public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r12-6488] i386: Introduce V2QImode vector cmove for -msse4.1 [PR103861]
@ 2022-01-11 18:24 Uros Bizjak
0 siblings, 0 replies; only message in thread
From: Uros Bizjak @ 2022-01-11 18:24 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:820ac79e8448ad6c631e1387ba51a93dcf2b4e89
commit r12-6488-g820ac79e8448ad6c631e1387ba51a93dcf2b4e89
Author: Uros Bizjak <ubizjak@gmail.com>
Date: Tue Jan 11 19:23:15 2022 +0100
i386: Introduce V2QImode vector cmove for -msse4.1 [PR103861]
This patch also moves V2HI and V4QImode vector conditional moves
to SSE4.1 targets. Vector cmoves are implemented with SSE logic functions
without -msse4.1, and they are hardly worthwile for narrow vector modes.
More important, we would like to keep vector logic functions for GPR
registers, and the current RTX description of 32-bit vector modes logic
insns does not include the necessary CC reg clobber. Solve these issues by
restricting vector cmove insns for these modes to -msse4.1, where logic
instructions are avoided, and pblend insn is used instead.
A follow-up patch will add clobbers and necessary splits to 32-bit
vector mode logic insns, and in a future patch, ix86_sse_movcc will be
improved to use expand_simple_{unop,binop} to emit logic insns, allowing
us to re-enable 16-bit and 32-bit narrow vector cmoves for -msse2.
2022-01-11 Uroš Bizjak <ubizjak@gmail.com>
gcc/ChangeLog:
PR target/103861
* config/i386/mmx.md (vcond<mode><mode>):
Use VI_16_32 mode iterator. Enable for TARGET_SSE4_1.
(vcondu<mode><mode>): Ditto.
(vcond_mask_<mode><mode>): Ditto.
(mmx_pblendvb_v8qi): Rename from mmx_pblendvb64.
(mmx_pblendvb_<mode>): Rename from mmx_pblendvb32.
Use VI_16_32 mode iterator.
* config/i386/i386-expand.c (ix86_expand_sse_movcc):
Update for rename. Handle V2QImode.
(expand_vec_perm_blend): Update for rename.
gcc/testsuite/ChangeLog:
PR target/103861
* g++.target/i386/pr100637-1b.C (dg-options):
Use -msse4 instead of -msse2.
* g++.target/i386/pr100637-1w.C (dg-options): Ditto.
* g++.target/i386/pr103861-1.C: New test.
* gcc.target/i386/pr100637-4b.c (dg-options):
Use -msse4 instead of -msse2.
* gcc.target/i386/pr103861-4.c: New test.
Diff:
---
gcc/config/i386/i386-expand.c | 16 ++++++---
gcc/config/i386/mmx.md | 54 ++++++++++++++---------------
gcc/testsuite/g++.target/i386/pr100637-1b.C | 2 +-
gcc/testsuite/g++.target/i386/pr100637-1w.C | 2 +-
gcc/testsuite/g++.target/i386/pr103861-1.C | 17 +++++++++
gcc/testsuite/gcc.target/i386/pr100637-4b.c | 2 +-
gcc/testsuite/gcc.target/i386/pr103861-4.c | 19 ++++++++++
7 files changed, 78 insertions(+), 34 deletions(-)
diff --git a/gcc/config/i386/i386-expand.c b/gcc/config/i386/i386-expand.c
index add748bcf40..8b1266fb9f1 100644
--- a/gcc/config/i386/i386-expand.c
+++ b/gcc/config/i386/i386-expand.c
@@ -3899,7 +3899,7 @@ ix86_expand_sse_movcc (rtx dest, rtx cmp, rtx op_true, rtx op_false)
{
op_true = force_reg (mode, op_true);
- gen = gen_mmx_pblendvb64;
+ gen = gen_mmx_pblendvb_v8qi;
if (mode != V8QImode)
d = gen_reg_rtx (V8QImode);
op_false = gen_lowpart (V8QImode, op_false);
@@ -3913,7 +3913,7 @@ ix86_expand_sse_movcc (rtx dest, rtx cmp, rtx op_true, rtx op_false)
{
op_true = force_reg (mode, op_true);
- gen = gen_mmx_pblendvb32;
+ gen = gen_mmx_pblendvb_v4qi;
if (mode != V4QImode)
d = gen_reg_rtx (V4QImode);
op_false = gen_lowpart (V4QImode, op_false);
@@ -3921,6 +3921,14 @@ ix86_expand_sse_movcc (rtx dest, rtx cmp, rtx op_true, rtx op_false)
cmp = gen_lowpart (V4QImode, cmp);
}
break;
+ case E_V2QImode:
+ if (TARGET_SSE4_1)
+ {
+ op_true = force_reg (mode, op_true);
+
+ gen = gen_mmx_pblendvb_v2qi;
+ }
+ break;
case E_V16QImode:
case E_V8HImode:
case E_V8HFmode:
@@ -18462,9 +18470,9 @@ expand_vec_perm_blend (struct expand_vec_perm_d *d)
vperm = force_reg (vmode, vperm);
if (GET_MODE_SIZE (vmode) == 4)
- emit_insn (gen_mmx_pblendvb32 (target, op0, op1, vperm));
+ emit_insn (gen_mmx_pblendvb_v4qi (target, op0, op1, vperm));
else if (GET_MODE_SIZE (vmode) == 8)
- emit_insn (gen_mmx_pblendvb64 (target, op0, op1, vperm));
+ emit_insn (gen_mmx_pblendvb_v8qi (target, op0, op1, vperm));
else if (GET_MODE_SIZE (vmode) == 16)
emit_insn (gen_sse4_1_pblendvb (target, op0, op1, vperm));
else
diff --git a/gcc/config/i386/mmx.md b/gcc/config/i386/mmx.md
index 91d642187be..fa67278e003 100644
--- a/gcc/config/i386/mmx.md
+++ b/gcc/config/i386/mmx.md
@@ -2580,14 +2580,14 @@
})
(define_expand "vcond<mode><mode>"
- [(set (match_operand:VI_32 0 "register_operand")
- (if_then_else:VI_32
+ [(set (match_operand:VI_16_32 0 "register_operand")
+ (if_then_else:VI_16_32
(match_operator 3 ""
- [(match_operand:VI_32 4 "register_operand")
- (match_operand:VI_32 5 "register_operand")])
- (match_operand:VI_32 1)
- (match_operand:VI_32 2)))]
- "TARGET_SSE2"
+ [(match_operand:VI_16_32 4 "register_operand")
+ (match_operand:VI_16_32 5 "register_operand")])
+ (match_operand:VI_16_32 1)
+ (match_operand:VI_16_32 2)))]
+ "TARGET_SSE4_1"
{
bool ok = ix86_expand_int_vcond (operands);
gcc_assert (ok);
@@ -2612,14 +2612,14 @@
})
(define_expand "vcondu<mode><mode>"
- [(set (match_operand:VI_32 0 "register_operand")
- (if_then_else:VI_32
+ [(set (match_operand:VI_16_32 0 "register_operand")
+ (if_then_else:VI_16_32
(match_operator 3 ""
- [(match_operand:VI_32 4 "register_operand")
- (match_operand:VI_32 5 "register_operand")])
- (match_operand:VI_32 1)
- (match_operand:VI_32 2)))]
- "TARGET_SSE2"
+ [(match_operand:VI_16_32 4 "register_operand")
+ (match_operand:VI_16_32 5 "register_operand")])
+ (match_operand:VI_16_32 1)
+ (match_operand:VI_16_32 2)))]
+ "TARGET_SSE4_1"
{
bool ok = ix86_expand_int_vcond (operands);
gcc_assert (ok);
@@ -2640,19 +2640,19 @@
})
(define_expand "vcond_mask_<mode><mode>"
- [(set (match_operand:VI_32 0 "register_operand")
- (vec_merge:VI_32
- (match_operand:VI_32 1 "register_operand")
- (match_operand:VI_32 2 "register_operand")
- (match_operand:VI_32 3 "register_operand")))]
- "TARGET_SSE2"
+ [(set (match_operand:VI_16_32 0 "register_operand")
+ (vec_merge:VI_16_32
+ (match_operand:VI_16_32 1 "register_operand")
+ (match_operand:VI_16_32 2 "register_operand")
+ (match_operand:VI_16_32 3 "register_operand")))]
+ "TARGET_SSE4_1"
{
ix86_expand_sse_movcc (operands[0], operands[3],
operands[1], operands[2]);
DONE;
})
-(define_insn "mmx_pblendvb64"
+(define_insn "mmx_pblendvb_v8qi"
[(set (match_operand:V8QI 0 "register_operand" "=Yr,*x,x")
(unspec:V8QI
[(match_operand:V8QI 1 "register_operand" "0,0,x")
@@ -2672,12 +2672,12 @@
(set_attr "btver2_decode" "vector")
(set_attr "mode" "TI")])
-(define_insn "mmx_pblendvb32"
- [(set (match_operand:V4QI 0 "register_operand" "=Yr,*x,x")
- (unspec:V4QI
- [(match_operand:V4QI 1 "register_operand" "0,0,x")
- (match_operand:V4QI 2 "register_operand" "Yr,*x,x")
- (match_operand:V4QI 3 "register_operand" "Yz,Yz,x")]
+(define_insn "mmx_pblendvb_<mode>"
+ [(set (match_operand:VI_16_32 0 "register_operand" "=Yr,*x,x")
+ (unspec:VI_16_32
+ [(match_operand:VI_16_32 1 "register_operand" "0,0,x")
+ (match_operand:VI_16_32 2 "register_operand" "Yr,*x,x")
+ (match_operand:VI_16_32 3 "register_operand" "Yz,Yz,x")]
UNSPEC_BLENDV))]
"TARGET_SSE4_1"
"@
diff --git a/gcc/testsuite/g++.target/i386/pr100637-1b.C b/gcc/testsuite/g++.target/i386/pr100637-1b.C
index 35b5df7c9dd..d602ac08b4d 100644
--- a/gcc/testsuite/g++.target/i386/pr100637-1b.C
+++ b/gcc/testsuite/g++.target/i386/pr100637-1b.C
@@ -1,6 +1,6 @@
/* PR target/100637 */
/* { dg-do compile } */
-/* { dg-options "-O2 -msse2" } */
+/* { dg-options "-O2 -msse4" } */
typedef unsigned char __attribute__((__vector_size__ (4))) __v4qu;
typedef char __attribute__((__vector_size__ (4))) __v4qi;
diff --git a/gcc/testsuite/g++.target/i386/pr100637-1w.C b/gcc/testsuite/g++.target/i386/pr100637-1w.C
index a3ed06fddee..c6056454897 100644
--- a/gcc/testsuite/g++.target/i386/pr100637-1w.C
+++ b/gcc/testsuite/g++.target/i386/pr100637-1w.C
@@ -1,6 +1,6 @@
/* PR target/100637 */
/* { dg-do compile } */
-/* { dg-options "-O2 -msse2" } */
+/* { dg-options "-O2 -msse4" } */
typedef unsigned short __attribute__((__vector_size__ (4))) __v2hu;
typedef short __attribute__((__vector_size__ (4))) __v2hi;
diff --git a/gcc/testsuite/g++.target/i386/pr103861-1.C b/gcc/testsuite/g++.target/i386/pr103861-1.C
new file mode 100644
index 00000000000..940c939e04f
--- /dev/null
+++ b/gcc/testsuite/g++.target/i386/pr103861-1.C
@@ -0,0 +1,17 @@
+/* PR target/103861 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -msse4" } */
+
+typedef unsigned char __attribute__((__vector_size__ (2))) __v2qu;
+typedef char __attribute__((__vector_size__ (2))) __v2qi;
+
+__v2qu au, bu;
+__v2qi as, bs;
+
+__v2qu uu (__v2qu a, __v2qu b) { return (a > b) ? au : bu; }
+__v2qu us (__v2qi a, __v2qi b) { return (a > b) ? au : bu; }
+__v2qi su (__v2qu a, __v2qu b) { return (a > b) ? as : bs; }
+__v2qi ss (__v2qi a, __v2qi b) { return (a > b) ? as : bs; }
+
+/* { dg-final { scan-assembler-times "pcmpeqb" 2 } } */
+/* { dg-final { scan-assembler-times "pcmpgtb" 2 } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr100637-4b.c b/gcc/testsuite/gcc.target/i386/pr100637-4b.c
index 198e3dd3352..add4506e4c1 100644
--- a/gcc/testsuite/gcc.target/i386/pr100637-4b.c
+++ b/gcc/testsuite/gcc.target/i386/pr100637-4b.c
@@ -1,6 +1,6 @@
/* PR target/100637 */
/* { dg-do compile } */
-/* { dg-options "-O2 -ftree-vectorize -msse2" } */
+/* { dg-options "-O2 -ftree-vectorize -msse4" } */
typedef char T;
diff --git a/gcc/testsuite/gcc.target/i386/pr103861-4.c b/gcc/testsuite/gcc.target/i386/pr103861-4.c
new file mode 100644
index 00000000000..54c1859b027
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr103861-4.c
@@ -0,0 +1,19 @@
+/* PR target/100637 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-vectorize -msse4" } */
+
+typedef char T;
+
+#define M 2
+
+extern T a[M], b[M], s1[M], s2[M], r[M];
+
+void foo (void)
+{
+ int j;
+
+ for (j = 0; j < M; j++)
+ r[j] = (a[j] < b[j]) ? s1[j] : s2[j];
+}
+
+/* { dg-final { scan-assembler "pcmpgtb" { xfail *-*-* } } } */
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2022-01-11 18:24 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-01-11 18:24 [gcc r12-6488] i386: Introduce V2QImode vector cmove for -msse4.1 [PR103861] Uros Bizjak
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).