Hi,
V3. :-)
[PATCH, rs6000] (v2) Gimple folding for vec_madd()
Add support for gimple folding of the vec_madd() (vector multiply-add)
intrinsics.
Renamed the define_insn of altivec_vmladduhm to fmav8hi4, Refreshed the
caller of gen_altivec_vmladduhm to call gen_fmav8hi, and updated the
rs6000-builtin.def entry for VMLADDUHM to point to the new name.
With this refresh I am no longer adding a define_expand.
Plus a few cosmetic tweaks per feedback.
Testcase coverage is provided by the existing tests as
gcc.target/powerpc/fold-vec-madd-*.c
Sniff-tests passed. Regtests will be kicked off shortly. OK for trunk?
Thanks,
-Will
[gcc]
2017-10-27 Will Schmidt <will_schm...@vnet.ibm.com>
* config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Add support for
gimple folding of vec_madd() intrinsics.
* config/rs6000/altivec.md: Rename altivec_vmladduhm to fmav8hi4
* config/rs6000/rs6000-builtin.def: Rename vmladduhm to fmav8hi4
diff --git a/gcc/config/rs6000/altivec.md b/gcc/config/rs6000/altivec.md
index 6ea529b..b2f173d 100644
--- a/gcc/config/rs6000/altivec.md
+++ b/gcc/config/rs6000/altivec.md
@@ -793,15 +793,16 @@
"TARGET_ALTIVEC"
{
rtx zero = gen_reg_rtx (V8HImode);
emit_insn (gen_altivec_vspltish (zero, const0_rtx));
- emit_insn (gen_altivec_vmladduhm(operands[0], operands[1], operands[2],
zero));
+ emit_insn (gen_fmav8hi4 (operands[0], operands[1], operands[2], zero));
DONE;
})
+
;; Fused multiply subtract
(define_insn "*altivec_vnmsubfp"
[(set (match_operand:V4SF 0 "register_operand" "=v")
(neg:V4SF
(fma:V4SF (match_operand:V4SF 1 "register_operand" "v")
@@ -934,11 +935,11 @@
(set (reg:SI VSCR_REGNO) (unspec:SI [(const_int 0)] UNSPEC_SET_VSCR))]
"TARGET_ALTIVEC"
"vmhraddshs %0,%1,%2,%3"
[(set_attr "type" "veccomplex")])
-(define_insn "altivec_vmladduhm"
+(define_insn "fmav8hi4"
[(set (match_operand:V8HI 0 "register_operand" "=v")
(plus:V8HI (mult:V8HI (match_operand:V8HI 1 "register_operand" "v")
(match_operand:V8HI 2 "register_operand" "v"))
(match_operand:V8HI 3 "register_operand" "v")))]
"TARGET_ALTIVEC"
diff --git a/gcc/config/rs6000/rs6000-builtin.def
b/gcc/config/rs6000/rs6000-builtin.def
index ac9ddae..7834bef 100644
--- a/gcc/config/rs6000/rs6000-builtin.def
+++ b/gcc/config/rs6000/rs6000-builtin.def
@@ -959,11 +959,11 @@ BU_SPECIAL_X (RS6000_BUILTIN_NONE, NULL, 0,
RS6000_BTC_MISC)
/* 3 argument Altivec builtins. */
BU_ALTIVEC_3 (VMADDFP, "vmaddfp", FP, fmav4sf4)
BU_ALTIVEC_3 (VMHADDSHS, "vmhaddshs", SAT,
altivec_vmhaddshs)
BU_ALTIVEC_3 (VMHRADDSHS, "vmhraddshs", SAT,
altivec_vmhraddshs)
-BU_ALTIVEC_3 (VMLADDUHM, "vmladduhm", CONST,
altivec_vmladduhm)
+BU_ALTIVEC_3 (VMLADDUHM, "vmladduhm", CONST, fmav8hi4)
BU_ALTIVEC_3 (VMSUMUBM, "vmsumubm", CONST,
altivec_vmsumubm)
BU_ALTIVEC_3 (VMSUMMBM, "vmsummbm", CONST,
altivec_vmsummbm)
BU_ALTIVEC_3 (VMSUMUHM, "vmsumuhm", CONST,
altivec_vmsumuhm)
BU_ALTIVEC_3 (VMSUMSHM, "vmsumshm", CONST,
altivec_vmsumshm)
BU_ALTIVEC_3 (VMSUMUHS, "vmsumuhs", SAT,
altivec_vmsumuhs)
diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 4837e14..aef34b7 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -16606,10 +16606,26 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi)
build_int_cst (arg2_type, 0)), arg0);
gimple_set_location (g, loc);
gsi_replace (gsi, g, true);
return true;
}
+
+ /* Vector Fused multiply-add (fma). */
+ case ALTIVEC_BUILTIN_VMADDFP:
+ case VSX_BUILTIN_XVMADDDP:
+ case ALTIVEC_BUILTIN_VMLADDUHM:
+ {
+ arg0 = gimple_call_arg (stmt, 0);
+ arg1 = gimple_call_arg (stmt, 1);
+ tree arg2 = gimple_call_arg (stmt, 2);
+ lhs = gimple_call_lhs (stmt);
+ gimple *g = gimple_build_assign (lhs, FMA_EXPR , arg0, arg1, arg2);
+ gimple_set_location (g, gimple_location (stmt));
+ gsi_replace (gsi, g, true);
+ return true;
+ }
+
default:
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "gimple builtin intrinsic not matched:%d %s %s\n",
fn_code, fn_name1, fn_name2);
break;
