ChangeLog:

	* config/spu/spu_internals.h (spu_sr, spu_sra, spu_srqw,
	spu_srqwbyte, spu_srqwbytebc): Define.
	* config/spu/spu-builtins.def (spu_sr, spu_sra, spu_srqw,
	spu_srqwbyte, spu_srqwbytebc): New overloaded builtins.
	* config/spu/spu.md ("shrqbybi_<mode>", "shrqbi_<mode>",
	"shrqby_<mode>"): New insn-and-split patterns.
	* config/spu/spu.c (expand_builtin_args): Determine and return
	number of operands using spu_builtin_description data.
	(spu_expand_builtin_1): Use it.

testsuite/ChangeLog:

	* gcc.target/spu/intrinsics-sr.c: New test.


git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@144178 138bc75d-0d04-0410-961f-82ee72b054a4

4 files changed, 128 insertions, 5 deletions

diff --git a/gcc/config/spu/spu-builtins.def b/gcc/config/spu/spu-builtins.def
index eecd337f73f..17ce6163b5a 100644
--- a/gcc/config/spu/spu-builtins.def
+++ b/gcc/config/spu/spu-builtins.def
@@ -670,6 +670,57 @@ DEF_BUILTIN (SPU_SLQWBYTEBC_6,     CODE_FOR_shlqbybi_ti,   "spu_slqwbytebc_6",
 DEF_BUILTIN (SPU_SLQWBYTEBC_7,     CODE_FOR_shlqbybi_ti,   "spu_slqwbytebc_7",     B_INTERNAL, _A3(SPU_BTI_UV16QI, SPU_BTI_UV16QI, SPU_BTI_UINTSI))
 DEF_BUILTIN (SPU_SLQWBYTEBC_8,     CODE_FOR_shlqbybi_ti,   "spu_slqwbytebc_8",     B_INTERNAL, _A3(SPU_BTI_V4SF,   SPU_BTI_V4SF,   SPU_BTI_UINTSI))
 DEF_BUILTIN (SPU_SLQWBYTEBC_9,     CODE_FOR_shlqbybi_ti,   "spu_slqwbytebc_9",     B_INTERNAL, _A3(SPU_BTI_V2DF,   SPU_BTI_V2DF,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SR,               CODE_FOR_nothing,       "spu_sr",               B_OVERLOAD, _A1(SPU_BTI_VOID))
+DEF_BUILTIN (SPU_SR_0,             CODE_FOR_vlshrv8hi3,    "spu_sr_0",             B_INTERNAL, _A3(SPU_BTI_UV8HI,  SPU_BTI_UV8HI,  SPU_BTI_UV8HI))
+DEF_BUILTIN (SPU_SR_1,             CODE_FOR_vlshrv8hi3,    "spu_sr_1",             B_INTERNAL, _A3(SPU_BTI_V8HI,   SPU_BTI_V8HI,   SPU_BTI_UV8HI))
+DEF_BUILTIN (SPU_SR_2,             CODE_FOR_vlshrv4si3,    "spu_sr_2",             B_INTERNAL, _A3(SPU_BTI_UV4SI,  SPU_BTI_UV4SI,  SPU_BTI_UV4SI))
+DEF_BUILTIN (SPU_SR_3,             CODE_FOR_vlshrv4si3,    "spu_sr_3",             B_INTERNAL, _A3(SPU_BTI_V4SI,   SPU_BTI_V4SI,   SPU_BTI_UV4SI))
+DEF_BUILTIN (SPU_SR_4,             CODE_FOR_vlshrv8hi3,    "spu_sr_4",             B_INTERNAL, _A3(SPU_BTI_UV8HI,  SPU_BTI_UV8HI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SR_5,             CODE_FOR_vlshrv8hi3,    "spu_sr_5",             B_INTERNAL, _A3(SPU_BTI_V8HI,   SPU_BTI_V8HI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SR_6,             CODE_FOR_vlshrv4si3,    "spu_sr_6",             B_INTERNAL, _A3(SPU_BTI_UV4SI,  SPU_BTI_UV4SI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SR_7,             CODE_FOR_vlshrv4si3,    "spu_sr_7",             B_INTERNAL, _A3(SPU_BTI_V4SI,   SPU_BTI_V4SI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRA,              CODE_FOR_nothing,       "spu_sra",              B_OVERLOAD, _A1(SPU_BTI_VOID))
+DEF_BUILTIN (SPU_SRA_0,            CODE_FOR_vashrv8hi3,    "spu_sra_0",            B_INTERNAL, _A3(SPU_BTI_UV8HI,  SPU_BTI_UV8HI,  SPU_BTI_UV8HI))
+DEF_BUILTIN (SPU_SRA_1,            CODE_FOR_vashrv8hi3,    "spu_sra_1",            B_INTERNAL, _A3(SPU_BTI_V8HI,   SPU_BTI_V8HI,   SPU_BTI_UV8HI))
+DEF_BUILTIN (SPU_SRA_2,            CODE_FOR_vashrv4si3,    "spu_sra_2",            B_INTERNAL, _A3(SPU_BTI_UV4SI,  SPU_BTI_UV4SI,  SPU_BTI_UV4SI))
+DEF_BUILTIN (SPU_SRA_3,            CODE_FOR_vashrv4si3,    "spu_sra_3",            B_INTERNAL, _A3(SPU_BTI_V4SI,   SPU_BTI_V4SI,   SPU_BTI_UV4SI))
+DEF_BUILTIN (SPU_SRA_4,            CODE_FOR_vashrv8hi3,    "spu_sra_4",            B_INTERNAL, _A3(SPU_BTI_UV8HI,  SPU_BTI_UV8HI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRA_5,            CODE_FOR_vashrv8hi3,    "spu_sra_5",            B_INTERNAL, _A3(SPU_BTI_V8HI,   SPU_BTI_V8HI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRA_6,            CODE_FOR_vashrv4si3,    "spu_sra_6",            B_INTERNAL, _A3(SPU_BTI_UV4SI,  SPU_BTI_UV4SI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRA_7,            CODE_FOR_vashrv4si3,    "spu_sra_7",            B_INTERNAL, _A3(SPU_BTI_V4SI,   SPU_BTI_V4SI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW,             CODE_FOR_nothing,       "spu_srqw",             B_OVERLOAD, _A1(SPU_BTI_VOID))
+DEF_BUILTIN (SPU_SRQW_0,           CODE_FOR_shrqbi_ti,     "spu_srqw_0",           B_INTERNAL, _A3(SPU_BTI_V2DI,   SPU_BTI_V2DI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW_1,           CODE_FOR_shrqbi_ti,     "spu_srqw_1",           B_INTERNAL, _A3(SPU_BTI_UV2DI,  SPU_BTI_UV2DI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW_2,           CODE_FOR_shrqbi_ti,     "spu_srqw_2",           B_INTERNAL, _A3(SPU_BTI_V4SI,   SPU_BTI_V4SI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW_3,           CODE_FOR_shrqbi_ti,     "spu_srqw_3",           B_INTERNAL, _A3(SPU_BTI_UV4SI,  SPU_BTI_UV4SI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW_4,           CODE_FOR_shrqbi_ti,     "spu_srqw_4",           B_INTERNAL, _A3(SPU_BTI_V8HI,   SPU_BTI_V8HI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW_5,           CODE_FOR_shrqbi_ti,     "spu_srqw_5",           B_INTERNAL, _A3(SPU_BTI_UV8HI,  SPU_BTI_UV8HI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW_6,           CODE_FOR_shrqbi_ti,     "spu_srqw_6",           B_INTERNAL, _A3(SPU_BTI_V16QI,  SPU_BTI_V16QI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW_7,           CODE_FOR_shrqbi_ti,     "spu_srqw_7",           B_INTERNAL, _A3(SPU_BTI_UV16QI, SPU_BTI_UV16QI, SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW_8,           CODE_FOR_shrqbi_ti,     "spu_srqw_8",           B_INTERNAL, _A3(SPU_BTI_V4SF,   SPU_BTI_V4SF,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQW_9,           CODE_FOR_shrqbi_ti,     "spu_srqw_9",           B_INTERNAL, _A3(SPU_BTI_V2DF,   SPU_BTI_V2DF,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE,         CODE_FOR_nothing,       "spu_srqwbyte",         B_OVERLOAD, _A1(SPU_BTI_VOID))
+DEF_BUILTIN (SPU_SRQWBYTE_0,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_0",       B_INTERNAL, _A3(SPU_BTI_V2DI,   SPU_BTI_V2DI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE_1,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_1",       B_INTERNAL, _A3(SPU_BTI_UV2DI,  SPU_BTI_UV2DI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE_2,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_2",       B_INTERNAL, _A3(SPU_BTI_V4SI,   SPU_BTI_V4SI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE_3,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_3",       B_INTERNAL, _A3(SPU_BTI_UV4SI,  SPU_BTI_UV4SI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE_4,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_4",       B_INTERNAL, _A3(SPU_BTI_V8HI,   SPU_BTI_V8HI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE_5,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_5",       B_INTERNAL, _A3(SPU_BTI_UV8HI,  SPU_BTI_UV8HI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE_6,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_6",       B_INTERNAL, _A3(SPU_BTI_V16QI,  SPU_BTI_V16QI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE_7,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_7",       B_INTERNAL, _A3(SPU_BTI_UV16QI, SPU_BTI_UV16QI, SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE_8,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_8",       B_INTERNAL, _A3(SPU_BTI_V4SF,   SPU_BTI_V4SF,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTE_9,       CODE_FOR_shrqby_ti,     "spu_srqwbyte_9",       B_INTERNAL, _A3(SPU_BTI_V2DF,   SPU_BTI_V2DF,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC,       CODE_FOR_nothing,       "spu_srqwbytebc",       B_OVERLOAD, _A1(SPU_BTI_VOID))
+DEF_BUILTIN (SPU_SRQWBYTEBC_0,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_0",     B_INTERNAL, _A3(SPU_BTI_V2DI,   SPU_BTI_V2DI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC_1,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_1",     B_INTERNAL, _A3(SPU_BTI_UV2DI,  SPU_BTI_UV2DI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC_2,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_2",     B_INTERNAL, _A3(SPU_BTI_V4SI,   SPU_BTI_V4SI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC_3,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_3",     B_INTERNAL, _A3(SPU_BTI_UV4SI,  SPU_BTI_UV4SI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC_4,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_4",     B_INTERNAL, _A3(SPU_BTI_V8HI,   SPU_BTI_V8HI,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC_5,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_5",     B_INTERNAL, _A3(SPU_BTI_UV8HI,  SPU_BTI_UV8HI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC_6,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_6",     B_INTERNAL, _A3(SPU_BTI_V16QI,  SPU_BTI_V16QI,  SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC_7,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_7",     B_INTERNAL, _A3(SPU_BTI_UV16QI, SPU_BTI_UV16QI, SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC_8,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_8",     B_INTERNAL, _A3(SPU_BTI_V4SF,   SPU_BTI_V4SF,   SPU_BTI_UINTSI))
+DEF_BUILTIN (SPU_SRQWBYTEBC_9,     CODE_FOR_shrqbybi_ti,   "spu_srqwbytebc_9",     B_INTERNAL, _A3(SPU_BTI_V2DF,   SPU_BTI_V2DF,   SPU_BTI_UINTSI))
 
 DEF_BUILTIN (SPU_SPLATS,           CODE_FOR_nothing,       "spu_splats",           B_OVERLOAD, _A1(SPU_BTI_VOID))
 DEF_BUILTIN (SPU_SPLATS_0,         CODE_FOR_spu_splats,    "spu_splats_0",         B_INTERNAL, _A2(SPU_BTI_UV16QI, SPU_BTI_UINTQI))
diff --git a/gcc/config/spu/spu.c b/gcc/config/spu/spu.c
index 75f785393f9..a008830d02b 100644
--- a/gcc/config/spu/spu.c
+++ b/gcc/config/spu/spu.c
@@ -5878,7 +5878,7 @@ spu_check_builtin_parm (struct spu_builtin_description *d, rtx op, int p)
 }
 
 
-static void
+static int
 expand_builtin_args (struct spu_builtin_description *d, tree exp,
 		     rtx target, rtx ops[])
 {
@@ -5890,13 +5890,18 @@ expand_builtin_args (struct spu_builtin_description *d, tree exp,
   if (d->parm[0] != SPU_BTI_VOID)
     ops[i++] = target;
 
-  for (a = 0; i < insn_data[icode].n_operands; i++, a++)
+  for (a = 0; d->parm[a+1] != SPU_BTI_END_OF_PARAMS; i++, a++)
     {
       tree arg = CALL_EXPR_ARG (exp, a);
       if (arg == 0)
 	abort ();
       ops[i] = expand_expr (arg, NULL_RTX, VOIDmode, 0);
     }
+
+  /* The insn pattern may have additional operands (SCRATCH).
+     Return the number of actual non-SCRATCH operands.  */
+  gcc_assert (i <= insn_data[icode].n_operands);
+  return i;
 }
 
 static rtx
@@ -5908,10 +5913,11 @@ spu_expand_builtin_1 (struct spu_builtin_description *d,
   enum insn_code icode = d->icode;
   enum machine_mode mode, tmode;
   int i, p;
+  int n_operands;
   tree return_type;
 
   /* Set up ops[] with values from arglist. */
-  expand_builtin_args (d, exp, target, ops);
+  n_operands = expand_builtin_args (d, exp, target, ops);
 
   /* Handle the target operand which must be operand 0. */
   i = 0;
@@ -5969,7 +5975,7 @@ spu_expand_builtin_1 (struct spu_builtin_description *d,
     return 0;
 
   /* Handle the rest of the operands. */
-  for (p = 1; i < insn_data[icode].n_operands; i++, p++)
+  for (p = 1; i < n_operands; i++, p++)
     {
       if (insn_data[d->icode].operand[i].mode != VOIDmode)
 	mode = insn_data[d->icode].operand[i].mode;
@@ -6009,7 +6015,7 @@ spu_expand_builtin_1 (struct spu_builtin_description *d,
 	ops[i] = spu_force_reg (mode, ops[i]);
     }
 
-  switch (insn_data[icode].n_operands)
+  switch (n_operands)
     {
     case 0:
       pat = GEN_FCN (icode) (0);
diff --git a/gcc/config/spu/spu.md b/gcc/config/spu/spu.md
index e7b48f56941..cb951558e7d 100644
--- a/gcc/config/spu/spu.md
+++ b/gcc/config/spu/spu.md
@@ -2401,6 +2401,27 @@
     emit_insn (gen_subsi3(operands[5], GEN_INT(7), operands[2]));
   })
 
+(define_insn_and_split "shrqbybi_<mode>"
+  [(set (match_operand:DTI 0 "spu_reg_operand" "=r,r")
+	(lshiftrt:DTI (match_operand:DTI 1 "spu_reg_operand" "r,r")
+		      (and:SI (match_operand:SI 2 "spu_nonmem_operand" "r,I")
+			      (const_int -8))))
+   (clobber (match_scratch:SI 3 "=&r,X"))]
+  ""
+  "#"
+  "reload_completed"
+  [(set (match_dup:DTI 0)
+	(lshiftrt:DTI (match_dup:DTI 1)
+		      (and:SI (neg:SI (and:SI (match_dup:SI 3) (const_int -8)))
+			      (const_int -8))))]
+  {
+    if (GET_CODE (operands[2]) == CONST_INT)
+      operands[3] = GEN_INT (7 - INTVAL (operands[2]));
+    else
+      emit_insn (gen_subsi3 (operands[3], GEN_INT (7), operands[2]));
+  }
+  [(set_attr "type" "shuf")])
+
 (define_insn "rotqmbybi_<mode>"
   [(set (match_operand:DTI 0 "spu_reg_operand" "=r,r")
 	(lshiftrt:DTI (match_operand:DTI 1 "spu_reg_operand" "r,r")
@@ -2413,6 +2434,26 @@
    rotqmbyi\t%0,%1,-%H2"
   [(set_attr "type" "shuf")])
 
+(define_insn_and_split "shrqbi_<mode>"
+  [(set (match_operand:DTI 0 "spu_reg_operand" "=r,r")
+	(lshiftrt:DTI (match_operand:DTI 1 "spu_reg_operand" "r,r")
+		      (and:SI (match_operand:SI 2 "spu_nonmem_operand" "r,I")
+			      (const_int 7))))
+   (clobber (match_scratch:SI 3 "=&r,X"))]
+  ""
+  "#"
+  "reload_completed"
+  [(set (match_dup:DTI 0)
+	(lshiftrt:DTI (match_dup:DTI 1)
+		      (and:SI (neg:SI (match_dup:SI 3)) (const_int 7))))]
+  {
+    if (GET_CODE (operands[2]) == CONST_INT)
+      operands[3] = GEN_INT (-INTVAL (operands[2]));
+    else
+      emit_insn (gen_subsi3 (operands[3], GEN_INT (0), operands[2]));
+  }
+  [(set_attr "type" "shuf")])
+
 (define_insn "rotqmbi_<mode>"
   [(set (match_operand:DTI 0 "spu_reg_operand" "=r,r")
 	(lshiftrt:DTI (match_operand:DTI 1 "spu_reg_operand" "r,r")
@@ -2424,6 +2465,26 @@
    rotqmbii\t%0,%1,-%E2"
   [(set_attr "type" "shuf")])
 
+(define_insn_and_split "shrqby_<mode>"
+  [(set (match_operand:DTI 0 "spu_reg_operand" "=r,r")
+	(lshiftrt:DTI (match_operand:DTI 1 "spu_reg_operand" "r,r")
+		      (mult:SI (match_operand:SI 2 "spu_nonmem_operand" "r,I")
+			       (const_int 8))))
+   (clobber (match_scratch:SI 3 "=&r,X"))]
+  ""
+  "#"
+  "reload_completed"
+  [(set (match_dup:DTI 0)
+	(lshiftrt:DTI (match_dup:DTI 1)
+		      (mult:SI (neg:SI (match_dup:SI 3)) (const_int 8))))]
+  {
+    if (GET_CODE (operands[2]) == CONST_INT)
+      operands[3] = GEN_INT (-INTVAL (operands[2]));
+    else
+      emit_insn (gen_subsi3 (operands[3], GEN_INT (0), operands[2]));
+  }
+  [(set_attr "type" "shuf")])
+
 (define_insn "rotqmby_<mode>"
   [(set (match_operand:DTI 0 "spu_reg_operand" "=r,r")
 	(lshiftrt:DTI (match_operand:DTI 1 "spu_reg_operand" "r,r")
diff --git a/gcc/config/spu/spu_internals.h b/gcc/config/spu/spu_internals.h
index fb42c870983..af66ecd3e32 100644
--- a/gcc/config/spu/spu_internals.h
+++ b/gcc/config/spu/spu_internals.h
@@ -337,6 +337,11 @@
 #define spu_slqw(ra,rb)           __builtin_spu_slqw(ra,rb) 
 #define spu_slqwbyte(ra,rb)       __builtin_spu_slqwbyte(ra,rb) 
 #define spu_slqwbytebc(ra,rb)     __builtin_spu_slqwbytebc(ra,rb) 
+#define spu_sr(ra,rb)             __builtin_spu_sr(ra,rb) 
+#define spu_sra(ra,rb)            __builtin_spu_sra(ra,rb) 
+#define spu_srqw(ra,rb)           __builtin_spu_srqw(ra,rb) 
+#define spu_srqwbyte(ra,rb)       __builtin_spu_srqwbyte(ra,rb) 
+#define spu_srqwbytebc(ra,rb)     __builtin_spu_srqwbytebc(ra,rb) 
 #define spu_extract(ra,pos)       __builtin_spu_extract(ra,pos) 
 #define spu_insert(scalar,ra,pos) __builtin_spu_insert(scalar,ra,pos) 
 #define spu_promote(scalar,pos)   __builtin_spu_promote(scalar,pos)