summaryrefslogtreecommitdiff
path: root/sysdeps/ia64/fpu/s_nextafterf.S
diff options
context:
space:
mode:
authorUlrich Drepper <drepper@redhat.com>2005-03-31 10:02:53 +0000
committerUlrich Drepper <drepper@redhat.com>2005-03-31 10:02:53 +0000
commitee6189855aab3a9be8f3c2d95ce2b2cd17db4ec2 (patch)
treecf3e2fe1f9be5b358033fd927a0bd922542e04a1 /sysdeps/ia64/fpu/s_nextafterf.S
parent4d6302cf51b16a129addf7687c91490c40a7225c (diff)
downloadglibc-ee6189855aab3a9be8f3c2d95ce2b2cd17db4ec2.tar.gz
* sysdeps/unix/sysv/linux/x86_64/getcontext.S: Use functionally
equivalent, but shorter instructions. * sysdeps/unix/sysv/linux/x86_64/sysdep.h: Likewise. * sysdeps/unix/sysv/linux/x86_64/setcontext.S: Likewise. * sysdeps/unix/sysv/linux/x86_64/clone.S: Likewise. * sysdeps/unix/sysv/linux/x86_64/swapcontext.S: Likewise. * sysdeps/unix/x86_64/sysdep.S: Likewise. * sysdeps/x86_64/strchr.S: Likewise. * sysdeps/x86_64/memset.S: Likewise. * sysdeps/x86_64/strcspn.S: Likewise. * sysdeps/x86_64/strcmp.S: Likewise. * sysdeps/x86_64/elf/start.S: Likewise. * sysdeps/x86_64/strspn.S: Likewise. * sysdeps/x86_64/dl-machine.h: Likewise. * sysdeps/x86_64/bsd-_setjmp.S: Likewise. * sysdeps/x86_64/bsd-setjmp.S: Likewise. * sysdeps/x86_64/strtok.S: Likewise.
Diffstat (limited to 'sysdeps/ia64/fpu/s_nextafterf.S')
-rw-r--r--sysdeps/ia64/fpu/s_nextafterf.S274
1 files changed, 138 insertions, 136 deletions
diff --git a/sysdeps/ia64/fpu/s_nextafterf.S b/sysdeps/ia64/fpu/s_nextafterf.S
index 6d2a92796d..0c269eca9e 100644
--- a/sysdeps/ia64/fpu/s_nextafterf.S
+++ b/sysdeps/ia64/fpu/s_nextafterf.S
@@ -1,7 +1,7 @@
.file "nextafterf.s"
-// Copyright (c) 2000 - 2003, Intel Corporation
+// Copyright (c) 2000 - 2004, Intel Corporation
// All rights reserved.
//
// Contributed 2000 by the Intel Numerics Group, Intel Corporation
@@ -51,6 +51,7 @@
// fixed flag settings for several cases
// 05/20/02 Cleaned up namespace and sf0 syntax
// 02/10/03 Reordered header: .section, .global, .proc, .align
+// 12/14/04 Added error handling on underflow.
//
// API
//==============================================================
@@ -60,21 +61,21 @@
//
// Registers used
//==============================================================
-nextafter_GR_max_pexp = r14
-nextafter_GR_min_pexp = r15
-nextafter_GR_exp = r16
-nextafter_GR_sig = r17
-nextafter_GR_lnorm_sig = r18
-nextafter_GR_sign_mask = r19
-nextafter_GR_exp_mask = r20
-nextafter_GR_sden_sig = r21
-nextafter_GR_new_sig = r22
-nextafter_GR_new_exp = r23
-nextafter_GR_lden_sig = r24
-nextafter_GR_snorm_sig = r25
-nextafter_GR_exp1 = r26
-nextafter_GR_x_exp = r27
-nextafter_GR_min_den_rexp = r28
+GR_max_pexp = r14
+GR_min_pexp = r15
+GR_exp = r16
+GR_sig = r17
+GR_lnorm_sig = r18
+GR_sign_mask = r19
+GR_exp_mask = r20
+GR_sden_sig = r21
+GR_new_sig = r22
+GR_new_exp = r23
+GR_lden_sig = r24
+GR_snorm_sig = r25
+GR_exp1 = r26
+GR_x_exp = r27
+GR_min_den_rexp = r28
// r36-39 parameters for libm_error_support
GR_SAVE_B0 = r34
@@ -84,20 +85,21 @@ GR_SAVE_PFS = r32
GR_Parameter_X = r36
GR_Parameter_Y = r37
GR_Parameter_RESULT = r38
-
-NEXTAFTER_lnorm_sig = f10
-NEXTAFTER_lnorm_exp = f11
-NEXTAFTER_lnorm = f12
-NEXTAFTER_sden_sig = f13
-NEXTAFTER_sden_exp = f14
-NEXTAFTER_sden = f15
-NEXTAFTER_save_f8 = f33
-NEXTAFTER_new_exp = f34
-NEXTAFTER_new_sig = f35
-NEXTAFTER_lden_sig = f36
-NEXTAFTER_snorm_sig = f37
-NEXTAFTER_exp1 = f38
-NEXTAFTER_tmp = f39
+GR_Parameter_TAG = r39
+
+FR_lnorm_sig = f10
+FR_lnorm_exp = f11
+FR_lnorm = f12
+FR_sden_sig = f13
+FR_sden_exp = f14
+FR_sden = f15
+FR_save_f8 = f33
+FR_new_exp = f34
+FR_new_sig = f35
+FR_lden_sig = f36
+FR_snorm_sig = f37
+FR_exp1 = f38
+FR_tmp = f39
//
// Overview of operation
@@ -112,21 +114,21 @@ GLOBAL_LIBM_ENTRY(nextafterf)
// Extract signexp from x
// Form smallest denormal significand = ulp size
{ .mlx
- getf.exp nextafter_GR_exp = f8
- movl nextafter_GR_sden_sig = 0x0000010000000000
+ getf.exp GR_exp = f8
+ movl GR_sden_sig = 0x0000010000000000
}
// Form largest normal exponent
// Is x < y ? p10 if yes, p11 if no
// Form smallest normal exponent
{ .mfi
- addl nextafter_GR_max_pexp = 0x1007e, r0
+ addl GR_max_pexp = 0x1007e, r0
fcmp.lt.s1 p10,p11 = f8, f9
- addl nextafter_GR_min_pexp = 0x0ff81, r0 ;;
+ addl GR_min_pexp = 0x0ff81, r0 ;;
}
// Is x=y?
{ .mfi
- getf.sig nextafter_GR_sig = f8
+ getf.sig GR_sig = f8
fcmp.eq.s0 p6,p0 = f8, f9
nop.i 0
}
@@ -134,14 +136,14 @@ GLOBAL_LIBM_ENTRY(nextafterf)
// Form largest normal significand
{ .mlx
nop.m 0
- movl nextafter_GR_lnorm_sig = 0xffffff0000000000 ;;
+ movl GR_lnorm_sig = 0xffffff0000000000 ;;
}
// Move largest normal significand to fp reg for special cases
{ .mfi
- setf.sig NEXTAFTER_lnorm_sig = nextafter_GR_lnorm_sig
+ setf.sig FR_lnorm_sig = GR_lnorm_sig
nop.f 0
- addl nextafter_GR_sign_mask = 0x20000, r0 ;;
+ addl GR_sign_mask = 0x20000, r0 ;;
}
// Move smallest denormal significand and signexp to fp regs
@@ -150,14 +152,14 @@ GLOBAL_LIBM_ENTRY(nextafterf)
// It increases (p12 set) if x<y and x>=0 or if x>y and x<0
// It decreases (p13 set) if x<y and x<0 or if x>y and x>=0
{ .mfi
- setf.sig NEXTAFTER_sden_sig = nextafter_GR_sden_sig
+ setf.sig FR_sden_sig = GR_sden_sig
fclass.m p8,p0 = f8, 0xc3
-(p10) cmp.lt p12,p13 = nextafter_GR_exp, nextafter_GR_sign_mask
+(p10) cmp.lt p12,p13 = GR_exp, GR_sign_mask
}
{ .mfi
- setf.exp NEXTAFTER_sden_exp = nextafter_GR_min_pexp
+ setf.exp FR_sden_exp = GR_min_pexp
nop.f 999
-(p11) cmp.ge p12,p13 = nextafter_GR_exp, nextafter_GR_sign_mask ;;
+(p11) cmp.ge p12,p13 = GR_exp, GR_sign_mask ;;
}
.pred.rel "mutex",p12,p13
@@ -166,33 +168,33 @@ GLOBAL_LIBM_ENTRY(nextafterf)
// If x=y set result to y
// Form smallest normal significand and largest denormal significand
{ .mfi
-(p12) add nextafter_GR_new_sig = nextafter_GR_sig, nextafter_GR_sden_sig
+(p12) add GR_new_sig = GR_sig, GR_sden_sig
(p6) fmerge.s f8=f9,f9
- dep.z nextafter_GR_snorm_sig = 1,63,1 // 0x8000000000000000
+ dep.z GR_snorm_sig = 1,63,1 // 0x8000000000000000
}
{ .mlx
-(p13) sub nextafter_GR_new_sig = nextafter_GR_sig, nextafter_GR_sden_sig
- movl nextafter_GR_lden_sig = 0x7fffff0000000000 ;;
+(p13) sub GR_new_sig = GR_sig, GR_sden_sig
+ movl GR_lden_sig = 0x7fffff0000000000 ;;
}
// Move expected result significand and signexp to fp regs
// Is y=nan?
// Form new exponent in case result exponent needs incrementing or decrementing
{ .mfi
- setf.exp NEXTAFTER_new_exp = nextafter_GR_exp
+ setf.exp FR_new_exp = GR_exp
fclass.m p9,p0 = f9, 0xc3
-(p12) add nextafter_GR_exp1 = 1, nextafter_GR_exp
+(p12) add GR_exp1 = 1, GR_exp
}
{ .mib
- setf.sig NEXTAFTER_new_sig = nextafter_GR_new_sig
-(p13) add nextafter_GR_exp1 = -1, nextafter_GR_exp
+ setf.sig FR_new_sig = GR_new_sig
+(p13) add GR_exp1 = -1, GR_exp
(p6) br.ret.spnt b0 ;; // Exit if x=y
}
// Move largest normal signexp to fp reg for special cases
// Is x=zero?
{ .mfi
- setf.exp NEXTAFTER_lnorm_exp = nextafter_GR_max_pexp
+ setf.exp FR_lnorm_exp = GR_max_pexp
fclass.m p7,p0 = f8, 0x7
nop.i 999
}
@@ -205,12 +207,12 @@ GLOBAL_LIBM_ENTRY(nextafterf)
// Move exp+-1 and smallest normal significand to fp regs for special cases
// Is x=inf?
{ .mfi
- setf.exp NEXTAFTER_exp1 = nextafter_GR_exp1
+ setf.exp FR_exp1 = GR_exp1
fclass.m p6,p0 = f8, 0x23
- addl nextafter_GR_exp_mask = 0x1ffff, r0
+ addl GR_exp_mask = 0x1ffff, r0
}
{ .mfb
- setf.sig NEXTAFTER_snorm_sig = nextafter_GR_snorm_sig
+ setf.sig FR_snorm_sig = GR_snorm_sig
(p9) fma.s0 f8 = f8,f1,f9
(p9) br.ret.spnt b0 ;; // Exit if y=nan
}
@@ -218,16 +220,16 @@ GLOBAL_LIBM_ENTRY(nextafterf)
// Move largest denormal significand to fp regs for special cases
// Save x
{ .mfb
- setf.sig NEXTAFTER_lden_sig = nextafter_GR_lden_sig
- mov NEXTAFTER_save_f8 = f8
-(p7) br.cond.spnt NEXTAFTER_ZERO ;; // Exit if x=0
+ setf.sig FR_lden_sig = GR_lden_sig
+ mov FR_save_f8 = f8
+(p7) br.cond.spnt NEXT_ZERO ;; // Exit if x=0
}
// Mask off the sign to get x_exp
{ .mfb
- and nextafter_GR_x_exp = nextafter_GR_exp_mask, nextafter_GR_exp
+ and GR_x_exp = GR_exp_mask, GR_exp
nop.f 999
-(p6) br.cond.spnt NEXTAFTER_INF ;; // Exit if x=inf
+(p6) br.cond.spnt NEXT_INF ;; // Exit if x=inf
}
// Check 6 special cases when significand rolls over:
@@ -246,35 +248,35 @@ GLOBAL_LIBM_ENTRY(nextafterf)
//
// Form exponent of smallest float denormal (if normalized register format)
{ .mmi
- adds nextafter_GR_min_den_rexp = -23, nextafter_GR_min_pexp
-(p12) cmp.eq.unc p6,p0 = nextafter_GR_new_sig, r0
-(p13) cmp.eq.unc p8,p10 = nextafter_GR_new_sig, nextafter_GR_lden_sig ;;
+ adds GR_min_den_rexp = -23, GR_min_pexp
+(p12) cmp.eq.unc p6,p0 = GR_new_sig, r0
+(p13) cmp.eq.unc p8,p10 = GR_new_sig, GR_lden_sig ;;
}
{ .mmi
-(p6) cmp.lt.unc p6,p7 = nextafter_GR_x_exp, nextafter_GR_max_pexp
-(p8) cmp.gt.unc p8,p9 = nextafter_GR_x_exp, nextafter_GR_min_pexp
-(p10) cmp.eq.unc p10,p0 = nextafter_GR_new_sig, r0 ;;
+(p6) cmp.lt.unc p6,p7 = GR_x_exp, GR_max_pexp
+(p8) cmp.gt.unc p8,p9 = GR_x_exp, GR_min_pexp
+(p10) cmp.eq.unc p10,p0 = GR_new_sig, r0 ;;
}
// Create small normal in case need to generate underflow flag
{ .mfi
-(p10) cmp.le.unc p10,p0 = nextafter_GR_x_exp, nextafter_GR_min_pexp
- fmerge.se NEXTAFTER_tmp = NEXTAFTER_sden_exp, NEXTAFTER_lnorm_sig
-(p9) cmp.gt.unc p9,p14 = nextafter_GR_x_exp, nextafter_GR_min_den_rexp
+(p10) cmp.le.unc p10,p0 = GR_x_exp, GR_min_pexp
+ fmerge.se FR_tmp = FR_sden_exp, FR_lnorm_sig
+(p9) cmp.gt.unc p9,p14 = GR_x_exp, GR_min_den_rexp
}
// Branch if cases 1, 2, 3
{ .bbb
-(p6) br.cond.spnt NEXTAFTER_EXPUP
-(p7) br.cond.spnt NEXTAFTER_OVERFLOW
-(p8) br.cond.spnt NEXTAFTER_EXPDOWN ;;
+(p6) br.cond.spnt NEXT_EXPUP
+(p7) br.cond.spnt NEXT_OVERFLOW
+(p8) br.cond.spnt NEXT_EXPDOWN ;;
}
// Branch if cases 4, 5, 6
{ .bbb
-(p9) br.cond.spnt NEXTAFTER_NORM_TO_DENORM
-(p10) br.cond.spnt NEXTAFTER_UNDERFLOW_TO_ZERO
-(p14) br.cond.spnt NEXTAFTER_UNDERFLOW_TO_ZERO ;;
+(p9) br.cond.spnt NEXT_NORM_TO_DENORM
+(p10) br.cond.spnt NEXT_UNDERFLOW_TO_ZERO
+(p14) br.cond.spnt NEXT_UNDERFLOW_TO_ZERO ;;
}
// Here if no special cases
@@ -282,68 +284,72 @@ GLOBAL_LIBM_ENTRY(nextafterf)
// Case 1: x_exp=min_exp, x_sig=unnormalized
// Case 2: x_exp<min_exp
{ .mfi
- cmp.lt p6,p7 = nextafter_GR_x_exp, nextafter_GR_min_pexp
- fmerge.se f8 = NEXTAFTER_new_exp, NEXTAFTER_new_sig
+ cmp.lt p6,p7 = GR_x_exp, GR_min_pexp
+ fmerge.se f8 = FR_new_exp, FR_new_sig
nop.i 999 ;;
}
{ .mfi
nop.m 999
nop.f 999
-(p7) tbit.z p6,p0 = nextafter_GR_new_sig, 63 ;;
+(p7) tbit.z p6,p0 = GR_new_sig, 63 ;;
}
-NEXTAFTER_COMMON_FINISH:
+NEXT_COMMON_FINISH:
// Force underflow and inexact if denormal result
{ .mfi
nop.m 999
-(p6) fma.s.s0 NEXTAFTER_tmp = NEXTAFTER_tmp,NEXTAFTER_tmp,f0
- nop.i 999 ;;
+(p6) fma.s.s0 FR_tmp = FR_tmp,FR_tmp,f0
+ nop.i 999
+}
+{ .mfb
+ nop.m 999
+ fnorm.s.s0 f8 = f8 // Final normalization to result precision
+(p6) br.cond.spnt NEXT_UNDERFLOW ;;
}
-// Final normalization to result precision and exit
{ .mfb
nop.m 999
- fnorm.s.s0 f8 = f8
+ nop.f 999
br.ret.sptk b0;;
}
//Special cases
-NEXTAFTER_EXPUP:
+NEXT_EXPUP:
{ .mfb
- cmp.lt p6,p7 = nextafter_GR_x_exp, nextafter_GR_min_pexp
- fmerge.se f8 = NEXTAFTER_exp1, NEXTAFTER_snorm_sig
- br.cond.sptk NEXTAFTER_COMMON_FINISH ;;
+ cmp.lt p6,p7 = GR_x_exp, GR_min_pexp
+ fmerge.se f8 = FR_exp1, FR_snorm_sig
+ br.cond.sptk NEXT_COMMON_FINISH ;;
}
-NEXTAFTER_EXPDOWN:
+NEXT_EXPDOWN:
{ .mfb
- cmp.lt p6,p7 = nextafter_GR_x_exp, nextafter_GR_min_pexp
- fmerge.se f8 = NEXTAFTER_exp1, NEXTAFTER_lnorm_sig
- br.cond.sptk NEXTAFTER_COMMON_FINISH ;;
+ cmp.lt p6,p7 = GR_x_exp, GR_min_pexp
+ fmerge.se f8 = FR_exp1, FR_lnorm_sig
+ br.cond.sptk NEXT_COMMON_FINISH ;;
}
-NEXTAFTER_NORM_TO_DENORM:
+NEXT_NORM_TO_DENORM:
{ .mfi
nop.m 999
- fmerge.se f8 = NEXTAFTER_new_exp, NEXTAFTER_lden_sig
+ fmerge.se f8 = FR_new_exp, FR_lden_sig
nop.i 999
}
// Force underflow and inexact
{ .mfb
nop.m 999
- fma.s.s0 NEXTAFTER_tmp = NEXTAFTER_tmp,NEXTAFTER_tmp,f0
- br.ret.sptk b0 ;;
+ fma.s.s0 FR_tmp = FR_tmp,FR_tmp,f0
+ br.cond.sptk NEXT_UNDERFLOW ;;
}
-NEXTAFTER_UNDERFLOW_TO_ZERO:
+NEXT_UNDERFLOW_TO_ZERO:
{ .mfb
cmp.eq p6,p0 = r0,r0
- fmerge.s f8 = NEXTAFTER_save_f8,f0
- br.cond.sptk NEXTAFTER_COMMON_FINISH ;;
+ fmerge.s f8 = FR_save_f8,f0
+ br.cond.sptk NEXT_COMMON_FINISH ;;
}
-NEXTAFTER_INF:
+NEXT_INF:
// Here if f8 is +- infinity
// INF
// if f8 is +inf, no matter what y is return largest float
@@ -351,17 +357,17 @@ NEXTAFTER_INF:
{ .mfi
nop.m 999
- fmerge.se NEXTAFTER_lnorm = NEXTAFTER_lnorm_exp,NEXTAFTER_lnorm_sig
+ fmerge.se FR_lnorm = FR_lnorm_exp,FR_lnorm_sig
nop.i 999 ;;
}
{ .mfb
nop.m 999
- fmerge.s f8 = f8,NEXTAFTER_lnorm
+ fmerge.s f8 = f8,FR_lnorm
br.ret.sptk b0 ;;
}
-NEXTAFTER_ZERO:
+NEXT_ZERO:
// Here if f8 is +- zero
// ZERO
@@ -370,76 +376,72 @@ NEXTAFTER_ZERO:
{ .mfi
nop.m 999
- fmerge.se NEXTAFTER_sden = NEXTAFTER_sden_exp,NEXTAFTER_sden_sig
+ fmerge.se FR_sden = FR_sden_exp,FR_sden_sig
nop.i 999 ;;
}
// Create small normal to generate underflow flag
{ .mfi
nop.m 999
- fmerge.se NEXTAFTER_tmp = NEXTAFTER_sden_exp, NEXTAFTER_lnorm_sig
+ fmerge.se FR_tmp = FR_sden_exp, FR_lnorm_sig
nop.i 999 ;;
}
// Add correct sign from direction arg
{ .mfi
nop.m 999
- fmerge.s f8 = f9,NEXTAFTER_sden
+ fmerge.s f8 = f9,FR_sden
nop.i 999 ;;
}
// Force underflow and inexact flags
{ .mfb
nop.m 999
- fma.s.s0 NEXTAFTER_tmp = NEXTAFTER_tmp,NEXTAFTER_tmp,f0
- br.ret.sptk b0 ;;
+ fma.s.s0 FR_tmp = FR_tmp,FR_tmp,f0
+ br.cond.sptk NEXT_UNDERFLOW ;;
}
-GLOBAL_LIBM_END(nextafterf)
-// Stack operations when calling error support.
-// (1) (2) (3) (call) (4)
-// sp -> + psp -> + psp -> + sp -> +
-// | | | |
-// | | <- GR_Y R3 ->| <- GR_RESULT | -> f8
-// | | | |
-// | <-GR_Y Y2->| Y2 ->| <- GR_Y |
-// | | | |
-// | | <- GR_X X1 ->| |
-// | | | |
-// sp-64 -> + sp -> + sp -> + +
-// save ar.pfs save b0 restore gp
-// save gp restore ar.pfs
-
-
+NEXT_UNDERFLOW:
+// Here if result is a denorm, or input is finite and result is zero
+// Call error support to report possible range error
+{ .mib
+ alloc r32=ar.pfs,2,2,4,0
+ mov GR_Parameter_TAG = 269 // Error code
+ br.cond.sptk __libm_error_region // Branch to error call
+}
+;;
-LOCAL_LIBM_ENTRY(__libm_error_region)
-NEXTAFTER_OVERFLOW:
-// Here if f8 is finite, but result will be infinite
+NEXT_OVERFLOW:
+// Here if input is finite, but result will be infinite
// Use frcpa to generate infinity of correct sign
// Call error support to report possible range error
-.prologue
-
{ .mfi
alloc r32=ar.pfs,2,2,4,0
- frcpa.s1 f8,p6 = NEXTAFTER_save_f8, f0
- nop.i 999
+ frcpa.s1 f8,p6 = FR_save_f8, f0
+ nop.i 999 ;;
}
-// Create largest float
+// Create largest double
{ .mfi
nop.m 999
- fmerge.se NEXTAFTER_lnorm = NEXTAFTER_lnorm_exp,NEXTAFTER_lnorm_sig
+ fmerge.se FR_lnorm = FR_lnorm_exp,FR_lnorm_sig
nop.i 999 ;;
}
// Force overflow and inexact flags to be set
-{ .mfi
- mov r39 = 155 // Error code
- fma.s.s0 NEXTAFTER_tmp = NEXTAFTER_lnorm,NEXTAFTER_lnorm,f0
- nop.i 999
+{ .mfb
+ mov GR_Parameter_TAG = 155 // Error code
+ fma.s.s0 FR_tmp = FR_lnorm,FR_lnorm,f0
+ br.cond.sptk __libm_error_region // Branch to error call
}
;;
+GLOBAL_LIBM_END(nextafterf)
+
+
+LOCAL_LIBM_ENTRY(__libm_error_region)
+.prologue
+
// (1)
{ .mfi
add GR_Parameter_Y=-32,sp // Parameter 2 value
@@ -466,7 +468,7 @@ NEXTAFTER_OVERFLOW:
.body
// (3)
{ .mib
- stfs [GR_Parameter_X] = NEXTAFTER_save_f8 // STORE Parameter 1 on stack
+ stfs [GR_Parameter_X] = FR_save_f8 // STORE Parameter 1 on stack
add GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address
nop.b 0
}