC_EXTRACT_S16)))
(set (match_dup 0)
(ior:SI (ashift:SI (match_dup 0) (const_int 16))
(const:SI (unspec:SI [(match_dup 1)
(const_int 0)] UNSPEC_EXTRACT_U16))))]
"TARGET_SHMEDIA && reload_completed
&& MOVI_SHORI_BASE_OPERAND_P (operands[1])"
{
if (GET_CODE (operands[1]) == LABEL_REF
&& GET_CODE (XEXP (operands[1], 0)) == CODE_LABEL)
LABEL_NUSES (XEXP (operands[1], 0)) += 2;
else if (GOTOFF_P (operands[1]))
{
rtx unspec = XEXP (operands[1], 0);
if (! UNSPEC_GOTOFF_P (unspec))
{
unspec = XEXP (unspec, 0);
if (! UNSPEC_GOTOFF_P (unspec))
abort ();
}
if (GET_CODE (XVECEXP (unspec , 0, 0)) == LABEL_REF
&& (GET_CODE (XEXP (XVECEXP (unspec, 0, 0), 0)) == CODE_LABEL))
LABEL_NUSES (XEXP (XVECEXP (unspec, 0, 0), 0)) += 2;
}
})
(define_expand "movsi_const_16bit"
[(set (match_operand:SI 0 "arith_reg_operand" "=r")
(const:SI (unspec:SI [(match_operand:DI 1 "immediate_operand" "s")
(const_int 0)] UNSPEC_EXTRACT_S16)))]
"TARGET_SHMEDIA && flag_pic && reload_completed
&& GET_CODE (operands[1]) == SYMBOL_REF"
"")
(define_split
[(set (match_operand:SI 0 "arith_reg_dest" "")
(match_operand:SI 1 "immediate_operand" ""))]
"TARGET_SHMEDIA && reload_completed
&& MOVI_SHORI_BASE_OPERAND_P (operands[1])"
[(const_int 0)]
{
rtx insn = emit_insn (gen_movsi_const (operands[0], operands[1]));
set_unique_reg_note (insn, REG_EQUAL, copy_rtx (operands[1]));
DONE;
})
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(match_operand:SI 1 "immediate_operand" ""))]
"TARGET_SHMEDIA && reload_completed
&& ((CONST_INT_P (operands[1])
&& ! satisfies_constraint_I16 (operands[1]))
|| GET_CODE (operands[1]) == CONST_DOUBLE)"
[(set (subreg:DI (match_dup 0) 0) (match_dup 1))])
(define_expand "movsi"
[(set (match_operand:SI 0 "general_movdst_operand" "")
(match_operand:SI 1 "general_movsrc_operand" ""))]
""
{
prepare_move_operands (operands, SImode);
})
(define_expand "ic_invalidate_line"
[(parallel [(unspec_volatile [(match_operand:SI 0 "register_operand" "+r")
(match_dup 1)] UNSPEC_ICACHE)
(clobber (scratch:SI))])]
"TARGET_HARD_SH4 || TARGET_SH5"
{
if (TARGET_SHMEDIA)
{
emit_insn (gen_ic_invalidate_line_media (operands[0]));
DONE;
}
else if (TARGET_SHCOMPACT)
{
operands[1] = function_symbol (NULL, "__ic_invalidate", SFUNC_STATIC);
operands[1] = force_reg (Pmode, operands[1]);
emit_insn (gen_ic_invalidate_line_compact (operands[0], operands[1]));
DONE;
}
else if (TARGET_SH4A || TARGET_SH4_300)
{
emit_insn (gen_ic_invalidate_line_sh4a (operands[0]));
DONE;
}
operands[0] = force_reg (Pmode, operands[0]);
operands[1] = force_reg (Pmode, GEN_INT (trunc_int_for_mode (0xf0000008,
Pmode)));
})
;; The address %0 is assumed to be 4-aligned at least. Thus, by ORing
;; 0xf0000008, we get the low-oder bits *1*00 (binary), which fits
;; the requirement *1*00 for associative address writes. The alignment of
;; %0 implies that its least significant bit is cleared,
;; thus we clear the V bit of a matching entry if there is one.
(define_insn "ic_invalidate_line_i"
[(unspec_volatile [(match_operand:SI 0 "register_operand" "r")
(match_operand:SI 1 "register_operand" "r")]
UNSPEC_ICACHE)
(clobber (match_scratch:SI 2 "=&r"))]
"TARGET_HARD_SH4"
{
return "ocbwb @%0" "\n"
" extu.w %0,%2" "\n"
" or %1,%2" "\n"
" mov.l %0,@%2";
}
[(set_attr "length" "8")
(set_attr "type" "cwb")])
(define_insn "ic_invalidate_line_sh4a"
[(unspec_volatile [(match_operand:SI 0 "register_operand" "r")]
UNSPEC_ICACHE)]
"TARGET_SH4A || TARGET_SH4_300"
{
return "ocbwb @%0" "\n"
" synco" "\n"
" icbi @%0";
}
[(set_attr "length" "16") ;; FIXME: Why 16 and not 6? Looks like typo.
(set_attr "type" "cwb")])
;; ??? could make arg 0 an offsettable memory operand to allow to save
;; an add in the code that calculates the address.
(define_insn "ic_invalidate_line_media"
[(unspec_volatile [(match_operand 0 "any_register_operand" "r")]
UNSPEC_ICACHE)]
"TARGET_SHMEDIA"
{
return "ocbwb %0,0" "\n"
" synco" "\n"
" icbi %0,0" "\n"
" synci";
}
[(set_attr "length" "16")
(set_attr "type" "invalidate_line_media")])
(define_insn "ic_invalidate_line_compact"
[(unspec_volatile [(match_operand:SI 0 "register_operand" "z")
(match_operand:SI 1 "register_operand" "r")]
UNSPEC_ICACHE)
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_expand "initialize_trampoline"
[(match_operand:SI 0 "" "")
(match_operand:SI 1 "" "")
(match_operand:SI 2 "" "")]
"TARGET_SHCOMPACT"
{
rtx sfun, tramp;
tramp = force_reg (Pmode, operands[0]);
sfun = force_reg (Pmode, function_symbol (NULL, "__init_trampoline",
SFUNC_STATIC));
emit_move_insn (gen_rtx_REG (SImode, R2_REG), operands[1]);
emit_move_insn (gen_rtx_REG (SImode, R3_REG), operands[2]);
emit_insn (gen_initialize_trampoline_compact (tramp, sfun));
DONE;
})
(define_insn "initialize_trampoline_compact"
[(unspec_volatile [(match_operand:SI 0 "register_operand" "z")
(match_operand:SI 1 "register_operand" "r")
(reg:SI R2_REG) (reg:SI R3_REG)]
UNSPEC_INIT_TRAMP)
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT"
"jsr @%1%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_expand "mov"
[(set (match_operand:QIHI 0 "general_movdst_operand")
(match_operand:QIHI 1 "general_movsrc_operand"))]
""
{
if (can_create_pseudo_p () && CONST_INT_P (operands[1])
&& REG_P (operands[0]) && REGNO (operands[0]) != R0_REG)
{
rtx reg = gen_reg_rtx(SImode);
emit_move_insn (reg, operands[1]);
operands[1] = gen_lowpart (mode, reg);
}
prepare_move_operands (operands, mode);
})
;; Specifying the displacement addressing load / store patterns separately
;; before the generic movqi / movhi pattern allows controlling the order
;; in which load / store insns are selected in a more fine grained way.
;; FIXME: The non-SH2A and SH2A variants should be combined by adding
;; "enabled" attribute as it is done in other targets.
(define_insn "*mov_store_mem_disp04"
[(set (mem:QIHI
(plus:SI (match_operand:SI 0 "arith_reg_operand" "%r,r")
(match_operand:SI 1 "const_int_operand" ",N")))
(match_operand:QIHI 2 "arith_reg_operand" "z,r"))]
"TARGET_SH1 && sh_legitimate_index_p (mode, operands[1], false, true)"
"@
mov. %2,@(%O1,%0)
mov. %2,@%0"
[(set_attr "type" "store")])
(define_insn "*mov_store_mem_disp12"
[(set (mem:QIHI
(plus:SI (match_operand:SI 0 "arith_reg_operand" "%r")
(match_operand:SI 1 "const_int_operand" "")))
(match_operand:QIHI 2 "arith_reg_operand" "r"))]
"TARGET_SH2A && sh_legitimate_index_p (mode, operands[1], true, true)"
"mov. %2,@(%O1,%0)"
[(set_attr "type" "store")
(set_attr "length" "4")])
(define_insn "*mov_load_mem_disp04"
[(set (match_operand:QIHI 0 "arith_reg_dest" "=z,r")
(mem:QIHI
(plus:SI (match_operand:SI 1 "arith_reg_operand" "%r,r")
(match_operand:SI 2 "const_int_operand" ",N"))))]
"TARGET_SH1 && ! TARGET_SH2A
&& sh_legitimate_index_p (mode, operands[2], false, true)"
"@
mov. @(%O2,%1),%0
mov. @%1,%0"
[(set_attr "type" "load")])
(define_insn "*mov_load_mem_disp12"
[(set (match_operand:QIHI 0 "arith_reg_dest" "=z,r,r")
(mem:QIHI
(plus:SI
(match_operand:SI 1 "arith_reg_operand" "%r,r,r")
(match_operand:SI 2 "const_int_operand" ",N,"))))]
"TARGET_SH2A && sh_legitimate_index_p (mode, operands[2], true, true)"
"@
mov. @(%O2,%1),%0
mov. @%1,%0
mov. @(%O2,%1),%0"
[(set_attr "type" "load")
(set_attr "length" "2,2,4")])
;; The order of the constraint alternatives is important here.
;; Q/r has to come first, otherwise PC relative loads might wrongly get
;; placed into delay slots. Since there is no QImode PC relative load, the
;; Q constraint and general_movsrc_operand will reject it for QImode.
;; The Snd alternatives should come before Sdd in order to avoid a preference
;; of using r0 als the register operand for addressing modes other than
;; displacement addressing.
;; The Sdd alternatives allow only r0 as register operand, even though on
;; SH2A any register could be allowed by switching to a 32 bit insn.
;; Generally sticking to the r0 is preferrable, since it generates smaller
;; code. Obvious r0 reloads can then be eliminated with a peephole on SH2A.
(define_insn "*mov"
[(set (match_operand:QIHI 0 "general_movdst_operand"
"=r,r,r,Snd,r, Sdd,z, r,l")
(match_operand:QIHI 1 "general_movsrc_operand"
"Q,r,i,r, Snd,z, Sdd,l,r"))]
"TARGET_SH1
&& (arith_reg_operand (operands[0], mode)
|| arith_reg_operand (operands[1], mode))"
"@
mov. %1,%0
mov %1,%0
mov %1,%0
mov. %1,%0
mov. %1,%0
mov. %1,%0
mov. %1,%0
sts %1,%0
lds %1,%0"
[(set_attr "type" "pcload,move,movi8,store,load,store,load,prget,prset")
(set (attr "length")
(cond [(and (match_operand 0 "displacement_mem_operand")
(not (match_operand 0 "short_displacement_mem_operand")))
(const_int 4)
(and (match_operand 1 "displacement_mem_operand")
(not (match_operand 1 "short_displacement_mem_operand")))
(const_int 4)]
(const_int 2)))])
(define_insn "*movqi_media"
[(set (match_operand:QI 0 "general_movdst_operand" "=r,r,r,m")
(match_operand:QI 1 "general_movsrc_operand" "r,I16Css,m,rZ"))]
"TARGET_SHMEDIA
&& (arith_reg_operand (operands[0], QImode)
|| extend_reg_or_0_operand (operands[1], QImode))"
"@
add.l %1, r63, %0
movi %1, %0
ld%M1.ub %m1, %0
st%M0.b %m0, %N1"
[(set_attr "type" "arith_media,arith_media,load_media,store_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_expand "reload_inqi"
[(set (match_operand:SI 2 "" "=&r")
(match_operand:QI 1 "inqhi_operand" ""))
(set (match_operand:QI 0 "arith_reg_operand" "=r")
(truncate:QI (match_dup 3)))]
"TARGET_SHMEDIA"
{
rtx inner = XEXP (operands[1], 0);
int regno = REGNO (inner);
regno += HARD_REGNO_NREGS (regno, GET_MODE (inner)) - 1;
operands[1] = gen_rtx_REG (SImode, regno);
operands[3] = gen_rtx_REG (DImode, REGNO (operands[2]));
})
(define_insn "*movhi_media"
[(set (match_operand:HI 0 "general_movdst_operand" "=r,r,r,r,m")
(match_operand:HI 1 "general_movsrc_operand" "r,I16Css,n,m,rZ"))]
"TARGET_SHMEDIA
&& (arith_reg_operand (operands[0], HImode)
|| arith_reg_or_0_operand (operands[1], HImode))"
"@
add.l %1, r63, %0
movi %1, %0
#
ld%M1.w %m1, %0
st%M0.w %m0, %N1"
[(set_attr "type" "arith_media,arith_media,*,load_media,store_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_split
[(set (match_operand:HI 0 "register_operand" "")
(match_operand:HI 1 "immediate_operand" ""))]
"TARGET_SHMEDIA && reload_completed
&& ! satisfies_constraint_I16 (operands[1])"
[(set (subreg:DI (match_dup 0) 0) (match_dup 1))])
(define_expand "reload_inhi"
[(set (match_operand:SI 2 "" "=&r")
(match_operand:HI 1 "inqhi_operand" ""))
(set (match_operand:HI 0 "arith_reg_operand" "=r")
(truncate:HI (match_dup 3)))]
"TARGET_SHMEDIA"
{
rtx inner = XEXP (operands[1], 0);
int regno = REGNO (inner);
regno += HARD_REGNO_NREGS (regno, GET_MODE (inner)) - 1;
operands[1] = gen_rtx_REG (SImode, regno);
operands[3] = gen_rtx_REG (DImode, REGNO (operands[2]));
})
;; x/r can be created by inlining/cse, e.g. for execute/961213-1.c
;; compiled with -m2 -ml -O3 -funroll-loops
(define_insn "*movdi_i"
[(set (match_operand:DI 0 "general_movdst_operand" "=r,r,r,m,r,r,r,*!x")
(match_operand:DI 1 "general_movsrc_operand" "Q,r,m,r,I08,i,x,r"))]
"TARGET_SH1
&& (arith_reg_operand (operands[0], DImode)
|| arith_reg_operand (operands[1], DImode))"
{
return output_movedouble (insn, operands, DImode);
}
[(set_attr "length" "4")
(set_attr "type" "pcload,move,load,store,move,pcload,move,move")])
;; If the output is a register and the input is memory or a register, we have
;; to be careful and see which word needs to be loaded first.
(define_split
[(set (match_operand:DI 0 "general_movdst_operand" "")
(match_operand:DI 1 "general_movsrc_operand" ""))]
"TARGET_SH1 && reload_completed"
[(set (match_dup 2) (match_dup 3))
(set (match_dup 4) (match_dup 5))]
{
int regno;
if ((MEM_P (operands[0])
&& GET_CODE (XEXP (operands[0], 0)) == PRE_DEC)
|| (MEM_P (operands[1])
&& GET_CODE (XEXP (operands[1], 0)) == POST_INC))
FAIL;
switch (GET_CODE (operands[0]))
{
case REG:
regno = REGNO (operands[0]);
break;
case SUBREG:
regno = subreg_regno (operands[0]);
break;
case MEM:
regno = -1;
break;
default:
gcc_unreachable ();
}
if (regno == -1
|| ! refers_to_regno_p (regno, regno + 1, operands[1], 0))
{
operands[2] = operand_subword (operands[0], 0, 0, DImode);
operands[3] = operand_subword (operands[1], 0, 0, DImode);
operands[4] = operand_subword (operands[0], 1, 0, DImode);
operands[5] = operand_subword (operands[1], 1, 0, DImode);
}
else
{
operands[2] = operand_subword (operands[0], 1, 0, DImode);
operands[3] = operand_subword (operands[1], 1, 0, DImode);
operands[4] = operand_subword (operands[0], 0, 0, DImode);
operands[5] = operand_subword (operands[1], 0, 0, DImode);
}
if (operands[2] == 0 || operands[3] == 0
|| operands[4] == 0 || operands[5] == 0)
FAIL;
})
;; The '?'s in the following constraints may not reflect the time taken
;; to perform the move. They are there to discourage the use of floating-
;; point registers for storing integer values.
(define_insn "*movdi_media"
[(set (match_operand:DI 0 "general_movdst_operand"
"=r,r,r,rl,m,f?,m,f?,r,f?,*b,r,*b")
(match_operand:DI 1 "general_movsrc_operand"
"r,I16Css,nCpgF,m,rlZ,m,f?,rZ,f?,f?,r,*b,Csy"))]
"TARGET_SHMEDIA_FPU
&& (register_operand (operands[0], DImode)
|| sh_register_operand (operands[1], DImode))"
"@
add %1, r63, %0
movi %1, %0
#
ld%M1.q %m1, %0
st%M0.q %m0, %N1
fld%M1.d %m1, %0
fst%M0.d %m0, %1
fmov.qd %N1, %0
fmov.dq %1, %0
fmov.d %1, %0
ptabs %1, %0
gettr %1, %0
pt %1, %0"
[(set_attr "type" "arith_media,arith_media,*,load_media,store_media,
fload_media,fstore_media,fload_media,dfpconv_media,
fmove_media,ptabs_media,gettr_media,pt_media")
(set_attr "length" "4,4,16,4,4,4,4,4,4,4,4,4,*")])
(define_insn "*movdi_media_nofpu"
[(set (match_operand:DI 0 "general_movdst_operand" "=r,r,r,rl,m,*b,r,*b");
(match_operand:DI 1 "general_movsrc_operand" "r,I16Css,nCpgF,m,rlZ,r,*b,Csy"))]
"TARGET_SHMEDIA
&& (register_operand (operands[0], DImode)
|| sh_register_operand (operands[1], DImode))"
"@
add %1, r63, %0
movi %1, %0
#
ld%M1.q %m1, %0
st%M0.q %m0, %N1
ptabs %1, %0
gettr %1, %0
pt %1, %0"
[(set_attr "type" "arith_media,arith_media,*,load_media,store_media,
ptabs_media,gettr_media,pt_media")
(set_attr "length" "4,4,16,4,4,4,4,*")])
(define_insn "*movdi_media_I16"
[(set (match_operand:DI 0 "ext_dest_operand" "=r")
(match_operand:DI 1 "const_int_operand" "I16"))]
"TARGET_SHMEDIA && reload_completed"
"movi %1, %0"
[(set_attr "type" "arith_media")
(set_attr "length" "4")])
(define_split
[(set (match_operand:DI 0 "arith_reg_dest" "")
(match_operand:DI 1 "immediate_operand" ""))]
"TARGET_SHMEDIA && reload_completed
&& MOVI_SHORI_BASE_OPERAND_P (operands[1])"
[(set (match_dup 0) (match_dup 1))]
{
rtx insn;
if (TARGET_SHMEDIA64)
insn = emit_insn (gen_movdi_const (operands[0], operands[1]));
else
insn = emit_insn (gen_movdi_const_32bit (operands[0], operands[1]));
set_unique_reg_note (insn, REG_EQUAL, copy_rtx (operands[1]));
DONE;
})
(define_expand "movdi_const"
[(set (match_operand:DI 0 "arith_reg_operand" "=r")
(const:DI (unspec:DI [(match_operand:DI 1 "immediate_operand" "s")
(const_int 48)] UNSPEC_EXTRACT_S16)))
(set (match_dup 0)
(ior:DI (ashift:DI (match_dup 0) (const_int 16))
(const:DI (unspec:DI [(match_dup 1)
(const_int 32)] UNSPEC_EXTRACT_U16))))
(set (match_dup 0)
(ior:DI (ashift:DI (match_dup 0) (const_int 16))
(const:DI (unspec:DI [(match_dup 1)
(const_int 16)] UNSPEC_EXTRACT_U16))))
(set (match_dup 0)
(ior:DI (ashift:DI (match_dup 0) (const_int 16))
(const:DI (unspec:DI [(match_dup 1)
(const_int 0)] UNSPEC_EXTRACT_U16))))]
"TARGET_SHMEDIA64 && reload_completed
&& MOVI_SHORI_BASE_OPERAND_P (operands[1])"
{
sh_mark_label (operands[1], 4);
})
(define_expand "movdi_const_32bit"
[(set (match_operand:DI 0 "arith_reg_operand" "=r")
(const:DI (unspec:DI [(match_operand:DI 1 "immediate_operand" "s")
(const_int 16)] UNSPEC_EXTRACT_S16)))
(set (match_dup 0)
(ior:DI (ashift:DI (match_dup 0) (const_int 16))
(const:DI (unspec:DI [(match_dup 1)
(const_int 0)] UNSPEC_EXTRACT_U16))))]
"TARGET_SHMEDIA32 && reload_completed
&& MOVI_SHORI_BASE_OPERAND_P (operands[1])"
{
sh_mark_label (operands[1], 2);
})
(define_expand "movdi_const_16bit"
[(set (match_operand:DI 0 "arith_reg_operand" "=r")
(const:DI (unspec:DI [(match_operand:DI 1 "immediate_operand" "s")
(const_int 0)] UNSPEC_EXTRACT_S16)))]
"TARGET_SHMEDIA && flag_pic && reload_completed
&& GET_CODE (operands[1]) == SYMBOL_REF"
"")
(define_split
[(set (match_operand:DI 0 "ext_dest_operand" "")
(match_operand:DI 1 "immediate_operand" ""))]
"TARGET_SHMEDIA && reload_completed
&& CONST_INT_P (operands[1])
&& ! satisfies_constraint_I16 (operands[1])"
[(set (match_dup 0) (match_dup 2))
(match_dup 1)]
{
unsigned HOST_WIDE_INT val = INTVAL (operands[1]);
unsigned HOST_WIDE_INT low = val;
unsigned HOST_WIDE_INT high = val;
unsigned HOST_WIDE_INT sign;
unsigned HOST_WIDE_INT val2 = val ^ (val-1);
/* Zero-extend the 16 least-significant bits. */
low &= 0xffff;
/* Arithmetic shift right the word by 16 bits. */
high >>= 16;
if (GET_CODE (operands[0]) == SUBREG
&& GET_MODE (SUBREG_REG (operands[0])) == SImode)
{
high &= 0xffff;
high ^= 0x8000;
high -= 0x8000;
}
else
{
sign = 1;
sign <<= (HOST_BITS_PER_WIDE_INT - 16 - 1);
high ^= sign;
high -= sign;
}
do
{
/* If we can't generate the constant with a two-insn movi / shori
sequence, try some other strategies. */
if (! CONST_OK_FOR_I16 (high))
{
/* Try constant load / left shift. We know VAL != 0. */
val2 = val ^ (val-1);
if (val2 > 0x1ffff)
{
int trailing_zeroes = exact_log2 ((val2 >> 16) + 1) + 15;
if (CONST_OK_FOR_I16 (val >> trailing_zeroes)
|| (! CONST_OK_FOR_I16 (high >> 16)
&& CONST_OK_FOR_I16 (val >> (trailing_zeroes + 16))))
{
val2 = (HOST_WIDE_INT) val >> trailing_zeroes;
operands[1] = gen_ashldi3_media (operands[0], operands[0],
GEN_INT (trailing_zeroes));
break;
}
}
/* Try constant load / right shift. */
val2 = (val >> 15) + 1;
if (val2 == (val2 & -val2))
{
int shift = 49 - exact_log2 (val2);
val2 = trunc_int_for_mode (val << shift, DImode);
if (CONST_OK_FOR_I16 (val2))
{
operands[1] = gen_lshrdi3_media (operands[0], operands[0],
GEN_INT (shift));
break;
}
}
/* Try mperm.w . */
val2 = val & 0xffff;
if ((val >> 16 & 0xffff) == val2
&& (val >> 32 & 0xffff) == val2
&& (val >> 48 & 0xffff) == val2)
{
val2 = (HOST_WIDE_INT) val >> 48;
operands[1] = gen_rtx_REG (V4HImode, true_regnum (operands[0]));
operands[1] = gen_mperm_w0 (operands[1], operands[1]);
break;
}
/* Try movi / mshflo.l */
val2 = (HOST_WIDE_INT) val >> 32;
if (val2 == ((unsigned HOST_WIDE_INT)
trunc_int_for_mode (val, SImode)))
{
operands[1] = gen_mshflo_l_di (operands[0], operands[0],
operands[0]);
break;
}
/* Try movi / mshflo.l w/ r63. */
val2 = val + ((HOST_WIDE_INT) -1 << 32);
if ((HOST_WIDE_INT) val2 < 0 && CONST_OK_FOR_I16 (val2))
{
operands[1] = gen_mshflo_l_di (operands[0], operands[0],
const0_rtx);
break;
}
}
val2 = high;
operands[1] = gen_shori_media (operands[0], operands[0], GEN_INT (low));
}
while (0);
operands[2] = GEN_INT (val2);
})
(define_split
[(set (match_operand:DI 0 "ext_dest_operand" "")
(match_operand:DI 1 "immediate_operand" ""))]
"TARGET_SHMEDIA && reload_completed
&& GET_CODE (operands[1]) == CONST_DOUBLE"
[(set (match_dup 0) (match_dup 2))
(set (match_dup 0)
(ior:DI (ashift:DI (match_dup 0) (const_int 16)) (match_dup 1)))]
{
unsigned HOST_WIDE_INT low = CONST_DOUBLE_LOW (operands[1]);
unsigned HOST_WIDE_INT high = CONST_DOUBLE_HIGH (operands[1]);
unsigned HOST_WIDE_INT val = low;
unsigned HOST_WIDE_INT sign;
/* Zero-extend the 16 least-significant bits. */
val &= 0xffff;
operands[1] = GEN_INT (val);
/* Arithmetic shift right the double-word by 16 bits. */
low >>= 16;
low |= (high & 0xffff) << (HOST_BITS_PER_WIDE_INT - 16);
high >>= 16;
sign = 1;
sign <<= (HOST_BITS_PER_WIDE_INT - 16 - 1);
high ^= sign;
high -= sign;
/* This will only be true if high is a sign-extension of low, i.e.,
it must be either 0 or (unsigned)-1, and be zero iff the
most-significant bit of low is set. */
if (high + (low >> (HOST_BITS_PER_WIDE_INT - 1)) == 0)
operands[2] = GEN_INT (low);
else
operands[2] = immed_double_const (low, high, DImode);
})
(define_insn "shori_media"
[(set (match_operand:DI 0 "ext_dest_operand" "=r,r")
(ior:DI (ashift:DI (match_operand:DI 1 "arith_reg_operand" "0,0")
(const_int 16))
(match_operand:DI 2 "immediate_operand" "K16Csu,nF")))]
"TARGET_SHMEDIA && (reload_completed || arith_reg_dest (operands[0], DImode))"
"@
shori %u2, %0
#"
[(set_attr "type" "arith_media,*")])
(define_insn "*shori_media_si"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(ior:SI (ashift:SI (match_operand:SI 1 "arith_reg_operand" "0")
(const_int 16))
(match_operand:SI 2 "immediate_operand" "K16Csu")))]
"TARGET_SHMEDIA"
"shori %u2, %0")
(define_expand "movdi"
[(set (match_operand:DI 0 "general_movdst_operand" "")
(match_operand:DI 1 "general_movsrc_operand" ""))]
""
{
prepare_move_operands (operands, DImode);
})
(define_insn "movdf_media"
[(set (match_operand:DF 0 "general_movdst_operand" "=f,f,r,r,r,f,m,r,m")
(match_operand:DF 1 "general_movsrc_operand" "f,rZ,f,r,F,m,f,m,rZ"))]
"TARGET_SHMEDIA_FPU
&& (register_operand (operands[0], DFmode)
|| sh_register_operand (operands[1], DFmode))"
"@
fmov.d %1, %0
fmov.qd %N1, %0
fmov.dq %1, %0
add %1, r63, %0
#
fld%M1.d %m1, %0
fst%M0.d %m0, %1
ld%M1.q %m1, %0
st%M0.q %m0, %N1"
[(set_attr "type" "fmove_media,fload_media,dfpconv_media,arith_media,*,
fload_media,fstore_media,load_media,store_media")])
(define_insn "movdf_media_nofpu"
[(set (match_operand:DF 0 "general_movdst_operand" "=r,r,r,m")
(match_operand:DF 1 "general_movsrc_operand" "r,F,m,rZ"))]
"TARGET_SHMEDIA
&& (register_operand (operands[0], DFmode)
|| sh_register_operand (operands[1], DFmode))"
"@
add %1, r63, %0
#
ld%M1.q %m1, %0
st%M0.q %m0, %N1"
[(set_attr "type" "arith_media,*,load_media,store_media")])
(define_split
[(set (match_operand:DF 0 "arith_reg_dest" "")
(match_operand:DF 1 "immediate_operand" ""))]
"TARGET_SHMEDIA && reload_completed"
[(set (match_dup 3) (match_dup 2))]
{
int endian = WORDS_BIG_ENDIAN ? 1 : 0;
long values[2];
REAL_VALUE_TYPE value;
REAL_VALUE_FROM_CONST_DOUBLE (value, operands[1]);
REAL_VALUE_TO_TARGET_DOUBLE (value, values);
if (HOST_BITS_PER_WIDE_INT >= 64)
operands[2] = immed_double_const ((unsigned long) values[endian]
| ((HOST_WIDE_INT) values[1 - endian]
<< 32), 0, DImode);
else
{
gcc_assert (HOST_BITS_PER_WIDE_INT == 32);
operands[2] = immed_double_const (values[endian], values[1 - endian],
DImode);
}
operands[3] = gen_rtx_REG (DImode, true_regnum (operands[0]));
})
;; FIXME: This should be a define_insn_and_split.
(define_insn "movdf_k"
[(set (match_operand:DF 0 "general_movdst_operand" "=r,r,r,m")
(match_operand:DF 1 "general_movsrc_operand" "r,FQ,m,r"))]
"TARGET_SH1
&& (! (TARGET_SH4 || TARGET_SH2A_DOUBLE) || reload_completed
/* ??? We provide some insn so that direct_{load,store}[DFmode] get set */
|| (REG_P (operands[0]) && REGNO (operands[0]) == 3)
|| (REG_P (operands[1]) && REGNO (operands[1]) == 3))
&& (arith_reg_operand (operands[0], DFmode)
|| arith_reg_operand (operands[1], DFmode))"
{
return output_movedouble (insn, operands, DFmode);
}
[(set_attr "length" "4")
(set_attr "type" "move,pcload,load,store")])
;; All alternatives of movdf_i4 are split for ! TARGET_FMOVD.
;; However, the d/F/c/z alternative cannot be split directly; it is converted
;; with special code in machine_dependent_reorg into a load of the R0_REG and
;; the d/m/c/X alternative, which is split later into single-precision
;; instructions. And when not optimizing, no splits are done before fixing
;; up pcloads, so we need usable length information for that.
(define_insn "movdf_i4"
[(set (match_operand:DF 0 "general_movdst_operand" "=d,r,d,d,m,r,r,m,!??r,!???d")
(match_operand:DF 1 "general_movsrc_operand" "d,r,F,m,d,FQ,m,r,d,r"))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2 "=X,X,&z,X,X,X,X,X,X,X"))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)
&& (arith_reg_operand (operands[0], DFmode)
|| arith_reg_operand (operands[1], DFmode))"
{
switch (which_alternative)
{
case 0:
if (TARGET_FMOVD)
return "fmov %1,%0";
else if (REGNO (operands[0]) != REGNO (operands[1]) + 1)
return "fmov %R1,%R0" "\n"
" fmov %S1,%S0";
else
return "fmov %S1,%S0" "\n"
" fmov %R1,%R0";
case 3:
case 4:
return "fmov.d %1,%0";
default:
return "#";
}
}
[(set_attr_alternative "length"
[(if_then_else (eq_attr "fmovd" "yes") (const_int 4) (const_int 8))
(const_int 4)
(if_then_else (eq_attr "fmovd" "yes") (const_int 4) (const_int 6))
(if_then_else (eq_attr "fmovd" "yes") (const_int 4) (const_int 6))
(if_then_else (eq_attr "fmovd" "yes") (const_int 4) (const_int 6))
(const_int 4)
(const_int 8) (const_int 8) ;; these need only 8 bytes for @(r0,rn)
;; We can't use 4-byte push/pop on SHcompact, so we have to
;; increment or decrement r15 explicitly.
(if_then_else
(match_test "TARGET_SHCOMPACT")
(const_int 10) (const_int 8))
(if_then_else
(match_test "TARGET_SHCOMPACT")
(const_int 10) (const_int 8))])
(set_attr "type" "fmove,move,pcfload,fload,fstore,pcload,load,store,load,fload")
(set_attr "late_fp_use" "*,*,*,*,yes,*,*,*,*,*")
(set (attr "fp_mode") (if_then_else (eq_attr "fmovd" "yes")
(const_string "double")
(const_string "none")))])
;; Moving DFmode between fp/general registers through memory
;; (the top of the stack) is faster than moving through fpul even for
;; little endian. Because the type of an instruction is important for its
;; scheduling, it is beneficial to split these operations, rather than
;; emitting them in one single chunk, even if this will expose a stack
;; use that will prevent scheduling of other stack accesses beyond this
;; instruction.
(define_split
[(set (match_operand:DF 0 "register_operand")
(match_operand:DF 1 "register_operand"))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) && reload_completed
&& (true_regnum (operands[0]) < 16) != (true_regnum (operands[1]) < 16)"
[(const_int 0)]
{
rtx insn, tos;
if (TARGET_SH5 && true_regnum (operands[1]) < 16)
{
emit_move_insn (stack_pointer_rtx,
plus_constant (Pmode, stack_pointer_rtx, -8));
tos = gen_tmp_stack_mem (DFmode, stack_pointer_rtx);
}
else
tos = gen_tmp_stack_mem (DFmode,
gen_rtx_PRE_DEC (Pmode, stack_pointer_rtx));
insn = emit_insn (gen_movdf_i4 (tos, operands[1]));
if (! (TARGET_SH5 && true_regnum (operands[1]) < 16))
add_reg_note (insn, REG_INC, stack_pointer_rtx);
if (TARGET_SH5 && true_regnum (operands[0]) < 16)
tos = gen_tmp_stack_mem (DFmode, stack_pointer_rtx);
else
tos = gen_tmp_stack_mem (DFmode,
gen_rtx_POST_INC (Pmode, stack_pointer_rtx));
insn = emit_insn (gen_movdf_i4 (operands[0], tos));
if (TARGET_SH5 && true_regnum (operands[0]) < 16)
emit_move_insn (stack_pointer_rtx,
plus_constant (Pmode, stack_pointer_rtx, 8));
else
add_reg_note (insn, REG_INC, stack_pointer_rtx);
DONE;
})
;; local-alloc sometimes allocates scratch registers even when not required,
;; so we must be prepared to handle these.
;; Remove the use and clobber from a movdf_i4 so that we can use movdf_k.
(define_split
[(set (match_operand:DF 0 "general_movdst_operand" "")
(match_operand:DF 1 "general_movsrc_operand" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)
&& reload_completed
&& true_regnum (operands[0]) < 16
&& true_regnum (operands[1]) < 16"
[(set (match_dup 0) (match_dup 1))]
{
/* If this was a reg <-> mem operation with base + index reg addressing,
we have to handle this in a special way. */
rtx mem = operands[0];
int store_p = 1;
if (! memory_operand (mem, DFmode))
{
mem = operands[1];
store_p = 0;
}
if (GET_CODE (mem) == SUBREG && SUBREG_BYTE (mem) == 0)
mem = SUBREG_REG (mem);
if (MEM_P (mem))
{
rtx addr = XEXP (mem, 0);
if (GET_CODE (addr) == PLUS
&& REG_P (XEXP (addr, 0))
&& REG_P (XEXP (addr, 1)))
{
int offset;
rtx reg0 = gen_rtx_REG (Pmode, 0);
rtx regop = operands[store_p], word0 ,word1;
if (GET_CODE (regop) == SUBREG)
alter_subreg (®op, true);
if (REGNO (XEXP (addr, 0)) == REGNO (XEXP (addr, 1)))
offset = 2;
else
offset = 4;
mem = copy_rtx (mem);
PUT_MODE (mem, SImode);
word0 = gen_rtx_SUBREG (SImode, regop, 0);
alter_subreg (&word0, true);
word1 = gen_rtx_SUBREG (SImode, regop, 4);
alter_subreg (&word1, true);
if (store_p || ! refers_to_regno_p (REGNO (word0),
REGNO (word0) + 1, addr, 0))
{
emit_insn (store_p
? gen_movsi_ie (mem, word0)
: gen_movsi_ie (word0, mem));
emit_insn (gen_addsi3 (reg0, reg0, GEN_INT (offset)));
mem = copy_rtx (mem);
emit_insn (store_p
? gen_movsi_ie (mem, word1)
: gen_movsi_ie (word1, mem));
emit_insn (gen_addsi3 (reg0, reg0, GEN_INT (-offset)));
}
else
{
emit_insn (gen_addsi3 (reg0, reg0, GEN_INT (offset)));
emit_insn (gen_movsi_ie (word1, mem));
emit_insn (gen_addsi3 (reg0, reg0, GEN_INT (-offset)));
mem = copy_rtx (mem);
emit_insn (gen_movsi_ie (word0, mem));
}
DONE;
}
}
})
;; Split away the clobber of r0 after machine_dependent_reorg has fixed pcloads.
(define_split
[(set (match_operand:DF 0 "register_operand" "")
(match_operand:DF 1 "memory_operand" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI R0_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) && reload_completed"
[(parallel [(set (match_dup 0) (match_dup 1))
(use (reg:SI FPSCR_MODES_REG))
(clobber (scratch:SI))])]
"")
(define_expand "reload_indf__frn"
[(parallel [(set (match_operand:DF 0 "register_operand" "=a")
(match_operand:DF 1 "immediate_operand" "FQ"))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_operand:SI 2 "register_operand" "=&z"))])]
"TARGET_SH1"
"")
(define_expand "reload_outdf__RnFRm"
[(parallel [(set (match_operand:DF 0 "register_operand" "=r,f")
(match_operand:DF 1 "register_operand" "af,r"))
(clobber (match_operand:SI 2 "register_operand" "=&y,y"))])]
"TARGET_SH1"
"")
;; Simplify no-op moves.
(define_split
[(set (match_operand:SF 0 "register_operand" "")
(match_operand:SF 1 "register_operand" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2))]
"TARGET_SH2E && reload_completed
&& true_regnum (operands[0]) == true_regnum (operands[1])"
[(set (match_dup 0) (match_dup 0))]
"")
;; fmovd substitute post-reload splits
(define_split
[(set (match_operand:DF 0 "register_operand" "")
(match_operand:DF 1 "register_operand" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2))]
"TARGET_SH4 && ! TARGET_FMOVD && reload_completed
&& FP_OR_XD_REGISTER_P (true_regnum (operands[0]))
&& FP_OR_XD_REGISTER_P (true_regnum (operands[1]))"
[(const_int 0)]
{
int dst = true_regnum (operands[0]), src = true_regnum (operands[1]);
emit_insn (gen_movsf_ie (gen_rtx_REG (SFmode, dst),
gen_rtx_REG (SFmode, src)));
emit_insn (gen_movsf_ie (gen_rtx_REG (SFmode, dst + 1),
gen_rtx_REG (SFmode, src + 1)));
DONE;
})
(define_split
[(set (match_operand:DF 0 "register_operand" "")
(mem:DF (match_operand:SI 1 "register_operand" "")))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) && ! TARGET_FMOVD && reload_completed
&& FP_OR_XD_REGISTER_P (true_regnum (operands[0]))
&& find_regno_note (insn, REG_DEAD, true_regnum (operands[1]))"
[(const_int 0)]
{
int regno = true_regnum (operands[0]);
rtx insn;
rtx mem = SET_SRC (XVECEXP (PATTERN (curr_insn), 0, 0));
rtx mem2
= change_address (mem, SFmode, gen_rtx_POST_INC (Pmode, operands[1]));
insn = emit_insn (gen_movsf_ie (gen_rtx_REG (SFmode,
regno + SH_REG_MSW_OFFSET),
mem2));
add_reg_note (insn, REG_INC, operands[1]);
insn = emit_insn (gen_movsf_ie (gen_rtx_REG (SFmode,
regno + SH_REG_LSW_OFFSET),
change_address (mem, SFmode, NULL_RTX)));
DONE;
})
(define_split
[(set (match_operand:DF 0 "register_operand" "")
(match_operand:DF 1 "memory_operand" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) && ! TARGET_FMOVD && reload_completed
&& FP_OR_XD_REGISTER_P (true_regnum (operands[0]))"
[(const_int 0)]
{
int regno = true_regnum (operands[0]);
rtx addr, insn;
rtx mem2 = change_address (operands[1], SFmode, NULL_RTX);
rtx reg0 = gen_rtx_REG (SFmode, regno + SH_REG_MSW_OFFSET);
rtx reg1 = gen_rtx_REG (SFmode, regno + SH_REG_LSW_OFFSET);
operands[1] = copy_rtx (mem2);
addr = XEXP (mem2, 0);
switch (GET_CODE (addr))
{
case REG:
/* This is complicated. If the register is an arithmetic register
we can just fall through to the REG+DISP case below. Otherwise
we have to use a combination of POST_INC and REG addressing... */
if (! arith_reg_operand (operands[1], SFmode))
{
XEXP (mem2, 0) = addr = gen_rtx_POST_INC (SImode, addr);
insn = emit_insn (gen_movsf_ie (reg0, mem2));
add_reg_note (insn, REG_INC, XEXP (addr, 0));
emit_insn (gen_movsf_ie (reg1, operands[1]));
/* If we have modified the stack pointer, the value that we have
read with post-increment might be modified by an interrupt,
so write it back. */
if (REGNO (XEXP (addr, 0)) == STACK_POINTER_REGNUM)
emit_insn (gen_push_e (reg0));
else
emit_insn (gen_addsi3 (XEXP (operands[1], 0), XEXP (operands[1], 0),
GEN_INT (-4)));
break;
}
/* Fall through. */
case PLUS:
emit_insn (gen_movsf_ie (reg0, operands[1]));
operands[1] = copy_rtx (operands[1]);
XEXP (operands[1], 0) = plus_constant (Pmode, addr, 4);
emit_insn (gen_movsf_ie (reg1, operands[1]));
break;
case POST_INC:
insn = emit_insn (gen_movsf_ie (reg0, operands[1]));
add_reg_note (insn, REG_INC, XEXP (addr, 0));
insn = emit_insn (gen_movsf_ie (reg1, operands[1]));
add_reg_note (insn, REG_INC, XEXP (addr, 0));
break;
default:
debug_rtx (addr);
gcc_unreachable ();
}
DONE;
})
(define_split
[(set (match_operand:DF 0 "memory_operand" "")
(match_operand:DF 1 "register_operand" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) && ! TARGET_FMOVD && reload_completed
&& FP_OR_XD_REGISTER_P (true_regnum (operands[1]))"
[(const_int 0)]
{
int regno = true_regnum (operands[1]);
rtx insn, addr;
rtx reg0 = gen_rtx_REG (SFmode, regno + SH_REG_MSW_OFFSET);
rtx reg1 = gen_rtx_REG (SFmode, regno + SH_REG_LSW_OFFSET);
operands[0] = copy_rtx (operands[0]);
PUT_MODE (operands[0], SFmode);
addr = XEXP (operands[0], 0);
switch (GET_CODE (addr))
{
case REG:
/* This is complicated. If the register is an arithmetic register
we can just fall through to the REG+DISP case below. Otherwise
we have to use a combination of REG and PRE_DEC addressing... */
if (! arith_reg_operand (operands[0], SFmode))
{
emit_insn (gen_addsi3 (addr, addr, GEN_INT (4)));
emit_insn (gen_movsf_ie (operands[0], reg1));
operands[0] = copy_rtx (operands[0]);
XEXP (operands[0], 0) = addr = gen_rtx_PRE_DEC (SImode, addr);
insn = emit_insn (gen_movsf_ie (operands[0], reg0));
add_reg_note (insn, REG_INC, XEXP (addr, 0));
break;
}
/* Fall through. */
case PLUS:
/* Since REG+DISP addressing has already been decided upon by gcc
we can rely upon it having chosen an arithmetic register as the
register component of the address. Just emit the lower numbered
register first, to the lower address, then the higher numbered
register to the higher address. */
emit_insn (gen_movsf_ie (operands[0], reg0));
operands[0] = copy_rtx (operands[0]);
XEXP (operands[0], 0) = plus_constant (Pmode, addr, 4);
emit_insn (gen_movsf_ie (operands[0], reg1));
break;
case PRE_DEC:
/* This is easy. Output the word to go to the higher address
first (ie the word in the higher numbered register) then the
word to go to the lower address. */
insn = emit_insn (gen_movsf_ie (operands[0], reg1));
add_reg_note (insn, REG_INC, XEXP (addr, 0));
insn = emit_insn (gen_movsf_ie (operands[0], reg0));
add_reg_note (insn, REG_INC, XEXP (addr, 0));
break;
default:
/* FAIL; */
debug_rtx (addr);
gcc_unreachable ();
}
DONE;
})
;; If the output is a register and the input is memory or a register, we have
;; to be careful and see which word needs to be loaded first.
(define_split
[(set (match_operand:DF 0 "general_movdst_operand" "")
(match_operand:DF 1 "general_movsrc_operand" ""))]
"TARGET_SH1 && reload_completed"
[(set (match_dup 2) (match_dup 3))
(set (match_dup 4) (match_dup 5))]
{
int regno;
if ((MEM_P (operands[0])
&& GET_CODE (XEXP (operands[0], 0)) == PRE_DEC)
|| (MEM_P (operands[1])
&& GET_CODE (XEXP (operands[1], 0)) == POST_INC))
FAIL;
switch (GET_CODE (operands[0]))
{
case REG:
regno = REGNO (operands[0]);
break;
case SUBREG:
regno = subreg_regno (operands[0]);
break;
case MEM:
regno = -1;
break;
default:
gcc_unreachable ();
}
if (regno == -1
|| ! refers_to_regno_p (regno, regno + 1, operands[1], 0))
{
operands[2] = operand_subword (operands[0], 0, 0, DFmode);
operands[3] = operand_subword (operands[1], 0, 0, DFmode);
operands[4] = operand_subword (operands[0], 1, 0, DFmode);
operands[5] = operand_subword (operands[1], 1, 0, DFmode);
}
else
{
operands[2] = operand_subword (operands[0], 1, 0, DFmode);
operands[3] = operand_subword (operands[1], 1, 0, DFmode);
operands[4] = operand_subword (operands[0], 0, 0, DFmode);
operands[5] = operand_subword (operands[1], 0, 0, DFmode);
}
if (operands[2] == 0 || operands[3] == 0
|| operands[4] == 0 || operands[5] == 0)
FAIL;
})
(define_expand "movdf"
[(set (match_operand:DF 0 "general_movdst_operand" "")
(match_operand:DF 1 "general_movsrc_operand" ""))]
""
{
prepare_move_operands (operands, DFmode);
if (TARGET_SHMEDIA)
{
if (TARGET_SHMEDIA_FPU)
emit_insn (gen_movdf_media (operands[0], operands[1]));
else
emit_insn (gen_movdf_media_nofpu (operands[0], operands[1]));
DONE;
}
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_movdf_i4 (operands[0], operands[1]));
DONE;
}
})
;;This is incompatible with the way gcc uses subregs.
;;(define_insn "movv2sf_i"
;; [(set (match_operand:V2SF 0 "nonimmediate_operand" "=f,f,m")
;; (match_operand:V2SF 1 "nonimmediate_operand" "f,m,f"))]
;; "TARGET_SHMEDIA_FPU
;; && (fp_arith_reg_operand (operands[0], V2SFmode)
;; || fp_arith_reg_operand (operands[1], V2SFmode))"
;; "@
;; #
;; fld%M1.p %m1, %0
;; fst%M0.p %m0, %1"
;; [(set_attr "type" "*,fload_media,fstore_media")])
(define_insn_and_split "movv2sf_i"
[(set (match_operand:V2SF 0 "general_movdst_operand" "=f,rf,r,m,mf")
(match_operand:V2SF 1 "general_operand" "fm,rfm?,F?,f,rfZ?"))]
"TARGET_SHMEDIA_FPU"
"#"
"TARGET_SHMEDIA_FPU && reload_completed"
[(set (match_dup 0) (match_dup 1))]
{
operands[0] = simplify_gen_subreg (DFmode, operands[0], V2SFmode, 0);
operands[1] = simplify_gen_subreg (DFmode, operands[1], V2SFmode, 0);
})
(define_expand "movv2sf"
[(set (match_operand:V2SF 0 "general_movdst_operand" "")
(match_operand:V2SF 1 "nonimmediate_operand" ""))]
"TARGET_SHMEDIA_FPU"
{
prepare_move_operands (operands, V2SFmode);
})
(define_expand "addv2sf3"
[(match_operand:V2SF 0 "fp_arith_reg_operand" "")
(match_operand:V2SF 1 "fp_arith_reg_operand" "")
(match_operand:V2SF 2 "fp_arith_reg_operand" "")]
"TARGET_SHMEDIA_FPU"
{
sh_expand_binop_v2sf (PLUS, operands[0], operands[1], operands[2]);
DONE;
})
(define_expand "subv2sf3"
[(match_operand:V2SF 0 "fp_arith_reg_operand" "")
(match_operand:V2SF 1 "fp_arith_reg_operand" "")
(match_operand:V2SF 2 "fp_arith_reg_operand" "")]
"TARGET_SHMEDIA_FPU"
{
sh_expand_binop_v2sf (MINUS, operands[0], operands[1], operands[2]);
DONE;
})
(define_expand "mulv2sf3"
[(match_operand:V2SF 0 "fp_arith_reg_operand" "")
(match_operand:V2SF 1 "fp_arith_reg_operand" "")
(match_operand:V2SF 2 "fp_arith_reg_operand" "")]
"TARGET_SHMEDIA_FPU"
{
sh_expand_binop_v2sf (MULT, operands[0], operands[1], operands[2]);
DONE;
})
(define_expand "divv2sf3"
[(match_operand:V2SF 0 "fp_arith_reg_operand" "")
(match_operand:V2SF 1 "fp_arith_reg_operand" "")
(match_operand:V2SF 2 "fp_arith_reg_operand" "")]
"TARGET_SHMEDIA_FPU"
{
sh_expand_binop_v2sf (DIV, operands[0], operands[1], operands[2]);
DONE;
})
(define_insn_and_split "*movv4sf_i"
[(set (match_operand:V4SF 0 "general_movdst_operand" "=f,rf,r,m,mf")
(match_operand:V4SF 1 "general_operand" "fm,rfm?,F?,f,rfZ?"))]
"TARGET_SHMEDIA_FPU"
"#"
"&& reload_completed"
[(const_int 0)]
{
for (int i = 0; i < 4/2; i++)
{
rtx x, y;
if (MEM_P (operands[0]))
x = adjust_address (operands[0], V2SFmode,
i * GET_MODE_SIZE (V2SFmode));
else
x = simplify_gen_subreg (V2SFmode, operands[0], V4SFmode, i * 8);
if (MEM_P (operands[1]))
y = adjust_address (operands[1], V2SFmode,
i * GET_MODE_SIZE (V2SFmode));
else
y = simplify_gen_subreg (V2SFmode, operands[1], V4SFmode, i * 8);
emit_insn (gen_movv2sf_i (x, y));
}
DONE;
}
[(set_attr "length" "8")])
(define_expand "movv4sf"
[(set (match_operand:V4SF 0 "nonimmediate_operand" "")
(match_operand:V4SF 1 "general_operand" ""))]
"TARGET_SHMEDIA_FPU"
{
prepare_move_operands (operands, V4SFmode);
})
(define_insn_and_split "*movv16sf_i"
[(set (match_operand:V16SF 0 "nonimmediate_operand" "=f,f,m")
(match_operand:V16SF 1 "nonimmediate_operand" "f,m,f"))]
"TARGET_SHMEDIA_FPU"
"#"
"&& reload_completed"
[(const_int 0)]
{
for (int i = 0; i < 16/2; i++)
{
rtx x, y;
if (MEM_P (operands[0]))
x = adjust_address (operands[0], V2SFmode,
i * GET_MODE_SIZE (V2SFmode));
else
{
x = gen_rtx_SUBREG (V2SFmode, operands[0], i * 8);
alter_subreg (&x, true);
}
if (MEM_P (operands[1]))
y = adjust_address (operands[1], V2SFmode,
i * GET_MODE_SIZE (V2SFmode));
else
{
y = gen_rtx_SUBREG (V2SFmode, operands[1], i * 8);
alter_subreg (&y, true);
}
emit_insn (gen_movv2sf_i (x, y));
}
DONE;
}
[(set_attr "length" "32")])
(define_expand "movv16sf"
[(set (match_operand:V16SF 0 "nonimmediate_operand" "=f,f,m")
(match_operand:V16SF 1 "nonimmediate_operand" "f,m,f"))]
"TARGET_SHMEDIA_FPU"
{
prepare_move_operands (operands, V16SFmode);
})
(define_insn "movsf_media"
[(set (match_operand:SF 0 "general_movdst_operand" "=f,f,r,r,r,f,m,r,m")
(match_operand:SF 1 "general_movsrc_operand" "f,rZ,f,r,F,m,f,m,rZ"))]
"TARGET_SHMEDIA_FPU
&& (register_operand (operands[0], SFmode)
|| sh_register_operand (operands[1], SFmode))"
"@
fmov.s %1, %0
fmov.ls %N1, %0
fmov.sl %1, %0
add.l %1, r63, %0
#
fld%M1.s %m1, %0
fst%M0.s %m0, %1
ld%M1.l %m1, %0
st%M0.l %m0, %N1"
[(set_attr "type" "fmove_media,fload_media,fpconv_media,arith_media,*,fload_media,fstore_media,load_media,store_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_insn "movsf_media_nofpu"
[(set (match_operand:SF 0 "general_movdst_operand" "=r,r,r,m")
(match_operand:SF 1 "general_movsrc_operand" "r,F,m,rZ"))]
"TARGET_SHMEDIA
&& (register_operand (operands[0], SFmode)
|| sh_register_operand (operands[1], SFmode))"
"@
add.l %1, r63, %0
#
ld%M1.l %m1, %0
st%M0.l %m0, %N1"
[(set_attr "type" "arith_media,*,load_media,store_media")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_split
[(set (match_operand:SF 0 "arith_reg_dest" "")
(match_operand:SF 1 "immediate_operand" ""))]
"TARGET_SHMEDIA && reload_completed
&& ! FP_REGISTER_P (true_regnum (operands[0]))"
[(set (match_dup 3) (match_dup 2))]
{
long values;
REAL_VALUE_TYPE value;
REAL_VALUE_FROM_CONST_DOUBLE (value, operands[1]);
REAL_VALUE_TO_TARGET_SINGLE (value, values);
operands[2] = GEN_INT (values);
operands[3] = gen_rtx_REG (DImode, true_regnum (operands[0]));
})
(define_insn "movsf_i"
[(set (match_operand:SF 0 "general_movdst_operand" "=r,r,r,r,m,l,r")
(match_operand:SF 1 "general_movsrc_operand" "r,G,FQ,mr,r,r,l"))]
"TARGET_SH1
&& (! TARGET_SH2E
/* ??? We provide some insn so that direct_{load,store}[SFmode] get set */
|| (REG_P (operands[0]) && REGNO (operands[0]) == 3)
|| (REG_P (operands[1]) && REGNO (operands[1]) == 3))
&& (arith_reg_operand (operands[0], SFmode)
|| arith_reg_operand (operands[1], SFmode))"
"@
mov %1,%0
mov #0,%0
mov.l %1,%0
mov.l %1,%0
mov.l %1,%0
lds %1,%0
sts %1,%0"
[(set_attr "type" "move,move,pcload,load,store,move,move")])
;; We may not split the ry/yr/XX alternatives to movsi_ie, since
;; update_flow_info would not know where to put REG_EQUAL notes
;; when the destination changes mode.
(define_insn "movsf_ie"
[(set (match_operand:SF 0 "general_movdst_operand"
"=f,r,f,f,fy,f,m,r,r,m,f,y,y,rf,r,y,<,y,y")
(match_operand:SF 1 "general_movsrc_operand"
"f,r,G,H,FQ,mf,f,FQ,mr,r,y,f,>,fr,y,r,y,>,y"))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2 "=X,X,Bsc,Bsc,&z,X,X,X,X,X,X,X,X,y,X,X,X,X,X"))]
"TARGET_SH2E
&& (arith_reg_operand (operands[0], SFmode) || fpul_operand (operands[0], SFmode)
|| arith_reg_operand (operands[1], SFmode) || fpul_operand (operands[1], SFmode)
|| arith_reg_operand (operands[2], SImode))"
"@
fmov %1,%0
mov %1,%0
fldi0 %0
fldi1 %0
#
fmov.s %1,%0
fmov.s %1,%0
mov.l %1,%0
mov.l %1,%0
mov.l %1,%0
fsts fpul,%0
flds %1,fpul
lds.l %1,%0
#
sts %1,%0
lds %1,%0
sts.l %1,%0
lds.l %1,%0
! move optimized away"
[(set_attr "type" "fmove,move,fmove,fmove,pcfload,fload,fstore,pcload,load,
store,fmove,fmove,load,*,fpul_gp,gp_fpul,fstore,load,nil")
(set_attr "late_fp_use" "*,*,*,*,*,*,yes,*,*,*,*,*,*,*,yes,*,yes,*,*")
(set_attr_alternative "length"
[(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 4)
(if_then_else
(match_test "TARGET_SH2A")
(const_int 4) (const_int 2))
(if_then_else
(match_test "TARGET_SH2A")
(const_int 4) (const_int 2))
(const_int 2)
(if_then_else
(match_test "TARGET_SH2A")
(const_int 4) (const_int 2))
(if_then_else
(match_test "TARGET_SH2A")
(const_int 4) (const_int 2))
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 4)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 2)
(const_int 0)])
(set_attr_alternative "fp_mode"
[(if_then_else (eq_attr "fmovd" "yes")
(const_string "single") (const_string "none"))
(const_string "none")
(const_string "single")
(const_string "single")
(const_string "none")
(if_then_else (eq_attr "fmovd" "yes")
(const_string "single") (const_string "none"))
(if_then_else (eq_attr "fmovd" "yes")
(const_string "single") (const_string "none"))
(const_string "none")
(const_string "none")
(const_string "none")
(const_string "none")
(const_string "none")
(const_string "none")
(const_string "none")
(const_string "none")
(const_string "none")
(const_string "none")
(const_string "none")
(const_string "none")])])
(define_split
[(set (match_operand:SF 0 "register_operand" "")
(match_operand:SF 1 "register_operand" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI FPUL_REG))]
"TARGET_SH1"
[(parallel [(set (reg:SF FPUL_REG) (match_dup 1))
(use (reg:SI FPSCR_MODES_REG))
(clobber (scratch:SI))])
(parallel [(set (match_dup 0) (reg:SF FPUL_REG))
(use (reg:SI FPSCR_MODES_REG))
(clobber (scratch:SI))])]
"")
(define_expand "movsf"
[(set (match_operand:SF 0 "general_movdst_operand" "")
(match_operand:SF 1 "general_movsrc_operand" ""))]
""
{
prepare_move_operands (operands, SFmode);
if (TARGET_SHMEDIA)
{
if (TARGET_SHMEDIA_FPU)
emit_insn (gen_movsf_media (operands[0], operands[1]));
else
emit_insn (gen_movsf_media_nofpu (operands[0], operands[1]));
DONE;
}
if (TARGET_SH2E)
{
emit_insn (gen_movsf_ie (operands[0], operands[1]));
DONE;
}
})
(define_insn "mov_nop"
[(set (match_operand 0 "any_register_operand" "") (match_dup 0))]
"TARGET_SH2E"
""
[(set_attr "length" "0")
(set_attr "type" "nil")])
(define_expand "reload_insf__frn"
[(parallel [(set (match_operand:SF 0 "register_operand" "=a")
(match_operand:SF 1 "immediate_operand" "FQ"))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_operand:SI 2 "register_operand" "=&z"))])]
"TARGET_SH1"
"")
(define_expand "reload_insi__i_fpul"
[(parallel [(set (match_operand:SI 0 "fpul_operand" "=y")
(match_operand:SI 1 "immediate_operand" "i"))
(clobber (match_operand:SI 2 "register_operand" "=&z"))])]
"TARGET_SH1"
"")
(define_expand "ptabs"
[(set (match_operand 0 "" "=b") (match_operand 1 "" "r"))]
"TARGET_SHMEDIA"
{
if (!TARGET_PT_FIXED)
{
rtx eq = operands[1];
/* ??? For canonical RTL we really should remove any CONST from EQ
before wrapping it in the AND, and finally wrap the EQ into a
const if is constant. However, for reload we must expose the
input register or symbolic constant, and we can't have
different insn structures outside of the operands for different
alternatives of the same pattern. */
eq = gen_rtx_EQ (SImode, gen_rtx_AND (Pmode, eq, GEN_INT (3)),
GEN_INT (3));
operands[1]
= (gen_rtx_IF_THEN_ELSE
(PDImode,
eq,
gen_rtx_MEM (PDImode, operands[1]),
gen_rtx_fmt_e (TARGET_SHMEDIA32 ? SIGN_EXTEND : TRUNCATE,
PDImode, operands[1])));
}
})
;; expanded by ptabs expander.
(define_insn "*extendsipdi_media"
[(set (match_operand:PDI 0 "target_reg_operand" "=b,b");
(if_then_else:PDI (eq (and:SI (match_operand:SI 1 "target_operand"
"r,Csy")
(const_int 3))
(const_int 3))
(mem:PDI (match_dup 1))
(sign_extend:PDI (match_dup 1))))]
"TARGET_SHMEDIA && !TARGET_PT_FIXED"
"@
ptabs %1, %0
pt %1, %0"
[(set_attr "type" "ptabs_media,pt_media")
(set_attr "length" "4,*")])
(define_insn "*truncdipdi_media"
[(set (match_operand:PDI 0 "target_reg_operand" "=b,b");
(if_then_else:PDI (eq (and:DI (match_operand:DI 1 "target_operand"
"r,Csy")
(const_int 3))
(const_int 3))
(mem:PDI (match_dup 1))
(truncate:PDI (match_dup 1))))]
"TARGET_SHMEDIA && !TARGET_PT_FIXED"
"@
ptabs %1, %0
pt %1, %0"
[(set_attr "type" "ptabs_media,pt_media")
(set_attr "length" "4,*")])
(define_insn "*movsi_y"
[(set (match_operand:SI 0 "register_operand" "=y,y")
(match_operand:SI 1 "immediate_operand" "Qi,I08"))
(clobber (match_scratch:SI 2 "=&z,r"))]
"TARGET_SH2E
&& (reload_in_progress || reload_completed)"
"#"
[(set_attr "length" "4")
(set_attr "type" "pcload,move")])
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(match_operand:SI 1 "immediate_operand" ""))
(clobber (match_operand:SI 2 "register_operand" ""))]
"TARGET_SH1"
[(set (match_dup 2) (match_dup 1))
(set (match_dup 0) (match_dup 2))]
"")
;; ------------------------------------------------------------------------
;; Define the real conditional branch instructions.
;; ------------------------------------------------------------------------
(define_expand "branch_true"
[(set (pc) (if_then_else (ne (reg:SI T_REG) (const_int 0))
(label_ref (match_operand 0))
(pc)))]
"TARGET_SH1")
(define_expand "branch_false"
[(set (pc) (if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_operand 0))
(pc)))]
"TARGET_SH1")
(define_insn_and_split "*cbranch_t"
[(set (pc) (if_then_else (match_operand 1 "cbranch_treg_value")
(label_ref (match_operand 0))
(pc)))]
"TARGET_SH1"
{
return output_branch (sh_eval_treg_value (operands[1]), insn, operands);
}
"&& 1"
[(const_int 0)]
{
/* Try to canonicalize the branch condition if it is not one of:
(ne (reg:SI T_REG) (const_int 0))
(eq (reg:SI T_REG) (const_int 0))
Instead of splitting out a new insn, we modify the current insn's
operands as needed. This preserves things such as REG_DEAD notes. */
if ((GET_CODE (operands[1]) == EQ || GET_CODE (operands[1]) == NE)
&& REG_P (XEXP (operands[1], 0)) && REGNO (XEXP (operands[1], 0)) == T_REG
&& XEXP (operands[1], 1) == const0_rtx)
DONE;
int branch_cond = sh_eval_treg_value (operands[1]);
rtx new_cond_rtx = NULL_RTX;
if (branch_cond == 0)
new_cond_rtx = gen_rtx_EQ (VOIDmode, get_t_reg_rtx (), const0_rtx);
else if (branch_cond == 1)
new_cond_rtx = gen_rtx_NE (VOIDmode, get_t_reg_rtx (), const0_rtx);
if (new_cond_rtx != NULL_RTX)
validate_change (curr_insn, &XEXP (XEXP (PATTERN (curr_insn), 1), 0),
new_cond_rtx, false);
DONE;
}
[(set_attr "type" "cbranch")])
;; Patterns to prevent reorg from re-combining a condbranch with a branch
;; which destination is too far away.
;; The const_int_operand is distinct for each branch target; it avoids
;; unwanted matches with redundant_insn.
(define_insn "block_branch_redirect"
[(set (pc) (unspec [(match_operand 0 "const_int_operand" "")] UNSPEC_BBR))]
"TARGET_SH1"
""
[(set_attr "length" "0")])
;; This one has the additional purpose to record a possible scratch register
;; for the following branch.
;; ??? Unfortunately, just setting the scratch register is not good enough,
;; because the insn then might be deemed dead and deleted. And we can't
;; make the use in the jump insn explicit because that would disable
;; delay slot scheduling from the target.
(define_insn "indirect_jump_scratch"
[(set (match_operand:SI 0 "register_operand" "=r")
(unspec:SI [(match_operand 1 "const_int_operand" "")] UNSPEC_BBR))
(set (pc) (unspec [(const_int 0)] UNSPEC_BBR))]
"TARGET_SH1"
""
[(set_attr "length" "0")])
;; This one is used to preemt an insn from beyond the bra / braf / jmp
;; being pulled into the delay slot of a condbranch that has been made to
;; jump around the unconditional jump because it was out of range.
(define_insn "stuff_delay_slot"
[(set (pc)
(unspec [(match_operand:SI 0 "const_int_operand" "") (pc)
(match_operand:SI 1 "const_int_operand" "")] UNSPEC_BBR))]
"TARGET_SH1"
""
[(set_attr "length" "0")
(set_attr "cond_delay_slot" "yes")])
;; Conditional branch insns
(define_expand "cbranchint4_media"
[(set (pc)
(if_then_else (match_operator 0 "shmedia_cbranch_comparison_operator"
[(match_operand 1 "" "")
(match_operand 2 "" "")])
(match_operand 3 "" "")
(pc)))]
"TARGET_SHMEDIA"
{
enum machine_mode mode = GET_MODE (operands[1]);
if (mode == VOIDmode)
mode = GET_MODE (operands[2]);
if (GET_CODE (operands[0]) == EQ || GET_CODE (operands[0]) == NE)
{
operands[1] = force_reg (mode, operands[1]);
if (CONSTANT_P (operands[2])
&& (! satisfies_constraint_I06 (operands[2])))
operands[2] = force_reg (mode, operands[2]);
}
else
{
if (operands[1] != const0_rtx)
operands[1] = force_reg (mode, operands[1]);
if (operands[2] != const0_rtx)
operands[2] = force_reg (mode, operands[2]);
}
switch (GET_CODE (operands[0]))
{
case LEU:
case LE:
case LTU:
case LT:
operands[0] = gen_rtx_fmt_ee (swap_condition (GET_CODE (operands[0])),
VOIDmode, operands[2], operands[1]);
operands[1] = XEXP (operands[0], 0);
operands[2] = XEXP (operands[0], 1);
break;
default:
operands[0] = gen_rtx_fmt_ee (GET_CODE (operands[0]),
VOIDmode, operands[1], operands[2]);
break;
}
operands[3] = gen_rtx_LABEL_REF (Pmode, operands[3]);
})
(define_expand "cbranchfp4_media"
[(set (pc)
(if_then_else (match_operator 0 "sh_float_comparison_operator"
[(match_operand 1 "" "")
(match_operand 2 "" "")])
(match_operand 3 "" "")
(pc)))]
"TARGET_SHMEDIA"
{
rtx tmp = gen_reg_rtx (SImode);
rtx cmp;
if (GET_CODE (operands[0]) == NE)
cmp = gen_rtx_EQ (SImode, operands[1], operands[2]);
else
cmp = gen_rtx_fmt_ee (GET_CODE (operands[0]), SImode,
operands[1], operands[2]);
emit_insn (gen_cstore4_media (tmp, cmp, operands[1], operands[2]));
if (GET_CODE (cmp) == GET_CODE (operands[0]))
operands[0] = gen_rtx_NE (VOIDmode, tmp, const0_rtx);
else
operands[0] = gen_rtx_EQ (VOIDmode, tmp, const0_rtx);
operands[1] = tmp;
operands[2] = const0_rtx;
operands[3] = gen_rtx_LABEL_REF (Pmode, operands[3]);
})
(define_insn "*beq_media_i"
[(set (pc)
(if_then_else (match_operator 3 "equality_comparison_operator"
[(match_operand:DI 1 "arith_reg_operand" "r,r")
(match_operand:DI 2 "arith_operand" "r,I06")])
(match_operand 0 "target_operand" "b,b")
(pc)))]
"TARGET_SHMEDIA"
"@
b%o3%' %1, %2, %0%>
b%o3i%' %1, %2, %0%>"
[(set_attr "type" "cbranch_media")])
(define_insn "*beq_media_i32"
[(set (pc)
(if_then_else (match_operator 3 "equality_comparison_operator"
[(match_operand:SI 1 "arith_reg_operand" "r,r")
(match_operand:SI 2 "arith_operand" "r,I06")])
(match_operand 0 "target_operand" "b,b")
(pc)))]
"TARGET_SHMEDIA"
"@
b%o3%' %1, %2, %0%>
b%o3i%' %1, %2, %0%>"
[(set_attr "type" "cbranch_media")])
(define_insn "*bgt_media_i"
[(set (pc)
(if_then_else (match_operator 3 "greater_comparison_operator"
[(match_operand:DI 1 "arith_reg_or_0_operand" "rN")
(match_operand:DI 2 "arith_reg_or_0_operand" "rN")])
(match_operand 0 "target_operand" "b")
(pc)))]
"TARGET_SHMEDIA"
"b%o3%' %N1, %N2, %0%>"
[(set_attr "type" "cbranch_media")])
(define_insn "*bgt_media_i32"
[(set (pc)
(if_then_else (match_operator 3 "greater_comparison_operator"
[(match_operand:SI 1 "arith_reg_or_0_operand" "rN")
(match_operand:SI 2 "arith_reg_or_0_operand" "rN")])
(match_operand 0 "target_operand" "b")
(pc)))]
"TARGET_SHMEDIA"
"b%o3%' %N1, %N2, %0%>"
[(set_attr "type" "cbranch_media")])
;; These are only needed to make invert_jump() happy - otherwise, jump
;; optimization will be silently disabled.
(define_insn "*blt_media_i"
[(set (pc)
(if_then_else (match_operator 3 "less_comparison_operator"
[(match_operand:DI 1 "arith_reg_or_0_operand" "rN")
(match_operand:DI 2 "arith_reg_or_0_operand" "rN")])
(match_operand 0 "target_operand" "b")
(pc)))]
"TARGET_SHMEDIA"
"b%o3%' %N2, %N1, %0%>"
[(set_attr "type" "cbranch_media")])
(define_insn "*blt_media_i32"
[(set (pc)
(if_then_else (match_operator 3 "less_comparison_operator"
[(match_operand:SI 1 "arith_reg_or_0_operand" "rN")
(match_operand:SI 2 "arith_reg_or_0_operand" "rN")])
(match_operand 0 "target_operand" "b")
(pc)))]
"TARGET_SHMEDIA"
"b%o3%' %N2, %N1, %0%>"
[(set_attr "type" "cbranch_media")])
;; combiner splitter for test-and-branch on single bit in register. This
;; is endian dependent because the non-paradoxical subreg looks different
;; on big endian.
(define_split
[(set (pc)
(if_then_else
(match_operator 3 "equality_comparison_operator"
[(subreg:SI
(zero_extract:DI
(subreg:DI (match_operand:SI 1 "extend_reg_operand" "") 0)
(const_int 1)
(match_operand 2 "const_int_operand" "")) 0)
(const_int 0)])
(match_operand 0 "target_operand" "")
(pc)))
(clobber (match_operand:SI 4 "arith_reg_dest" ""))]
"TARGET_SHMEDIA && TARGET_LITTLE_ENDIAN"
[(set (match_dup 4) (ashift:SI (match_dup 1) (match_dup 5)))
(set (pc) (if_then_else (match_dup 6) (match_dup 0) (pc)))]
{
operands[5] = GEN_INT (31 - INTVAL (operands[2]));
operands[6] = (GET_CODE (operands[3]) == EQ
? gen_rtx_GE (VOIDmode, operands[4], const0_rtx)
: gen_rtx_GT (VOIDmode, const0_rtx, operands[4]));
})
; operand 0 is the loop count pseudo register
; operand 1 is the label to jump to at the top of the loop
(define_expand "doloop_end"
[(parallel [(set (pc)
(if_then_else (ne:SI (match_operand:SI 0 "" "")
(const_int 1))
(label_ref (match_operand 1 "" ""))
(pc)))
(set (match_dup 0)
(plus:SI (match_dup 0) (const_int -1)))
(clobber (reg:SI T_REG))])]
"TARGET_SH2"
{
if (GET_MODE (operands[0]) != SImode)
FAIL;
emit_jump_insn (gen_doloop_end_split (operands[0], operands[1], operands[0]));
DONE;
})
(define_insn_and_split "doloop_end_split"
[(set (pc)
(if_then_else (ne:SI (match_operand:SI 2 "arith_reg_dest" "0")
(const_int 1))
(label_ref (match_operand 1 "" ""))
(pc)))
(set (match_operand:SI 0 "arith_reg_dest" "=r")
(plus:SI (match_dup 2) (const_int -1)))
(clobber (reg:SI T_REG))]
"TARGET_SH2"
"#"
""
[(parallel [(set (reg:SI T_REG)
(eq:SI (match_dup 2) (const_int 1)))
(set (match_dup 0) (plus:SI (match_dup 2) (const_int -1)))])
(set (pc) (if_then_else (eq (reg:SI T_REG) (const_int 0))
(label_ref (match_dup 1))
(pc)))]
""
[(set_attr "type" "cbranch")])
;; ------------------------------------------------------------------------
;; Jump and linkage insns
;; ------------------------------------------------------------------------
(define_insn "jump_compact"
[(set (pc)
(label_ref (match_operand 0 "" "")))]
"TARGET_SH1 && !CROSSING_JUMP_P (insn)"
{
/* The length is 16 if the delay slot is unfilled. */
if (get_attr_length(insn) > 4)
return output_far_jump(insn, operands[0]);
else
return "bra %l0%#";
}
[(set_attr "type" "jump")
(set_attr "needs_delay_slot" "yes")])
;; ??? It would be much saner to explicitly use the scratch register
;; in the jump insn, and have indirect_jump_scratch only set it,
;; but fill_simple_delay_slots would refuse to do delay slot filling
;; from the target then, as it uses simplejump_p.
;;(define_insn "jump_compact_far"
;; [(set (pc)
;; (label_ref (match_operand 0 "" "")))
;; (use (match_operand 1 "register_operand" "r")]
;; "TARGET_SH1"
;; "* return output_far_jump(insn, operands[0], operands[1]);"
;; [(set_attr "type" "jump")
;; (set_attr "needs_delay_slot" "yes")])
(define_insn "jump_media"
[(set (pc)
(match_operand 0 "target_operand" "b"))]
"TARGET_SHMEDIA"
"blink %0, r63%>"
[(set_attr "type" "jump_media")])
(define_expand "jump"
[(set (pc)
(label_ref (match_operand 0 "" "")))]
""
{
if (TARGET_SH1)
emit_jump_insn (gen_jump_compact (operands[0]));
else if (TARGET_SHMEDIA)
{
if (reload_in_progress || reload_completed)
FAIL;
emit_jump_insn (gen_jump_media (gen_rtx_LABEL_REF (Pmode, operands[0])));
}
DONE;
})
(define_insn "force_mode_for_call"
[(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SHCOMPACT"
""
[(set_attr "length" "0")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))])
(define_insn "calli"
[(call (mem:SI (match_operand:SI 0 "arith_reg_operand" "r"))
(match_operand 1 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SH1"
{
if (TARGET_SH2A && (dbr_sequence_length () == 0))
return "jsr/n @%0";
else
return "jsr @%0%#";
}
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")
(set_attr "fp_set" "unknown")])
;; This is TBR relative jump instruction for SH2A architecture.
;; Its use is enabled by assigning an attribute "function_vector"
;; and the vector number to a function during its declaration.
(define_insn "calli_tbr_rel"
[(call (mem (match_operand:SI 0 "symbol_ref_operand" ""))
(match_operand 1 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SH2A && sh2a_is_function_vector_call (operands[0])"
{
unsigned HOST_WIDE_INT vect_num;
vect_num = sh2a_get_function_vector_number (operands[0]);
operands[2] = GEN_INT (vect_num * 4);
return "jsr/n @@(%O2,tbr)";
}
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "no")
(set_attr "fp_set" "unknown")])
;; This is a pc-rel call, using bsrf, for use with PIC.
(define_insn "calli_pcrel"
[(call (mem:SI (match_operand:SI 0 "arith_reg_operand" "r"))
(match_operand 1 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(use (reg:SI PIC_REG))
(use (match_operand 2 "" ""))
(clobber (reg:SI PR_REG))]
"TARGET_SH2"
{
return "bsrf %0" "\n"
"%O2:%#";
}
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")
(set_attr "fp_set" "unknown")])
(define_insn_and_split "call_pcrel"
[(call (mem:SI (match_operand:SI 0 "symbol_ref_operand" ""))
(match_operand 1 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(use (reg:SI PIC_REG))
(clobber (reg:SI PR_REG))
(clobber (match_scratch:SI 2 "=r"))]
"TARGET_SH2"
"#"
"reload_completed"
[(const_int 0)]
{
rtx lab = PATTERN (gen_call_site ());
if (SYMBOL_REF_LOCAL_P (operands[0]))
emit_insn (gen_sym_label2reg (operands[2], operands[0], lab));
else
emit_insn (gen_symPLT_label2reg (operands[2], operands[0], lab));
emit_call_insn (gen_calli_pcrel (operands[2], operands[1], copy_rtx (lab)));
DONE;
}
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")
(set_attr "fp_set" "unknown")])
(define_insn "call_compact"
[(call (mem:SI (match_operand:SI 0 "arith_reg_operand" "r"))
(match_operand 1 "" ""))
(match_operand 2 "immediate_operand" "n")
(use (reg:SI R0_REG))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT && ! (INTVAL (operands[2]) & CALL_COOKIE_RET_TRAMP (1))"
"jsr @%0%#"
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")])
(define_insn "call_compact_rettramp"
[(call (mem:SI (match_operand:SI 0 "arith_reg_operand" "r"))
(match_operand 1 "" ""))
(match_operand 2 "immediate_operand" "n")
(use (reg:SI R0_REG))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI R10_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT && (INTVAL (operands[2]) & CALL_COOKIE_RET_TRAMP (1))"
"jsr @%0%#"
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")])
(define_insn "call_media"
[(call (mem:DI (match_operand 0 "target_reg_operand" "b"))
(match_operand 1 "" ""))
(clobber (reg:DI PR_MEDIA_REG))]
"TARGET_SHMEDIA"
"blink %0, r18"
[(set_attr "type" "jump_media")])
(define_insn "call_valuei"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "arith_reg_operand" "r"))
(match_operand 2 "" "")))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SH1"
{
if (TARGET_SH2A && (dbr_sequence_length () == 0))
return "jsr/n @%1";
else
return "jsr @%1%#";
}
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")
(set_attr "fp_set" "unknown")])
;; This is TBR relative jump instruction for SH2A architecture.
;; Its use is enabled by assigning an attribute "function_vector"
;; and the vector number to a function during its declaration.
(define_insn "call_valuei_tbr_rel"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "symbol_ref_operand" ""))
(match_operand 2 "" "")))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SH2A && sh2a_is_function_vector_call (operands[1])"
{
unsigned HOST_WIDE_INT vect_num;
vect_num = sh2a_get_function_vector_number (operands[1]);
operands[3] = GEN_INT (vect_num * 4);
return "jsr/n @@(%O3,tbr)";
}
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "no")
(set_attr "fp_set" "unknown")])
(define_insn "call_valuei_pcrel"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "arith_reg_operand" "r"))
(match_operand 2 "" "")))
(use (reg:SI FPSCR_MODES_REG))
(use (reg:SI PIC_REG))
(use (match_operand 3 "" ""))
(clobber (reg:SI PR_REG))]
"TARGET_SH2"
{
return "bsrf %1" "\n"
"%O3:%#";
}
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")
(set_attr "fp_set" "unknown")])
(define_insn_and_split "call_value_pcrel"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "symbol_ref_operand" ""))
(match_operand 2 "" "")))
(use (reg:SI FPSCR_MODES_REG))
(use (reg:SI PIC_REG))
(clobber (reg:SI PR_REG))
(clobber (match_scratch:SI 3 "=r"))]
"TARGET_SH2"
"#"
"reload_completed"
[(const_int 0)]
{
rtx lab = PATTERN (gen_call_site ());
if (SYMBOL_REF_LOCAL_P (operands[1]))
emit_insn (gen_sym_label2reg (operands[3], operands[1], lab));
else
emit_insn (gen_symPLT_label2reg (operands[3], operands[1], lab));
emit_call_insn (gen_call_valuei_pcrel (operands[0], operands[3],
operands[2], copy_rtx (lab)));
DONE;
}
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")
(set_attr "fp_set" "unknown")])
(define_insn "call_value_compact"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "arith_reg_operand" "r"))
(match_operand 2 "" "")))
(match_operand 3 "immediate_operand" "n")
(use (reg:SI R0_REG))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT && ! (INTVAL (operands[3]) & CALL_COOKIE_RET_TRAMP (1))"
"jsr @%1%#"
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")])
(define_insn "call_value_compact_rettramp"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "arith_reg_operand" "r"))
(match_operand 2 "" "")))
(match_operand 3 "immediate_operand" "n")
(use (reg:SI R0_REG))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI R10_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT && (INTVAL (operands[3]) & CALL_COOKIE_RET_TRAMP (1))"
"jsr @%1%#"
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")])
(define_insn "call_value_media"
[(set (match_operand 0 "" "=rf")
(call (mem:DI (match_operand 1 "target_reg_operand" "b"))
(match_operand 2 "" "")))
(clobber (reg:DI PR_MEDIA_REG))]
"TARGET_SHMEDIA"
"blink %1, r18"
[(set_attr "type" "jump_media")])
(define_expand "call"
[(parallel [(call (mem:SI (match_operand 0 "arith_reg_operand" ""))
(match_operand 1 "" ""))
(match_operand 2 "" "")
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))])]
""
{
if (TARGET_SHMEDIA)
{
operands[0] = shmedia_prepare_call_address (operands[0], 0);
emit_call_insn (gen_call_media (operands[0], operands[1]));
DONE;
}
else if (TARGET_SHCOMPACT && operands[2] && INTVAL (operands[2]))
{
rtx cookie_rtx = operands[2];
long cookie = INTVAL (cookie_rtx);
rtx func = XEXP (operands[0], 0);
rtx r0, r1;
if (flag_pic)
{
if (GET_CODE (func) == SYMBOL_REF && ! SYMBOL_REF_LOCAL_P (func))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOTPLT2reg (reg, func));
func = reg;
}
else
func = legitimize_pic_address (func, Pmode, 0);
}
r0 = gen_rtx_REG (SImode, R0_REG);
r1 = gen_rtx_REG (SImode, R1_REG);
/* Since such a call function may use all call-clobbered
registers, we force a mode switch earlier, so that we don't
run out of registers when adjusting fpscr for the call. */
emit_insn (gen_force_mode_for_call ());
operands[0]
= function_symbol (NULL, "__GCC_shcompact_call_trampoline", SFUNC_GOT);
operands[0] = force_reg (SImode, operands[0]);
emit_move_insn (r0, func);
emit_move_insn (r1, cookie_rtx);
if (cookie & CALL_COOKIE_RET_TRAMP (1))
emit_call_insn (gen_call_compact_rettramp (operands[0], operands[1],
operands[2]));
else
emit_call_insn (gen_call_compact (operands[0], operands[1],
operands[2]));
DONE;
}
else if (TARGET_SHCOMPACT && flag_pic
&& GET_CODE (XEXP (operands[0], 0)) == SYMBOL_REF
&& ! SYMBOL_REF_LOCAL_P (XEXP (operands[0], 0)))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOTPLT2reg (reg, XEXP (operands[0], 0)));
XEXP (operands[0], 0) = reg;
}
if (!flag_pic && TARGET_SH2A
&& MEM_P (operands[0])
&& GET_CODE (XEXP (operands[0], 0)) == SYMBOL_REF)
{
if (sh2a_is_function_vector_call (XEXP (operands[0], 0)))
{
emit_call_insn (gen_calli_tbr_rel (XEXP (operands[0], 0),
operands[1]));
DONE;
}
}
if (flag_pic && TARGET_SH2
&& MEM_P (operands[0])
&& GET_CODE (XEXP (operands[0], 0)) == SYMBOL_REF)
{
emit_call_insn (gen_call_pcrel (XEXP (operands[0], 0), operands[1]));
DONE;
}
else
{
operands[0] = force_reg (SImode, XEXP (operands[0], 0));
operands[1] = operands[2];
}
emit_call_insn (gen_calli (operands[0], operands[1]));
DONE;
})
(define_insn "call_pop_compact"
[(call (mem:SI (match_operand:SI 0 "arith_reg_operand" "r"))
(match_operand 1 "" ""))
(match_operand 2 "immediate_operand" "n")
(set (reg:SI SP_REG) (plus:SI (reg:SI SP_REG)
(match_operand 3 "immediate_operand" "n")))
(use (reg:SI R0_REG))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT && ! (INTVAL (operands[2]) & CALL_COOKIE_RET_TRAMP (1))"
"jsr @%0%#"
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")])
(define_insn "call_pop_compact_rettramp"
[(call (mem:SI (match_operand:SI 0 "arith_reg_operand" "r"))
(match_operand 1 "" ""))
(match_operand 2 "immediate_operand" "n")
(set (reg:SI SP_REG) (plus:SI (reg:SI SP_REG)
(match_operand 3 "immediate_operand" "n")))
(use (reg:SI R0_REG))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI R10_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT && (INTVAL (operands[2]) & CALL_COOKIE_RET_TRAMP (1))"
"jsr @%0%#"
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")])
(define_expand "call_pop"
[(parallel [(call (mem:SI (match_operand 0 "arith_reg_operand" ""))
(match_operand 1 "" ""))
(match_operand 2 "" "")
(set (reg:SI SP_REG) (plus:SI (reg:SI SP_REG)
(match_operand 3 "" "")))])]
"TARGET_SHCOMPACT"
{
rtx cookie_rtx;
long cookie;
rtx func;
rtx r0, r1;
gcc_assert (operands[2] && INTVAL (operands[2]));
cookie_rtx = operands[2];
cookie = INTVAL (cookie_rtx);
func = XEXP (operands[0], 0);
if (flag_pic)
{
if (GET_CODE (func) == SYMBOL_REF && ! SYMBOL_REF_LOCAL_P (func))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOTPLT2reg (reg, func));
func = reg;
}
else
func = legitimize_pic_address (func, Pmode, 0);
}
r0 = gen_rtx_REG (SImode, R0_REG);
r1 = gen_rtx_REG (SImode, R1_REG);
/* Since such a call function may use all call-clobbered
registers, we force a mode switch earlier, so that we don't
run out of registers when adjusting fpscr for the call. */
emit_insn (gen_force_mode_for_call ());
operands[0] = function_symbol (NULL, "__GCC_shcompact_call_trampoline",
SFUNC_GOT);
operands[0] = force_reg (SImode, operands[0]);
emit_move_insn (r0, func);
emit_move_insn (r1, cookie_rtx);
if (cookie & CALL_COOKIE_RET_TRAMP (1))
emit_call_insn (gen_call_pop_compact_rettramp
(operands[0], operands[1], operands[2], operands[3]));
else
emit_call_insn (gen_call_pop_compact
(operands[0], operands[1], operands[2], operands[3]));
DONE;
})
(define_expand "call_value"
[(parallel [(set (match_operand 0 "arith_reg_operand" "")
(call (mem:SI (match_operand 1 "arith_reg_operand" ""))
(match_operand 2 "" "")))
(match_operand 3 "" "")
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))])]
""
{
if (TARGET_SHMEDIA)
{
operands[1] = shmedia_prepare_call_address (operands[1], 0);
emit_call_insn (gen_call_value_media (operands[0], operands[1],
operands[2]));
DONE;
}
else if (TARGET_SHCOMPACT && operands[3] && INTVAL (operands[3]))
{
rtx cookie_rtx = operands[3];
long cookie = INTVAL (cookie_rtx);
rtx func = XEXP (operands[1], 0);
rtx r0, r1;
if (flag_pic)
{
if (GET_CODE (func) == SYMBOL_REF && ! SYMBOL_REF_LOCAL_P (func))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOTPLT2reg (reg, func));
func = reg;
}
else
func = legitimize_pic_address (func, Pmode, 0);
}
r0 = gen_rtx_REG (SImode, R0_REG);
r1 = gen_rtx_REG (SImode, R1_REG);
/* Since such a call function may use all call-clobbered
registers, we force a mode switch earlier, so that we don't
run out of registers when adjusting fpscr for the call. */
emit_insn (gen_force_mode_for_call ());
operands[1]
= function_symbol (NULL, "__GCC_shcompact_call_trampoline", SFUNC_GOT);
operands[1] = force_reg (SImode, operands[1]);
emit_move_insn (r0, func);
emit_move_insn (r1, cookie_rtx);
if (cookie & CALL_COOKIE_RET_TRAMP (1))
emit_call_insn (gen_call_value_compact_rettramp (operands[0],
operands[1],
operands[2],
operands[3]));
else
emit_call_insn (gen_call_value_compact (operands[0], operands[1],
operands[2], operands[3]));
DONE;
}
else if (TARGET_SHCOMPACT && flag_pic
&& GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF
&& ! SYMBOL_REF_LOCAL_P (XEXP (operands[1], 0)))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOTPLT2reg (reg, XEXP (operands[1], 0)));
XEXP (operands[1], 0) = reg;
}
if (!flag_pic && TARGET_SH2A
&& MEM_P (operands[1])
&& GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF)
{
if (sh2a_is_function_vector_call (XEXP (operands[1], 0)))
{
emit_call_insn (gen_call_valuei_tbr_rel (operands[0],
XEXP (operands[1], 0), operands[2]));
DONE;
}
}
if (flag_pic && TARGET_SH2
&& MEM_P (operands[1])
&& GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF)
{
emit_call_insn (gen_call_value_pcrel (operands[0], XEXP (operands[1], 0),
operands[2]));
DONE;
}
else
operands[1] = force_reg (SImode, XEXP (operands[1], 0));
emit_call_insn (gen_call_valuei (operands[0], operands[1], operands[2]));
DONE;
})
(define_insn "sibcalli"
[(call (mem:SI (match_operand:SI 0 "register_operand" "k"))
(match_operand 1 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(return)]
"TARGET_SH1"
"jmp @%0%#"
[(set_attr "needs_delay_slot" "yes")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "type" "jump_ind")])
(define_insn "sibcalli_pcrel"
[(call (mem:SI (match_operand:SI 0 "arith_reg_operand" "k"))
(match_operand 1 "" ""))
(use (match_operand 2 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(return)]
"TARGET_SH2"
{
return "braf %0" "\n"
"%O2:%#";
}
[(set_attr "needs_delay_slot" "yes")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "type" "jump_ind")])
;; This uses an unspec to describe that the symbol_ref is very close.
(define_insn "sibcalli_thunk"
[(call (mem:SI (unspec:SI [(match_operand:SI 0 "symbol_ref_operand" "")]
UNSPEC_THUNK))
(match_operand 1 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(return)]
"TARGET_SH1"
"bra %O0"
[(set_attr "needs_delay_slot" "yes")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "type" "jump")
(set_attr "length" "2")])
(define_insn_and_split "sibcall_pcrel"
[(call (mem:SI (match_operand:SI 0 "symbol_ref_operand" ""))
(match_operand 1 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 2 "=k"))
(return)]
"TARGET_SH2"
"#"
"reload_completed"
[(const_int 0)]
{
rtx lab = PATTERN (gen_call_site ());
rtx call_insn;
emit_insn (gen_sym_label2reg (operands[2], operands[0], lab));
call_insn = emit_call_insn (gen_sibcalli_pcrel (operands[2], operands[1],
copy_rtx (lab)));
SIBLING_CALL_P (call_insn) = 1;
DONE;
}
[(set_attr "needs_delay_slot" "yes")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "type" "jump_ind")])
(define_insn "sibcall_compact"
[(call (mem:SI (match_operand:SI 0 "register_operand" "k,k"))
(match_operand 1 "" ""))
(return)
(use (match_operand:SI 2 "register_operand" "z,x"))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
;; We want to make sure the `x' above will only match MACH_REG
;; because sibcall_epilogue may clobber MACL_REG.
(clobber (reg:SI MACL_REG))]
"TARGET_SHCOMPACT"
{
static const char* alt[] =
{
"jmp @%0%#",
"jmp @%0" "\n"
" sts %2,r0"
};
return alt[which_alternative];
}
[(set_attr "needs_delay_slot" "yes,no")
(set_attr "length" "2,4")
(set (attr "fp_mode") (const_string "single"))
(set_attr "type" "jump_ind")])
(define_insn "sibcall_media"
[(call (mem:DI (match_operand 0 "target_reg_operand" "k"))
(match_operand 1 "" ""))
(use (reg:SI PR_MEDIA_REG))
(return)]
"TARGET_SHMEDIA"
"blink %0, r63"
[(set_attr "type" "jump_media")])
(define_expand "sibcall"
[(parallel
[(call (mem:SI (match_operand 0 "arith_reg_operand" ""))
(match_operand 1 "" ""))
(match_operand 2 "" "")
(use (reg:SI FPSCR_MODES_REG))
(return)])]
""
{
if (TARGET_SHMEDIA)
{
operands[0] = shmedia_prepare_call_address (operands[0], 1);
emit_call_insn (gen_sibcall_media (operands[0], operands[1]));
DONE;
}
else if (TARGET_SHCOMPACT && operands[2]
&& (INTVAL (operands[2]) & ~ CALL_COOKIE_RET_TRAMP (1)))
{
rtx cookie_rtx = operands[2];
long cookie = INTVAL (cookie_rtx);
rtx func = XEXP (operands[0], 0);
rtx mach, r1;
if (flag_pic)
{
if (GET_CODE (func) == SYMBOL_REF && ! SYMBOL_REF_LOCAL_P (func))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOT2reg (reg, func));
func = reg;
}
else
func = legitimize_pic_address (func, Pmode, 0);
}
/* FIXME: if we could tell whether all argument registers are
already taken, we could decide whether to force the use of
MACH_REG or to stick to R0_REG. Unfortunately, there's no
simple way to tell. We could use the CALL_COOKIE, but we
can't currently tell a register used for regular argument
passing from one that is unused. If we leave it up to reload
to decide which register to use, it seems to always choose
R0_REG, which leaves no available registers in SIBCALL_REGS
to hold the address of the trampoline. */
mach = gen_rtx_REG (SImode, MACH_REG);
r1 = gen_rtx_REG (SImode, R1_REG);
/* Since such a call function may use all call-clobbered
registers, we force a mode switch earlier, so that we don't
run out of registers when adjusting fpscr for the call. */
emit_insn (gen_force_mode_for_call ());
operands[0]
= function_symbol (NULL, "__GCC_shcompact_call_trampoline", SFUNC_GOT);
operands[0] = force_reg (SImode, operands[0]);
/* We don't need a return trampoline, since the callee will
return directly to the upper caller. */
if (cookie & CALL_COOKIE_RET_TRAMP (1))
{
cookie &= ~ CALL_COOKIE_RET_TRAMP (1);
cookie_rtx = GEN_INT (cookie);
}
emit_move_insn (mach, func);
emit_move_insn (r1, cookie_rtx);
emit_call_insn (gen_sibcall_compact (operands[0], operands[1], mach));
DONE;
}
else if (TARGET_SHCOMPACT && flag_pic
&& GET_CODE (XEXP (operands[0], 0)) == SYMBOL_REF
&& ! SYMBOL_REF_LOCAL_P (XEXP (operands[0], 0)))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOT2reg (reg, XEXP (operands[0], 0)));
XEXP (operands[0], 0) = reg;
}
if (flag_pic && TARGET_SH2
&& MEM_P (operands[0])
&& GET_CODE (XEXP (operands[0], 0)) == SYMBOL_REF
/* The PLT needs the PIC register, but the epilogue would have
to restore it, so we can only use PC-relative PIC calls for
static functions. */
&& SYMBOL_REF_LOCAL_P (XEXP (operands[0], 0)))
{
emit_call_insn (gen_sibcall_pcrel (XEXP (operands[0], 0), operands[1]));
DONE;
}
else
operands[0] = force_reg (SImode, XEXP (operands[0], 0));
emit_call_insn (gen_sibcalli (operands[0], operands[1]));
DONE;
})
(define_insn "sibcall_valuei"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "register_operand" "k"))
(match_operand 2 "" "")))
(use (reg:SI FPSCR_MODES_REG))
(return)]
"TARGET_SH1"
"jmp @%1%#"
[(set_attr "needs_delay_slot" "yes")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "type" "jump_ind")])
(define_insn "sibcall_valuei_pcrel"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "arith_reg_operand" "k"))
(match_operand 2 "" "")))
(use (match_operand 3 "" ""))
(use (reg:SI FPSCR_MODES_REG))
(return)]
"TARGET_SH2"
{
return "braf %1" "\n"
"%O3:%#";
}
[(set_attr "needs_delay_slot" "yes")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "type" "jump_ind")])
(define_insn_and_split "sibcall_value_pcrel"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "symbol_ref_operand" ""))
(match_operand 2 "" "")))
(use (reg:SI FPSCR_MODES_REG))
(clobber (match_scratch:SI 3 "=k"))
(return)]
"TARGET_SH2"
"#"
"reload_completed"
[(const_int 0)]
{
rtx lab = PATTERN (gen_call_site ());
rtx call_insn;
emit_insn (gen_sym_label2reg (operands[3], operands[1], lab));
call_insn = emit_call_insn (gen_sibcall_valuei_pcrel (operands[0],
operands[3],
operands[2],
copy_rtx (lab)));
SIBLING_CALL_P (call_insn) = 1;
DONE;
}
[(set_attr "needs_delay_slot" "yes")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "type" "jump_ind")])
(define_insn "sibcall_value_compact"
[(set (match_operand 0 "" "=rf,rf")
(call (mem:SI (match_operand:SI 1 "register_operand" "k,k"))
(match_operand 2 "" "")))
(return)
(use (match_operand:SI 3 "register_operand" "z,x"))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
;; We want to make sure the `x' above will only match MACH_REG
;; because sibcall_epilogue may clobber MACL_REG.
(clobber (reg:SI MACL_REG))]
"TARGET_SHCOMPACT"
{
static const char* alt[] =
{
"jmp @%1%#",
"jmp @%1" "\n"
" sts %3,r0"
};
return alt[which_alternative];
}
[(set_attr "needs_delay_slot" "yes,no")
(set_attr "length" "2,4")
(set (attr "fp_mode") (const_string "single"))
(set_attr "type" "jump_ind")])
(define_insn "sibcall_value_media"
[(set (match_operand 0 "" "=rf")
(call (mem:DI (match_operand 1 "target_reg_operand" "k"))
(match_operand 2 "" "")))
(use (reg:SI PR_MEDIA_REG))
(return)]
"TARGET_SHMEDIA"
"blink %1, r63"
[(set_attr "type" "jump_media")])
(define_expand "sibcall_value"
[(parallel
[(set (match_operand 0 "arith_reg_operand" "")
(call (mem:SI (match_operand 1 "arith_reg_operand" ""))
(match_operand 2 "" "")))
(match_operand 3 "" "")
(use (reg:SI FPSCR_MODES_REG))
(return)])]
""
{
if (TARGET_SHMEDIA)
{
operands[1] = shmedia_prepare_call_address (operands[1], 1);
emit_call_insn (gen_sibcall_value_media (operands[0], operands[1],
operands[2]));
DONE;
}
else if (TARGET_SHCOMPACT && operands[3]
&& (INTVAL (operands[3]) & ~ CALL_COOKIE_RET_TRAMP (1)))
{
rtx cookie_rtx = operands[3];
long cookie = INTVAL (cookie_rtx);
rtx func = XEXP (operands[1], 0);
rtx mach, r1;
if (flag_pic)
{
if (GET_CODE (func) == SYMBOL_REF && ! SYMBOL_REF_LOCAL_P (func))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOT2reg (reg, func));
func = reg;
}
else
func = legitimize_pic_address (func, Pmode, 0);
}
/* FIXME: if we could tell whether all argument registers are
already taken, we could decide whether to force the use of
MACH_REG or to stick to R0_REG. Unfortunately, there's no
simple way to tell. We could use the CALL_COOKIE, but we
can't currently tell a register used for regular argument
passing from one that is unused. If we leave it up to reload
to decide which register to use, it seems to always choose
R0_REG, which leaves no available registers in SIBCALL_REGS
to hold the address of the trampoline. */
mach = gen_rtx_REG (SImode, MACH_REG);
r1 = gen_rtx_REG (SImode, R1_REG);
/* Since such a call function may use all call-clobbered
registers, we force a mode switch earlier, so that we don't
run out of registers when adjusting fpscr for the call. */
emit_insn (gen_force_mode_for_call ());
operands[1]
= function_symbol (NULL, "__GCC_shcompact_call_trampoline", SFUNC_GOT);
operands[1] = force_reg (SImode, operands[1]);
/* We don't need a return trampoline, since the callee will
return directly to the upper caller. */
if (cookie & CALL_COOKIE_RET_TRAMP (1))
{
cookie &= ~ CALL_COOKIE_RET_TRAMP (1);
cookie_rtx = GEN_INT (cookie);
}
emit_move_insn (mach, func);
emit_move_insn (r1, cookie_rtx);
emit_call_insn (gen_sibcall_value_compact (operands[0], operands[1],
operands[2], mach));
DONE;
}
else if (TARGET_SHCOMPACT && flag_pic
&& GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF
&& ! SYMBOL_REF_LOCAL_P (XEXP (operands[1], 0)))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOT2reg (reg, XEXP (operands[1], 0)));
XEXP (operands[1], 0) = reg;
}
if (flag_pic && TARGET_SH2
&& MEM_P (operands[1])
&& GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF
/* The PLT needs the PIC register, but the epilogue would have
to restore it, so we can only use PC-relative PIC calls for
static functions. */
&& SYMBOL_REF_LOCAL_P (XEXP (operands[1], 0)))
{
emit_call_insn (gen_sibcall_value_pcrel (operands[0],
XEXP (operands[1], 0),
operands[2]));
DONE;
}
else
operands[1] = force_reg (SImode, XEXP (operands[1], 0));
emit_call_insn (gen_sibcall_valuei (operands[0], operands[1], operands[2]));
DONE;
})
(define_insn "call_value_pop_compact"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "arith_reg_operand" "r"))
(match_operand 2 "" "")))
(set (reg:SI SP_REG) (plus:SI (reg:SI SP_REG)
(match_operand 4 "immediate_operand" "n")))
(match_operand 3 "immediate_operand" "n")
(use (reg:SI R0_REG))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT && ! (INTVAL (operands[3]) & CALL_COOKIE_RET_TRAMP (1))"
"jsr @%1%#"
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")])
(define_insn "call_value_pop_compact_rettramp"
[(set (match_operand 0 "" "=rf")
(call (mem:SI (match_operand:SI 1 "arith_reg_operand" "r"))
(match_operand 2 "" "")))
(set (reg:SI SP_REG) (plus:SI (reg:SI SP_REG)
(match_operand 4 "immediate_operand" "n")))
(match_operand 3 "immediate_operand" "n")
(use (reg:SI R0_REG))
(use (reg:SI R1_REG))
(use (reg:SI FPSCR_MODES_REG))
(clobber (reg:SI R10_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT && (INTVAL (operands[3]) & CALL_COOKIE_RET_TRAMP (1))"
"jsr @%1%#"
[(set_attr "type" "call")
(set (attr "fp_mode")
(if_then_else (eq_attr "fpu_single" "yes")
(const_string "single") (const_string "double")))
(set_attr "needs_delay_slot" "yes")])
(define_expand "call_value_pop"
[(parallel [(set (match_operand 0 "arith_reg_operand" "")
(call (mem:SI (match_operand 1 "arith_reg_operand" ""))
(match_operand 2 "" "")))
(match_operand 3 "" "")
(set (reg:SI SP_REG) (plus:SI (reg:SI SP_REG)
(match_operand 4 "" "")))])]
"TARGET_SHCOMPACT"
{
rtx cookie_rtx;
long cookie;
rtx func;
rtx r0, r1;
gcc_assert (TARGET_SHCOMPACT && operands[3] && INTVAL (operands[3]));
cookie_rtx = operands[3];
cookie = INTVAL (cookie_rtx);
func = XEXP (operands[1], 0);
if (flag_pic)
{
if (GET_CODE (func) == SYMBOL_REF && ! SYMBOL_REF_LOCAL_P (func))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_symGOTPLT2reg (reg, func));
func = reg;
}
else
func = legitimize_pic_address (func, Pmode, 0);
}
r0 = gen_rtx_REG (SImode, R0_REG);
r1 = gen_rtx_REG (SImode, R1_REG);
/* Since such a call function may use all call-clobbered
registers, we force a mode switch earlier, so that we don't
run out of registers when adjusting fpscr for the call. */
emit_insn (gen_force_mode_for_call ());
operands[1] = function_symbol (NULL, "__GCC_shcompact_call_trampoline",
SFUNC_GOT);
operands[1] = force_reg (SImode, operands[1]);
emit_move_insn (r0, func);
emit_move_insn (r1, cookie_rtx);
if (cookie & CALL_COOKIE_RET_TRAMP (1))
emit_call_insn (gen_call_value_pop_compact_rettramp
(operands[0], operands[1], operands[2],
operands[3], operands[4]));
else
emit_call_insn (gen_call_value_pop_compact
(operands[0], operands[1], operands[2],
operands[3], operands[4]));
DONE;
})
(define_expand "sibcall_epilogue"
[(return)]
""
{
sh_expand_epilogue (true);
if (TARGET_SHCOMPACT)
{
rtx_insn *insn;
rtx set;
/* If epilogue clobbers r0, preserve it in macl. */
for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
if ((set = single_set (insn))
&& REG_P (SET_DEST (set))
&& REGNO (SET_DEST (set)) == R0_REG)
{
rtx r0 = gen_rtx_REG (SImode, R0_REG);
rtx tmp = gen_rtx_REG (SImode, MACL_REG);
/* We can't tell at this point whether the sibcall is a
sibcall_compact and, if it is, whether it uses r0 or
mach as operand 2, so let the instructions that
preserve r0 be optimized away if r0 turns out to be
dead. */
emit_insn_before (gen_rtx_SET (SImode, tmp, r0), insn);
emit_move_insn (r0, tmp);
break;
}
}
DONE;
})
(define_insn "indirect_jump_compact"
[(set (pc)
(match_operand:SI 0 "arith_reg_operand" "r"))]
"TARGET_SH1"
"jmp @%0%#"
[(set_attr "needs_delay_slot" "yes")
(set_attr "type" "jump_ind")])
(define_expand "indirect_jump"
[(set (pc)
(match_operand 0 "register_operand" ""))]
""
{
if (GET_MODE (operands[0]) != Pmode)
operands[0] = gen_rtx_SUBREG (Pmode, operands[0], 0);
})
;; The use of operand 1 / 2 helps us distinguish case table jumps
;; which can be present in structured code from indirect jumps which can not
;; be present in structured code. This allows -fprofile-arcs to work.
;; For SH1 processors.
(define_insn "casesi_jump_1"
[(set (pc)
(match_operand:SI 0 "register_operand" "r"))
(use (label_ref (match_operand 1 "" "")))]
"TARGET_SH1"
"jmp @%0%#"
[(set_attr "needs_delay_slot" "yes")
(set_attr "type" "jump_ind")])
;; For all later processors.
(define_insn "casesi_jump_2"
[(set (pc) (plus:SI (match_operand:SI 0 "register_operand" "r")
(label_ref (match_operand 1 "" ""))))
(use (label_ref (match_operand 2 "" "")))]
"TARGET_SH2
&& (! INSN_UID (operands[1]) || prev_real_insn (operands[1]) == insn)"
"braf %0%#"
[(set_attr "needs_delay_slot" "yes")
(set_attr "type" "jump_ind")])
(define_insn "casesi_jump_media"
[(set (pc) (match_operand 0 "target_reg_operand" "b"))
(use (label_ref (match_operand 1 "" "")))]
"TARGET_SHMEDIA"
"blink %0, r63"
[(set_attr "type" "jump_media")])
;; Call subroutine returning any type.
;; ??? This probably doesn't work.
(define_expand "untyped_call"
[(parallel [(call (match_operand 0 "" "")
(const_int 0))
(match_operand 1 "" "")
(match_operand 2 "" "")])]
"(TARGET_SH2E || TARGET_SH2A) || TARGET_SHMEDIA"
{
emit_call_insn (gen_call (operands[0], const0_rtx, const0_rtx));
for (int i = 0; i < XVECLEN (operands[2], 0); i++)
{
rtx set = XVECEXP (operands[2], 0, i);
emit_move_insn (SET_DEST (set), SET_SRC (set));
}
/* The optimizer does not know that the call sets the function value
registers we stored in the result block. We avoid problems by
claiming that all hard registers are used and clobbered at this
point. */
emit_insn (gen_blockage ());
DONE;
})
;; ------------------------------------------------------------------------
;; Misc insns
;; ------------------------------------------------------------------------
(define_insn "dect"
[(set (reg:SI T_REG)
(eq:SI (match_operand:SI 1 "arith_reg_dest" "0") (const_int 1)))
(set (match_operand:SI 0 "arith_reg_dest" "=r")
(plus:SI (match_dup 1) (const_int -1)))]
"TARGET_SH2"
"dt %0"
[(set_attr "type" "arith")])
(define_insn "nop"
[(const_int 0)]
""
"nop")
;; Load address of a label. This is only generated by the casesi expand,
;; and by machine_dependent_reorg (fixing up fp moves).
;; This must use unspec, because this only works for labels that are
;; within range.
(define_insn "mova"
[(set (reg:SI R0_REG)
(unspec:SI [(label_ref (match_operand 0 "" ""))] UNSPEC_MOVA))]
"TARGET_SH1"
"mova %O0,r0"
[(set_attr "in_delay_slot" "no")
(set_attr "type" "arith")])
;; machine_dependent_reorg will make this a `mova'.
(define_insn "mova_const"
[(set (reg:SI R0_REG)
(unspec:SI [(match_operand 0 "immediate_operand" "i")] UNSPEC_MOVA))]
"TARGET_SH1"
"#"
[(set_attr "in_delay_slot" "no")
(set_attr "type" "arith")])
(define_expand "GOTaddr2picreg"
[(set (reg:SI R0_REG)
(unspec:SI [(const:SI (unspec:SI [(match_dup 1)] UNSPEC_PIC))]
UNSPEC_MOVA))
(set (match_dup 0) (const:SI (unspec:SI [(match_dup 1)] UNSPEC_PIC)))
(set (match_dup 0) (plus:SI (match_dup 0) (reg:SI R0_REG)))]
""
{
if (TARGET_VXWORKS_RTP)
{
rtx gott_base = gen_rtx_SYMBOL_REF (Pmode, VXWORKS_GOTT_BASE);
rtx gott_index = gen_rtx_SYMBOL_REF (Pmode, VXWORKS_GOTT_INDEX);
emit_insn (gen_vxworks_picreg (gott_base, gott_index));
DONE;
}
operands[0] = gen_rtx_REG (Pmode, PIC_REG);
operands[1] = gen_rtx_SYMBOL_REF (VOIDmode, GOT_SYMBOL_NAME);
if (TARGET_SHMEDIA)
{
rtx tr = gen_rtx_REG (Pmode, TR0_REG);
rtx pic = operands[0];
rtx lab = PATTERN (gen_call_site ());
rtx insn, equiv;
equiv = operands[1];
operands[1] = gen_rtx_UNSPEC (Pmode, gen_rtvec (2, operands[1], lab),
UNSPEC_PCREL_SYMOFF);
operands[1] = gen_rtx_CONST (Pmode, operands[1]);
if (Pmode == SImode)
{
emit_insn (gen_movsi_const (pic, operands[1]));
emit_insn (gen_ptrel_si (tr, pic, copy_rtx (lab)));
}
else
{
emit_insn (gen_movdi_const (pic, operands[1]));
emit_insn (gen_ptrel_di (tr, pic, copy_rtx (lab)));
}
insn = emit_move_insn (operands[0], tr);
set_unique_reg_note (insn, REG_EQUAL, equiv);
DONE;
}
})
;; A helper for GOTaddr2picreg to finish up the initialization of the
;; PIC register.
(define_expand "vxworks_picreg"
[(set (reg:SI PIC_REG)
(const:SI (unspec:SI [(match_operand:SI 0 "" "")] UNSPEC_PIC)))
(set (reg:SI R0_REG)
(const:SI (unspec:SI [(match_operand:SI 1 "" "")] UNSPEC_PIC)))
(set (reg:SI PIC_REG)
(mem:SI (reg:SI PIC_REG)))
(set (reg:SI PIC_REG)
(mem:SI (plus:SI (reg:SI PIC_REG)
(reg:SI R0_REG))))]
"TARGET_VXWORKS_RTP")
(define_insn "*ptb"
[(set (match_operand 0 "target_reg_operand" "=b")
(const (unspec [(match_operand 1 "" "Csy")]
UNSPEC_DATALABEL)))]
"TARGET_SHMEDIA && flag_pic
&& satisfies_constraint_Csy (operands[1])"
"ptb/u datalabel %1, %0"
[(set_attr "type" "ptabs_media")
(set_attr "length" "*")])
(define_insn "ptrel_si"
[(set (match_operand:SI 0 "target_reg_operand" "=b")
(plus:SI (match_operand:SI 1 "register_operand" "r")
(pc)))
(match_operand:SI 2 "" "")]
"TARGET_SHMEDIA"
"%O2: ptrel/u %1, %0"
[(set_attr "type" "ptabs_media")])
(define_insn "ptrel_di"
[(set (match_operand:DI 0 "target_reg_operand" "=b")
(plus:DI (match_operand:DI 1 "register_operand" "r")
(pc)))
(match_operand:DI 2 "" "")]
"TARGET_SHMEDIA"
"%O2: ptrel/u %1, %0"
[(set_attr "type" "ptabs_media")])
(define_expand "builtin_setjmp_receiver"
[(match_operand 0 "" "")]
"flag_pic"
{
emit_insn (gen_GOTaddr2picreg ());
DONE;
})
(define_expand "call_site"
[(unspec [(match_dup 0)] UNSPEC_CALLER)]
"TARGET_SH1"
{
static HOST_WIDE_INT i = 0;
operands[0] = GEN_INT (i);
i++;
})
;; op0 = op1 + r12 but hide it before reload completed. See the comment
;; in symGOT_load expand.
(define_insn_and_split "chk_guard_add"
[(set (match_operand:SI 0 "register_operand" "=&r")
(unspec:SI [(match_operand:SI 1 "register_operand" "r")
(reg:SI PIC_REG)]
UNSPEC_CHKADD))]
"TARGET_SH1"
"#"
"TARGET_SH1 && reload_completed"
[(set (match_dup 0) (reg:SI PIC_REG))
(set (match_dup 0) (plus:SI (match_dup 0) (match_dup 1)))]
""
[(set_attr "type" "arith")])
(define_expand "sym_label2reg"
[(set (match_operand:SI 0 "" "")
(const:SI (unspec:SI [(match_operand:SI 1 "" "")
(const (plus:SI (match_operand:SI 2 "" "")
(const_int 2)))]
UNSPEC_SYMOFF)))]
"TARGET_SH1" "")
(define_expand "symGOT_load"
[(set (match_dup 2) (match_operand 1 "" ""))
(set (match_dup 3) (plus (match_dup 2) (reg PIC_REG)))
(set (match_operand 0 "" "") (mem (match_dup 3)))]
""
{
rtx mem;
operands[2] = !can_create_pseudo_p () ? operands[0] : gen_reg_rtx (Pmode);
operands[3] = !can_create_pseudo_p () ? operands[0] : gen_reg_rtx (Pmode);
if (TARGET_SHMEDIA)
{
rtx reg = operands[2];
if (Pmode == DImode)
{
if (flag_pic > 1)
emit_insn (gen_movdi_const_32bit (reg, operands[1]));
else
emit_insn (gen_movdi_const_16bit (reg, operands[1]));
}
else
{
if (flag_pic > 1)
emit_insn (gen_movsi_const (reg, operands[1]));
else
emit_insn (gen_movsi_const_16bit (reg, operands[1]));
}
}
else
emit_move_insn (operands[2], operands[1]);
/* When stack protector inserts codes after the result is set to
R0, @(rX, r12) will cause a spill failure for R0. Use a unspec
insn to avoid combining (set A (plus rX r12)) and (set op0 (mem A))
when rX is a GOT address for the guard symbol. Ugly but doesn't
matter because this is a rare situation. */
if (!TARGET_SHMEDIA
&& flag_stack_protect
&& GET_CODE (operands[1]) == CONST
&& GET_CODE (XEXP (operands[1], 0)) == UNSPEC
&& GET_CODE (XVECEXP (XEXP (operands[1], 0), 0, 0)) == SYMBOL_REF
&& strcmp (XSTR (XVECEXP (XEXP (operands[1], 0), 0, 0), 0),
"__stack_chk_guard") == 0)
emit_insn (gen_chk_guard_add (operands[3], operands[2]));
else
emit_move_insn (operands[3], gen_rtx_PLUS (Pmode, operands[2],
gen_rtx_REG (Pmode, PIC_REG)));
/* N.B. This is not constant for a GOTPLT relocation. */
mem = gen_rtx_MEM (Pmode, operands[3]);
MEM_NOTRAP_P (mem) = 1;
/* ??? Should we have a special alias set for the GOT? */
emit_move_insn (operands[0], mem);
DONE;
})
(define_expand "sym2GOT"
[(const (unspec [(match_operand 0 "" "")] UNSPEC_GOT))]
""
"")
(define_expand "symGOT2reg"
[(match_operand 0 "" "") (match_operand 1 "" "")]
""
{
rtx gotsym, insn;
gotsym = gen_sym2GOT (operands[1]);
PUT_MODE (gotsym, Pmode);
insn = emit_insn (gen_symGOT_load (operands[0], gotsym));
MEM_READONLY_P (SET_SRC (PATTERN (insn))) = 1;
DONE;
})
(define_expand "symGOTPLT2reg"
[(match_operand 0 "" "") (match_operand 1 "" "")]
""
{
rtx pltsym = gen_rtx_CONST (Pmode,
gen_rtx_UNSPEC (Pmode,
gen_rtvec (1, operands[1]),
UNSPEC_GOTPLT));
emit_insn (gen_symGOT_load (operands[0], pltsym));
DONE;
})
(define_expand "sym2GOTOFF"
[(const (unspec [(match_operand 0 "" "")] UNSPEC_GOTOFF))]
""
"")
(define_expand "symGOTOFF2reg"
[(match_operand 0 "" "") (match_operand 1 "" "")]
""
{
rtx gotoffsym, insn;
rtx t = (!can_create_pseudo_p ()
? operands[0]
: gen_reg_rtx (GET_MODE (operands[0])));
gotoffsym = gen_sym2GOTOFF (operands[1]);
PUT_MODE (gotoffsym, Pmode);
emit_move_insn (t, gotoffsym);
insn = emit_move_insn (operands[0],
gen_rtx_PLUS (Pmode, t,
gen_rtx_REG (Pmode, PIC_REG)));
set_unique_reg_note (insn, REG_EQUAL, operands[1]);
DONE;
})
(define_expand "symPLT_label2reg"
[(set (match_operand:SI 0 "" "")
(const:SI
(unspec:SI
[(const:SI (unspec:SI [(match_operand:SI 1 "" "")] UNSPEC_PLT))
(const:SI (plus:SI (match_operand:SI 2 "" "")
(const_int 2)))] UNSPEC_PCREL_SYMOFF)))
;; Even though the PIC register is not really used by the call
;; sequence in which this is expanded, the PLT code assumes the PIC
;; register is set, so we must not skip its initialization. Since
;; we only use this expand as part of calling sequences, and never
;; to take the address of a function, this is the best point to
;; insert the (use). Using the PLT to take the address of a
;; function would be wrong, not only because the PLT entry could
;; then be called from a function that doesn't initialize the PIC
;; register to the proper GOT, but also because pointers to the
;; same function might not compare equal, should they be set by
;; different shared libraries.
(use (reg:SI PIC_REG))]
"TARGET_SH1"
"")
(define_expand "sym2PIC"
[(const (unspec [(match_operand:SI 0 "" "")] UNSPEC_PIC))]
""
"")
;; -------------------------------------------------------------------------
;; TLS code generation.
;; FIXME: The multi-insn asm blocks should be converted to use
;; define_insn_and_split.
;; See the thread [PATCH/RFA] SH TLS support on gcc-patches
;;
;; for details.
(define_insn "tls_global_dynamic"
[(set (match_operand:SI 0 "register_operand" "=&z")
(call:SI (mem:SI (unspec:SI [(match_operand:SI 1 "" "")]
UNSPEC_TLSGD))
(const_int 0)))
(use (reg:SI FPSCR_MODES_REG))
(use (reg:SI PIC_REG))
(clobber (reg:SI PR_REG))
(clobber (scratch:SI))]
"TARGET_SH1"
{
return "mov.l 1f,r4" "\n"
" mova 2f,r0" "\n"
" mov.l 2f,r1" "\n"
" add r0,r1" "\n"
" jsr @r1" "\n"
" add r12,r4" "\n"
" bra 3f" "\n"
" nop" "\n"
" .align 2" "\n"
"1: .long %a1@TLSGD" "\n"
"2: .long __tls_get_addr@PLT" "\n"
"3:";
}
[(set_attr "type" "tls_load")
(set_attr "length" "26")])
(define_insn "tls_local_dynamic"
[(set (match_operand:SI 0 "register_operand" "=&z")
(call:SI (mem:SI (unspec:SI [(match_operand:SI 1 "" "")]
UNSPEC_TLSLDM))
(const_int 0)))
(use (reg:SI FPSCR_MODES_REG))
(use (reg:SI PIC_REG))
(clobber (reg:SI PR_REG))
(clobber (scratch:SI))]
"TARGET_SH1"
{
return "mov.l 1f,r4" "\n"
" mova 2f,r0" "\n"
" mov.l 2f,r1" "\n"
" add r0,r1" "\n"
" jsr @r1" "\n"
" add r12,r4" "\n"
" bra 3f" "\n"
" nop" "\n"
" .align 2" "\n"
"1: .long %a1@TLSLDM" "\n"
"2: .long __tls_get_addr@PLT" "\n"
"3:";
}
[(set_attr "type" "tls_load")
(set_attr "length" "26")])
(define_expand "sym2DTPOFF"
[(const (unspec [(match_operand 0 "" "")] UNSPEC_DTPOFF))]
""
"")
(define_expand "symDTPOFF2reg"
[(match_operand 0 "" "") (match_operand 1 "" "") (match_operand 2 "" "")]
""
{
rtx dtpoffsym;
rtx t = (!can_create_pseudo_p ()
? operands[0]
: gen_reg_rtx (GET_MODE (operands[0])));
dtpoffsym = gen_sym2DTPOFF (operands[1]);
PUT_MODE (dtpoffsym, Pmode);
emit_move_insn (t, dtpoffsym);
emit_move_insn (operands[0], gen_rtx_PLUS (Pmode, t, operands[2]));
DONE;
})
(define_expand "sym2GOTTPOFF"
[(const (unspec [(match_operand 0 "" "")] UNSPEC_GOTTPOFF))]
""
"")
(define_insn "tls_initial_exec"
[(set (match_operand:SI 0 "register_operand" "=&r")
(unspec:SI [(match_operand:SI 1 "" "")]
UNSPEC_TLSIE))
(use (reg:SI GBR_REG))
(use (reg:SI PIC_REG))
(clobber (reg:SI R0_REG))]
""
{
return "mov.l 1f,r0" "\n"
" stc gbr,%0" "\n"
" mov.l @(r0,r12),r0" "\n"
" bra 2f" "\n"
" add r0,%0" "\n"
" .align 2" "\n"
"1: .long %a1" "\n"
"2:";
}
[(set_attr "type" "tls_load")
(set_attr "length" "16")])
(define_expand "sym2TPOFF"
[(const (unspec [(match_operand 0 "" "")] UNSPEC_TPOFF))]
""
"")
(define_expand "symTPOFF2reg"
[(match_operand 0 "" "") (match_operand 1 "" "")]
""
{
rtx tpoffsym;
tpoffsym = gen_sym2TPOFF (operands[1]);
PUT_MODE (tpoffsym, Pmode);
emit_move_insn (operands[0], tpoffsym);
DONE;
})
;;------------------------------------------------------------------------------
;; Thread pointer getter and setter.
;;
;; On SH the thread pointer is kept in the GBR.
;; These patterns are usually expanded from the respective built-in functions.
(define_expand "get_thread_pointersi"
[(set (match_operand:SI 0 "register_operand") (reg:SI GBR_REG))]
"TARGET_SH1")
;; The store_gbr insn can also be used on !TARGET_SH1 for doing TLS accesses.
(define_insn "store_gbr"
[(set (match_operand:SI 0 "register_operand" "=r") (reg:SI GBR_REG))]
""
"stc gbr,%0"
[(set_attr "type" "tls_load")])
(define_expand "set_thread_pointersi"
[(set (reg:SI GBR_REG)
(unspec_volatile:SI [(match_operand:SI 0 "register_operand")]
UNSPECV_GBR))]
"TARGET_SH1")
(define_insn "load_gbr"
[(set (reg:SI GBR_REG)
(unspec_volatile:SI [(match_operand:SI 0 "register_operand" "r")]
UNSPECV_GBR))]
"TARGET_SH1"
"ldc %0,gbr"
[(set_attr "type" "move")])
;;------------------------------------------------------------------------------
;; Thread pointer relative memory loads and stores.
;;
;; On SH there are GBR displacement address modes which can be utilized to
;; access memory behind the thread pointer.
;; Since we do not allow using GBR for general purpose memory accesses, these
;; GBR addressing modes are formed by the combine pass.
;; This could be done with fewer patterns than below by using a mem predicate
;; for the GBR mem, but then reload would try to reload addresses with a
;; zero displacement for some strange reason.
(define_insn "*mov_gbr_load"
[(set (match_operand:QIHISI 0 "register_operand" "=z")
(mem:QIHISI (plus:SI (reg:SI GBR_REG)
(match_operand:QIHISI 1 "gbr_displacement"))))]
"TARGET_SH1"
"mov. @(%O1,gbr),%0"
[(set_attr "type" "load")])
(define_insn "*mov_gbr_load"
[(set (match_operand:QIHISI 0 "register_operand" "=z")
(mem:QIHISI (reg:SI GBR_REG)))]
"TARGET_SH1"
"mov. @(0,gbr),%0"
[(set_attr "type" "load")])
(define_insn "*mov_gbr_load"
[(set (match_operand:SI 0 "register_operand" "=z")
(sign_extend:SI
(mem:QIHI (plus:SI (reg:SI GBR_REG)
(match_operand:QIHI 1 "gbr_displacement")))))]
"TARGET_SH1"
"mov. @(%O1,gbr),%0"
[(set_attr "type" "load")])
(define_insn "*mov_gbr_load"
[(set (match_operand:SI 0 "register_operand" "=z")
(sign_extend:SI (mem:QIHI (reg:SI GBR_REG))))]
"TARGET_SH1"
"mov. @(0,gbr),%0"
[(set_attr "type" "load")])
(define_insn "*mov_gbr_store"
[(set (mem:QIHISI (plus:SI (reg:SI GBR_REG)
(match_operand:QIHISI 0 "gbr_displacement")))
(match_operand:QIHISI 1 "register_operand" "z"))]
"TARGET_SH1"
"mov. %1,@(%O0,gbr)"
[(set_attr "type" "store")])
(define_insn "*mov_gbr_store"
[(set (mem:QIHISI (reg:SI GBR_REG))
(match_operand:QIHISI 0 "register_operand" "z"))]
"TARGET_SH1"
"mov. %0,@(0,gbr)"
[(set_attr "type" "store")])
;; DImode memory accesses have to be split in two SImode accesses.
;; Split them before reload, so that it gets a better chance to figure out
;; how to deal with the R0 restriction for the individual SImode accesses.
;; Do not match this insn during or after reload because it can't be split
;; afterwards.
(define_insn_and_split "*movdi_gbr_load"
[(set (match_operand:DI 0 "register_operand")
(match_operand:DI 1 "gbr_address_mem"))]
"TARGET_SH1 && can_create_pseudo_p ()"
"#"
"&& 1"
[(set (match_dup 3) (match_dup 5))
(set (match_dup 4) (match_dup 6))]
{
/* Swap low/high part load order on little endian, so that the result reg
of the second load can be used better. */
int off = TARGET_LITTLE_ENDIAN ? 1 : 0;
operands[3 + off] = gen_lowpart (SImode, operands[0]);
operands[5 + off] = gen_lowpart (SImode, operands[1]);
operands[4 - off] = gen_highpart (SImode, operands[0]);
operands[6 - off] = gen_highpart (SImode, operands[1]);
})
(define_insn_and_split "*movdi_gbr_store"
[(set (match_operand:DI 0 "gbr_address_mem")
(match_operand:DI 1 "register_operand"))]
"TARGET_SH1 && can_create_pseudo_p ()"
"#"
"&& 1"
[(set (match_dup 3) (match_dup 5))
(set (match_dup 4) (match_dup 6))]
{
/* Swap low/high part store order on big endian, so that stores of function
call results can save a reg copy. */
int off = TARGET_LITTLE_ENDIAN ? 0 : 1;
operands[3 + off] = gen_lowpart (SImode, operands[0]);
operands[5 + off] = gen_lowpart (SImode, operands[1]);
operands[4 - off] = gen_highpart (SImode, operands[0]);
operands[6 - off] = gen_highpart (SImode, operands[1]);
})
;; Sometimes memory accesses do not get combined with the store_gbr insn,
;; in particular when the displacements are in the range of the regular move
;; insns. Thus, in the first split pass after the combine pass we search
;; for missed opportunities and try to fix them up ourselves.
;; If an equivalent GBR address can be determined the load / store is split
;; into one of the GBR load / store patterns.
;; All of that must happen before reload (GBR address modes use R0 as the
;; other operand) and there's no point of doing it if the GBR is not
;; referenced in a function at all.
(define_split
[(set (match_operand:QIHISIDI 0 "register_operand")
(match_operand:QIHISIDI 1 "memory_operand"))]
"TARGET_SH1 && !reload_in_progress && !reload_completed
&& df_regs_ever_live_p (GBR_REG)"
[(set (match_dup 0) (match_dup 1))]
{
rtx gbr_mem = sh_find_equiv_gbr_addr (curr_insn, operands[1]);
if (gbr_mem != NULL_RTX)
operands[1] = replace_equiv_address (operands[1], gbr_mem);
else
FAIL;
})
(define_split
[(set (match_operand:SI 0 "register_operand")
(sign_extend:SI (match_operand:QIHI 1 "memory_operand")))]
"TARGET_SH1 && !reload_in_progress && !reload_completed
&& df_regs_ever_live_p (GBR_REG)"
[(set (match_dup 0) (sign_extend:SI (match_dup 1)))]
{
rtx gbr_mem = sh_find_equiv_gbr_addr (curr_insn, operands[1]);
if (gbr_mem != NULL_RTX)
operands[1] = replace_equiv_address (operands[1], gbr_mem);
else
FAIL;
})
;; On SH2A we've got movu.b and movu.w for doing zero-extending mem loads.
;; Split those so that a GBR load can be used.
(define_split
[(set (match_operand:SI 0 "register_operand")
(zero_extend:SI (match_operand:QIHI 1 "memory_operand")))]
"TARGET_SH2A && !reload_in_progress && !reload_completed
&& df_regs_ever_live_p (GBR_REG)"
[(set (match_dup 2) (match_dup 1))
(set (match_dup 0) (zero_extend:SI (match_dup 2)))]
{
rtx gbr_mem = sh_find_equiv_gbr_addr (curr_insn, operands[1]);
if (gbr_mem != NULL_RTX)
{
operands[2] = gen_reg_rtx (GET_MODE (operands[1]));
operands[1] = replace_equiv_address (operands[1], gbr_mem);
}
else
FAIL;
})
(define_split
[(set (match_operand:QIHISIDI 0 "memory_operand")
(match_operand:QIHISIDI 1 "register_operand"))]
"TARGET_SH1 && !reload_in_progress && !reload_completed
&& df_regs_ever_live_p (GBR_REG)"
[(set (match_dup 0) (match_dup 1))]
{
rtx gbr_mem = sh_find_equiv_gbr_addr (curr_insn, operands[0]);
if (gbr_mem != NULL_RTX)
operands[0] = replace_equiv_address (operands[0], gbr_mem);
else
FAIL;
})
;;------------------------------------------------------------------------------
;; case instruction for switch statements.
;; operand 0 is index
;; operand 1 is the minimum bound
;; operand 2 is the maximum bound - minimum bound + 1
;; operand 3 is CODE_LABEL for the table;
;; operand 4 is the CODE_LABEL to go to if index out of range.
(define_expand "casesi"
[(match_operand:SI 0 "arith_reg_operand" "")
(match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "arith_reg_operand" "")
(match_operand 3 "" "") (match_operand 4 "" "")]
""
{
rtx reg = gen_reg_rtx (SImode);
rtx reg2 = gen_reg_rtx (SImode);
if (TARGET_SHMEDIA)
{
rtx reg = gen_reg_rtx (DImode);
rtx reg2 = gen_reg_rtx (DImode);
rtx reg3 = gen_reg_rtx (Pmode);
rtx reg4 = gen_reg_rtx (Pmode);
rtx reg5 = gen_reg_rtx (Pmode);
rtx load, test;
operands[0] = convert_modes (DImode, SImode, operands[0], 0);
operands[1] = convert_modes (DImode, SImode, operands[1], 0);
operands[2] = convert_modes (DImode, SImode, operands[2], 1);
test = gen_rtx_GT (VOIDmode, operands[1], operands[0]);
emit_jump_insn (gen_cbranchdi4 (test, operands[1], operands[0],
operands[4]));
emit_move_insn (reg, gen_rtx_MINUS (DImode, operands[0], operands[1]));
test = gen_rtx_GTU (VOIDmode, reg, operands[2]);
emit_jump_insn (gen_cbranchdi4 (test, reg, operands[2], operands[4]));
emit_insn (gen_casesi_shift_media (reg2, reg, operands[3]));
emit_move_insn (reg3, gen_datalabel_ref (gen_rtx_LABEL_REF
(Pmode, operands[3])));
/* Messy: can we subreg to clean this up? */
if (Pmode == DImode)
load = gen_casesi_load_media (reg4, reg3, reg2, operands[3]);
else
load = gen_casesi_load_media (reg4,
gen_rtx_SUBREG (DImode, reg3, 0),
reg2, operands[3]);
PUT_MODE (SET_SRC (load), Pmode);
emit_insn (load);
/* ??? The following add could be eliminated if we used ptrel. */
emit_move_insn (reg5, gen_rtx_PLUS (Pmode, reg3, reg4));
emit_jump_insn (gen_casesi_jump_media (reg5, operands[3]));
emit_barrier ();
DONE;
}
operands[1] = copy_to_mode_reg (SImode, operands[1]);
operands[2] = copy_to_mode_reg (SImode, operands[2]);
/* If optimizing, casesi_worker depends on the mode of the instruction
before label it 'uses' - operands[3]. */
emit_insn (gen_casesi_0 (operands[0], operands[1], operands[2], operands[4],
reg));
emit_insn (gen_casesi_worker_0 (reg2, reg, operands[3]));
if (TARGET_SH2)
emit_jump_insn (gen_casesi_jump_2 (reg2, gen_label_rtx (), operands[3]));
else
emit_jump_insn (gen_casesi_jump_1 (reg2, operands[3]));
/* For SH2 and newer, the ADDR_DIFF_VEC is not actually relative to
operands[3], but to lab. We will fix this up in
machine_dependent_reorg. */
emit_barrier ();
DONE;
})
(define_expand "casesi_0"
[(set (match_operand:SI 4 "" "") (match_operand:SI 0 "arith_reg_operand" ""))
(set (match_dup 4) (minus:SI (match_dup 4)
(match_operand:SI 1 "arith_operand" "")))
(set (reg:SI T_REG)
(gtu:SI (match_dup 4)
(match_operand:SI 2 "arith_reg_operand" "")))
(set (pc)
(if_then_else (ne (reg:SI T_REG)
(const_int 0))
(label_ref (match_operand 3 "" ""))
(pc)))]
"TARGET_SH1"
"")
;; ??? reload might clobber r0 if we use it explicitly in the RTL before
;; reload; using a R0_REGS pseudo reg is likely to give poor code.
;; So we keep the use of r0 hidden in a R0_REGS clobber until after reload.
(define_insn "casesi_worker_0"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(unspec:SI [(match_operand:SI 1 "register_operand" "0,r")
(label_ref (match_operand 2 "" ""))] UNSPEC_CASESI))
(clobber (match_scratch:SI 3 "=X,1"))
(clobber (match_scratch:SI 4 "=&z,z"))]
"TARGET_SH1"
"#")
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(unspec:SI [(match_operand:SI 1 "register_operand" "")
(label_ref (match_operand 2 "" ""))] UNSPEC_CASESI))
(clobber (match_scratch:SI 3 ""))
(clobber (match_scratch:SI 4 ""))]
"TARGET_SH1 && ! TARGET_SH2 && reload_completed"
[(set (reg:SI R0_REG) (unspec:SI [(label_ref (match_dup 2))] UNSPEC_MOVA))
(parallel [(set (match_dup 0)
(unspec:SI [(reg:SI R0_REG) (match_dup 1)
(label_ref (match_dup 2))] UNSPEC_CASESI))
(clobber (match_dup 3))])
(set (match_dup 0) (plus:SI (match_dup 0) (reg:SI R0_REG)))]
{
if (GET_CODE (operands[2]) == CODE_LABEL)
LABEL_NUSES (operands[2])++;
})
(define_split
[(set (match_operand:SI 0 "register_operand" "")
(unspec:SI [(match_operand:SI 1 "register_operand" "")
(label_ref (match_operand 2 "" ""))] UNSPEC_CASESI))
(clobber (match_scratch:SI 3 ""))
(clobber (match_scratch:SI 4 ""))]
"TARGET_SH2 && reload_completed"
[(set (reg:SI R0_REG) (unspec:SI [(label_ref (match_dup 2))] UNSPEC_MOVA))
(parallel [(set (match_dup 0)
(unspec:SI [(reg:SI R0_REG) (match_dup 1)
(label_ref (match_dup 2))] UNSPEC_CASESI))
(clobber (match_dup 3))])]
{
if (GET_CODE (operands[2]) == CODE_LABEL)
LABEL_NUSES (operands[2])++;
})
(define_insn "casesi_worker_1"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(unspec:SI [(reg:SI R0_REG)
(match_operand:SI 1 "register_operand" "0,r")
(label_ref (match_operand 2 "" ""))] UNSPEC_CASESI))
(clobber (match_scratch:SI 3 "=X,1"))]
"TARGET_SH1"
{
rtx diff_vec = PATTERN (NEXT_INSN (as_a (operands[2])));
gcc_assert (GET_CODE (diff_vec) == ADDR_DIFF_VEC);
switch (GET_MODE (diff_vec))
{
case SImode:
return "shll2 %1" "\n"
" mov.l @(r0,%1),%0";
case HImode:
return "add %1,%1" "\n"
" mov.w @(r0,%1),%0";
case QImode:
if (ADDR_DIFF_VEC_FLAGS (diff_vec).offset_unsigned)
return "mov.b @(r0,%1),%0" "\n"
" extu.b %0,%0";
else
return "mov.b @(r0,%1),%0";
default:
gcc_unreachable ();
}
}
[(set_attr "length" "4")])
(define_insn "casesi_worker_2"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(unspec:SI [(reg:SI R0_REG)
(match_operand:SI 1 "register_operand" "0,r")
(label_ref (match_operand 2 "" ""))
(label_ref (match_operand 3 "" ""))] UNSPEC_CASESI))
(clobber (match_operand:SI 4 "" "=X,1"))]
"TARGET_SH2 && reload_completed && flag_pic"
{
rtx diff_vec = PATTERN (NEXT_INSN (as_a (operands[2])));
gcc_assert (GET_CODE (diff_vec) == ADDR_DIFF_VEC);
switch (GET_MODE (diff_vec))
{
case SImode:
return "shll2 %1" "\n"
" add r0,%1" "\n"
" mova %O3,r0" "\n"
" mov.l @(r0,%1),%0";
case HImode:
return "add %1,%1" "\n"
" add r0,%1" "\n"
" mova %O3,r0" "\n"
" mov.w @(r0,%1),%0";
case QImode:
if (ADDR_DIFF_VEC_FLAGS (diff_vec).offset_unsigned)
return "add r0,%1" "\n"
" mova %O3,r0" "\n"
" mov.b @(r0,%1),%0" "\n"
" extu.b %0,%0";
else
return "add r0,%1" "\n"
" mova %O3,r0" "\n"
" mov.b @(r0,%1),%0";
default:
gcc_unreachable ();
}
}
[(set_attr "length" "8")])
(define_insn "casesi_shift_media"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ashift:DI (match_operand:DI 1 "arith_reg_operand" "r")
(unspec:DI [(label_ref:DI (match_operand 2 "" ""))]
UNSPEC_CASESI)))]
"TARGET_SHMEDIA"
{
rtx diff_vec = PATTERN (NEXT_INSN (as_a (operands[2])));
gcc_assert (GET_CODE (diff_vec) == ADDR_DIFF_VEC);
switch (GET_MODE (diff_vec))
{
case SImode:
return "shlli %1, 2, %0";
case HImode:
return "shlli %1, 1, %0";
case QImode:
if (rtx_equal_p (operands[0], operands[1]))
return "";
return "add %1, r63, %0";
default:
gcc_unreachable ();
}
}
[(set_attr "type" "arith_media")])
(define_insn "casesi_load_media"
[(set (match_operand 0 "any_arith_reg_dest" "=r")
(mem (unspec [(match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "arith_reg_operand" "r")
(label_ref:DI (match_operand 3 "" ""))] UNSPEC_CASESI)))]
"TARGET_SHMEDIA"
{
rtx diff_vec = PATTERN (NEXT_INSN (as_a (operands[3])));
gcc_assert (GET_CODE (diff_vec) == ADDR_DIFF_VEC);
switch (GET_MODE (diff_vec))
{
case SImode:
return "ldx.l %1, %2, %0";
case HImode:
#if 0
if (ADDR_DIFF_VEC_FLAGS (diff_vec).offset_unsigned)
return "ldx.uw %1, %2, %0";
#endif
return "ldx.w %1, %2, %0";
case QImode:
if (ADDR_DIFF_VEC_FLAGS (diff_vec).offset_unsigned)
return "ldx.ub %1, %2, %0";
return "ldx.b %1, %2, %0";
default:
gcc_unreachable ();
}
}
[(set_attr "type" "load_media")])
(define_expand "simple_return"
[(simple_return)]
"sh_can_use_simple_return_p ()")
(define_expand "return"
[(return)]
"reload_completed && epilogue_completed"
{
if (TARGET_SHMEDIA)
{
emit_jump_insn (gen_return_media ());
DONE;
}
if (TARGET_SHCOMPACT
&& (crtl->args.info.call_cookie & CALL_COOKIE_RET_TRAMP (1)))
{
emit_jump_insn (gen_shcompact_return_tramp ());
DONE;
}
})
(define_insn "*_i"
[(any_return)]
"TARGET_SH1 && ! (TARGET_SHCOMPACT
&& (crtl->args.info.call_cookie
& CALL_COOKIE_RET_TRAMP (1)))
&& reload_completed
&& ! sh_cfun_trap_exit_p ()"
{
if (TARGET_SH2A && (dbr_sequence_length () == 0)
&& !current_function_interrupt)
return "rts/n";
else
return "%@ %#";
}
[(set_attr "type" "return")
(set_attr "needs_delay_slot" "yes")])
;; trapa has no delay slot.
(define_insn "*return_trapa"
[(return)]
"TARGET_SH1 && !TARGET_SHCOMPACT
&& reload_completed"
"%@"
[(set_attr "type" "return")])
(define_expand "shcompact_return_tramp"
[(return)]
"TARGET_SHCOMPACT
&& (crtl->args.info.call_cookie & CALL_COOKIE_RET_TRAMP (1))"
{
rtx reg = gen_rtx_REG (Pmode, R0_REG);
function_symbol (reg, "__GCC_shcompact_return_trampoline", SFUNC_STATIC);
emit_jump_insn (gen_shcompact_return_tramp_i ());
DONE;
})
(define_insn "shcompact_return_tramp_i"
[(parallel [(return) (use (reg:SI R0_REG))])]
"TARGET_SHCOMPACT
&& (crtl->args.info.call_cookie & CALL_COOKIE_RET_TRAMP (1))"
"jmp @r0%#"
[(set_attr "type" "jump_ind")
(set_attr "needs_delay_slot" "yes")])
(define_insn "return_media_i"
[(parallel [(return) (use (match_operand 0 "target_reg_operand" "k"))])]
"TARGET_SHMEDIA && reload_completed"
"blink %0, r63"
[(set_attr "type" "jump_media")])
(define_insn "return_media_rte"
[(return)]
"TARGET_SHMEDIA && reload_completed && current_function_interrupt"
"rte"
[(set_attr "type" "jump_media")])
(define_expand "return_media"
[(return)]
"TARGET_SHMEDIA && reload_completed"
{
int tr_regno = sh_media_register_for_return ();
rtx tr;
if (current_function_interrupt)
{
emit_jump_insn (gen_return_media_rte ());
DONE;
}
if (tr_regno < 0)
{
rtx r18 = gen_rtx_REG (Pmode, PR_MEDIA_REG);
gcc_assert (call_really_used_regs[TR0_REG] && !fixed_regs[TR0_REG]);
tr_regno = TR0_REG;
tr = gen_rtx_REG (Pmode, tr_regno);
emit_move_insn (tr, r18);
}
else
tr = gen_rtx_REG (Pmode, tr_regno);
emit_jump_insn (gen_return_media_i (tr));
DONE;
})
(define_insn "shcompact_preserve_incoming_args"
[(set (match_operand:SI 0 "register_operand" "+r")
(unspec:SI [(match_dup 0)] UNSPEC_COMPACT_ARGS))]
"TARGET_SHCOMPACT"
""
[(set_attr "length" "0")])
(define_insn "shcompact_incoming_args"
[(set (reg:SI R2_REG) (unspec:SI [(reg:SI R2_REG)] UNSPEC_COMPACT_ARGS))
(set (reg:SI R3_REG) (unspec:SI [(reg:SI R3_REG)] UNSPEC_COMPACT_ARGS))
(set (reg:SI R4_REG) (unspec:SI [(reg:SI R4_REG)] UNSPEC_COMPACT_ARGS))
(set (reg:SI R5_REG) (unspec:SI [(reg:SI R5_REG)] UNSPEC_COMPACT_ARGS))
(set (reg:SI R6_REG) (unspec:SI [(reg:SI R6_REG)] UNSPEC_COMPACT_ARGS))
(set (reg:SI R7_REG) (unspec:SI [(reg:SI R7_REG)] UNSPEC_COMPACT_ARGS))
(set (reg:SI R8_REG) (unspec:SI [(reg:SI R8_REG)] UNSPEC_COMPACT_ARGS))
(set (reg:SI R9_REG) (unspec:SI [(reg:SI R9_REG)] UNSPEC_COMPACT_ARGS))
(set (mem:BLK (reg:SI MACL_REG))
(unspec:BLK [(reg:SI MACH_REG)] UNSPEC_COMPACT_ARGS))
(use (reg:SI R0_REG))
(clobber (reg:SI R0_REG))
(clobber (reg:SI MACL_REG))
(clobber (reg:SI MACH_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT"
"jsr @r0%#"
[(set_attr "needs_delay_slot" "yes")])
(define_insn "shmedia_save_restore_regs_compact"
[(set (reg:SI SP_REG)
(plus:SI (reg:SI SP_REG)
(match_operand:SI 0 "immediate_operand" "i")))
(use (reg:SI R0_REG))
(clobber (reg:SI PR_REG))]
"TARGET_SHCOMPACT
&& (INTVAL (operands[0]) == SHMEDIA_REGS_STACK_ADJUST ()
|| INTVAL (operands[0]) == - SHMEDIA_REGS_STACK_ADJUST ())"
"jsr @r0%#"
[(set_attr "needs_delay_slot" "yes")])
(define_expand "prologue"
[(const_int 0)]
""
{
sh_expand_prologue ();
DONE;
})
(define_expand "epilogue"
[(return)]
""
{
sh_expand_epilogue (false);
if (TARGET_SHMEDIA
|| (TARGET_SHCOMPACT
&& (crtl->args.info.call_cookie & CALL_COOKIE_RET_TRAMP (1))))
{
emit_jump_insn (gen_return ());
DONE;
}
})
(define_expand "eh_return"
[(use (match_operand 0 "register_operand" ""))]
""
{
rtx ra = operands[0];
if (TARGET_SHMEDIA64)
emit_insn (gen_eh_set_ra_di (ra));
else
emit_insn (gen_eh_set_ra_si (ra));
DONE;
})
;; Clobber the return address on the stack. We can't expand this
;; until we know where it will be put in the stack frame.
(define_insn "eh_set_ra_si"
[(unspec_volatile [(match_operand:SI 0 "register_operand" "r")]
UNSPECV_EH_RETURN)
(clobber (match_scratch:SI 1 "=&r"))]
"! TARGET_SHMEDIA64"
"#")
(define_insn "eh_set_ra_di"
[(unspec_volatile [(match_operand:DI 0 "register_operand" "r")]
UNSPECV_EH_RETURN)
(clobber (match_scratch:DI 1 "=&r"))]
"TARGET_SHMEDIA64"
"#")
(define_split
[(unspec_volatile [(match_operand 0 "register_operand" "")]
UNSPECV_EH_RETURN)
(clobber (match_scratch 1 ""))]
"reload_completed"
[(const_int 0)]
{
sh_set_return_address (operands[0], operands[1]);
DONE;
})
(define_insn "blockage"
[(unspec_volatile [(const_int 0)] UNSPECV_BLOCKAGE)]
""
""
[(set_attr "length" "0")])
;; Define movml instructions for SH2A target. Currently they are
;; used to push and pop all banked registers only.
(define_insn "movml_push_banked"
[(set (match_operand:SI 0 "register_operand" "=r")
(plus (match_dup 0) (const_int -32)))
(set (mem:SI (plus:SI (match_dup 0) (const_int 28))) (reg:SI R7_REG))
(set (mem:SI (plus:SI (match_dup 0) (const_int 24))) (reg:SI R6_REG))
(set (mem:SI (plus:SI (match_dup 0) (const_int 20))) (reg:SI R5_REG))
(set (mem:SI (plus:SI (match_dup 0) (const_int 16))) (reg:SI R4_REG))
(set (mem:SI (plus:SI (match_dup 0) (const_int 12))) (reg:SI R3_REG))
(set (mem:SI (plus:SI (match_dup 0) (const_int 8))) (reg:SI R2_REG))
(set (mem:SI (plus:SI (match_dup 0) (const_int 4))) (reg:SI R1_REG))
(set (mem:SI (plus:SI (match_dup 0) (const_int 0))) (reg:SI R0_REG))]
"TARGET_SH2A && REGNO (operands[0]) == 15"
"movml.l r7,@-r15"
[(set_attr "in_delay_slot" "no")])
(define_insn "movml_pop_banked"
[(set (match_operand:SI 0 "register_operand" "=r")
(plus (match_dup 0) (const_int 32)))
(set (reg:SI R0_REG) (mem:SI (plus:SI (match_dup 0) (const_int -32))))
(set (reg:SI R1_REG) (mem:SI (plus:SI (match_dup 0) (const_int -28))))
(set (reg:SI R2_REG) (mem:SI (plus:SI (match_dup 0) (const_int -24))))
(set (reg:SI R3_REG) (mem:SI (plus:SI (match_dup 0) (const_int -20))))
(set (reg:SI R4_REG) (mem:SI (plus:SI (match_dup 0) (const_int -16))))
(set (reg:SI R5_REG) (mem:SI (plus:SI (match_dup 0) (const_int -12))))
(set (reg:SI R6_REG) (mem:SI (plus:SI (match_dup 0) (const_int -8))))
(set (reg:SI R7_REG) (mem:SI (plus:SI (match_dup 0) (const_int -4))))]
"TARGET_SH2A && REGNO (operands[0]) == 15"
"movml.l @r15+,r7"
[(set_attr "in_delay_slot" "no")])
;; ------------------------------------------------------------------------
;; Scc instructions
;; ------------------------------------------------------------------------
(define_insn "movt"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(match_operand:SI 1 "t_reg_operand"))]
"TARGET_SH1"
"movt %0"
[(set_attr "type" "arith")])
(define_insn "movrt"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(xor:SI (match_operand:SI 1 "t_reg_operand" "") (const_int 1)))]
"TARGET_SH2A"
"movrt %0"
[(set_attr "type" "arith")])
(define_expand "cstore4_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(match_operator:SI 1 "sh_float_comparison_operator"
[(match_operand 2 "logical_operand" "")
(match_operand 3 "cmp_operand" "")]))]
"TARGET_SHMEDIA"
{
enum machine_mode mode = GET_MODE (operands[2]);
enum rtx_code code = GET_CODE (operands[1]);
bool invert, swap;
if (mode == VOIDmode)
mode = GET_MODE (operands[3]);
if (operands[2] == const0_rtx)
{
if (code == EQ || code == NE)
operands[2] = operands[3], operands[3] = const0_rtx;
}
else
operands[2] = force_reg (mode, operands[2]);
if (operands[3] != const0_rtx)
operands[3] = force_reg (mode, operands[3]);
switch (code)
{
case GEU:
case GE:
swap = invert = !FLOAT_MODE_P (mode);
break;
case LEU:
case LE:
swap = FLOAT_MODE_P (mode), invert = !swap;
break;
case LTU:
case LT:
swap = true, invert = false;
break;
case GTU:
case GT:
case EQ:
case UNORDERED:
swap = invert = false;
break;
case NE:
swap = invert = true;
break;
default:
gcc_unreachable ();
}
if (swap)
{
rtx tem = operands[2];
operands[2] = operands[3];
operands[3] = tem;
code = swap_condition (code);
}
if (invert)
{
rtx tem = can_create_pseudo_p () ? gen_reg_rtx (SImode) : operands[0];
code = reverse_condition (code);
operands[1] = gen_rtx_fmt_ee (code, VOIDmode, operands[2], operands[3]);
emit_insn (gen_cstore4_media (tem, operands[1],
operands[2], operands[3]));
code = EQ;
operands[2] = tem;
operands[3] = const0_rtx;
}
operands[1] = gen_rtx_fmt_ee (code, VOIDmode, operands[2], operands[3]);
})
(define_expand "cstoresi4"
[(set (match_operand:SI 0 "register_operand" "=r")
(match_operator:SI 1 "comparison_operator"
[(match_operand:SI 2 "cmpsi_operand" "")
(match_operand:SI 3 "arith_operand" "")]))]
"TARGET_SH1 || TARGET_SHMEDIA"
{
if (TARGET_SHMEDIA)
{
emit_insn (gen_cstore4_media (operands[0], operands[1],
operands[2], operands[3]));
DONE;
}
if (sh_expand_t_scc (operands))
DONE;
if (! currently_expanding_to_rtl)
FAIL;
sh_emit_compare_and_set (operands, SImode);
DONE;
})
(define_expand "cstoredi4"
[(set (match_operand:SI 0 "register_operand" "=r")
(match_operator:SI 1 "comparison_operator"
[(match_operand:DI 2 "arith_operand" "")
(match_operand:DI 3 "arith_operand" "")]))]
"TARGET_SH2 || TARGET_SHMEDIA"
{
if (TARGET_SHMEDIA)
{
emit_insn (gen_cstore4_media (operands[0], operands[1],
operands[2], operands[3]));
DONE;
}
if (sh_expand_t_scc (operands))
DONE;
if (! currently_expanding_to_rtl)
FAIL;
sh_emit_compare_and_set (operands, DImode);
DONE;
})
;; Move the complement of the T reg to a reg.
;; On SH2A the movrt insn can be used.
;; On anything else than SH2A this has to be done with multiple instructions.
;; One obvious way would be:
;; cmp/eq ...
;; movt r0
;; xor #1,r0
;;
;; However, this puts pressure on r0 in most cases and thus the following is
;; more appealing:
;; cmp/eq ...
;; mov #-1,temp
;; negc temp,dest
;;
;; If the constant -1 can be CSE-ed or lifted out of a loop it effectively
;; becomes a one instruction operation. Moreover, care must be taken that
;; the insn can still be combined with inverted compare and branch code
;; around it. On the other hand, if a function returns the complement of
;; a previous comparison result in the T bit, the xor #1,r0 approach might
;; lead to better code.
(define_expand "movnegt"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(xor:SI (match_operand:SI 1 "t_reg_operand" "") (const_int 1)))]
"TARGET_SH1"
{
if (TARGET_SH2A)
emit_insn (gen_movrt (operands[0], operands[1]));
else
{
rtx val = force_reg (SImode, gen_int_mode (-1, SImode));
emit_insn (gen_movrt_negc (operands[0], operands[1], val));
}
DONE;
})
(define_insn "movrt_negc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(xor:SI (match_operand:SI 1 "t_reg_operand" "") (const_int 1)))
(set (reg:SI T_REG) (const_int 1))
(use (match_operand:SI 2 "arith_reg_operand" "r"))]
"TARGET_SH1"
"negc %2,%0"
[(set_attr "type" "arith")])
;; The -1 constant will not be CSE-ed for the *movrt_negc pattern, but the
;; pattern can be used by the combine pass. Using a scratch reg for the
;; -1 constant results in slightly better register allocations compared to
;; generating a pseudo reg before reload.
(define_insn_and_split "*movrt_negc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(xor:SI (match_operand:SI 1 "t_reg_operand" "") (const_int 1)))
(clobber (match_scratch:SI 2 "=r"))
(clobber (reg:SI T_REG))]
"TARGET_SH1 && ! TARGET_SH2A"
"#"
"&& reload_completed"
[(set (match_dup 2) (const_int -1))
(parallel
[(set (match_dup 0) (xor:SI (match_dup 1) (const_int 1)))
(set (reg:SI T_REG) (const_int 1))
(use (match_dup 2))])])
;; Store the negated T bit in a reg using r0 and xor. This one doesn't
;; clobber the T bit, which is useful when storing the T bit and the
;; negated T bit in parallel. On SH2A the movrt insn can be used for that.
;; Usually we don't want this insn to be matched, except for cases where the
;; T bit clobber is really not appreciated. Hence the extra use on T_REG.
(define_insn_and_split "movrt_xor"
[(set (match_operand:SI 0 "arith_reg_dest" "=z")
(xor:SI (match_operand:SI 1 "t_reg_operand") (const_int 1)))
(use (reg:SI T_REG))]
"TARGET_SH1 && !TARGET_SH2A"
"#"
"&& reload_completed"
[(set (match_dup 0) (reg:SI T_REG))
(set (match_dup 0) (xor:SI (match_dup 0) (const_int 1)))])
;; Store the T bit and the negated T bit in two regs in parallel. There is
;; no real insn to do that, but specifying this pattern will give combine
;; some opportunities.
(define_insn_and_split "*movt_movrt"
[(parallel [(set (match_operand:SI 0 "arith_reg_dest")
(match_operand:SI 1 "negt_reg_operand"))
(set (match_operand:SI 2 "arith_reg_dest")
(match_operand:SI 3 "t_reg_operand"))])]
"TARGET_SH1"
"#"
"&& 1"
[(const_int 0)]
{
rtx i = TARGET_SH2A
? gen_movrt (operands[0], get_t_reg_rtx ())
: gen_movrt_xor (operands[0], get_t_reg_rtx ());
emit_insn (i);
emit_insn (gen_movt (operands[2], get_t_reg_rtx ()));
DONE;
})
(define_insn_and_split "*movt_movrt"
[(parallel [(set (match_operand:SI 0 "arith_reg_dest")
(match_operand:SI 1 "t_reg_operand"))
(set (match_operand:SI 2 "arith_reg_dest")
(match_operand:SI 3 "negt_reg_operand"))])]
"TARGET_SH1"
"#"
"&& 1"
[(parallel [(set (match_dup 2) (match_dup 3))
(set (match_dup 0) (match_dup 1))])])
;; Use negc to store the T bit in a MSB of a reg in the following way:
;; T = 1: 0x80000000 -> reg
;; T = 0: 0x7FFFFFFF -> reg
;; This works because 0 - 0x80000000 = 0x80000000.
;;
;; This insn must not match again after it has been split into the constant
;; load and negc. This is accomplished by the special negc insn that
;; has a use on the operand.
(define_insn_and_split "*mov_t_msb_neg"
[(set (match_operand:SI 0 "arith_reg_dest")
(minus:SI (const_int -2147483648) ;; 0x80000000
(match_operand 1 "t_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(set (match_dup 2) (const_int -2147483648))
(parallel [(set (match_dup 0) (minus:SI (neg:SI (match_dup 2))
(reg:SI T_REG)))
(clobber (reg:SI T_REG))
(use (match_dup 2))])]
{
operands[2] = gen_reg_rtx (SImode);
})
(define_insn "*mov_t_msb_neg_negc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(minus:SI (neg:SI (match_operand:SI 1 "arith_reg_operand" "r"))
(match_operand:SI 2 "t_reg_operand")))
(clobber (reg:SI T_REG))
(use (match_dup 1))]
"TARGET_SH1"
"negc %1,%0"
[(set_attr "type" "arith")])
;; These are essentially the same as above, but with the inverted T bit.
;; Combine recognizes the split patterns, but does not take them sometimes
;; if the T_REG clobber is specified. Instead it tries to split out the
;; T bit negation. Since these splits are supposed to be taken only by
;; combine, it will see the T_REG clobber of the *mov_t_msb_neg insn, so this
;; should be fine.
(define_split
[(set (match_operand:SI 0 "arith_reg_dest")
(plus:SI (match_operand 1 "negt_reg_operand")
(const_int 2147483647)))] ;; 0x7fffffff
"TARGET_SH1 && can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(minus:SI (const_int -2147483648) (reg:SI T_REG)))
(clobber (reg:SI T_REG))])])
(define_split
[(set (match_operand:SI 0 "arith_reg_dest")
(if_then_else:SI (match_operand 1 "t_reg_operand")
(const_int 2147483647) ;; 0x7fffffff
(const_int -2147483648)))] ;; 0x80000000
"TARGET_SH1 && can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(minus:SI (const_int -2147483648) (reg:SI T_REG)))
(clobber (reg:SI T_REG))])])
;; The *negnegt pattern helps the combine pass to figure out how to fold
;; an explicit double T bit negation.
(define_insn_and_split "*negnegt"
[(set (reg:SI T_REG)
(eq:SI (match_operand 0 "negt_reg_operand" "") (const_int 0)))]
"TARGET_SH1"
"#"
""
[(const_int 0)])
;; Store T bit as all zeros or ones in a reg.
(define_insn "mov_neg_si_t"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(neg:SI (match_operand 1 "t_reg_operand" "")))]
"TARGET_SH1"
"subc %0,%0"
[(set_attr "type" "arith")])
;; Store negated T bit as all zeros or ones in a reg.
;; Use the following sequence:
;; subc Rn,Rn ! Rn = Rn - Rn - T; T = T
;; not Rn,Rn ! Rn = 0 - Rn
(define_split
[(set (match_operand:SI 0 "arith_reg_dest" "")
(neg:SI (match_operand 1 "negt_reg_operand" "")))]
"TARGET_SH1"
[(set (match_dup 0) (neg:SI (reg:SI T_REG)))
(set (match_dup 0) (not:SI (match_dup 0)))])
;; The *movtt pattern eliminates redundant T bit to T bit moves / tests.
(define_insn_and_split "*movtt"
[(set (reg:SI T_REG)
(eq:SI (match_operand 0 "t_reg_operand" "") (const_int 1)))]
"TARGET_SH1"
"#"
""
[(const_int 0)])
;; Invert the T bit.
;; On SH2A we can use the nott insn. On anything else this must be done with
;; multiple insns like:
;; movt Rn
;; tst Rn,Rn
;; This requires an additional pseudo. The SH specific sh_treg_combine RTL
;; pass will look for this insn. Disallow using it if pseudos can't be
;; created.
(define_insn_and_split "nott"
[(set (reg:SI T_REG)
(xor:SI (match_operand:SI 0 "t_reg_operand") (const_int 1)))]
"TARGET_SH2A || (TARGET_SH1 && can_create_pseudo_p ())"
{
gcc_assert (TARGET_SH2A);
return "nott";
}
"! TARGET_SH2A && can_create_pseudo_p ()"
[(set (match_dup 0) (reg:SI T_REG))
(set (reg:SI T_REG) (eq:SI (match_dup 0) (const_int 0)))]
{
operands[0] = gen_reg_rtx (SImode);
})
;; Store T bit as MSB in a reg.
;; T = 0: 0x00000000 -> reg
;; T = 1: 0x80000000 -> reg
(define_insn_and_split "*movt_msb"
[(set (match_operand:SI 0 "arith_reg_dest")
(mult:SI (match_operand:SI 1 "t_reg_operand")
(const_int -2147483648))) ;; 0xffffffff80000000
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
"&& 1"
[(set (match_dup 0) (ashift:SI (reg:SI T_REG) (const_int 31)))])
;; Store inverted T bit as MSB in a reg.
;; T = 0: 0x80000000 -> reg
;; T = 1: 0x00000000 -> reg
;; On SH2A we can get away without clobbering the T_REG using the movrt insn.
;; On non SH2A we resort to the following sequence:
;; movt Rn
;; tst Rn,Rn
;; rotcr Rn
;; The T bit value will be modified during the sequence, but the rotcr insn
;; will restore its original value.
(define_insn_and_split "*negt_msb"
[(set (match_operand:SI 0 "arith_reg_dest")
(match_operand:SI 1 "negt_reg_shl31_operand"))]
"TARGET_SH1"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
rtx tmp = gen_reg_rtx (SImode);
if (TARGET_SH2A)
{
emit_insn (gen_movrt (tmp, get_t_reg_rtx ()));
emit_insn (gen_rotrsi3 (operands[0], tmp, const1_rtx));
}
else
{
emit_move_insn (tmp, get_t_reg_rtx ());
emit_insn (gen_cmpeqsi_t (tmp, const0_rtx));
emit_insn (gen_rotcr (operands[0], tmp, get_t_reg_rtx ()));
}
DONE;
})
;; The *cset_zero patterns convert optimizations such as
;; "if (test) x = 0;"
;; to
;; "x &= -(test == 0);"
;; back to conditional branch sequences if zero-displacement branches
;; are enabled.
;; FIXME: These patterns can be removed when conditional execution patterns
;; are implemented, since ifcvt will not perform these optimizations if
;; conditional execution is supported.
(define_insn "*cset_zero"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(and:SI (plus:SI (match_operand:SI 1 "t_reg_operand")
(const_int -1))
(match_operand:SI 2 "arith_reg_operand" "0")))]
"TARGET_SH1 && TARGET_ZDCBRANCH"
{
return "bf 0f" "\n"
" mov #0,%0" "\n"
"0:";
}
[(set_attr "type" "arith") ;; poor approximation
(set_attr "length" "4")])
(define_insn "*cset_zero"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(if_then_else:SI (match_operand:SI 1 "cbranch_treg_value")
(match_operand:SI 2 "arith_reg_operand" "0")
(const_int 0)))]
"TARGET_SH1 && TARGET_ZDCBRANCH"
{
int tval = sh_eval_treg_value (operands[1]);
if (tval == true)
return "bt 0f" "\n"
" mov #0,%0" "\n"
"0:";
else if (tval == false)
return "bf 0f" "\n"
" mov #0,%0" "\n"
"0:";
else
gcc_unreachable ();
}
[(set_attr "type" "arith") ;; poor approximation
(set_attr "length" "4")])
(define_expand "cstoresf4"
[(set (match_operand:SI 0 "register_operand" "=r")
(match_operator:SI 1 "sh_float_comparison_operator"
[(match_operand:SF 2 "arith_operand" "")
(match_operand:SF 3 "arith_operand" "")]))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU"
{
if (TARGET_SHMEDIA)
{
emit_insn (gen_cstore4_media (operands[0], operands[1],
operands[2], operands[3]));
DONE;
}
if (! currently_expanding_to_rtl)
FAIL;
sh_emit_compare_and_set (operands, SFmode);
DONE;
})
(define_expand "cstoredf4"
[(set (match_operand:SI 0 "register_operand" "=r")
(match_operator:SI 1 "sh_float_comparison_operator"
[(match_operand:DF 2 "arith_operand" "")
(match_operand:DF 3 "arith_operand" "")]))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SHMEDIA)
{
emit_insn (gen_cstore4_media (operands[0], operands[1],
operands[2], operands[3]));
DONE;
}
if (! currently_expanding_to_rtl)
FAIL;
sh_emit_compare_and_set (operands, DFmode);
DONE;
})
;; -------------------------------------------------------------------------
;; Instructions to cope with inline literal tables
;; -------------------------------------------------------------------------
;; 2 byte integer in line
(define_insn "consttable_2"
[(unspec_volatile [(match_operand:SI 0 "general_operand" "=g")
(match_operand 1 "" "")]
UNSPECV_CONST2)]
""
{
if (operands[1] != const0_rtx)
assemble_integer (operands[0], 2, BITS_PER_UNIT * 2, 1);
return "";
}
[(set_attr "length" "2")
(set_attr "in_delay_slot" "no")])
;; 4 byte integer in line
(define_insn "consttable_4"
[(unspec_volatile [(match_operand:SI 0 "general_operand" "=g")
(match_operand 1 "" "")]
UNSPECV_CONST4)]
""
{
if (operands[1] != const0_rtx)
{
assemble_integer (operands[0], 4, BITS_PER_UNIT * 4, 1);
mark_symbol_refs_as_used (operands[0]);
}
return "";
}
[(set_attr "length" "4")
(set_attr "in_delay_slot" "no")])
;; 8 byte integer in line
(define_insn "consttable_8"
[(unspec_volatile [(match_operand:SI 0 "general_operand" "=g")
(match_operand 1 "" "")]
UNSPECV_CONST8)]
""
{
if (operands[1] != const0_rtx)
assemble_integer (operands[0], 8, BITS_PER_UNIT * 8, 1);
return "";
}
[(set_attr "length" "8")
(set_attr "in_delay_slot" "no")])
;; 4 byte floating point
(define_insn "consttable_sf"
[(unspec_volatile [(match_operand:SF 0 "general_operand" "=g")
(match_operand 1 "" "")]
UNSPECV_CONST4)]
""
{
if (operands[1] != const0_rtx)
{
REAL_VALUE_TYPE d;
REAL_VALUE_FROM_CONST_DOUBLE (d, operands[0]);
assemble_real (d, SFmode, GET_MODE_ALIGNMENT (SFmode));
}
return "";
}
[(set_attr "length" "4")
(set_attr "in_delay_slot" "no")])
;; 8 byte floating point
(define_insn "consttable_df"
[(unspec_volatile [(match_operand:DF 0 "general_operand" "=g")
(match_operand 1 "" "")]
UNSPECV_CONST8)]
""
{
if (operands[1] != const0_rtx)
{
REAL_VALUE_TYPE d;
REAL_VALUE_FROM_CONST_DOUBLE (d, operands[0]);
assemble_real (d, DFmode, GET_MODE_ALIGNMENT (DFmode));
}
return "";
}
[(set_attr "length" "8")
(set_attr "in_delay_slot" "no")])
;; Alignment is needed for some constant tables; it may also be added for
;; Instructions at the start of loops, or after unconditional branches.
;; ??? We would get more accurate lengths if we did instruction
;; alignment based on the value of INSN_CURRENT_ADDRESS; the approach used
;; here is too conservative.
;; align to a two byte boundary
(define_expand "align_2"
[(unspec_volatile [(const_int 1)] UNSPECV_ALIGN)]
""
"")
;; Align to a four byte boundary.
;; align_4 and align_log are instructions for the starts of loops, or
;; after unconditional branches, which may take up extra room.
(define_expand "align_4"
[(unspec_volatile [(const_int 2)] UNSPECV_ALIGN)]
""
"")
;; Align to a cache line boundary.
(define_insn "align_log"
[(unspec_volatile [(match_operand 0 "const_int_operand" "")] UNSPECV_ALIGN)]
""
""
[(set_attr "length" "0")
(set_attr "in_delay_slot" "no")])
;; Emitted at the end of the literal table, used to emit the
;; 32bit branch labels if needed.
(define_insn "consttable_end"
[(unspec_volatile [(const_int 0)] UNSPECV_CONST_END)]
""
{
return output_jump_label_table ();
}
[(set_attr "in_delay_slot" "no")])
;; Emitted at the end of the window in the literal table.
(define_insn "consttable_window_end"
[(unspec_volatile [(match_operand 0 "" "")] UNSPECV_WINDOW_END)]
""
""
[(set_attr "length" "0")
(set_attr "in_delay_slot" "no")])
;; -------------------------------------------------------------------------
;; Minimum / maximum operations.
;; -------------------------------------------------------------------------
;; The SH2A clips.b and clips.w insns do a signed min-max function. If smin
;; and smax standard name patterns are defined, they will be used during
;; initial expansion and combine will then be able to form the actual min-max
;; pattern.
;; The clips.b and clips.w set the SR.CS bit if the value in the register is
;; clipped, but there is currently no way of making use of this information.
;; The only way to read or reset the SR.CS bit is by accessing the SR.
(define_expand "si3"
[(parallel [(set (match_operand:SI 0 "arith_reg_dest")
(SMIN_SMAX:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand 2 "const_int_operand")))
(clobber (reg:SI T_REG))])]
"TARGET_SH2A"
{
/* Force the comparison value into a register, because greater-than
comparisons can work only on registers. Combine will be able to pick up
the constant value from the REG_EQUAL note when trying to form a min-max
pattern. */
operands[2] = force_reg (SImode, operands[2]);
})
;; Convert
;; smax (smin (...))
;; to
;; smin (smax (...))
(define_insn_and_split "*clips"
[(set (match_operand:SI 0 "arith_reg_dest")
(smax:SI (smin:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand 2 "clips_max_const_int"))
(match_operand 3 "clips_min_const_int")))]
"TARGET_SH2A"
"#"
"&& 1"
[(set (match_dup 0)
(smin:SI (smax:SI (match_dup 1) (match_dup 3)) (match_dup 2)))])
(define_insn "*clips"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(smin:SI (smax:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand 2 "clips_min_const_int"))
(match_operand 3 "clips_max_const_int")))]
"TARGET_SH2A"
{
if (INTVAL (operands[3]) == 127)
return "clips.b %0";
else if (INTVAL (operands[3]) == 32767)
return "clips.w %0";
else
gcc_unreachable ();
}
[(set_attr "type" "arith")])
;; If the expanded smin or smax patterns were not combined, split them into
;; a compare and branch sequence, because there are no real smin or smax
;; insns.
(define_insn_and_split "*si3"
[(set (match_operand:SI 0 "arith_reg_dest")
(SMIN_SMAX:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand:SI 2 "arith_reg_or_0_or_1_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH2A && can_create_pseudo_p ()"
"#"
"&& 1"
[(const_int 0)]
{
rtx skip_label = gen_label_rtx ();
emit_move_insn (operands[0], operands[1]);
rtx cmp_val = operands[2];
if (satisfies_constraint_M (cmp_val))
cmp_val = const0_rtx;
emit_insn (gen_cmpgtsi_t (operands[0], cmp_val));
emit_jump_insn ( == SMIN
? gen_branch_false (skip_label)
: gen_branch_true (skip_label));
emit_label_after (skip_label, emit_move_insn (operands[0], operands[2]));
DONE;
})
;; The SH2A clipu.b and clipu.w insns can be used to implement a min function
;; with a register and a constant.
;; The clipu.b and clipu.w set the SR.CS bit if the value in the register is
;; clipped, but there is currently no way of making use of this information.
;; The only way to read or reset the SR.CS bit is by accessing the SR.
(define_expand "uminsi3"
[(set (match_operand:SI 0 "arith_reg_dest")
(umin:SI (match_operand:SI 1 "arith_reg_operand")
(match_operand 2 "const_int_operand")))]
"TARGET_SH2A"
{
if (INTVAL (operands[2]) == 1)
{
emit_insn (gen_clipu_one (operands[0], operands[1]));
DONE;
}
else if (! clipu_max_const_int (operands[2], VOIDmode))
FAIL;
})
(define_insn "*clipu"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(umin:SI (match_operand:SI 1 "arith_reg_operand" "0")
(match_operand 2 "clipu_max_const_int")))]
"TARGET_SH2A"
{
if (INTVAL (operands[2]) == 255)
return "clipu.b %0";
else if (INTVAL (operands[2]) == 65535)
return "clipu.w %0";
else
gcc_unreachable ();
}
[(set_attr "type" "arith")])
(define_insn_and_split "clipu_one"
[(set (match_operand:SI 0 "arith_reg_dest")
(umin:SI (match_operand:SI 1 "arith_reg_operand") (const_int 1)))
(clobber (reg:SI T_REG))]
"TARGET_SH2A"
"#"
"&& can_create_pseudo_p ()"
[(const_int 0)]
{
emit_insn (gen_cmpeqsi_t (operands[1], const0_rtx));
emit_insn (gen_movnegt (operands[0], get_t_reg_rtx ()));
DONE;
})
;; -------------------------------------------------------------------------
;; Misc
;; -------------------------------------------------------------------------
;; String/block move insn.
(define_expand "movmemsi"
[(parallel [(set (mem:BLK (match_operand:BLK 0))
(mem:BLK (match_operand:BLK 1)))
(use (match_operand:SI 2 "nonmemory_operand"))
(use (match_operand:SI 3 "immediate_operand"))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R4_REG))
(clobber (reg:SI R5_REG))
(clobber (reg:SI R0_REG))])]
"TARGET_SH1 && ! TARGET_SH5"
{
if (expand_block_move (operands))
DONE;
else
FAIL;
})
(define_insn "block_move_real"
[(parallel [(set (mem:BLK (reg:SI R4_REG))
(mem:BLK (reg:SI R5_REG)))
(use (match_operand:SI 0 "arith_reg_operand" "r"))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R0_REG))])]
"TARGET_SH1 && ! TARGET_HARD_SH4"
"jsr @%0%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_insn "block_lump_real"
[(parallel [(set (mem:BLK (reg:SI R4_REG))
(mem:BLK (reg:SI R5_REG)))
(use (match_operand:SI 0 "arith_reg_operand" "r"))
(use (reg:SI R6_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI T_REG))
(clobber (reg:SI R4_REG))
(clobber (reg:SI R5_REG))
(clobber (reg:SI R6_REG))
(clobber (reg:SI R0_REG))])]
"TARGET_SH1 && ! TARGET_HARD_SH4"
"jsr @%0%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_insn "block_move_real_i4"
[(parallel [(set (mem:BLK (reg:SI R4_REG))
(mem:BLK (reg:SI R5_REG)))
(use (match_operand:SI 0 "arith_reg_operand" "r"))
(clobber (reg:SI PR_REG))
(clobber (reg:SI R0_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R2_REG))])]
"TARGET_HARD_SH4"
"jsr @%0%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
(define_insn "block_lump_real_i4"
[(parallel [(set (mem:BLK (reg:SI R4_REG))
(mem:BLK (reg:SI R5_REG)))
(use (match_operand:SI 0 "arith_reg_operand" "r"))
(use (reg:SI R6_REG))
(clobber (reg:SI PR_REG))
(clobber (reg:SI T_REG))
(clobber (reg:SI R4_REG))
(clobber (reg:SI R5_REG))
(clobber (reg:SI R6_REG))
(clobber (reg:SI R0_REG))
(clobber (reg:SI R1_REG))
(clobber (reg:SI R2_REG))
(clobber (reg:SI R3_REG))])]
"TARGET_HARD_SH4"
"jsr @%0%#"
[(set_attr "type" "sfunc")
(set_attr "needs_delay_slot" "yes")])
;; byte compare pattern
;; temp = a ^ b;
;; !((temp & 0xF000) && (temp & 0x0F00) && (temp & 0x00F0) && (temp & 0x000F))
(define_insn "cmpstr_t"
[(set (reg:SI T_REG)
(eq:SI (and:SI
(and:SI
(and:SI
(zero_extract:SI
(xor:SI (match_operand:SI 0 "arith_reg_operand" "r")
(match_operand:SI 1 "arith_reg_operand" "r"))
(const_int 8) (const_int 0))
(zero_extract:SI (xor:SI (match_dup 0) (match_dup 1))
(const_int 8) (const_int 8)))
(zero_extract:SI (xor:SI (match_dup 0) (match_dup 1))
(const_int 8) (const_int 16)))
(zero_extract:SI (xor:SI (match_dup 0) (match_dup 1))
(const_int 8) (const_int 24)))
(const_int 0)))]
"TARGET_SH1"
"cmp/str %0,%1"
[(set_attr "type" "mt_group")])
(define_expand "cmpstrsi"
[(set (match_operand:SI 0 "register_operand")
(compare:SI (match_operand:BLK 1 "memory_operand")
(match_operand:BLK 2 "memory_operand")))
(use (match_operand 3 "immediate_operand"))]
"TARGET_SH1 && optimize"
{
if (! optimize_insn_for_size_p () && sh_expand_cmpstr (operands))
DONE;
else
FAIL;
})
(define_expand "cmpstrnsi"
[(set (match_operand:SI 0 "register_operand")
(compare:SI (match_operand:BLK 1 "memory_operand")
(match_operand:BLK 2 "memory_operand")))
(use (match_operand:SI 3 "immediate_operand"))
(use (match_operand:SI 4 "immediate_operand"))]
"TARGET_SH1 && optimize"
{
if (! optimize_insn_for_size_p () && sh_expand_cmpnstr (operands))
DONE;
else
FAIL;
})
(define_expand "strlensi"
[(set (match_operand:SI 0 "register_operand")
(unspec:SI [(match_operand:BLK 1 "memory_operand")
(match_operand:SI 2 "immediate_operand")
(match_operand:SI 3 "immediate_operand")]
UNSPEC_BUILTIN_STRLEN))]
"TARGET_SH1 && optimize"
{
if (! optimize_insn_for_size_p () && sh_expand_strlen (operands))
DONE;
else
FAIL;
})
(define_expand "setmemqi"
[(parallel [(set (match_operand:BLK 0 "memory_operand")
(match_operand 2 "const_int_operand"))
(use (match_operand:QI 1 "const_int_operand"))
(use (match_operand:QI 3 "const_int_operand"))])]
"TARGET_SH1 && optimize"
{
if (optimize_insn_for_size_p ())
FAIL;
sh_expand_setmem (operands);
DONE;
})
;; -------------------------------------------------------------------------
;; Floating point instructions.
;; -------------------------------------------------------------------------
;; FIXME: For now we disallow any memory operands for fpscr loads/stores,
;; except for post-inc loads and pre-dec stores for push/pop purposes.
;; This avoids problems with reload. As a consequence, user initiated fpscr
;; stores to memory will always be ferried through a general register.
;; User initiated fpscr loads always have to undergo bit masking to preserve
;; the current fpu mode settings for the compiler generated code. Thus such
;; fpscr loads will always have to go through general registers anyways.
(define_insn "lds_fpscr"
[(set (reg:SI FPSCR_REG)
(match_operand:SI 0 "fpscr_movsrc_operand" "r,>"))
(set (reg:SI FPSCR_STAT_REG)
(unspec_volatile:SI [(const_int 0)] UNSPECV_FPSCR_STAT))
(set (reg:SI FPSCR_MODES_REG)
(unspec_volatile:SI [(const_int 0)] UNSPECV_FPSCR_MODES))]
"TARGET_FPU_ANY"
"@
lds %0,fpscr
lds.l %0,fpscr"
[(set_attr "type" "gp_fpscr,mem_fpscr")])
;; A move fpscr -> reg schedules like a move mac -> reg. Thus we use mac_gp
;; type for it.
(define_insn "sts_fpscr"
[(set (match_operand:SI 0 "fpscr_movdst_operand" "=r,<")
(reg:SI FPSCR_REG))
(use (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_FPU_ANY"
"@
sts fpscr,%0
sts.l fpscr,%0"
[(set_attr "type" "mac_gp,fstore")])
(define_expand "set_fpscr"
[(parallel [(set (reg:SI FPSCR_REG)
(match_operand:SI 0 "general_operand"))
(set (reg:SI FPSCR_STAT_REG)
(unspec_volatile:SI [(const_int 0)] UNSPECV_FPSCR_MODES))])]
"TARGET_FPU_ANY"
{
/* We have to mask out the FR, SZ and PR bits. To do that, we need to
get the current FPSCR value first.
(a & ~mask) | (b & mask) = a ^ ((a ^ b) & mask) */
rtx mask = force_reg (SImode, GEN_INT (FPSCR_FR | FPSCR_SZ | FPSCR_PR));
rtx a = force_reg (SImode, operands[0]);
rtx b = gen_reg_rtx (SImode);
emit_insn (gen_sts_fpscr (b));
rtx a_xor_b = gen_reg_rtx (SImode);
emit_insn (gen_xorsi3 (a_xor_b, a, b));
rtx a_xor_b_and_mask = gen_reg_rtx (SImode);
emit_insn (gen_andsi3 (a_xor_b_and_mask, a_xor_b, mask));
rtx r = gen_reg_rtx (SImode);
emit_insn (gen_xorsi3 (r, a_xor_b_and_mask, a));
emit_insn (gen_lds_fpscr (r));
DONE;
})
;; ??? This uses the fp unit, but has no type indicating that.
;; If we did that, this would either give a bogus latency or introduce
;; a bogus FIFO constraint.
;; Since this insn is currently only used for prologues/epilogues,
;; it is probably best to claim no function unit, which matches the
;; current setting.
(define_insn "toggle_sz"
[(set (reg:SI FPSCR_REG)
(xor:SI (reg:SI FPSCR_REG) (const_int FPSCR_SZ)))
(set (reg:SI FPSCR_MODES_REG)
(unspec_volatile:SI [(const_int 0)] UNSPECV_FPSCR_MODES))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fschg"
[(set_attr "type" "fpscr_toggle") (set_attr "fp_set" "unknown")])
;; Toggle FPU precision PR mode.
(define_insn "toggle_pr"
[(set (reg:SI FPSCR_REG)
(xor:SI (reg:SI FPSCR_REG) (const_int FPSCR_PR)))
(set (reg:SI FPSCR_MODES_REG)
(unspec_volatile:SI [(const_int 0)] UNSPECV_FPSCR_MODES))]
"TARGET_SH4A_FP"
"fpchg"
[(set_attr "type" "fpscr_toggle")])
(define_expand "addsf3"
[(set (match_operand:SF 0 "fp_arith_reg_operand")
(plus:SF (match_operand:SF 1 "fp_arith_reg_operand")
(match_operand:SF 2 "fp_arith_reg_operand")))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH2E)
{
emit_insn (gen_addsf3_i (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*addsf3_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(plus:SF (match_operand:SF 1 "fp_arith_reg_operand" "%f")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fadd.s %1, %2, %0"
[(set_attr "type" "fparith_media")])
(define_insn_and_split "unary_sf_op"
[(set (match_operand:V2SF 0 "fp_arith_reg_operand" "=f")
(vec_select:V2SF
(vec_concat:V2SF
(vec_select:SF
(match_dup 0)
(parallel [(not:BI (match_operand 3 "const_int_operand" "n"))]))
(match_operator:SF 2 "unary_float_operator"
[(vec_select:SF (match_operand:V2SF 1 "fp_arith_reg_operand" "f")
(parallel [(match_operand 4
"const_int_operand" "n")]))]))
(parallel [(not:BI (match_dup 3)) (match_dup 3)])))]
"TARGET_SHMEDIA_FPU"
"#"
"TARGET_SHMEDIA_FPU && reload_completed"
[(set (match_dup 5) (match_dup 6))]
{
int endian = TARGET_LITTLE_ENDIAN ? 0 : 1;
rtx op1 = gen_rtx_REG (SFmode,
(true_regnum (operands[1])
+ (INTVAL (operands[4]) ^ endian)));
operands[7] = gen_rtx_REG (SFmode,
(true_regnum (operands[0])
+ (INTVAL (operands[3]) ^ endian)));
operands[6] = gen_rtx_fmt_e (GET_CODE (operands[2]), SFmode, op1);
}
[(set_attr "type" "fparith_media")])
(define_insn_and_split "binary_sf_op0"
[(set (match_operand:V2SF 0 "fp_arith_reg_operand" "=f")
(vec_concat:V2SF
(match_operator:SF 3 "binary_float_operator"
[(vec_select:SF (match_operand:V2SF 1 "fp_arith_reg_operand" "f")
(parallel [(const_int 0)]))
(vec_select:SF (match_operand:V2SF 2 "fp_arith_reg_operand" "f")
(parallel [(const_int 0)]))])
(vec_select:SF
(match_dup 0)
(parallel [(const_int 1)]))))]
"TARGET_SHMEDIA_FPU"
"#"
"&& reload_completed"
[(set (match_dup 4) (match_dup 5))]
{
int endian = TARGET_LITTLE_ENDIAN ? 0 : 1;
rtx op1 = gen_rtx_REG (SFmode,
true_regnum (operands[1]) + endian);
rtx op2 = gen_rtx_REG (SFmode,
true_regnum (operands[2]) + endian);
operands[4] = gen_rtx_REG (SFmode,
true_regnum (operands[0]) + endian);
operands[5] = gen_rtx_fmt_ee (GET_CODE (operands[3]), SFmode, op1, op2);
}
[(set_attr "type" "fparith_media")])
(define_insn_and_split "binary_sf_op1"
[(set (match_operand:V2SF 0 "fp_arith_reg_operand" "=f")
(vec_concat:V2SF
(vec_select:SF
(match_dup 0)
(parallel [(const_int 0)]))
(match_operator:SF 3 "binary_float_operator"
[(vec_select:SF (match_operand:V2SF 1 "fp_arith_reg_operand" "f")
(parallel [(const_int 1)]))
(vec_select:SF (match_operand:V2SF 2 "fp_arith_reg_operand" "f")
(parallel [(const_int 1)]))])))]
"TARGET_SHMEDIA_FPU"
"#"
"&& reload_completed"
[(set (match_dup 4) (match_dup 5))]
{
int endian = TARGET_LITTLE_ENDIAN ? 0 : 1;
rtx op1 = gen_rtx_REG (SFmode, true_regnum (operands[1]) + (1 ^ endian));
rtx op2 = gen_rtx_REG (SFmode, true_regnum (operands[2]) + (1 ^ endian));
operands[4] = gen_rtx_REG (SFmode, true_regnum (operands[0]) + (1 ^ endian));
operands[5] = gen_rtx_fmt_ee (GET_CODE (operands[3]), SFmode, op1, op2);
}
[(set_attr "type" "fparith_media")])
(define_insn "addsf3_i"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(plus:SF (match_operand:SF 1 "fp_arith_reg_operand" "%0")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E"
"fadd %2,%0"
[(set_attr "type" "fp")
(set_attr "fp_mode" "single")])
(define_expand "subsf3"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "")
(minus:SF (match_operand:SF 1 "fp_arith_reg_operand" "")
(match_operand:SF 2 "fp_arith_reg_operand" "")))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH2E)
{
emit_insn (gen_subsf3_i (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*subsf3_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(minus:SF (match_operand:SF 1 "fp_arith_reg_operand" "f")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fsub.s %1, %2, %0"
[(set_attr "type" "fparith_media")])
(define_insn "subsf3_i"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(minus:SF (match_operand:SF 1 "fp_arith_reg_operand" "0")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E"
"fsub %2,%0"
[(set_attr "type" "fp")
(set_attr "fp_mode" "single")])
(define_expand "mulsf3"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "")
(mult:SF (match_operand:SF 1 "fp_arith_reg_operand" "")
(match_operand:SF 2 "fp_arith_reg_operand" "")))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH2E)
{
emit_insn (gen_mulsf3_i (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*mulsf3_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(mult:SF (match_operand:SF 1 "fp_arith_reg_operand" "%f")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fmul.s %1, %2, %0"
[(set_attr "type" "fparith_media")])
(define_insn "mulsf3_i"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(mult:SF (match_operand:SF 1 "fp_arith_reg_operand" "%0")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E"
"fmul %2,%0"
[(set_attr "type" "fp")
(set_attr "fp_mode" "single")])
;; FMA (fused multiply-add) patterns
(define_expand "fmasf4"
[(set (match_operand:SF 0 "fp_arith_reg_operand")
(fma:SF (match_operand:SF 1 "fp_arith_reg_operand")
(match_operand:SF 2 "fp_arith_reg_operand")
(match_operand:SF 3 "fp_arith_reg_operand")))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH2E)
{
emit_insn (gen_fmasf4_i (operands[0], operands[1], operands[2],
operands[3]));
DONE;
}
})
(define_insn "fmasf4_i"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(fma:SF (match_operand:SF 1 "fp_arith_reg_operand" "w")
(match_operand:SF 2 "fp_arith_reg_operand" "f")
(match_operand:SF 3 "fp_arith_reg_operand" "0")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E"
"fmac %1,%2,%0"
[(set_attr "type" "fp")
(set_attr "fp_mode" "single")])
(define_insn "fmasf4_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(fma:SF (match_operand:SF 1 "fp_arith_reg_operand" "f")
(match_operand:SF 2 "fp_arith_reg_operand" "f")
(match_operand:SF 3 "fp_arith_reg_operand" "0")))]
"TARGET_SHMEDIA_FPU"
"fmac.s %1, %2, %0"
[(set_attr "type" "fparith_media")])
;; For some cases such as 'a * b + a' the FMA pattern is not generated by
;; previous transformations. If FMA is generally allowed, let the combine
;; pass utilize it.
(define_insn_and_split "*fmasf4"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(plus:SF (mult:SF (match_operand:SF 1 "fp_arith_reg_operand" "%w")
(match_operand:SF 2 "fp_arith_reg_operand" "f"))
(match_operand:SF 3 "arith_reg_operand" "0")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E && flag_fp_contract_mode != FP_CONTRACT_OFF"
"fmac %1,%2,%0"
"&& can_create_pseudo_p ()"
[(parallel [(set (match_dup 0)
(fma:SF (match_dup 1) (match_dup 2) (match_dup 3)))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))])]
{
/* Change 'b * a + a' into 'a * b + a'.
This is better for register allocation. */
if (REGNO (operands[2]) == REGNO (operands[3]))
{
rtx tmp = operands[1];
operands[1] = operands[2];
operands[2] = tmp;
}
}
[(set_attr "type" "fp")
(set_attr "fp_mode" "single")])
(define_insn "*fmasf4_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(plus:SF (mult:SF (match_operand:SF 1 "fp_arith_reg_operand" "%f")
(match_operand:SF 2 "fp_arith_reg_operand" "f"))
(match_operand:SF 3 "fp_arith_reg_operand" "0")))]
"TARGET_SHMEDIA_FPU && flag_fp_contract_mode != FP_CONTRACT_OFF"
"fmac.s %1, %2, %0"
[(set_attr "type" "fparith_media")])
(define_expand "divsf3"
[(set (match_operand:SF 0 "fp_arith_reg_operand")
(div:SF (match_operand:SF 1 "fp_arith_reg_operand")
(match_operand:SF 2 "fp_arith_reg_operand")))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH2E)
{
emit_insn (gen_divsf3_i (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*divsf3_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(div:SF (match_operand:SF 1 "fp_arith_reg_operand" "f")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fdiv.s %1, %2, %0"
[(set_attr "type" "fdiv_media")])
(define_insn "divsf3_i"
[(set (match_operand:SF 0 "fp_arith_reg_dest" "=f")
(div:SF (match_operand:SF 1 "fp_arith_reg_operand" "0")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E"
"fdiv %2,%0"
[(set_attr "type" "fdiv")
(set_attr "fp_mode" "single")])
(define_insn "floatdisf2"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(float:SF (match_operand:DI 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"float.qs %1, %0"
[(set_attr "type" "fpconv_media")])
(define_expand "floatsisf2"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "")
(float:SF (match_operand:SI 1 "fpul_operand" "")))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU"
{
if (!TARGET_SHMEDIA_FPU)
{
emit_insn (gen_floatsisf2_i4 (operands[0], operands[1]));
DONE;
}
})
(define_insn "*floatsisf2_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(float:SF (match_operand:SI 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"float.ls %1, %0"
[(set_attr "type" "fpconv_media")])
(define_insn "floatsisf2_i4"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(float:SF (match_operand:SI 1 "fpul_operand" "y")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E"
"float %1,%0"
[(set_attr "type" "fp")
(set_attr "fp_mode" "single")])
(define_insn "fix_truncsfdi2"
[(set (match_operand:DI 0 "fp_arith_reg_dest" "=f")
(fix:DI (match_operand:SF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"ftrc.sq %1, %0"
[(set_attr "type" "fpconv_media")])
(define_expand "fix_truncsfsi2"
[(set (match_operand:SI 0 "fpul_operand" "=y")
(fix:SI (match_operand:SF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU"
{
if (!TARGET_SHMEDIA_FPU)
{
emit_insn (gen_fix_truncsfsi2_i4 (operands[0], operands[1]));
DONE;
}
})
(define_insn "*fix_truncsfsi2_media"
[(set (match_operand:SI 0 "fp_arith_reg_operand" "=f")
(fix:SI (match_operand:SF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"ftrc.sl %1, %0"
[(set_attr "type" "fpconv_media")])
(define_insn "fix_truncsfsi2_i4"
[(set (match_operand:SI 0 "fpul_operand" "=y")
(fix:SI (match_operand:SF 1 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E"
"ftrc %1,%0"
[(set_attr "type" "ftrc_s")
(set_attr "fp_mode" "single")])
(define_insn "cmpgtsf_t"
[(set (reg:SI T_REG)
(gt:SI (match_operand:SF 0 "fp_arith_reg_operand" "f")
(match_operand:SF 1 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E || TARGET_SH4 || TARGET_SH2A_SINGLE"
"fcmp/gt %1,%0"
[(set_attr "type" "fp_cmp")
(set_attr "fp_mode" "single")])
(define_insn "cmpeqsf_t"
[(set (reg:SI T_REG)
(eq:SI (match_operand:SF 0 "fp_arith_reg_operand" "f")
(match_operand:SF 1 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH2E || TARGET_SH4 || TARGET_SH2A_SINGLE"
"fcmp/eq %1,%0"
[(set_attr "type" "fp_cmp")
(set_attr "fp_mode" "single")])
(define_insn "ieee_ccmpeqsf_t"
[(set (reg:SI T_REG)
(ior:SI (reg:SI T_REG)
(eq:SI (match_operand:SF 0 "fp_arith_reg_operand" "f")
(match_operand:SF 1 "fp_arith_reg_operand" "f"))))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_IEEE && TARGET_SH2E"
{
return output_ieee_ccmpeq (insn, operands);
}
[(set_attr "length" "4")
(set_attr "fp_mode" "single")])
(define_insn "cmpeqsf_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(eq:SI (match_operand:SF 1 "fp_arith_reg_operand" "f")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcmpeq.s %1, %2, %0"
[(set_attr "type" "fcmp_media")])
(define_insn "cmpgtsf_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(gt:SI (match_operand:SF 1 "fp_arith_reg_operand" "f")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcmpgt.s %1, %2, %0"
[(set_attr "type" "fcmp_media")])
(define_insn "cmpgesf_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(ge:SI (match_operand:SF 1 "fp_arith_reg_operand" "f")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcmpge.s %1, %2, %0"
[(set_attr "type" "fcmp_media")])
(define_insn "cmpunsf_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(unordered:SI (match_operand:SF 1 "fp_arith_reg_operand" "f")
(match_operand:SF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcmpun.s %1, %2, %0"
[(set_attr "type" "fcmp_media")])
(define_expand "cbranchsf4"
[(set (pc)
(if_then_else (match_operator 0 "sh_float_comparison_operator"
[(match_operand:SF 1 "arith_operand" "")
(match_operand:SF 2 "arith_operand" "")])
(match_operand 3 "" "")
(pc)))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU"
{
if (TARGET_SHMEDIA)
emit_jump_insn (gen_cbranchfp4_media (operands[0], operands[1], operands[2],
operands[3]));
else
sh_emit_compare_and_branch (operands, SFmode);
DONE;
})
(define_expand "negsf2"
[(set (match_operand:SF 0 "fp_arith_reg_operand")
(neg:SF (match_operand:SF 1 "fp_arith_reg_operand")))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU")
(define_insn "*negsf2_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(neg:SF (match_operand:SF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fneg.s %1, %0"
[(set_attr "type" "fmove_media")])
(define_insn "*negsf2_i"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(neg:SF (match_operand:SF 1 "fp_arith_reg_operand" "0")))]
"TARGET_SH2E"
"fneg %0"
[(set_attr "type" "fmove")])
(define_expand "sqrtsf2"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "")
(sqrt:SF (match_operand:SF 1 "fp_arith_reg_operand" "")))]
"TARGET_SH3E || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH3E)
{
emit_insn (gen_sqrtsf2_i (operands[0], operands[1]));
DONE;
}
})
(define_insn "*sqrtsf2_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(sqrt:SF (match_operand:SF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fsqrt.s %1, %0"
[(set_attr "type" "fdiv_media")])
(define_insn "sqrtsf2_i"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(sqrt:SF (match_operand:SF 1 "fp_arith_reg_operand" "0")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_SH3E"
"fsqrt %0"
[(set_attr "type" "fdiv")
(set_attr "fp_mode" "single")])
(define_insn "rsqrtsf2"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(div:SF (match_operand:SF 1 "immediate_operand" "i")
(sqrt:SF (match_operand:SF 2 "fp_arith_reg_operand" "0"))))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_FPU_ANY && TARGET_FSRRA
&& operands[1] == CONST1_RTX (SFmode)"
"fsrra %0"
[(set_attr "type" "fsrra")
(set_attr "fp_mode" "single")])
;; When the sincos pattern is defined, the builtin functions sin and cos
;; will be expanded to the sincos pattern and one of the output values will
;; remain unused.
(define_expand "sincossf3"
[(set (match_operand:SF 0 "nonimmediate_operand")
(unspec:SF [(match_operand:SF 2 "fp_arith_reg_operand")] UNSPEC_FCOSA))
(set (match_operand:SF 1 "nonimmediate_operand")
(unspec:SF [(match_dup 2)] UNSPEC_FSINA))]
"TARGET_FPU_ANY && TARGET_FSCA"
{
rtx scaled = gen_reg_rtx (SFmode);
rtx truncated = gen_reg_rtx (SImode);
rtx fsca = gen_reg_rtx (V2SFmode);
rtx scale_reg = force_reg (SFmode, sh_fsca_sf2int ());
emit_insn (gen_mulsf3 (scaled, operands[2], scale_reg));
emit_insn (gen_fix_truncsfsi2 (truncated, scaled));
emit_insn (gen_fsca (fsca, truncated, sh_fsca_int2sf ()));
emit_move_insn (operands[0], gen_rtx_SUBREG (SFmode, fsca, 4));
emit_move_insn (operands[1], gen_rtx_SUBREG (SFmode, fsca, 0));
DONE;
})
(define_insn_and_split "fsca"
[(set (match_operand:V2SF 0 "fp_arith_reg_operand" "=f")
(vec_concat:V2SF
(unspec:SF [(mult:SF
(float:SF (match_operand:SI 1 "fpul_fsca_operand" "y"))
(match_operand:SF 2 "fsca_scale_factor" "i"))
] UNSPEC_FSINA)
(unspec:SF [(mult:SF (float:SF (match_dup 1)) (match_dup 2))
] UNSPEC_FCOSA)))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_FPU_ANY && TARGET_FSCA"
"fsca fpul,%d0"
"&& !fpul_operand (operands[1], SImode)"
[(const_int 0)]
{
/* If operands[1] is something like (fix:SF (float:SF (reg:SI))) reduce it
to a simple reg, otherwise reload will have trouble reloading the
pseudo into fpul. */
rtx x = XEXP (operands[1], 0);
while (x != NULL_RTX && !fpul_operand (x, SImode))
{
gcc_assert (GET_CODE (x) == FIX || GET_CODE (x) == FLOAT);
x = XEXP (x, 0);
}
gcc_assert (x != NULL_RTX && fpul_operand (x, SImode));
emit_insn (gen_fsca (operands[0], x, operands[2]));
DONE;
}
[(set_attr "type" "fsca")
(set_attr "fp_mode" "single")])
(define_expand "abssf2"
[(set (match_operand:SF 0 "fp_arith_reg_operand")
(abs:SF (match_operand:SF 1 "fp_arith_reg_operand")))]
"TARGET_SH2E || TARGET_SHMEDIA_FPU")
(define_insn "*abssf2_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(abs:SF (match_operand:SF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fabs.s %1, %0"
[(set_attr "type" "fmove_media")])
(define_insn "*abssf2_i"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(abs:SF (match_operand:SF 1 "fp_arith_reg_operand" "0")))]
"TARGET_SH2E"
"fabs %0"
[(set_attr "type" "fmove")])
(define_expand "adddf3"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "")
(plus:DF (match_operand:DF 1 "fp_arith_reg_operand" "")
(match_operand:DF 2 "fp_arith_reg_operand" "")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_adddf3_i (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*adddf3_media"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(plus:DF (match_operand:DF 1 "fp_arith_reg_operand" "%f")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fadd.d %1, %2, %0"
[(set_attr "type" "dfparith_media")])
(define_insn "adddf3_i"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(plus:DF (match_operand:DF 1 "fp_arith_reg_operand" "%0")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fadd %2,%0"
[(set_attr "type" "dfp_arith")
(set_attr "fp_mode" "double")])
(define_expand "subdf3"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "")
(minus:DF (match_operand:DF 1 "fp_arith_reg_operand" "")
(match_operand:DF 2 "fp_arith_reg_operand" "")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_subdf3_i (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*subdf3_media"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(minus:DF (match_operand:DF 1 "fp_arith_reg_operand" "f")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fsub.d %1, %2, %0"
[(set_attr "type" "dfparith_media")])
(define_insn "subdf3_i"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(minus:DF (match_operand:DF 1 "fp_arith_reg_operand" "0")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fsub %2,%0"
[(set_attr "type" "dfp_arith")
(set_attr "fp_mode" "double")])
(define_expand "muldf3"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "")
(mult:DF (match_operand:DF 1 "fp_arith_reg_operand" "")
(match_operand:DF 2 "fp_arith_reg_operand" "")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_muldf3_i (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*muldf3_media"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(mult:DF (match_operand:DF 1 "fp_arith_reg_operand" "%f")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fmul.d %1, %2, %0"
[(set_attr "type" "dfmul_media")])
(define_insn "muldf3_i"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(mult:DF (match_operand:DF 1 "fp_arith_reg_operand" "%0")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fmul %2,%0"
[(set_attr "type" "dfp_mul")
(set_attr "fp_mode" "double")])
(define_expand "divdf3"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "")
(div:DF (match_operand:DF 1 "fp_arith_reg_operand" "")
(match_operand:DF 2 "fp_arith_reg_operand" "")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_divdf3_i (operands[0], operands[1], operands[2]));
DONE;
}
})
(define_insn "*divdf3_media"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(div:DF (match_operand:DF 1 "fp_arith_reg_operand" "f")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fdiv.d %1, %2, %0"
[(set_attr "type" "dfdiv_media")])
(define_insn "divdf3_i"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(div:DF (match_operand:DF 1 "fp_arith_reg_operand" "0")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fdiv %2,%0"
[(set_attr "type" "dfdiv")
(set_attr "fp_mode" "double")])
(define_insn "floatdidf2"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(float:DF (match_operand:DI 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"float.qd %1, %0"
[(set_attr "type" "dfpconv_media")])
(define_expand "floatsidf2"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "")
(float:DF (match_operand:SI 1 "fpul_operand" "")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_floatsidf2_i (operands[0], operands[1]));
DONE;
}
})
(define_insn "*floatsidf2_media"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(float:DF (match_operand:SI 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"float.ld %1, %0"
[(set_attr "type" "dfpconv_media")])
(define_insn "floatsidf2_i"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(float:DF (match_operand:SI 1 "fpul_operand" "y")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"float %1,%0"
[(set_attr "type" "dfp_conv")
(set_attr "fp_mode" "double")])
(define_insn "fix_truncdfdi2"
[(set (match_operand:DI 0 "fp_arith_reg_dest" "=f")
(fix:DI (match_operand:DF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"ftrc.dq %1, %0"
[(set_attr "type" "dfpconv_media")])
(define_expand "fix_truncdfsi2"
[(set (match_operand:SI 0 "fpul_operand" "")
(fix:SI (match_operand:DF 1 "fp_arith_reg_operand" "")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_fix_truncdfsi2_i (operands[0], operands[1]));
DONE;
}
})
(define_insn "*fix_truncdfsi2_media"
[(set (match_operand:SI 0 "fp_arith_reg_operand" "=f")
(fix:SI (match_operand:DF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"ftrc.dl %1, %0"
[(set_attr "type" "dfpconv_media")])
(define_insn "fix_truncdfsi2_i"
[(set (match_operand:SI 0 "fpul_operand" "=y")
(fix:SI (match_operand:DF 1 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"ftrc %1,%0"
[(set_attr "type" "dfp_conv")
(set_attr "dfp_comp" "no")
(set_attr "fp_mode" "double")])
(define_insn "cmpgtdf_t"
[(set (reg:SI T_REG)
(gt:SI (match_operand:DF 0 "fp_arith_reg_operand" "f")
(match_operand:DF 1 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fcmp/gt %1,%0"
[(set_attr "type" "dfp_cmp")
(set_attr "fp_mode" "double")])
(define_insn "cmpeqdf_t"
[(set (reg:SI T_REG)
(eq:SI (match_operand:DF 0 "fp_arith_reg_operand" "f")
(match_operand:DF 1 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fcmp/eq %1,%0"
[(set_attr "type" "dfp_cmp")
(set_attr "fp_mode" "double")])
(define_insn "*ieee_ccmpeqdf_t"
[(set (reg:SI T_REG)
(ior:SI (reg:SI T_REG)
(eq:SI (match_operand:DF 0 "fp_arith_reg_operand" "f")
(match_operand:DF 1 "fp_arith_reg_operand" "f"))))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"TARGET_IEEE && (TARGET_SH4 || TARGET_SH2A_DOUBLE)"
{
return output_ieee_ccmpeq (insn, operands);
}
[(set_attr "length" "4")
(set_attr "fp_mode" "double")])
(define_insn "cmpeqdf_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(eq:SI (match_operand:DF 1 "fp_arith_reg_operand" "f")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcmpeq.d %1,%2,%0"
[(set_attr "type" "fcmp_media")])
(define_insn "cmpgtdf_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(gt:SI (match_operand:DF 1 "fp_arith_reg_operand" "f")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcmpgt.d %1,%2,%0"
[(set_attr "type" "fcmp_media")])
(define_insn "cmpgedf_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(ge:SI (match_operand:DF 1 "fp_arith_reg_operand" "f")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcmpge.d %1,%2,%0"
[(set_attr "type" "fcmp_media")])
(define_insn "cmpundf_media"
[(set (match_operand:SI 0 "register_operand" "=r")
(unordered:SI (match_operand:DF 1 "fp_arith_reg_operand" "f")
(match_operand:DF 2 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcmpun.d %1,%2,%0"
[(set_attr "type" "fcmp_media")])
(define_expand "cbranchdf4"
[(set (pc)
(if_then_else (match_operator 0 "sh_float_comparison_operator"
[(match_operand:DF 1 "arith_operand" "")
(match_operand:DF 2 "arith_operand" "")])
(match_operand 3 "" "")
(pc)))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SHMEDIA)
emit_jump_insn (gen_cbranchfp4_media (operands[0], operands[1], operands[2],
operands[3]));
else
sh_emit_compare_and_branch (operands, DFmode);
DONE;
})
(define_expand "negdf2"
[(set (match_operand:DF 0 "fp_arith_reg_operand")
(neg:DF (match_operand:DF 1 "fp_arith_reg_operand")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU")
(define_insn "*negdf2_media"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(neg:DF (match_operand:DF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fneg.d %1, %0"
[(set_attr "type" "fmove_media")])
(define_insn "*negdf2_i"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(neg:DF (match_operand:DF 1 "fp_arith_reg_operand" "0")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fneg %0"
[(set_attr "type" "fmove")])
(define_expand "sqrtdf2"
[(set (match_operand:DF 0 "fp_arith_reg_operand")
(sqrt:DF (match_operand:DF 1 "fp_arith_reg_operand")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_sqrtdf2_i (operands[0], operands[1]));
DONE;
}
})
(define_insn "*sqrtdf2_media"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(sqrt:DF (match_operand:DF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fsqrt.d %1, %0"
[(set_attr "type" "dfdiv_media")])
(define_insn "sqrtdf2_i"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(sqrt:DF (match_operand:DF 1 "fp_arith_reg_operand" "0")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fsqrt %0"
[(set_attr "type" "dfdiv")
(set_attr "fp_mode" "double")])
(define_expand "absdf2"
[(set (match_operand:DF 0 "fp_arith_reg_operand")
(abs:DF (match_operand:DF 1 "fp_arith_reg_operand")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU")
(define_insn "*absdf2_media"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(abs:DF (match_operand:DF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fabs.d %1, %0"
[(set_attr "type" "fmove_media")])
(define_insn "*absdf2_i"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(abs:DF (match_operand:DF 1 "fp_arith_reg_operand" "0")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fabs %0"
[(set_attr "type" "fmove")])
(define_expand "extendsfdf2"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "")
(float_extend:DF (match_operand:SF 1 "fpul_operand" "")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_extendsfdf2_i4 (operands[0], operands[1]));
DONE;
}
})
(define_insn "*extendsfdf2_media"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(float_extend:DF (match_operand:SF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcnv.sd %1, %0"
[(set_attr "type" "dfpconv_media")])
(define_insn "extendsfdf2_i4"
[(set (match_operand:DF 0 "fp_arith_reg_operand" "=f")
(float_extend:DF (match_operand:SF 1 "fpul_operand" "y")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fcnvsd %1,%0"
[(set_attr "type" "fp")
(set_attr "fp_mode" "double")])
(define_expand "truncdfsf2"
[(set (match_operand:SF 0 "fpul_operand" "")
(float_truncate:SF (match_operand:DF 1 "fp_arith_reg_operand" "")))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE) || TARGET_SHMEDIA_FPU"
{
if (TARGET_SH4 || TARGET_SH2A_DOUBLE)
{
emit_insn (gen_truncdfsf2_i4 (operands[0], operands[1]));
DONE;
}
})
(define_insn "*truncdfsf2_media"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(float_truncate:SF (match_operand:DF 1 "fp_arith_reg_operand" "f")))]
"TARGET_SHMEDIA_FPU"
"fcnv.ds %1, %0"
[(set_attr "type" "dfpconv_media")])
(define_insn "truncdfsf2_i4"
[(set (match_operand:SF 0 "fpul_operand" "=y")
(float_truncate:SF (match_operand:DF 1 "fp_arith_reg_operand" "f")))
(clobber (reg:SI FPSCR_STAT_REG))
(use (reg:SI FPSCR_MODES_REG))]
"(TARGET_SH4 || TARGET_SH2A_DOUBLE)"
"fcnvds %1,%0"
[(set_attr "type" "fp")
(set_attr "fp_mode" "double")])
;; -------------------------------------------------------------------------
;; Bit field extract patterns.
;; -------------------------------------------------------------------------
;; These give better code for packed bitfields, because they allow
;; auto-increment addresses to be generated.
(define_expand "insv"
[(set (zero_extract:SI (match_operand:QI 0 "memory_operand" "")
(match_operand:SI 1 "immediate_operand" "")
(match_operand:SI 2 "immediate_operand" ""))
(match_operand:SI 3 "general_operand" ""))]
"TARGET_SH1 && TARGET_BIG_ENDIAN"
{
rtx addr_target, orig_address, shift_reg, qi_val;
HOST_WIDE_INT bitsize, size, v = 0;
rtx x = operands[3];
if (TARGET_SH2A && TARGET_BITOPS
&& (satisfies_constraint_Sbw (operands[0])
|| satisfies_constraint_Sbv (operands[0]))
&& satisfies_constraint_M (operands[1])
&& satisfies_constraint_K03 (operands[2]))
{
if (satisfies_constraint_N (operands[3]))
{
emit_insn (gen_bclr_m2a (operands[0], operands[2]));
DONE;
}
else if (satisfies_constraint_M (operands[3]))
{
emit_insn (gen_bset_m2a (operands[0], operands[2]));
DONE;
}
else if ((REG_P (operands[3]) && REGNO (operands[3]) == T_REG)
&& satisfies_constraint_M (operands[1]))
{
emit_insn (gen_bst_m2a (operands[0], operands[2]));
DONE;
}
else if (REG_P (operands[3])
&& satisfies_constraint_M (operands[1]))
{
emit_insn (gen_bld_reg (operands[3], const0_rtx));
emit_insn (gen_bst_m2a (operands[0], operands[2]));
DONE;
}
}
/* ??? expmed doesn't care for non-register predicates. */
if (! memory_operand (operands[0], VOIDmode)
|| ! immediate_operand (operands[1], VOIDmode)
|| ! immediate_operand (operands[2], VOIDmode)
|| ! general_operand (x, VOIDmode))
FAIL;
/* If this isn't a 16 / 24 / 32 bit field, or if
it doesn't start on a byte boundary, then fail. */
bitsize = INTVAL (operands[1]);
if (bitsize < 16 || bitsize > 32 || bitsize % 8 != 0
|| (INTVAL (operands[2]) % 8) != 0)
FAIL;
size = bitsize / 8;
orig_address = XEXP (operands[0], 0);
shift_reg = gen_reg_rtx (SImode);
if (CONST_INT_P (x))
{
v = INTVAL (x);
qi_val = force_reg (QImode, GEN_INT (trunc_int_for_mode (v, QImode)));
}
else
{
emit_insn (gen_movsi (shift_reg, operands[3]));
qi_val = gen_rtx_SUBREG (QImode, shift_reg, 3);
}
addr_target = copy_addr_to_reg (plus_constant (Pmode,
orig_address, size - 1));
operands[0] = replace_equiv_address (operands[0], addr_target);
emit_insn (gen_movqi (operands[0], qi_val));
while (size -= 1)
{
if (CONST_INT_P (x))
qi_val
= force_reg (QImode, GEN_INT (trunc_int_for_mode (v >>= 8, QImode)));
else
{
emit_insn (gen_lshrsi3_k (shift_reg, shift_reg, GEN_INT (8)));
qi_val = gen_rtx_SUBREG (QImode, shift_reg, 3);
}
emit_insn (gen_addsi3 (addr_target, addr_target, constm1_rtx));
emit_insn (gen_movqi (operands[0], qi_val));
}
DONE;
})
(define_insn "movua"
[(set (match_operand:SI 0 "register_operand" "=z")
(unspec:SI [(match_operand:BLK 1 "unaligned_load_operand" "Sua>")]
UNSPEC_MOVUA))]
"TARGET_SH4A"
"movua.l %1,%0"
[(set_attr "type" "movua")])
;; We shouldn't need this, but cse replaces increments with references
;; to other regs before flow has a chance to create post_inc
;; addressing modes, and only postreload's cse_move2add brings the
;; increments back to a usable form.
(define_peephole2
[(set (match_operand:SI 0 "register_operand" "")
(sign_extract:SI (mem:SI (match_operand:SI 1 "register_operand" ""))
(const_int 32) (const_int 0)))
(set (match_dup 1) (plus:SI (match_dup 1) (const_int 4)))]
"TARGET_SH4A && REGNO (operands[0]) != REGNO (operands[1])"
[(set (match_operand:SI 0 "register_operand" "")
(sign_extract:SI (mem:SI (post_inc:SI
(match_operand:SI 1 "register_operand" "")))
(const_int 32) (const_int 0)))]
"")
(define_expand "extv"
[(set (match_operand:SI 0 "register_operand" "")
(sign_extract:SI (match_operand:QI 1 "unaligned_load_operand" "")
(match_operand 2 "const_int_operand" "")
(match_operand 3 "const_int_operand" "")))]
"TARGET_SH4A || TARGET_SH2A"
{
if (TARGET_SH2A && TARGET_BITOPS
&& (satisfies_constraint_Sbw (operands[1])
|| satisfies_constraint_Sbv (operands[1]))
&& satisfies_constraint_M (operands[2])
&& satisfies_constraint_K03 (operands[3]))
{
emit_insn (gen_bldsign_m2a (operands[1], operands[3]));
if (REGNO (operands[0]) != T_REG)
emit_insn (gen_movsi (operands[0], gen_rtx_REG (SImode, T_REG)));
DONE;
}
if (TARGET_SH4A
&& INTVAL (operands[2]) == 32
&& INTVAL (operands[3]) == 0
&& MEM_P (operands[1]) && MEM_ALIGN (operands[1]) < 32)
{
rtx src = adjust_address (operands[1], BLKmode, 0);
set_mem_size (src, 4);
emit_insn (gen_movua (operands[0], src));
DONE;
}
FAIL;
})
(define_expand "extzv"
[(set (match_operand:SI 0 "register_operand" "")
(zero_extract:SI (match_operand:QI 1 "unaligned_load_operand" "")
(match_operand 2 "const_int_operand" "")
(match_operand 3 "const_int_operand" "")))]
"TARGET_SH4A || TARGET_SH2A"
{
if (TARGET_SH2A && TARGET_BITOPS
&& (satisfies_constraint_Sbw (operands[1])
|| satisfies_constraint_Sbv (operands[1]))
&& satisfies_constraint_M (operands[2])
&& satisfies_constraint_K03 (operands[3]))
{
emit_insn (gen_bld_m2a (operands[1], operands[3]));
if (REGNO (operands[0]) != T_REG)
emit_insn (gen_movsi (operands[0], gen_rtx_REG (SImode, T_REG)));
DONE;
}
if (TARGET_SH4A
&& INTVAL (operands[2]) == 32
&& INTVAL (operands[3]) == 0
&& MEM_P (operands[1]) && MEM_ALIGN (operands[1]) < 32)
{
rtx src = adjust_address (operands[1], BLKmode, 0);
set_mem_size (src, 4);
emit_insn (gen_movua (operands[0], src));
DONE;
}
FAIL;
})
;; SH2A instructions for bitwise operations.
;; FIXME: Convert multiple instruction insns to insn_and_split.
;; FIXME: Use iterators to fold at least and,xor,or insn variations.
;; Clear a bit in a memory location.
(define_insn "bclr_m2a"
[(set (match_operand:QI 0 "bitwise_memory_operand" "+Sbw,Sbv")
(and:QI
(not:QI (ashift:QI (const_int 1)
(match_operand:QI 1 "const_int_operand" "K03,K03")))
(match_dup 0)))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[1])"
"@
bclr.b %1,%0
bclr.b %1,@(0,%t0)"
[(set_attr "length" "4,4")])
(define_insn "bclrmem_m2a"
[(set (match_operand:QI 0 "bitwise_memory_operand" "+Sbw,Sbv")
(and:QI (match_dup 0)
(match_operand:QI 1 "const_int_operand" "Psz,Psz")))]
"TARGET_SH2A && satisfies_constraint_Psz (operands[1]) && TARGET_BITOPS"
"@
bclr.b %W1,%0
bclr.b %W1,@(0,%t0)"
[(set_attr "length" "4,4")])
;; Set a bit in a memory location.
(define_insn "bset_m2a"
[(set (match_operand:QI 0 "bitwise_memory_operand" "+Sbw,Sbv")
(ior:QI
(ashift:QI (const_int 1)
(match_operand:QI 1 "const_int_operand" "K03,K03"))
(match_dup 0)))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[1])"
"@
bset.b %1,%0
bset.b %1,@(0,%t0)"
[(set_attr "length" "4,4")])
(define_insn "bsetmem_m2a"
[(set (match_operand:QI 0 "bitwise_memory_operand" "+Sbw,Sbv")
(ior:QI (match_dup 0)
(match_operand:QI 1 "const_int_operand" "Pso,Pso")))]
"TARGET_SH2A && satisfies_constraint_Pso (operands[1]) && TARGET_BITOPS"
"@
bset.b %V1,%0
bset.b %V1,@(0,%t0)"
[(set_attr "length" "4,4")])
;;; Transfer the contents of the T bit to a specified bit of memory.
(define_insn "bst_m2a"
[(set (match_operand:QI 0 "bitwise_memory_operand" "+Sbw,m")
(if_then_else (eq (reg:SI T_REG) (const_int 0))
(and:QI
(not:QI (ashift:QI (const_int 1)
(match_operand:QI 1 "const_int_operand" "K03,K03")))
(match_dup 0))
(ior:QI
(ashift:QI (const_int 1) (match_dup 1))
(match_dup 0))))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[1])"
"@
bst.b %1,%0
bst.b %1,@(0,%t0)"
[(set_attr "length" "4")])
;; Store a specified bit of memory in the T bit.
(define_insn "bld_m2a"
[(set (reg:SI T_REG)
(zero_extract:SI
(match_operand:QI 0 "bitwise_memory_operand" "Sbw,Sbv")
(const_int 1)
(match_operand 1 "const_int_operand" "K03,K03")))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[1])"
"@
bld.b %1,%0
bld.b %1,@(0,%t0)"
[(set_attr "length" "4,4")])
;; Store a specified bit of memory in the T bit.
(define_insn "bldsign_m2a"
[(set (reg:SI T_REG)
(sign_extract:SI
(match_operand:QI 0 "bitwise_memory_operand" "Sbw,m")
(const_int 1)
(match_operand 1 "const_int_operand" "K03,K03")))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[1])"
"@
bld.b %1,%0
bld.b %1,@(0,%t0)"
[(set_attr "length" "4,4")])
;; Store a specified bit of the LSB 8 bits of a register in the T bit.
(define_insn "bld_reg"
[(set (reg:SI T_REG)
(zero_extract:SI (match_operand:SI 0 "arith_reg_operand" "r")
(const_int 1)
(match_operand 1 "const_int_operand" "K03")))]
"TARGET_SH2A && satisfies_constraint_K03 (operands[1])"
"bld %1,%0")
(define_insn "*bld_regqi"
[(set (reg:SI T_REG)
(zero_extract:SI (match_operand:QI 0 "arith_reg_operand" "r")
(const_int 1)
(match_operand 1 "const_int_operand" "K03")))]
"TARGET_SH2A && satisfies_constraint_K03 (operands[1])"
"bld %1,%0")
;; Take logical and of a specified bit of memory with the T bit and
;; store its result in the T bit.
(define_insn "band_m2a"
[(set (reg:SI T_REG)
(and:SI (reg:SI T_REG)
(zero_extract:SI
(match_operand:QI 0 "bitwise_memory_operand" "Sbw,m")
(const_int 1)
(match_operand 1 "const_int_operand" "K03,K03"))))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[1])"
"@
band.b %1,%0
band.b %1,@(0,%t0)"
[(set_attr "length" "4,4")])
(define_insn "bandreg_m2a"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(and:SI (zero_extract:SI
(match_operand:QI 1 "bitwise_memory_operand" "Sbw,Sbv")
(const_int 1)
(match_operand 2 "const_int_operand" "K03,K03"))
(match_operand:SI 3 "register_operand" "r,r")))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[2])"
{
static const char* alt[] =
{
"band.b %2,%1" "\n"
" movt %0",
"band.b %2,@(0,%t1)" "\n"
" movt %0"
};
return alt[which_alternative];
}
[(set_attr "length" "6,6")])
;; Take logical or of a specified bit of memory with the T bit and
;; store its result in the T bit.
(define_insn "bor_m2a"
[(set (reg:SI T_REG)
(ior:SI (reg:SI T_REG)
(zero_extract:SI
(match_operand:QI 0 "bitwise_memory_operand" "Sbw,m")
(const_int 1)
(match_operand 1 "const_int_operand" "K03,K03"))))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[1])"
"@
bor.b %1,%0
bor.b %1,@(0,%t0)"
[(set_attr "length" "4,4")])
(define_insn "borreg_m2a"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(ior:SI (zero_extract:SI
(match_operand:QI 1 "bitwise_memory_operand" "Sbw,Sbv")
(const_int 1)
(match_operand 2 "const_int_operand" "K03,K03"))
(match_operand:SI 3 "register_operand" "=r,r")))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[2])"
{
static const char* alt[] =
{
"bor.b %2,%1" "\n"
" movt %0",
"bor.b %2,@(0,%t1)" "\n"
" movt %0"
};
return alt[which_alternative];
}
[(set_attr "length" "6,6")])
;; Take exclusive or of a specified bit of memory with the T bit and
;; store its result in the T bit.
(define_insn "bxor_m2a"
[(set (reg:SI T_REG)
(xor:SI (reg:SI T_REG)
(zero_extract:SI
(match_operand:QI 0 "bitwise_memory_operand" "Sbw,m")
(const_int 1)
(match_operand 1 "const_int_operand" "K03,K03"))))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[1])"
"@
bxor.b %1,%0
bxor.b %1,@(0,%t0)"
[(set_attr "length" "4,4")])
(define_insn "bxorreg_m2a"
[(set (match_operand:SI 0 "register_operand" "=r,r")
(xor:SI (zero_extract:SI
(match_operand:QI 1 "bitwise_memory_operand" "Sbw,Sbv")
(const_int 1)
(match_operand 2 "const_int_operand" "K03,K03"))
(match_operand:SI 3 "register_operand" "=r,r")))]
"TARGET_SH2A && TARGET_BITOPS && satisfies_constraint_K03 (operands[2])"
{
static const char* alt[] =
{
"bxor.b %2,%1" "\n"
" movt %0",
"bxor.b %2,@(0,%t1)" "\n"
" movt %0"
};
return alt[which_alternative];
}
[(set_attr "length" "6,6")])
;; -------------------------------------------------------------------------
;; Peepholes
;; -------------------------------------------------------------------------
;; This matches cases where the bit in a memory location is set.
(define_peephole2
[(set (match_operand:SI 0 "register_operand")
(sign_extend:SI (match_operand:QI 1 "bitwise_memory_operand")))
(set (match_dup 0)
(ior:SI (match_dup 0)
(match_operand:SI 2 "const_int_operand")))
(set (match_dup 1)
(match_operand 3 "arith_reg_operand"))]
"TARGET_SH2A && TARGET_BITOPS
&& satisfies_constraint_Pso (operands[2])
&& REGNO (operands[0]) == REGNO (operands[3])"
[(set (match_dup 1)
(ior:QI (match_dup 1) (match_dup 2)))]
"")
;; This matches cases where the bit in a memory location is cleared.
(define_peephole2
[(set (match_operand:SI 0 "register_operand")
(sign_extend:SI (match_operand:QI 1 "bitwise_memory_operand")))
(set (match_dup 0)
(and:SI (match_dup 0)
(match_operand:SI 2 "const_int_operand")))
(set (match_dup 1)
(match_operand 3 "arith_reg_operand"))]
"TARGET_SH2A && TARGET_BITOPS
&& satisfies_constraint_Psz (operands[2])
&& REGNO (operands[0]) == REGNO (operands[3])"
[(set (match_dup 1)
(and:QI (match_dup 1) (match_dup 2)))]
"")
;; This matches cases where a stack pointer increment at the start of the
;; epilogue combines with a stack slot read loading the return value.
(define_peephole
[(set (match_operand:SI 0 "arith_reg_operand" "")
(mem:SI (match_operand:SI 1 "arith_reg_operand" "")))
(set (match_dup 1) (plus:SI (match_dup 1) (const_int 4)))]
"TARGET_SH1 && REGNO (operands[1]) != REGNO (operands[0])"
"mov.l @%1+,%0")
;; See the comment on the dt combiner pattern above.
(define_peephole
[(set (match_operand:SI 0 "arith_reg_operand" "=r")
(plus:SI (match_dup 0)
(const_int -1)))
(set (reg:SI T_REG)
(eq:SI (match_dup 0) (const_int 0)))]
"TARGET_SH2"
"dt %0")
;; The following peepholes fold load sequences for which reload was not
;; able to generate a displacement addressing move insn.
;; This can happen when reload has to transform a move insn
;; without displacement into one with displacement. Or when reload can't
;; fit a displacement into the insn's constraints. In the latter case, the
;; load destination reg remains at r0, which reload compensates by inserting
;; another mov insn.
;; Fold sequence:
;; mov #54,r0
;; mov.{b,w} @(r0,r15),r0
;; mov r0,r3
;; into:
;; mov.{b,w} @(54,r15),r3
;;
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(match_operand:SI 1 "const_int_operand" ""))
(set (match_operand:SI 2 "arith_reg_dest" "")
(sign_extend:SI
(mem:QI (plus:SI (match_dup 0)
(match_operand:SI 3 "arith_reg_operand" "")))))
(set (match_operand:QI 4 "arith_reg_dest" "")
(match_operand:QI 5 "arith_reg_operand" ""))]
"TARGET_SH2A
&& sh_legitimate_index_p (QImode, operands[1], true, true)
&& REGNO (operands[2]) == REGNO (operands[5])
&& peep2_reg_dead_p (3, operands[5])"
[(set (match_dup 4) (mem:QI (plus:SI (match_dup 3) (match_dup 1))))]
"")
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(match_operand:SI 1 "const_int_operand" ""))
(set (match_operand:SI 2 "arith_reg_dest" "")
(sign_extend:SI
(mem:HI (plus:SI (match_dup 0)
(match_operand:SI 3 "arith_reg_operand" "")))))
(set (match_operand:HI 4 "arith_reg_dest" "")
(match_operand:HI 5 "arith_reg_operand" ""))]
"TARGET_SH2A
&& sh_legitimate_index_p (HImode, operands[1], true, true)
&& REGNO (operands[2]) == REGNO (operands[5])
&& peep2_reg_dead_p (3, operands[5])"
[(set (match_dup 4) (mem:HI (plus:SI (match_dup 3) (match_dup 1))))]
"")
;; Fold sequence:
;; mov #54,r0
;; mov.{b,w} @(r0,r15),r1
;; into:
;; mov.{b,w} @(54,r15),r1
;;
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(match_operand:SI 1 "const_int_operand" ""))
(set (match_operand:SI 2 "arith_reg_dest" "")
(sign_extend:SI
(mem:QI (plus:SI (match_dup 0)
(match_operand:SI 3 "arith_reg_operand" "")))))]
"TARGET_SH2A
&& sh_legitimate_index_p (QImode, operands[1], true, true)
&& (peep2_reg_dead_p (2, operands[0])
|| REGNO (operands[0]) == REGNO (operands[2]))"
[(set (match_dup 2)
(sign_extend:SI (mem:QI (plus:SI (match_dup 3) (match_dup 1)))))]
"")
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(match_operand:SI 1 "const_int_operand" ""))
(set (match_operand:SI 2 "arith_reg_dest" "")
(sign_extend:SI
(mem:HI (plus:SI (match_dup 0)
(match_operand:SI 3 "arith_reg_operand" "")))))]
"TARGET_SH2A
&& sh_legitimate_index_p (HImode, operands[1], true, true)
&& (peep2_reg_dead_p (2, operands[0])
|| REGNO (operands[0]) == REGNO (operands[2]))"
[(set (match_dup 2)
(sign_extend:SI (mem:HI (plus:SI (match_dup 3) (match_dup 1)))))]
"")
;; Fold sequence:
;; mov.{b,w} @(r0,r15),r0
;; mov r0,r3
;; into:
;; mov.{b,w} @(r0,r15),r3
;;
;; This can happen when initially a displacement address is picked, where
;; the destination reg is fixed to r0, and then the address is transformed
;; into 'r0 + reg'.
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(sign_extend:SI
(mem:QI (plus:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "arith_reg_operand" "")))))
(set (match_operand:QI 3 "arith_reg_dest" "")
(match_operand:QI 4 "arith_reg_operand" ""))]
"TARGET_SH1
&& REGNO (operands[0]) == REGNO (operands[4])
&& peep2_reg_dead_p (2, operands[0])"
[(set (match_dup 3)
(mem:QI (plus:SI (match_dup 1) (match_dup 2))))]
"")
(define_peephole2
[(set (match_operand:SI 0 "arith_reg_dest" "")
(sign_extend:SI
(mem:HI (plus:SI (match_operand:SI 1 "arith_reg_operand" "")
(match_operand:SI 2 "arith_reg_operand" "")))))
(set (match_operand:HI 3 "arith_reg_dest" "")
(match_operand:HI 4 "arith_reg_operand" ""))]
"TARGET_SH1
&& REGNO (operands[0]) == REGNO (operands[4])
&& peep2_reg_dead_p (2, operands[0])"
[(set (match_dup 3)
(mem:HI (plus:SI (match_dup 1) (match_dup 2))))]
"")
(define_peephole
[(set (match_operand:SI 0 "register_operand" "=r")
(plus:SI (match_dup 0) (match_operand:SI 1 "register_operand" "r")))
(set (mem:SF (match_dup 0))
(match_operand:SF 2 "general_movsrc_operand" ""))]
"TARGET_SH1 && REGNO (operands[0]) == 0
&& ((REG_P (operands[2]) && REGNO (operands[2]) < 16)
|| (GET_CODE (operands[2]) == SUBREG
&& REGNO (SUBREG_REG (operands[2])) < 16))
&& reg_unused_after (operands[0], insn)"
"mov.l %2,@(%0,%1)")
(define_peephole
[(set (match_operand:SI 0 "register_operand" "=r")
(plus:SI (match_dup 0) (match_operand:SI 1 "register_operand" "r")))
(set (match_operand:SF 2 "general_movdst_operand" "")
(mem:SF (match_dup 0)))]
"TARGET_SH1 && REGNO (operands[0]) == 0
&& ((REG_P (operands[2]) && REGNO (operands[2]) < 16)
|| (GET_CODE (operands[2]) == SUBREG
&& REGNO (SUBREG_REG (operands[2])) < 16))
&& reg_unused_after (operands[0], insn)"
"mov.l @(%0,%1),%2")
(define_peephole
[(set (match_operand:SI 0 "register_operand" "=r")
(plus:SI (match_dup 0) (match_operand:SI 1 "register_operand" "r")))
(set (mem:SF (match_dup 0))
(match_operand:SF 2 "general_movsrc_operand" ""))]
"TARGET_SH2E && REGNO (operands[0]) == 0
&& ((REG_P (operands[2])
&& FP_OR_XD_REGISTER_P (REGNO (operands[2])))
|| (GET_CODE (operands[2]) == SUBREG
&& FP_OR_XD_REGISTER_P (REGNO (SUBREG_REG (operands[2])))))
&& reg_unused_after (operands[0], insn)"
"fmov{.s|} %2,@(%0,%1)")
(define_peephole
[(set (match_operand:SI 0 "register_operand" "=r")
(plus:SI (match_dup 0) (match_operand:SI 1 "register_operand" "r")))
(set (match_operand:SF 2 "general_movdst_operand" "")
(mem:SF (match_dup 0)))]
"TARGET_SH2E && REGNO (operands[0]) == 0
&& ((REG_P (operands[2])
&& FP_OR_XD_REGISTER_P (REGNO (operands[2])))
|| (GET_CODE (operands[2]) == SUBREG
&& FP_OR_XD_REGISTER_P (REGNO (SUBREG_REG (operands[2])))))
&& reg_unused_after (operands[0], insn)"
"fmov{.s|} @(%0,%1),%2")
;; Switch to a new stack with its address in sp_switch (a SYMBOL_REF).
(define_insn "sp_switch_1"
[(set (reg:SI SP_REG) (unspec_volatile [(match_operand:SI 0 "" "")]
UNSPECV_SP_SWITCH_B))]
"TARGET_SH1"
{
return "mov.l r0,@-r15" "\n"
" mov.l %0,r0" "\n"
" mov.l @r0,r0" "\n"
" mov.l r15,@-r0" "\n"
" mov r0,r15";
}
[(set_attr "length" "10")])
;; Switch back to the original stack for interrupt functions with the
;; sp_switch attribute.
(define_insn "sp_switch_2"
[(unspec_volatile [(const_int 0)]
UNSPECV_SP_SWITCH_E)]
"TARGET_SH1"
{
return "mov.l @r15,r15" "\n"
" mov.l @r15+,r0";
}
[(set_attr "length" "4")])
;; -------------------------------------------------------------------------
;; Integer vector moves
;; -------------------------------------------------------------------------
(define_expand "movv8qi"
[(set (match_operand:V8QI 0 "general_movdst_operand" "")
(match_operand:V8QI 1 "general_movsrc_operand" ""))]
"TARGET_SHMEDIA"
{
prepare_move_operands (operands, V8QImode);
})
(define_insn "movv8qi_i"
[(set (match_operand:V8QI 0 "general_movdst_operand" "=r,r,r,rl,m")
(match_operand:V8QI 1 "general_movsrc_operand" "r,I16CssZ,nW,m,rlZ"))]
"TARGET_SHMEDIA
&& (register_operand (operands[0], V8QImode)
|| sh_register_operand (operands[1], V8QImode))"
"@
add %1, r63, %0
movi %1, %0
#
ld%M1.q %m1, %0
st%M0.q %m0, %N1"
[(set_attr "type" "arith_media,arith_media,*,load_media,store_media")
(set_attr "length" "4,4,16,4,4")])
(define_split
[(set (match_operand:V8QI 0 "arith_reg_dest" "")
(subreg:V8QI (const_int 0) 0))]
"TARGET_SHMEDIA"
[(set (match_dup 0)
(const_vector:V8QI [(const_int 0) (const_int 0) (const_int 0)
(const_int 0) (const_int 0) (const_int 0)
(const_int 0) (const_int 0)]))])
(define_split
[(set (match_operand 0 "arith_reg_dest" "")
(match_operand 1 "sh_rep_vec" ""))]
"TARGET_SHMEDIA && reload_completed
&& GET_MODE (operands[0]) == GET_MODE (operands[1])
&& sh_vector_mode_supported_p (GET_MODE (operands[0]))
&& GET_MODE_SIZE (GET_MODE (operands[0])) == 8
&& (XVECEXP (operands[1], 0, 0) != const0_rtx
|| XVECEXP (operands[1], 0, 1) != const0_rtx)
&& (XVECEXP (operands[1], 0, 0) != constm1_rtx
|| XVECEXP (operands[1], 0, 1) != constm1_rtx)"
[(set (match_dup 0) (match_dup 1))
(match_dup 2)]
{
int unit_size = GET_MODE_UNIT_SIZE (GET_MODE (operands[1]));
rtx elt1 = XVECEXP (operands[1], 0, 1);
if (unit_size > 2)
operands[2] = gen_mshflo_l (operands[0], operands[0], operands[0]);
else
{
if (unit_size < 2)
operands[0] = gen_rtx_REG (V4HImode, true_regnum (operands[0]));
operands[2] = gen_mperm_w0 (operands[0], operands[0]);
}
operands[0] = gen_rtx_REG (DImode, true_regnum (operands[0]));
operands[1] = XVECEXP (operands[1], 0, 0);
if (unit_size < 2)
{
if (CONST_INT_P (operands[1]) && CONST_INT_P (elt1))
operands[1]
= GEN_INT (TARGET_LITTLE_ENDIAN
? (INTVAL (operands[1]) & 0xff) + (INTVAL (elt1) << 8)
: (INTVAL (operands[1]) << 8) + (INTVAL (elt1) & 0xff));
else
{
operands[0] = gen_rtx_REG (V2QImode, true_regnum (operands[0]));
operands[1]
= gen_rtx_PARALLEL (VOIDmode, gen_rtvec (2, operands[1], elt1));
}
}
})
(define_split
[(set (match_operand 0 "arith_reg_dest" "")
(match_operand 1 "sh_const_vec" ""))]
"TARGET_SHMEDIA && reload_completed
&& GET_MODE (operands[0]) == GET_MODE (operands[1])
&& sh_vector_mode_supported_p (GET_MODE (operands[0]))"
[(set (match_dup 0) (match_dup 1))]
{
rtx v = operands[1];
enum machine_mode new_mode
= mode_for_size (GET_MODE_BITSIZE (GET_MODE (v)), MODE_INT, 0);
operands[0] = gen_rtx_REG (new_mode, true_regnum (operands[0]));
operands[1]
= simplify_subreg (new_mode, operands[1], GET_MODE (operands[1]), 0);
})
(define_expand "movv2hi"
[(set (match_operand:V2HI 0 "general_movdst_operand" "")
(match_operand:V2HI 1 "general_movsrc_operand" ""))]
"TARGET_SHMEDIA"
{
prepare_move_operands (operands, V2HImode);
})
(define_insn "movv2hi_i"
[(set (match_operand:V2HI 0 "general_movdst_operand" "=r,r,r,rl,m")
(match_operand:V2HI 1 "general_movsrc_operand" "r,I16CssZ,nW,m,rlZ"))]
"TARGET_SHMEDIA
&& (register_operand (operands[0], V2HImode)
|| sh_register_operand (operands[1], V2HImode))"
"@
add.l %1, r63, %0
movi %1, %0
#
ld%M1.l %m1, %0
st%M0.l %m0, %N1"
[(set_attr "type" "arith_media,arith_media,*,load_media,store_media")
(set_attr "length" "4,4,16,4,4")
(set (attr "highpart")
(cond [(match_test "sh_contains_memref_p (insn)")
(const_string "user")]
(const_string "ignore")))])
(define_expand "movv4hi"
[(set (match_operand:V4HI 0 "general_movdst_operand" "")
(match_operand:V4HI 1 "general_movsrc_operand" ""))]
"TARGET_SHMEDIA"
{
prepare_move_operands (operands, V4HImode);
})
(define_insn "movv4hi_i"
[(set (match_operand:V4HI 0 "general_movdst_operand" "=r,r,r,rl,m")
(match_operand:V4HI 1 "general_movsrc_operand" "r,I16CssZ,nW,m,rlZ"))]
"TARGET_SHMEDIA
&& (register_operand (operands[0], V4HImode)
|| sh_register_operand (operands[1], V4HImode))"
"@
add %1, r63, %0
movi %1, %0
#
ld%M1.q %m1, %0
st%M0.q %m0, %N1"
[(set_attr "type" "arith_media,arith_media,*,load_media,store_media")
(set_attr "length" "4,4,16,4,4")
(set_attr "highpart" "depend")])
(define_expand "movv2si"
[(set (match_operand:V2SI 0 "general_movdst_operand" "")
(match_operand:V2SI 1 "general_movsrc_operand" ""))]
"TARGET_SHMEDIA"
{
prepare_move_operands (operands, V2SImode);
})
(define_insn "movv2si_i"
[(set (match_operand:V2SI 0 "general_movdst_operand" "=r,r,r,rl,m")
(match_operand:V2SI 1 "general_movsrc_operand" "r,I16CssZ,nW,m,rlZ"))]
"TARGET_SHMEDIA
&& (register_operand (operands[0], V2SImode)
|| sh_register_operand (operands[1], V2SImode))"
"@
add %1, r63, %0
#
#
ld%M1.q %m1, %0
st%M0.q %m0, %N1"
[(set_attr "type" "arith_media,arith_media,*,load_media,store_media")
(set_attr "length" "4,4,16,4,4")
(set_attr "highpart" "depend")])
;; -------------------------------------------------------------------------
;; Multimedia Intrinsics
;; -------------------------------------------------------------------------
(define_insn "absv2si2"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(abs:V2SI (match_operand:V2SI 1 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mabs.l %1, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "absv4hi2"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(abs:V4HI (match_operand:V4HI 1 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mabs.w %1, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "addv2si3"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(plus:V2SI (match_operand:V2SI 1 "arith_reg_operand" "%r")
(match_operand:V2SI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"madd.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn "addv4hi3"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(plus:V4HI (match_operand:V4HI 1 "arith_reg_operand" "%r")
(match_operand:V4HI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"madd.w %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn_and_split "addv2hi3"
[(set (match_operand:V2HI 0 "arith_reg_dest" "=r")
(plus:V2HI (match_operand:V2HI 1 "extend_reg_operand" "%r")
(match_operand:V2HI 2 "extend_reg_operand" "r")))]
"TARGET_SHMEDIA"
"#"
"TARGET_SHMEDIA"
[(const_int 0)]
{
rtx src0 = simplify_gen_subreg (V4HImode, operands[1], V2HImode, 0);
rtx src1 = simplify_gen_subreg (V4HImode, operands[2], V2HImode, 0);
rtx v4hi_dst = simplify_gen_subreg (V4HImode, operands[0], V2HImode, 0);
rtx di_dst = simplify_gen_subreg (DImode, operands[0], V2HImode, 0);
rtx si_dst = simplify_gen_subreg (SImode, operands[0], V2HImode, 0);
emit_insn (gen_addv4hi3 (v4hi_dst, src0, src1));
emit_insn (gen_truncdisi2 (si_dst, di_dst));
DONE;
}
[(set_attr "highpart" "must_split")])
(define_insn "ssaddv2si3"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(ss_plus:V2SI (match_operand:V2SI 1 "arith_reg_operand" "%r")
(match_operand:V2SI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"madds.l %1, %2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "usaddv8qi3"
[(set (match_operand:V8QI 0 "arith_reg_dest" "=r")
(us_plus:V8QI (match_operand:V8QI 1 "arith_reg_operand" "%r")
(match_operand:V8QI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"madds.ub %1, %2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "ssaddv4hi3"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(ss_plus:V4HI (match_operand:V4HI 1 "arith_reg_operand" "%r")
(match_operand:V4HI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"madds.w %1, %2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "negcmpeqv8qi"
[(set (match_operand:V8QI 0 "arith_reg_dest" "=r")
(neg:V8QI (eq:V8QI
(match_operand:V8QI 1 "arith_reg_or_0_operand" "%rZ")
(match_operand:V8QI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mcmpeq.b %N1, %N2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "negcmpeqv2si"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(neg:V2SI (eq:V2SI
(match_operand:V2SI 1 "arith_reg_or_0_operand" "%rZ")
(match_operand:V2SI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mcmpeq.l %N1, %N2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "negcmpeqv4hi"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(neg:V4HI (eq:V4HI
(match_operand:V4HI 1 "arith_reg_or_0_operand" "%rZ")
(match_operand:V4HI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mcmpeq.w %N1, %N2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "negcmpgtuv8qi"
[(set (match_operand:V8QI 0 "arith_reg_dest" "=r")
(neg:V8QI (gtu:V8QI
(match_operand:V8QI 1 "arith_reg_or_0_operand" "%rZ")
(match_operand:V8QI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mcmpgt.ub %N1, %N2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "negcmpgtv2si"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(neg:V2SI (gt:V2SI
(match_operand:V2SI 1 "arith_reg_or_0_operand" "%rZ")
(match_operand:V2SI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mcmpgt.l %N1, %N2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "negcmpgtv4hi"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(neg:V4HI (gt:V4HI
(match_operand:V4HI 1 "arith_reg_or_0_operand" "%rZ")
(match_operand:V4HI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mcmpgt.w %N1, %N2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "mcmv"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ior:DI (and:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:DI 2 "arith_reg_operand" "r"))
(and:DI (match_operand:DI 3 "arith_reg_operand" "0")
(not:DI (match_dup 2)))))]
"TARGET_SHMEDIA"
"mcmv %N1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn "mcnvs_lw"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(vec_concat:V4HI
(ss_truncate:V2HI (match_operand:V2SI 1 "arith_reg_or_0_operand" "rZ"))
(ss_truncate:V2HI
(match_operand:V2SI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mcnvs.lw %N1, %N2, %0"
[(set_attr "type" "mcmp_media")])
(define_insn "mcnvs_wb"
[(set (match_operand:V8QI 0 "arith_reg_dest" "=r")
(vec_concat:V8QI
(ss_truncate:V4QI (match_operand:V4HI 1 "arith_reg_or_0_operand" "rZ"))
(ss_truncate:V4QI
(match_operand:V4HI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mcnvs.wb %N1, %N2, %0"
[(set_attr "type" "mcmp_media")])
(define_insn "mcnvs_wub"
[(set (match_operand:V8QI 0 "arith_reg_dest" "=r")
(vec_concat:V8QI
(us_truncate:V4QI (match_operand:V4HI 1 "arith_reg_or_0_operand" "rZ"))
(us_truncate:V4QI
(match_operand:V4HI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mcnvs.wub %N1, %N2, %0"
[(set_attr "type" "mcmp_media")])
(define_insn "mextr_rl"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ior:DI (lshiftrt:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:HI 3 "mextr_bit_offset" "i"))
(ashift:DI (match_operand:DI 2 "arith_reg_or_0_operand" "rZ")
(match_operand:HI 4 "mextr_bit_offset" "i"))))]
"TARGET_SHMEDIA && INTVAL (operands[3]) + INTVAL (operands[4]) == 64"
{
static char templ[21];
sprintf (templ, "mextr%d %%N1, %%N2, %%0",
(int) INTVAL (operands[3]) >> 3);
return templ;
}
[(set_attr "type" "arith_media")])
(define_insn "*mextr_lr"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ior:DI (ashift:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:HI 3 "mextr_bit_offset" "i"))
(lshiftrt:DI (match_operand:DI 2 "arith_reg_or_0_operand" "rZ")
(match_operand:HI 4 "mextr_bit_offset" "i"))))]
"TARGET_SHMEDIA && INTVAL (operands[3]) + INTVAL (operands[4]) == 64"
{
static char templ[21];
sprintf (templ, "mextr%d %%N2, %%N1, %%0",
(int) INTVAL (operands[4]) >> 3);
return templ;
}
[(set_attr "type" "arith_media")])
; mextrN can be modelled with vec_select / vec_concat, but the selection
; vector then varies depending on endianness.
(define_expand "mextr1"
[(match_operand:DI 0 "arith_reg_dest" "")
(match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:DI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mextr_rl (operands[0], operands[1], operands[2],
GEN_INT (1 * 8), GEN_INT (7 * 8)));
DONE;
})
(define_expand "mextr2"
[(match_operand:DI 0 "arith_reg_dest" "")
(match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:DI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mextr_rl (operands[0], operands[1], operands[2],
GEN_INT (2 * 8), GEN_INT (6 * 8)));
DONE;
})
(define_expand "mextr3"
[(match_operand:DI 0 "arith_reg_dest" "")
(match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:DI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mextr_rl (operands[0], operands[1], operands[2],
GEN_INT (3 * 8), GEN_INT (5 * 8)));
DONE;
})
(define_expand "mextr4"
[(match_operand:DI 0 "arith_reg_dest" "")
(match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:DI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mextr_rl (operands[0], operands[1], operands[2],
GEN_INT (4 * 8), GEN_INT (4 * 8)));
DONE;
})
(define_expand "mextr5"
[(match_operand:DI 0 "arith_reg_dest" "")
(match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:DI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mextr_rl (operands[0], operands[1], operands[2],
GEN_INT (5 * 8), GEN_INT (3 * 8)));
DONE;
})
(define_expand "mextr6"
[(match_operand:DI 0 "arith_reg_dest" "")
(match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:DI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mextr_rl (operands[0], operands[1], operands[2],
GEN_INT (6 * 8), GEN_INT (2 * 8)));
DONE;
})
(define_expand "mextr7"
[(match_operand:DI 0 "arith_reg_dest" "")
(match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:DI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mextr_rl (operands[0], operands[1], operands[2],
GEN_INT (7 * 8), GEN_INT (1 * 8)));
DONE;
})
(define_expand "mmacfx_wl"
[(match_operand:V2SI 0 "arith_reg_dest" "")
(match_operand:V2HI 1 "extend_reg_operand" "")
(match_operand:V2HI 2 "extend_reg_operand" "")
(match_operand:V2SI 3 "arith_reg_operand" "")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mmacfx_wl_i (operands[0], operands[3],
operands[1], operands[2]));
DONE;
})
;; This could be highpart ignore if it only had inputs 2 or 3, but input 1
;; is depend
(define_insn "mmacfx_wl_i"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(ss_plus:V2SI
(match_operand:V2SI 1 "arith_reg_operand" "0")
(ss_truncate:V2SI
(ashift:V2DI
(sign_extend:V2DI
(mult:V2SI
(sign_extend:V2SI (match_operand:V2HI 2 "extend_reg_operand" "r"))
(sign_extend:V2SI (match_operand:V2HI 3 "extend_reg_operand" "r"))))
(const_int 1)))))]
"TARGET_SHMEDIA"
"mmacfx.wl %2, %3, %0"
[(set_attr "type" "mac_media")
(set_attr "highpart" "depend")])
(define_expand "mmacnfx_wl"
[(match_operand:V2SI 0 "arith_reg_dest" "")
(match_operand:V2HI 1 "extend_reg_operand" "")
(match_operand:V2HI 2 "extend_reg_operand" "")
(match_operand:V2SI 3 "arith_reg_operand" "")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mmacnfx_wl_i (operands[0], operands[3],
operands[1], operands[2]));
DONE;
})
(define_insn "mmacnfx_wl_i"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(ss_minus:V2SI
(match_operand:V2SI 1 "arith_reg_operand" "0")
(ss_truncate:V2SI
(ashift:V2DI
(sign_extend:V2DI
(mult:V2SI
(sign_extend:V2SI (match_operand:V2HI 2 "extend_reg_operand" "r"))
(sign_extend:V2SI (match_operand:V2HI 3 "extend_reg_operand" "r"))))
(const_int 1)))))]
"TARGET_SHMEDIA"
"mmacnfx.wl %2, %3, %0"
[(set_attr "type" "mac_media")
(set_attr "highpart" "depend")])
(define_insn "mulv2si3"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(mult:V2SI (match_operand:V2SI 1 "arith_reg_operand" "r")
(match_operand:V2SI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mmul.l %1, %2, %0"
[(set_attr "type" "d2mpy_media")
(set_attr "highpart" "depend")])
(define_insn "mulv4hi3"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(mult:V4HI (match_operand:V4HI 1 "arith_reg_operand" "r")
(match_operand:V4HI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mmul.w %1, %2, %0"
[(set_attr "type" "dmpy_media")
(set_attr "highpart" "depend")])
(define_insn "mmulfx_l"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(ss_truncate:V2SI
(ashiftrt:V2DI
(mult:V2DI
(sign_extend:V2DI (match_operand:V2SI 1 "arith_reg_operand" "r"))
(sign_extend:V2DI (match_operand:V2SI 2 "arith_reg_operand" "r")))
(const_int 31))))]
"TARGET_SHMEDIA"
"mmulfx.l %1, %2, %0"
[(set_attr "type" "d2mpy_media")
(set_attr "highpart" "depend")])
(define_insn "mmulfx_w"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(ss_truncate:V4HI
(ashiftrt:V4SI
(mult:V4SI
(sign_extend:V4SI (match_operand:V4HI 1 "arith_reg_operand" "r"))
(sign_extend:V4SI (match_operand:V4HI 2 "arith_reg_operand" "r")))
(const_int 15))))]
"TARGET_SHMEDIA"
"mmulfx.w %1, %2, %0"
[(set_attr "type" "dmpy_media")
(set_attr "highpart" "depend")])
(define_insn "mmulfxrp_w"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(ss_truncate:V4HI
(ashiftrt:V4SI
(plus:V4SI
(mult:V4SI
(sign_extend:V4SI (match_operand:V4HI 1 "arith_reg_operand" "r"))
(sign_extend:V4SI (match_operand:V4HI 2 "arith_reg_operand" "r")))
(const_int 16384))
(const_int 15))))]
"TARGET_SHMEDIA"
"mmulfxrp.w %1, %2, %0"
[(set_attr "type" "dmpy_media")
(set_attr "highpart" "depend")])
(define_expand "mmulhi_wl"
[(match_operand:V2SI 0 "arith_reg_dest" "")
(match_operand:V4HI 1 "arith_reg_operand" "")
(match_operand:V4HI 2 "arith_reg_operand" "")]
"TARGET_SHMEDIA"
{
emit_insn ((TARGET_LITTLE_ENDIAN ? gen_mmul23_wl : gen_mmul01_wl)
(operands[0], operands[1], operands[2]));
DONE;
})
(define_expand "mmullo_wl"
[(match_operand:V2SI 0 "arith_reg_dest" "")
(match_operand:V4HI 1 "arith_reg_operand" "")
(match_operand:V4HI 2 "arith_reg_operand" "")]
"TARGET_SHMEDIA"
{
emit_insn ((TARGET_LITTLE_ENDIAN ? gen_mmul01_wl : gen_mmul23_wl)
(operands[0], operands[1], operands[2]));
DONE;
})
(define_insn "mmul23_wl"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(vec_select:V2SI
(mult:V4SI
(sign_extend:V4SI (match_operand:V4HI 1 "arith_reg_operand" "r"))
(sign_extend:V4SI (match_operand:V4HI 2 "arith_reg_operand" "r")))
(parallel [(const_int 2) (const_int 3)])))]
"TARGET_SHMEDIA"
{
return (TARGET_LITTLE_ENDIAN
? "mmulhi.wl %1, %2, %0"
: "mmullo.wl %1, %2, %0");
}
[(set_attr "type" "dmpy_media")
(set (attr "highpart")
(cond [(eq_attr "endian" "big") (const_string "ignore")]
(const_string "user")))])
(define_insn "mmul01_wl"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(vec_select:V2SI
(mult:V4SI
(sign_extend:V4SI (match_operand:V4HI 1 "arith_reg_operand" "r"))
(sign_extend:V4SI (match_operand:V4HI 2 "arith_reg_operand" "r")))
(parallel [(const_int 0) (const_int 1)])))]
"TARGET_SHMEDIA"
{
return (TARGET_LITTLE_ENDIAN
? "mmullo.wl %1, %2, %0"
: "mmulhi.wl %1, %2, %0");
}
[(set_attr "type" "dmpy_media")
(set (attr "highpart")
(cond [(eq_attr "endian" "little") (const_string "ignore")]
(const_string "user")))])
(define_expand "mmulsum_wq"
[(match_operand:DI 0 "arith_reg_dest" "")
(match_operand:V4HI 1 "arith_reg_operand" "")
(match_operand:V4HI 2 "arith_reg_operand" "")
(match_operand:DI 3 "arith_reg_operand" "")]
"TARGET_SHMEDIA"
{
emit_insn (gen_mmulsum_wq_i (operands[0], operands[3],
operands[1], operands[2]));
DONE;
})
(define_insn "mmulsum_wq_i"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(plus:DI (match_operand:DI 1 "arith_reg_operand" "0")
(plus:DI
(plus:DI
(vec_select:DI
(mult:V4DI
(sign_extend:V4DI (match_operand:V4HI 2 "arith_reg_operand" "r"))
(sign_extend:V4DI (match_operand:V4HI 3 "arith_reg_operand" "r")))
(parallel [(const_int 0)]))
(vec_select:DI (mult:V4DI (sign_extend:V4DI (match_dup 2))
(sign_extend:V4DI (match_dup 3)))
(parallel [(const_int 1)])))
(plus:DI
(vec_select:DI (mult:V4DI (sign_extend:V4DI (match_dup 2))
(sign_extend:V4DI (match_dup 3)))
(parallel [(const_int 2)]))
(vec_select:DI (mult:V4DI (sign_extend:V4DI (match_dup 2))
(sign_extend:V4DI (match_dup 3)))
(parallel [(const_int 3)]))))))]
"TARGET_SHMEDIA"
"mmulsum.wq %2, %3, %0"
[(set_attr "type" "mac_media")])
(define_expand "mperm_w"
[(match_operand:V4HI 0 "arith_reg_dest" "=r")
(match_operand:V4HI 1 "arith_reg_operand" "r")
(match_operand:QI 2 "extend_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn ((TARGET_LITTLE_ENDIAN ? gen_mperm_w_little : gen_mperm_w_big)
(operands[0], operands[1], operands[2]));
DONE;
})
; This use of vec_select isn't exactly correct according to rtl.texi
; (because not constant), but it seems a straightforward extension.
(define_insn "mperm_w_little"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(vec_select:V4HI
(match_operand:V4HI 1 "arith_reg_operand" "r")
(parallel
[(zero_extract:QI (match_operand:QI 2 "extend_reg_or_0_operand" "rZ")
(const_int 2) (const_int 0))
(zero_extract:QI (match_dup 2) (const_int 2) (const_int 2))
(zero_extract:QI (match_dup 2) (const_int 2) (const_int 4))
(zero_extract:QI (match_dup 2) (const_int 2) (const_int 6))])))]
"TARGET_SHMEDIA && TARGET_LITTLE_ENDIAN"
"mperm.w %1, %N2, %0"
[(set_attr "type" "arith_media")])
(define_insn "mperm_w_big"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(vec_select:V4HI
(match_operand:V4HI 1 "arith_reg_operand" "r")
(parallel
[(zero_extract:QI (not:QI (match_operand:QI 2
"extend_reg_or_0_operand" "rZ"))
(const_int 2) (const_int 0))
(zero_extract:QI (not:QI (match_dup 2)) (const_int 2) (const_int 2))
(zero_extract:QI (not:QI (match_dup 2)) (const_int 2) (const_int 4))
(zero_extract:QI (not:QI (match_dup 2))
(const_int 2) (const_int 6))])))]
"TARGET_SHMEDIA && TARGET_BIG_ENDIAN"
"mperm.w %1, %N2, %0"
[(set_attr "type" "arith_media")])
(define_insn "mperm_w0"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(vec_duplicate:V4HI (truncate:HI (match_operand 1
"trunc_hi_operand" "r"))))]
"TARGET_SHMEDIA"
"mperm.w %1, r63, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_expand "msad_ubq"
[(match_operand:DI 0 "arith_reg_dest" "")
(match_operand:V8QI 1 "arith_reg_or_0_operand" "")
(match_operand:V8QI 2 "arith_reg_or_0_operand" "")
(match_operand:DI 3 "arith_reg_operand" "")]
"TARGET_SHMEDIA"
{
emit_insn (gen_msad_ubq_i (operands[0], operands[3],
operands[1], operands[2]));
DONE;
})
(define_insn "msad_ubq_i"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(plus:DI
(plus:DI
(plus:DI
(plus:DI
(match_operand:DI 1 "arith_reg_operand" "0")
(abs:DI (vec_select:DI
(minus:V8DI
(zero_extend:V8DI
(match_operand:V8QI 2 "arith_reg_or_0_operand" "rZ"))
(zero_extend:V8DI
(match_operand:V8QI 3 "arith_reg_or_0_operand" "rZ")))
(parallel [(const_int 0)]))))
(abs:DI (vec_select:DI (minus:V8DI (zero_extend:V8DI (match_dup 2))
(zero_extend:V8DI (match_dup 3)))
(parallel [(const_int 1)]))))
(plus:DI
(abs:DI (vec_select:DI (minus:V8DI (zero_extend:V8DI (match_dup 2))
(zero_extend:V8DI (match_dup 3)))
(parallel [(const_int 2)])))
(abs:DI (vec_select:DI (minus:V8DI (zero_extend:V8DI (match_dup 2))
(zero_extend:V8DI (match_dup 3)))
(parallel [(const_int 3)])))))
(plus:DI
(plus:DI
(abs:DI (vec_select:DI (minus:V8DI (zero_extend:V8DI (match_dup 2))
(zero_extend:V8DI (match_dup 3)))
(parallel [(const_int 4)])))
(abs:DI (vec_select:DI (minus:V8DI (zero_extend:V8DI (match_dup 2))
(zero_extend:V8DI (match_dup 3)))
(parallel [(const_int 5)]))))
(plus:DI
(abs:DI (vec_select:DI (minus:V8DI (zero_extend:V8DI (match_dup 2))
(zero_extend:V8DI (match_dup 3)))
(parallel [(const_int 6)])))
(abs:DI (vec_select:DI (minus:V8DI (zero_extend:V8DI (match_dup 2))
(zero_extend:V8DI (match_dup 3)))
(parallel [(const_int 7)])))))))]
"TARGET_SHMEDIA"
"msad.ubq %N2, %N3, %0"
[(set_attr "type" "mac_media")])
(define_insn "mshalds_l"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(ss_truncate:V2SI
(ashift:V2DI
(sign_extend:V2DI (match_operand:V2SI 1 "arith_reg_operand" "r"))
(and:DI (match_operand:DI 2 "arith_reg_operand" "r")
(const_int 31)))))]
"TARGET_SHMEDIA"
"mshalds.l %1, %2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "mshalds_w"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(ss_truncate:V4HI
(ashift:V4SI
(sign_extend:V4SI (match_operand:V4HI 1 "arith_reg_operand" "r"))
(and:DI (match_operand:DI 2 "arith_reg_operand" "r")
(const_int 15)))))]
"TARGET_SHMEDIA"
"mshalds.w %1, %2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "ashrv2si3"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(ashiftrt:V2SI (match_operand:V2SI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mshard.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn "ashrv4hi3"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(ashiftrt:V4HI (match_operand:V4HI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mshard.w %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn "mshards_q"
[(set (match_operand:HI 0 "arith_reg_dest" "=r")
(ss_truncate:HI
(ashiftrt:DI (match_operand:DI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "arith_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mshards.q %1, %N2, %0"
[(set_attr "type" "mcmp_media")])
(define_expand "mshfhi_b"
[(match_operand:V8QI 0 "arith_reg_dest" "")
(match_operand:V8QI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V8QI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn ((TARGET_LITTLE_ENDIAN ? gen_mshf4_b : gen_mshf0_b)
(operands[0], operands[1], operands[2]));
DONE;
})
(define_expand "mshflo_b"
[(match_operand:V8QI 0 "arith_reg_dest" "")
(match_operand:V8QI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V8QI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn ((TARGET_LITTLE_ENDIAN ? gen_mshf0_b : gen_mshf4_b)
(operands[0], operands[1], operands[2]));
DONE;
})
(define_insn "mshf4_b"
[(set
(match_operand:V8QI 0 "arith_reg_dest" "=r")
(vec_select:V8QI
(vec_concat:V16QI (match_operand:V8QI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V8QI 2 "arith_reg_or_0_operand" "rZ"))
(parallel [(const_int 4) (const_int 12) (const_int 5) (const_int 13)
(const_int 6) (const_int 14) (const_int 7) (const_int 15)])))]
"TARGET_SHMEDIA"
{
return (TARGET_LITTLE_ENDIAN
? "mshfhi.b %N1, %N2, %0"
: "mshflo.b %N1, %N2, %0");
}
[(set_attr "type" "arith_media")
(set (attr "highpart")
(cond [(eq_attr "endian" "big") (const_string "ignore")]
(const_string "user")))])
(define_insn "mshf0_b"
[(set
(match_operand:V8QI 0 "arith_reg_dest" "=r")
(vec_select:V8QI
(vec_concat:V16QI (match_operand:V8QI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V8QI 2 "arith_reg_or_0_operand" "rZ"))
(parallel [(const_int 0) (const_int 8) (const_int 1) (const_int 9)
(const_int 2) (const_int 10) (const_int 3) (const_int 11)])))]
"TARGET_SHMEDIA"
{
return (TARGET_LITTLE_ENDIAN
? "mshflo.b %N1, %N2, %0"
: "mshfhi.b %N1, %N2, %0");
}
[(set_attr "type" "arith_media")
(set (attr "highpart")
(cond [(eq_attr "endian" "little") (const_string "ignore")]
(const_string "user")))])
(define_expand "mshfhi_l"
[(match_operand:V2SI 0 "arith_reg_dest" "")
(match_operand:V2SI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V2SI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn ((TARGET_LITTLE_ENDIAN ? gen_mshf4_l : gen_mshf0_l)
(operands[0], operands[1], operands[2]));
DONE;
})
(define_expand "mshflo_l"
[(match_operand:V2SI 0 "arith_reg_dest" "")
(match_operand:V2SI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V2SI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn ((TARGET_LITTLE_ENDIAN ? gen_mshf0_l : gen_mshf4_l)
(operands[0], operands[1], operands[2]));
DONE;
})
(define_insn "mshf4_l"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(vec_select:V2SI
(vec_concat:V4SI (match_operand:V2SI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V2SI 2 "arith_reg_or_0_operand" "rZ"))
(parallel [(const_int 1) (const_int 3)])))]
"TARGET_SHMEDIA"
{
return (TARGET_LITTLE_ENDIAN
? "mshfhi.l %N1, %N2, %0"
: "mshflo.l %N1, %N2, %0");
}
[(set_attr "type" "arith_media")
(set (attr "highpart")
(cond [(eq_attr "endian" "big") (const_string "ignore")]
(const_string "user")))])
(define_insn "mshf0_l"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(vec_select:V2SI
(vec_concat:V4SI (match_operand:V2SI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V2SI 2 "arith_reg_or_0_operand" "rZ"))
(parallel [(const_int 0) (const_int 2)])))]
"TARGET_SHMEDIA"
{
return (TARGET_LITTLE_ENDIAN
? "mshflo.l %N1, %N2, %0"
: "mshfhi.l %N1, %N2, %0");
}
[(set_attr "type" "arith_media")
(set (attr "highpart")
(cond [(eq_attr "endian" "little") (const_string "ignore")]
(const_string "user")))])
(define_expand "mshfhi_w"
[(match_operand:V4HI 0 "arith_reg_dest" "")
(match_operand:V4HI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V4HI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn ((TARGET_LITTLE_ENDIAN ? gen_mshf4_w : gen_mshf0_w)
(operands[0], operands[1], operands[2]));
DONE;
})
(define_expand "mshflo_w"
[(match_operand:V4HI 0 "arith_reg_dest" "")
(match_operand:V4HI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V4HI 2 "arith_reg_or_0_operand" "rZ")]
"TARGET_SHMEDIA"
{
emit_insn ((TARGET_LITTLE_ENDIAN ? gen_mshf0_w : gen_mshf4_w)
(operands[0], operands[1], operands[2]));
DONE;
})
(define_insn "mshf4_w"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(vec_select:V4HI
(vec_concat:V8HI (match_operand:V4HI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V4HI 2 "arith_reg_or_0_operand" "rZ"))
(parallel [(const_int 2) (const_int 6) (const_int 3) (const_int 7)])))]
"TARGET_SHMEDIA"
{
return (TARGET_LITTLE_ENDIAN
? "mshfhi.w %N1, %N2, %0"
: "mshflo.w %N1, %N2, %0");
}
[(set_attr "type" "arith_media")
(set (attr "highpart")
(cond [(eq_attr "endian" "big") (const_string "ignore")]
(const_string "user")))])
(define_insn "mshf0_w"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(vec_select:V4HI
(vec_concat:V8HI (match_operand:V4HI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V4HI 2 "arith_reg_or_0_operand" "rZ"))
(parallel [(const_int 0) (const_int 4) (const_int 1) (const_int 5)])))]
"TARGET_SHMEDIA"
{
return (TARGET_LITTLE_ENDIAN
? "mshflo.w %N1, %N2, %0"
: "mshfhi.w %N1, %N2, %0");
}
[(set_attr "type" "arith_media")
(set (attr "highpart")
(cond [(eq_attr "endian" "little") (const_string "ignore")]
(const_string "user")))])
(define_insn "mshflo_w_x"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(vec_select:V4HI
(vec_concat:V4HI (match_operand:V2HI 1 "extend_reg_or_0_operand" "rZ")
(match_operand:V2HI 2 "extend_reg_or_0_operand" "rZ"))
(parallel [(const_int 2) (const_int 0) (const_int 3) (const_int 1)])))]
"TARGET_SHMEDIA"
"mshflo.w %N1, %N2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
;; These are useful to expand ANDs and as combiner patterns.
(define_insn_and_split "mshfhi_l_di"
[(set (match_operand:DI 0 "arith_reg_dest" "=r,f")
(ior:DI (lshiftrt:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rZ,f")
(const_int 32))
(and:DI (match_operand:DI 2 "arith_reg_or_0_operand" "rZ,?f")
(const_int -4294967296))))]
"TARGET_SHMEDIA"
"@
mshfhi.l %N1, %N2, %0
#"
"TARGET_SHMEDIA && reload_completed
&& ! GENERAL_REGISTER_P (true_regnum (operands[0]))"
[(set (match_dup 3) (match_dup 4))
(set (match_dup 5) (match_dup 6))]
{
operands[3] = gen_lowpart (SImode, operands[0]);
operands[4] = gen_highpart (SImode, operands[1]);
operands[5] = gen_highpart (SImode, operands[0]);
operands[6] = gen_highpart (SImode, operands[2]);
}
[(set_attr "type" "arith_media")])
(define_insn "*mshfhi_l_di_rev"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ior:DI (and:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(const_int -4294967296))
(lshiftrt:DI (match_operand:DI 2 "arith_reg_or_0_operand" "rZ")
(const_int 32))))]
"TARGET_SHMEDIA"
"mshfhi.l %N2, %N1, %0"
[(set_attr "type" "arith_media")])
(define_split
[(set (match_operand:DI 0 "arith_reg_dest" "")
(ior:DI (zero_extend:DI (match_operand:SI 1
"extend_reg_or_0_operand" ""))
(and:DI (match_operand:DI 2 "arith_reg_or_0_operand" "")
(const_int -4294967296))))
(clobber (match_operand:DI 3 "arith_reg_dest" ""))]
"TARGET_SHMEDIA"
[(const_int 0)]
{
emit_insn (gen_ashldi3_media (operands[3],
simplify_gen_subreg (DImode, operands[1],
SImode, 0),
GEN_INT (32)));
emit_insn (gen_mshfhi_l_di (operands[0], operands[3], operands[2]));
DONE;
})
(define_insn "mshflo_l_di"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ior:DI (and:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(const_int 4294967295))
(ashift:DI (match_operand:DI 2 "arith_reg_or_0_operand" "rZ")
(const_int 32))))]
"TARGET_SHMEDIA"
"mshflo.l %N1, %N2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn "*mshflo_l_di_rev"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ior:DI (ashift:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(const_int 32))
(and:DI (match_operand:DI 2 "arith_reg_or_0_operand" "rZ")
(const_int 4294967295))))]
"TARGET_SHMEDIA"
"mshflo.l %N2, %N1, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
;; Combiner pattern for trampoline initialization.
(define_insn_and_split "*double_shori"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ior:DI (ashift:DI (match_operand:DI 1 "arith_reg_operand" "0")
(const_int 32))
(match_operand:DI 2 "const_int_operand" "n")))]
"TARGET_SHMEDIA
&& ! (INTVAL (operands[2]) & ~(unsigned HOST_WIDE_INT) 0xffffffffUL)"
"#"
"rtx_equal_p (operands[0], operands[1])"
[(const_int 0)]
{
HOST_WIDE_INT v = INTVAL (operands[2]);
emit_insn (gen_shori_media (operands[0], operands[0], GEN_INT (v >> 16)));
emit_insn (gen_shori_media (operands[0], operands[0], GEN_INT (v & 65535)));
DONE;
}
[(set_attr "highpart" "ignore")])
(define_insn "*mshflo_l_di_x"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ior:DI (zero_extend:DI (match_operand:SI 1 "extend_reg_or_0_operand"
"rZ"))
(ashift:DI (match_operand:DI 2 "arith_reg_or_0_operand" "rZ")
(const_int 32))))]
"TARGET_SHMEDIA"
"mshflo.l %N1, %N2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn_and_split "concat_v2sf"
[(set (match_operand:V2SF 0 "register_operand" "=r,f,f?")
;; (vec_concat:V2SF (match_operand:SF 1 "register_operand" "rZ,0,f")
(vec_concat:V2SF (match_operand:SF 1 "register_operand" "rZ,f,f")
(match_operand:SF 2 "register_operand" "rZ,f,f")))]
"TARGET_SHMEDIA"
"@
mshflo.l %N1, %N2, %0
#
#"
"TARGET_SHMEDIA && reload_completed
&& ! GENERAL_REGISTER_P (true_regnum (operands[0]))"
[(set (match_dup 3) (match_dup 1))
(set (match_dup 4) (match_dup 2))]
{
operands[3] = simplify_gen_subreg (SFmode, operands[0], V2SFmode, 0);
operands[4] = simplify_gen_subreg (SFmode, operands[0], V2SFmode, 4);
}
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn "*mshflo_l_di_x_rev"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(ior:DI (ashift:DI (match_operand:DI 1 "arith_reg_or_0_operand" "rZ")
(const_int 32))
(zero_extend:DI
(match_operand:SI 2 "extend_reg_or_0_operand" "rZ"))))]
"TARGET_SHMEDIA"
"mshflo.l %N2, %N1, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "ignore")])
(define_insn "ashlv2si3"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(ashift:V2SI (match_operand:V2SI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "shift_count_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mshlld.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_split
[(set (match_operand 0 "any_register_operand" "")
(match_operator 3 "shift_operator"
[(match_operand 1 "any_register_operand" "")
(match_operand 2 "shift_count_reg_operand" "")]))]
"TARGET_SHMEDIA && ! register_operand (operands[2], VOIDmode)"
[(set (match_dup 0) (match_dup 3))]
{
rtx count = operands[2];
enum machine_mode outer_mode = GET_MODE (operands[2]), inner_mode;
while (GET_CODE (count) == ZERO_EXTEND || GET_CODE (count) == SIGN_EXTEND
|| (GET_CODE (count) == SUBREG && SUBREG_BYTE (count) == 0)
|| GET_CODE (count) == TRUNCATE)
count = XEXP (count, 0);
inner_mode = GET_MODE (count);
count = simplify_gen_subreg (outer_mode, count, inner_mode,
subreg_lowpart_offset (outer_mode, inner_mode));
operands[3] = gen_rtx_fmt_ee (GET_CODE (operands[3]), GET_MODE (operands[3]),
operands[1], count);
})
(define_insn "ashlv4hi3"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(ashift:V4HI (match_operand:V4HI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "shift_count_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mshlld.w %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn "lshrv2si3"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(lshiftrt:V2SI (match_operand:V2SI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "shift_count_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mshlrd.l %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn "lshrv4hi3"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(lshiftrt:V4HI (match_operand:V4HI 1 "arith_reg_operand" "r")
(match_operand:DI 2 "shift_count_reg_operand" "r")))]
"TARGET_SHMEDIA"
"mshlrd.w %1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn "subv2si3"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(minus:V2SI (match_operand:V2SI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V2SI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"msub.l %N1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn "subv4hi3"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(minus:V4HI (match_operand:V4HI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V4HI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"msub.w %N1, %2, %0"
[(set_attr "type" "arith_media")
(set_attr "highpart" "depend")])
(define_insn_and_split "subv2hi3"
[(set (match_operand:V2HI 0 "arith_reg_dest" "=r")
(minus:V2HI (match_operand:V2HI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V2HI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"#"
"TARGET_SHMEDIA"
[(const_int 0)]
{
rtx src0 = simplify_gen_subreg (V4HImode, operands[1], V2HImode, 0);
rtx src1 = simplify_gen_subreg (V4HImode, operands[2], V2HImode, 0);
rtx v4hi_dst = simplify_gen_subreg (V4HImode, operands[0], V2HImode, 0);
rtx di_dst = simplify_gen_subreg (DImode, operands[0], V2HImode, 0);
rtx si_dst = simplify_gen_subreg (SImode, operands[0], V2HImode, 0);
emit_insn (gen_subv4hi3 (v4hi_dst, src0, src1));
emit_insn (gen_truncdisi2 (si_dst, di_dst));
DONE;
}
[(set_attr "highpart" "must_split")])
(define_insn "sssubv2si3"
[(set (match_operand:V2SI 0 "arith_reg_dest" "=r")
(ss_minus:V2SI (match_operand:V2SI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V2SI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"msubs.l %N1, %2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "ussubv8qi3"
[(set (match_operand:V8QI 0 "arith_reg_dest" "=r")
(us_minus:V8QI (match_operand:V8QI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V8QI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"msubs.ub %N1, %2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
(define_insn "sssubv4hi3"
[(set (match_operand:V4HI 0 "arith_reg_dest" "=r")
(ss_minus:V4HI (match_operand:V4HI 1 "arith_reg_or_0_operand" "rZ")
(match_operand:V4HI 2 "arith_reg_operand" "r")))]
"TARGET_SHMEDIA"
"msubs.w %N1, %2, %0"
[(set_attr "type" "mcmp_media")
(set_attr "highpart" "depend")])
;; -------------------------------------------------------------------------
;; Floating Point Intrinsics
;; -------------------------------------------------------------------------
(define_insn "fcosa_s"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(unspec:SF [(match_operand:SI 1 "fp_arith_reg_operand" "f")]
UNSPEC_FCOSA))]
"TARGET_SHMEDIA"
"fcosa.s %1, %0"
[(set_attr "type" "atrans_media")])
(define_insn "fsina_s"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(unspec:SF [(match_operand:SI 1 "fp_arith_reg_operand" "f")]
UNSPEC_FSINA))]
"TARGET_SHMEDIA"
"fsina.s %1, %0"
[(set_attr "type" "atrans_media")])
(define_insn "fipr"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(plus:SF (plus:SF (vec_select:SF (mult:V4SF (match_operand:V4SF 1
"fp_arith_reg_operand" "f")
(match_operand:V4SF 2
"fp_arith_reg_operand" "f"))
(parallel [(const_int 0)]))
(vec_select:SF (mult:V4SF (match_dup 1) (match_dup 2))
(parallel [(const_int 1)])))
(plus:SF (vec_select:SF (mult:V4SF (match_dup 1) (match_dup 2))
(parallel [(const_int 2)]))
(vec_select:SF (mult:V4SF (match_dup 1) (match_dup 2))
(parallel [(const_int 3)])))))]
"TARGET_SHMEDIA"
"fipr.s %1, %2, %0"
[(set_attr "type" "fparith_media")])
(define_insn "fsrra_s"
[(set (match_operand:SF 0 "fp_arith_reg_operand" "=f")
(unspec:SF [(match_operand:SF 1 "fp_arith_reg_operand" "f")]
UNSPEC_FSRRA))]
"TARGET_SHMEDIA"
"fsrra.s %1, %0"
[(set_attr "type" "atrans_media")])
(define_insn "ftrv"
[(set (match_operand:V4SF 0 "fp_arith_reg_operand" "=f")
(plus:V4SF
(plus:V4SF
(mult:V4SF
(vec_select:V4SF (match_operand:V16SF 1 "fp_arith_reg_operand" "f")
(parallel [(const_int 0) (const_int 5)
(const_int 10) (const_int 15)]))
(match_operand:V4SF 2 "fp_arith_reg_operand" "f"))
(mult:V4SF
(vec_select:V4SF (match_dup 1)
(parallel [(const_int 4) (const_int 9)
(const_int 14) (const_int 3)]))
(vec_select:V4SF (match_dup 2)
(parallel [(const_int 1) (const_int 2)
(const_int 3) (const_int 0)]))))
(plus:V4SF
(mult:V4SF
(vec_select:V4SF (match_dup 1)
(parallel [(const_int 8) (const_int 13)
(const_int 2) (const_int 7)]))
(vec_select:V4SF (match_dup 2)
(parallel [(const_int 2) (const_int 3)
(const_int 0) (const_int 1)])))
(mult:V4SF
(vec_select:V4SF (match_dup 1)
(parallel [(const_int 12) (const_int 1)
(const_int 6) (const_int 11)]))
(vec_select:V4SF (match_dup 2)
(parallel [(const_int 3) (const_int 0)
(const_int 1) (const_int 2)]))))))]
"TARGET_SHMEDIA"
"ftrv.s %1, %2, %0"
[(set_attr "type" "fparith_media")])
(define_insn "ldhi_l"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(zero_extract:SI
(mem:SI (plus:SI (ior:SI (match_operand:QI 1 "ua_address_operand" "p")
(const_int 3))
(const_int -3)))
(plus:SI (and:SI (match_dup 1) (const_int 3)) (const_int 1))
(const_int 0)))]
"TARGET_SHMEDIA32"
"ldhi.l %U1, %0"
[(set_attr "type" "load_media")])
(define_insn "ldhi_q"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extract:DI
(mem:DI (plus:SI (ior:SI (match_operand:QI 1 "ua_address_operand" "p")
(const_int 7))
(const_int -7)))
(plus:SI (and:SI (match_dup 1) (const_int 7)) (const_int 1))
(const_int 0)))]
"TARGET_SHMEDIA32"
"ldhi.q %U1, %0"
[(set_attr "type" "load_media")])
(define_insn_and_split "*ldhi_q_comb0"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extract:DI
(mem:DI (plus:SI (ior:SI (plus:SI
(match_operand:SI 1 "register_operand" "r")
(match_operand:SI 2 "ua_offset" "I06"))
(const_int 7))
(const_int -7)))
(plus:SI (and:SI (match_dup 1) (const_int 7))
(const_int 1))
(const_int 0)))]
"TARGET_SHMEDIA32 && (INTVAL (operands[2]) & 7) == 0"
"#"
""
[(pc)]
{
emit_insn (gen_ldhi_q (operands[0],
gen_rtx_PLUS (SImode, operands[1], operands[2])));
DONE;
})
(define_insn_and_split "*ldhi_q_comb1"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extract:DI
(mem:DI (plus:SI (ior:SI (plus:SI
(match_operand:SI 1 "register_operand" "r")
(match_operand:SI 2 "ua_offset" "I06"))
(const_int 7))
(const_int -7)))
(plus:SI (and:SI (plus:SI (match_dup 1)
(match_operand:SI 3 "ua_offset" "I06"))
(const_int 7))
(const_int 1))
(const_int 0)))]
"TARGET_SHMEDIA32 && (INTVAL (operands[2]) & -8)
&& (INTVAL (operands[2]) & 7) == INTVAL (operands[3])"
"#"
""
[(pc)]
{
emit_insn (gen_ldhi_q (operands[0],
gen_rtx_PLUS (SImode, operands[1], operands[2])));
DONE;
})
(define_insn "ldlo_l"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(zero_extract:SI
(mem:SI (and:SI (match_operand:QI 1 "ua_address_operand" "p")
(const_int -4)))
(minus:SI (const_int 4) (and:SI (match_dup 1) (const_int 3)))
(and:SI (match_dup 1) (const_int 3))))]
"TARGET_SHMEDIA32"
"ldlo.l %U1, %0"
[(set_attr "type" "load_media")])
(define_insn "ldlo_q"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extract:DI
(mem:DI (and:SI (match_operand:QI 1 "ua_address_operand" "p")
(const_int -8)))
(minus:SI (const_int 8) (and:SI (match_dup 1) (const_int 7)))
(and:SI (match_dup 1) (const_int 7))))]
"TARGET_SHMEDIA32"
"ldlo.q %U1, %0"
[(set_attr "type" "load_media")])
(define_insn_and_split "*ldlo_q_comb0"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extract:DI
(mem:DI (and:SI (plus:SI (match_operand:SI 1 "register_operand" "r")
(match_operand:SI 2 "ua_offset" "I06"))
(const_int -8)))
(minus:SI (const_int 8) (and:SI (match_dup 1) (const_int 7)))
(and:SI (match_dup 1) (const_int 7))))]
"TARGET_SHMEDIA32 && (INTVAL (operands[2]) & 7) == 0"
"#"
""
[(pc)]
{
emit_insn (gen_ldlo_q (operands[0],
gen_rtx_PLUS (SImode, operands[1], operands[2])));
DONE;
})
(define_insn_and_split "*ldlo_q_comb1"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extract:DI
(mem:DI (and:SI (plus:SI (match_operand:SI 1 "register_operand" "r")
(match_operand:SI 2 "ua_offset" "I06"))
(const_int -8)))
(minus:SI (const_int 8)
(and:SI (plus:SI (match_dup 1)
(match_operand:SI 3 "ua_offset" "I06"))
(const_int 7)))
(and:SI (plus:SI (match_dup 1) (match_dup 3)) (const_int 7))))]
"TARGET_SHMEDIA32 && (INTVAL (operands[2]) & -8)
&& (INTVAL (operands[2]) & 7) == INTVAL (operands[3])"
"#"
""
[(pc)]
{
emit_insn (gen_ldlo_q (operands[0],
gen_rtx_PLUS (SImode, operands[1], operands[2])));
DONE;
})
(define_insn "sthi_l"
[(set (zero_extract:SI
(mem:SI (plus:SI (ior:SI (match_operand:QI 0 "ua_address_operand" "p")
(const_int 3))
(const_int -3)))
(plus:SI (and:SI (match_dup 0) (const_int 3)) (const_int 1))
(const_int 0))
(match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA32"
"sthi.l %U0, %1"
[(set_attr "type" "ustore_media")])
;; All unaligned stores are considered to be 'narrow' because they typically
;; operate on less that a quadword, and when they operate on a full quadword,
;; the vanilla store high / store low sequence will cause a stall if not
;; scheduled apart.
(define_insn "sthi_q"
[(set (zero_extract:DI
(mem:DI (plus:SI (ior:SI (match_operand:QI 0 "ua_address_operand" "p")
(const_int 7))
(const_int -7)))
(plus:SI (and:SI (match_dup 0) (const_int 7)) (const_int 1))
(const_int 0))
(match_operand:DI 1 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA32"
"sthi.q %U0, %1"
[(set_attr "type" "ustore_media")])
(define_insn_and_split "*sthi_q_comb0"
[(set (zero_extract:DI
(mem:DI (plus:SI (ior:SI (plus:SI
(match_operand:SI 0 "register_operand" "r")
(match_operand:SI 1 "ua_offset" "I06"))
(const_int 7))
(const_int -7)))
(plus:SI (and:SI (match_dup 0) (const_int 7)) (const_int 1))
(const_int 0))
(match_operand:DI 2 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA32 && (INTVAL (operands[1]) & 7) == 0"
"#"
""
[(pc)]
{
emit_insn (gen_sthi_q (gen_rtx_PLUS (SImode, operands[0], operands[1]),
operands[2]));
DONE;
})
(define_insn_and_split "*sthi_q_comb1"
[(set (zero_extract:DI
(mem:DI (plus:SI (ior:SI (plus:SI
(match_operand:SI 0 "register_operand" "r")
(match_operand:SI 1 "ua_offset" "I06"))
(const_int 7))
(const_int -7)))
(plus:SI (and:SI (plus:SI (match_dup 0)
(match_operand:SI 2 "ua_offset" "I06"))
(const_int 7))
(const_int 1))
(const_int 0))
(match_operand:DI 3 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA32 && (INTVAL (operands[1]) & -8)
&& (INTVAL (operands[1]) & 7) == INTVAL (operands[2])"
"#"
""
[(pc)]
{
emit_insn (gen_sthi_q (gen_rtx_PLUS (SImode, operands[0], operands[1]),
operands[3]));
DONE;
})
;; This is highpart user because the address is used as full 64 bit.
(define_insn "stlo_l"
[(set (zero_extract:SI
(mem:SI (and:SI (match_operand:QI 0 "ua_address_operand" "p")
(const_int -4)))
(minus:SI (const_int 4) (and:SI (match_dup 0) (const_int 3)))
(and:SI (match_dup 0) (const_int 3)))
(match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA32"
"stlo.l %U0, %1"
[(set_attr "type" "ustore_media")])
(define_insn "stlo_q"
[(set (zero_extract:DI
(mem:DI (and:SI (match_operand:QI 0 "ua_address_operand" "p")
(const_int -8)))
(minus:SI (const_int 8) (and:SI (match_dup 0) (const_int 7)))
(and:SI (match_dup 0) (const_int 7)))
(match_operand:DI 1 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA32"
"stlo.q %U0, %1"
[(set_attr "type" "ustore_media")])
(define_insn_and_split "*stlo_q_comb0"
[(set (zero_extract:DI
(mem:DI (and:SI (plus:SI (match_operand:SI 0 "register_operand" "r")
(match_operand:SI 1 "ua_offset" "I06"))
(const_int -8)))
(minus:SI (const_int 8) (and:SI (match_dup 0) (const_int 7)))
(and:SI (match_dup 0) (const_int 7)))
(match_operand:DI 2 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA32 && (INTVAL (operands[1]) & 7) == 0"
"#"
""
[(pc)]
{
emit_insn (gen_stlo_q (gen_rtx_PLUS (SImode, operands[0], operands[1]),
operands[2]));
DONE;
})
(define_insn_and_split "*stlo_q_comb1"
[(set (zero_extract:DI
(mem:DI (and:SI (plus:SI (match_operand:SI 0 "register_operand" "r")
(match_operand:SI 1 "ua_offset" "I06"))
(const_int -8)))
(minus:SI (const_int 8)
(and:SI (plus:SI (match_dup 0)
(match_operand:SI 2 "ua_offset" "I06"))
(const_int 7)))
(and:SI (plus:SI (match_dup 0) (match_dup 2)) (const_int 7)))
(match_operand:DI 3 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA32 && (INTVAL (operands[1]) & 7) == INTVAL (operands[2])"
"#"
""
[(pc)]
{
emit_insn (gen_stlo_q (gen_rtx_PLUS (SImode, operands[0], operands[1]),
operands[3]));
DONE;
})
(define_insn "ldhi_l64"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(zero_extract:SI
(mem:SI (plus:DI (ior:DI (match_operand:QI 1 "ua_address_operand" "p")
(const_int 3))
(const_int -3)))
(plus:DI (and:DI (match_dup 1) (const_int 3)) (const_int 1))
(const_int 0)))]
"TARGET_SHMEDIA64"
"ldhi.l %U1, %0"
[(set_attr "type" "load_media")])
(define_insn "ldhi_q64"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extract:DI
(mem:DI (plus:DI (ior:DI (match_operand:QI 1 "ua_address_operand" "p")
(const_int 7))
(const_int -7)))
(plus:DI (and:DI (match_dup 1) (const_int 7)) (const_int 1))
(const_int 0)))]
"TARGET_SHMEDIA64"
"ldhi.q %U1, %0"
[(set_attr "type" "load_media")])
(define_insn "ldlo_l64"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(zero_extract:SI
(mem:SI (and:DI (match_operand:QI 1 "ua_address_operand" "p")
(const_int -4)))
(minus:DI (const_int 4) (and:DI (match_dup 1) (const_int 3)))
(and:DI (match_dup 1) (const_int 3))))]
"TARGET_SHMEDIA64"
"ldlo.l %U1, %0"
[(set_attr "type" "load_media")])
(define_insn "ldlo_q64"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extract:DI
(mem:DI (and:DI (match_operand:QI 1 "ua_address_operand" "p")
(const_int -8)))
(minus:DI (const_int 8) (and:DI (match_dup 1) (const_int 7)))
(and:DI (match_dup 1) (const_int 7))))]
"TARGET_SHMEDIA64"
"ldlo.q %U1, %0"
[(set_attr "type" "load_media")])
(define_insn "sthi_l64"
[(set (zero_extract:SI
(mem:SI (plus:DI (ior:DI (match_operand:QI 0 "ua_address_operand" "p")
(const_int 3))
(const_int -3)))
(plus:DI (and:DI (match_dup 0) (const_int 3)) (const_int 1))
(const_int 0))
(match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA64"
"sthi.l %U0, %1"
[(set_attr "type" "ustore_media")])
(define_insn "sthi_q64"
[(set (zero_extract:DI
(mem:DI (plus:DI (ior:DI (match_operand:QI 0 "ua_address_operand" "p")
(const_int 7))
(const_int -7)))
(plus:DI (and:DI (match_dup 0) (const_int 7)) (const_int 1))
(const_int 0))
(match_operand:DI 1 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA64"
"sthi.q %U0, %1"
[(set_attr "type" "ustore_media")])
(define_insn "stlo_l64"
[(set (zero_extract:SI
(mem:SI (and:DI (match_operand:QI 0 "ua_address_operand" "p")
(const_int -4)))
(minus:DI (const_int 4) (and:DI (match_dup 0) (const_int 3)))
(and:DI (match_dup 0) (const_int 3)))
(match_operand:SI 1 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA64"
"stlo.l %U0, %1"
[(set_attr "type" "ustore_media")])
(define_insn "stlo_q64"
[(set (zero_extract:DI
(mem:DI (and:DI (match_operand:QI 0 "ua_address_operand" "p")
(const_int -8)))
(minus:DI (const_int 8) (and:DI (match_dup 0) (const_int 7)))
(and:DI (match_dup 0) (const_int 7)))
(match_operand:DI 1 "arith_reg_operand" "r"))]
"TARGET_SHMEDIA64"
"stlo.q %U0, %1"
[(set_attr "type" "ustore_media")])
(define_insn "nsb"
[(set (match_operand:QI 0 "arith_reg_dest" "=r")
(unspec:QI [(match_operand:DI 1 "arith_reg_operand" "r")]
UNSPEC_NSB))]
"TARGET_SHMEDIA"
"nsb %1, %0"
[(set_attr "type" "arith_media")])
(define_insn "nsbsi"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(zero_extend:SI
(unspec:QI [(match_operand:DI 1 "arith_reg_operand" "r")]
UNSPEC_NSB)))]
"TARGET_SHMEDIA"
"nsb %1, %0"
[(set_attr "type" "arith_media")])
(define_insn "nsbdi"
[(set (match_operand:DI 0 "arith_reg_dest" "=r")
(zero_extend:DI
(unspec:QI [(match_operand:DI 1 "arith_reg_operand" "r")]
UNSPEC_NSB)))]
"TARGET_SHMEDIA"
"nsb %1, %0"
[(set_attr "type" "arith_media")])
(define_expand "ffsdi2"
[(set (match_operand:DI 0 "arith_reg_dest" "")
(ffs:DI (match_operand:DI 1 "arith_reg_operand" "")))]
"TARGET_SHMEDIA"
{
rtx scratch = gen_reg_rtx (DImode);
rtx last;
emit_insn (gen_adddi3 (scratch, operands[1], constm1_rtx));
emit_insn (gen_xordi3 (scratch, operands[1], scratch));
emit_insn (gen_lshrdi3_media (scratch, scratch, const1_rtx));
emit_insn (gen_nsbdi (scratch, scratch));
emit_insn (gen_adddi3 (scratch, scratch, GEN_INT (-64)));
emit_insn (gen_movdicc_false (scratch, operands[1], const0_rtx, scratch));
last = emit_insn (gen_subdi3 (operands[0], const0_rtx, scratch));
set_unique_reg_note (last, REG_EQUAL, gen_rtx_FFS (DImode, operands[0]));
DONE;
})
(define_expand "ffssi2"
[(set (match_operand:SI 0 "arith_reg_dest" "")
(ffs:SI (match_operand:SI 1 "arith_reg_operand" "")))]
"TARGET_SHMEDIA"
{
rtx scratch = gen_reg_rtx (SImode);
rtx discratch = gen_reg_rtx (DImode);
rtx last;
emit_insn (gen_adddi3 (discratch,
simplify_gen_subreg (DImode, operands[1], SImode, 0),
constm1_rtx));
emit_insn (gen_andcdi3 (discratch,
simplify_gen_subreg (DImode, operands[1], SImode, 0),
discratch));
emit_insn (gen_nsbsi (scratch, discratch));
last = emit_insn (gen_subsi3 (operands[0],
force_reg (SImode, GEN_INT (63)), scratch));
set_unique_reg_note (last, REG_EQUAL, gen_rtx_FFS (SImode, operands[0]));
DONE;
})
(define_insn "byterev"
[(set (match_operand:V8QI 0 "arith_reg_dest" "=r")
(vec_select:V8QI (match_operand:V8QI 1 "arith_reg_operand" "r")
(parallel [(const_int 7) (const_int 6) (const_int 5)
(const_int 4) (const_int 3) (const_int 2)
(const_int 1) (const_int 0)])))]
"TARGET_SHMEDIA"
"byterev %1, %0"
[(set_attr "type" "arith_media")])
;; In user mode, the "pref" instruction will raise a RADDERR exception
;; for accesses to [0x80000000,0xffffffff]. This makes it an unsuitable
;; implementation of __builtin_prefetch for VxWorks RTPs.
(define_expand "prefetch"
[(prefetch (match_operand 0 "address_operand" "")
(match_operand:SI 1 "const_int_operand" "")
(match_operand:SI 2 "const_int_operand" ""))]
"(TARGET_SH2A || TARGET_SH3 || TARGET_SH5)
&& (TARGET_SHMEDIA || ! TARGET_VXWORKS_RTP)")
(define_insn "*prefetch"
[(prefetch (match_operand:SI 0 "register_operand" "r")
(match_operand:SI 1 "const_int_operand" "n")
(match_operand:SI 2 "const_int_operand" "n"))]
"(TARGET_SH2A || TARGET_SH3 || TARGET_SHCOMPACT) && ! TARGET_VXWORKS_RTP"
"pref @%0"
[(set_attr "type" "other")])
(define_insn "*prefetch_media"
[(prefetch (match_operand:QI 0 "address_operand" "p")
(match_operand:SI 1 "const_int_operand" "n")
(match_operand:SI 2 "const_int_operand" "n"))]
"TARGET_SHMEDIA"
{
operands[0] = gen_rtx_MEM (QImode, operands[0]);
output_asm_insn ("ld%M0.b %m0,r63", operands);
return "";
}
[(set_attr "type" "other")])
(define_insn "alloco_i"
[(set (mem:BLK (match_operand:QI 0 "cache_address_operand" "p"))
(unspec:BLK [(const_int 0)] UNSPEC_ALLOCO))]
"TARGET_SHMEDIA32"
{
rtx xops[2];
if (GET_CODE (operands[0]) == PLUS)
{
xops[0] = XEXP (operands[0], 0);
xops[1] = XEXP (operands[0], 1);
}
else
{
xops[0] = operands[0];
xops[1] = const0_rtx;
}
output_asm_insn ("alloco %0, %1", xops);
return "";
}
[(set_attr "type" "other")])
(define_split
[(set (match_operand 0 "any_register_operand" "")
(match_operand 1 "" ""))]
"TARGET_SHMEDIA && reload_completed"
[(set (match_dup 0) (match_dup 1))]
{
if (!shmedia_cleanup_truncate (operands[1]))
FAIL;
})
;; -------------------------------------------------------------------------
;; Stack Protector Patterns
;; -------------------------------------------------------------------------
(define_expand "stack_protect_set"
[(set (match_operand 0 "memory_operand" "")
(match_operand 1 "memory_operand" ""))]
""
{
if (TARGET_SHMEDIA)
{
if (TARGET_SHMEDIA64)
emit_insn (gen_stack_protect_set_di_media (operands[0], operands[1]));
else
emit_insn (gen_stack_protect_set_si_media (operands[0], operands[1]));
}
else
emit_insn (gen_stack_protect_set_si (operands[0], operands[1]));
DONE;
})
(define_insn "stack_protect_set_si"
[(set (match_operand:SI 0 "memory_operand" "=m")
(unspec:SI [(match_operand:SI 1 "memory_operand" "m")] UNSPEC_SP_SET))
(set (match_scratch:SI 2 "=&r") (const_int 0))]
"!TARGET_SHMEDIA"
{
return "mov.l %1,%2" "\n"
" mov.l %2,%0" "\n"
" mov #0,%2";
}
[(set_attr "type" "other")
(set_attr "length" "6")])
(define_insn "stack_protect_set_si_media"
[(set (match_operand:SI 0 "memory_operand" "=m")
(unspec:SI [(match_operand:SI 1 "memory_operand" "m")] UNSPEC_SP_SET))
(set (match_scratch:SI 2 "=&r") (const_int 0))]
"TARGET_SHMEDIA"
{
return "ld%M1.l %m1,%2" "\n"
" st%M0.l %m0,%2" "\n"
" movi 0,%2";
}
[(set_attr "type" "other")
(set_attr "length" "12")])
(define_insn "stack_protect_set_di_media"
[(set (match_operand:DI 0 "memory_operand" "=m")
(unspec:DI [(match_operand:DI 1 "memory_operand" "m")] UNSPEC_SP_SET))
(set (match_scratch:DI 2 "=&r") (const_int 0))]
"TARGET_SHMEDIA64"
{
return "ld%M1.q %m1,%2" "\n"
" st%M0.q %m0,%2" "\n"
" movi 0,%2";
}
[(set_attr "type" "other")
(set_attr "length" "12")])
(define_expand "stack_protect_test"
[(match_operand 0 "memory_operand" "")
(match_operand 1 "memory_operand" "")
(match_operand 2 "" "")]
""
{
if (TARGET_SHMEDIA)
{
rtx tmp = gen_reg_rtx (GET_MODE (operands[0]));
rtx test;
test = gen_rtx_NE (VOIDmode, tmp, const0_rtx);
if (TARGET_SHMEDIA64)
{
emit_insn (gen_stack_protect_test_di_media (tmp, operands[0],
operands[1]));
emit_jump_insn (gen_cbranchdi4 (test, tmp, const0_rtx, operands[2]));
}
else
{
emit_insn (gen_stack_protect_test_si_media (tmp, operands[0],
operands[1]));
emit_jump_insn (gen_cbranchsi4 (test, tmp, const0_rtx, operands[2]));
}
}
else
{
emit_insn (gen_stack_protect_test_si (operands[0], operands[1]));
emit_jump_insn (gen_branch_true (operands[2]));
}
DONE;
})
(define_insn "stack_protect_test_si"
[(set (reg:SI T_REG)
(unspec:SI [(match_operand:SI 0 "memory_operand" "m")
(match_operand:SI 1 "memory_operand" "m")]
UNSPEC_SP_TEST))
(set (match_scratch:SI 2 "=&r") (const_int 0))
(set (match_scratch:SI 3 "=&r") (const_int 0))]
"!TARGET_SHMEDIA"
{
return "mov.l %0,%2" "\n"
" mov.l %1,%3" "\n"
" cmp/eq %2,%3" "\n"
" mov #0,%2" "\n"
" mov #0,%3";
}
[(set_attr "type" "other")
(set_attr "length" "10")])
(define_insn "stack_protect_test_si_media"
[(set (match_operand:SI 0 "register_operand" "=&r")
(unspec:SI [(match_operand:SI 1 "memory_operand" "m")
(match_operand:SI 2 "memory_operand" "m")]
UNSPEC_SP_TEST))
(set (match_scratch:SI 3 "=&r") (const_int 0))]
"TARGET_SHMEDIA"
{
return "ld%M1.l %m1,%0" "\n"
" ld%M2.l %m2,%3" "\n"
" cmpeq %0,%3,%0" "\n"
" movi 0,%3";
}
[(set_attr "type" "other")
(set_attr "length" "16")])
(define_insn "stack_protect_test_di_media"
[(set (match_operand:DI 0 "register_operand" "=&r")
(unspec:DI [(match_operand:DI 1 "memory_operand" "m")
(match_operand:DI 2 "memory_operand" "m")]
UNSPEC_SP_TEST))
(set (match_scratch:DI 3 "=&r") (const_int 0))]
"TARGET_SHMEDIA64"
{
return "ld%M1.q %m1,%0" "\n"
" ld%M2.q %m2,%3" "\n"
" cmpeq %0,%3,%0" "\n"
" movi 0,%3";
}
[(set_attr "type" "other")
(set_attr "length" "16")])
;; -------------------------------------------------------------------------
;; Atomic operations
;; -------------------------------------------------------------------------
(include "sync.md")