* [PATCH] [i386][avx512]Add combine splitter to transform vpternlogd/vpcmpeqd/vpxor/vblendvps to vblendvps for ~op0
@ 2021-12-15 1:25 Haochen Jiang
2021-12-17 5:45 ` Hongtao Liu
0 siblings, 1 reply; 2+ messages in thread
From: Haochen Jiang @ 2021-12-15 1:25 UTC (permalink / raw)
To: gcc-patches; +Cc: hongtao.liu
Hi all,
This patch fix the regression previously reported on the combine splitter under '-m32 -march=cascadelake' options.
Regtested on x86_64-pc-linux-gnu.
BRs,
Haochen
gcc/ChangeLog:
PR target/100738
* config/i386/sse.md (*avx_cmp<mode>3_lt, *avx_cmp<mode>3_ltint):
Remove MEM_P restriction and add force_reg for operands[2].
(*avx_cmp<mode>3_ltint_not): Add new define_insn_and_split.
gcc/testsuite/ChangeLog:
PR target/100738
* g++.target/i386/avx512vl-pr100738-1.C: New test.
---
gcc/config/i386/sse.md | 44 +++++++++++++++++--
.../g++.target/i386/avx512vl-pr100738-1.C | 8 ++++
2 files changed, 48 insertions(+), 4 deletions(-)
create mode 100755 gcc/testsuite/g++.target/i386/avx512vl-pr100738-1.C
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 5421fb51684..8ec9fb075d0 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -3528,8 +3528,7 @@
UNSPEC_PCMP)))]
"TARGET_AVX512VL && ix86_pre_reload_split ()
/* LT or GE 0 */
- && ((INTVAL (operands[5]) == 1 && !MEM_P (operands[2]))
- || (INTVAL (operands[5]) == 5 && !MEM_P (operands[1])))"
+ && ((INTVAL (operands[5]) == 1) || (INTVAL (operands[5]) == 5))"
"#"
"&& 1"
[(set (match_dup 0)
@@ -3543,6 +3542,7 @@
{
if (INTVAL (operands[5]) == 5)
std::swap (operands[1], operands[2]);
+ operands[2] = force_reg (<MODE>mode, operands[2]);
})
(define_insn_and_split "*avx_cmp<mode>3_ltint"
@@ -3557,8 +3557,7 @@
UNSPEC_PCMP)))]
"TARGET_AVX512VL && ix86_pre_reload_split ()
/* LT or GE 0 */
- && ((INTVAL (operands[5]) == 1 && !MEM_P (operands[2]))
- || (INTVAL (operands[5]) == 5 && !MEM_P (operands[1])))"
+ && ((INTVAL (operands[5]) == 1) || (INTVAL (operands[5]) == 5))"
"#"
"&& 1"
[(set (match_dup 0)
@@ -3575,7 +3574,44 @@
std::swap (operands[1], operands[2]);
operands[0] = gen_lowpart (<ssebytemode>mode, operands[0]);
operands[1] = gen_lowpart (<ssebytemode>mode, operands[1]);
+ operands[2] = force_reg (<ssebytemode>mode,
+ gen_lowpart (<ssebytemode>mode, operands[2]));
+})
+
+(define_insn_and_split "*avx_cmp<mode>3_ltint_not"
+ [(set (match_operand:VI48_AVX 0 "register_operand")
+ (vec_merge:VI48_AVX
+ (match_operand:VI48_AVX 1 "vector_operand")
+ (match_operand:VI48_AVX 2 "vector_operand")
+ (unspec:<avx512fmaskmode>
+ [(subreg:VI48_AVX
+ (not:<ssebytemode>
+ (match_operand:<ssebytemode> 3 "vector_operand")) 0)
+ (match_operand:VI48_AVX 4 "const0_operand")
+ (match_operand:SI 5 "const_0_to_7_operand")]
+ UNSPEC_PCMP)))]
+ "TARGET_AVX512VL && ix86_pre_reload_split ()
+ /* not LT or GE 0 */
+ && ((INTVAL (operands[5]) == 1) || (INTVAL (operands[5]) == 5))"
+ "#"
+ "&& 1"
+ [(set (match_dup 0)
+ (unspec:<ssebytemode>
+ [(match_dup 1)
+ (match_dup 2)
+ (subreg:<ssebytemode>
+ (lt:VI48_AVX
+ (match_dup 3)
+ (match_dup 4)) 0)]
+ UNSPEC_BLENDV))]
+{
+ if (INTVAL (operands[5]) == 5)
+ std::swap (operands[1], operands[2]);
+ operands[0] = gen_lowpart (<ssebytemode>mode, operands[0]);
+ operands[1] = force_reg (<ssebytemode>mode,
+ gen_lowpart (<ssebytemode>mode, operands[1]));
operands[2] = gen_lowpart (<ssebytemode>mode, operands[2]);
+ operands[3] = lowpart_subreg (<MODE>mode, operands[3], <ssebytemode>mode);
})
(define_insn "avx_vmcmp<mode>3"
diff --git a/gcc/testsuite/g++.target/i386/avx512vl-pr100738-1.C b/gcc/testsuite/g++.target/i386/avx512vl-pr100738-1.C
new file mode 100755
index 00000000000..ac4d62b94d1
--- /dev/null
+++ b/gcc/testsuite/g++.target/i386/avx512vl-pr100738-1.C
@@ -0,0 +1,8 @@
+/* { dg-do compile } */
+/* { dg-options "-Ofast -march=cascadelake" } */
+/* { dg-final {scan-assembler-times "vblendvps\[ \\t\]" 2 } } */
+/* { dg-final {scan-assembler-not "vpcmpeqd\[ \\t\]" } } */
+/* { dg-final {scan-assembler-not "vpxor\[ \\t\]" } } */
+/* { dg-final {scan-assembler-not "vpternlogd\[ \\t\]" } } */
+
+#include "pr100738-1.C"
--
2.18.1
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] [i386][avx512]Add combine splitter to transform vpternlogd/vpcmpeqd/vpxor/vblendvps to vblendvps for ~op0
2021-12-15 1:25 [PATCH] [i386][avx512]Add combine splitter to transform vpternlogd/vpcmpeqd/vpxor/vblendvps to vblendvps for ~op0 Haochen Jiang
@ 2021-12-17 5:45 ` Hongtao Liu
0 siblings, 0 replies; 2+ messages in thread
From: Hongtao Liu @ 2021-12-17 5:45 UTC (permalink / raw)
To: Haochen Jiang; +Cc: GCC Patches, Liu, Hongtao
On Wed, Dec 15, 2021 at 9:26 AM Haochen Jiang via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> Hi all,
>
> This patch fix the regression previously reported on the combine splitter under '-m32 -march=cascadelake' options.
>
> Regtested on x86_64-pc-linux-gnu.
Ok.
>
> BRs,
> Haochen
>
> gcc/ChangeLog:
>
> PR target/100738
> * config/i386/sse.md (*avx_cmp<mode>3_lt, *avx_cmp<mode>3_ltint):
> Remove MEM_P restriction and add force_reg for operands[2].
> (*avx_cmp<mode>3_ltint_not): Add new define_insn_and_split.
>
> gcc/testsuite/ChangeLog:
>
> PR target/100738
> * g++.target/i386/avx512vl-pr100738-1.C: New test.
> ---
> gcc/config/i386/sse.md | 44 +++++++++++++++++--
> .../g++.target/i386/avx512vl-pr100738-1.C | 8 ++++
> 2 files changed, 48 insertions(+), 4 deletions(-)
> create mode 100755 gcc/testsuite/g++.target/i386/avx512vl-pr100738-1.C
>
> diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> index 5421fb51684..8ec9fb075d0 100644
> --- a/gcc/config/i386/sse.md
> +++ b/gcc/config/i386/sse.md
> @@ -3528,8 +3528,7 @@
> UNSPEC_PCMP)))]
> "TARGET_AVX512VL && ix86_pre_reload_split ()
> /* LT or GE 0 */
> - && ((INTVAL (operands[5]) == 1 && !MEM_P (operands[2]))
> - || (INTVAL (operands[5]) == 5 && !MEM_P (operands[1])))"
> + && ((INTVAL (operands[5]) == 1) || (INTVAL (operands[5]) == 5))"
> "#"
> "&& 1"
> [(set (match_dup 0)
> @@ -3543,6 +3542,7 @@
> {
> if (INTVAL (operands[5]) == 5)
> std::swap (operands[1], operands[2]);
> + operands[2] = force_reg (<MODE>mode, operands[2]);
> })
>
> (define_insn_and_split "*avx_cmp<mode>3_ltint"
> @@ -3557,8 +3557,7 @@
> UNSPEC_PCMP)))]
> "TARGET_AVX512VL && ix86_pre_reload_split ()
> /* LT or GE 0 */
> - && ((INTVAL (operands[5]) == 1 && !MEM_P (operands[2]))
> - || (INTVAL (operands[5]) == 5 && !MEM_P (operands[1])))"
> + && ((INTVAL (operands[5]) == 1) || (INTVAL (operands[5]) == 5))"
> "#"
> "&& 1"
> [(set (match_dup 0)
> @@ -3575,7 +3574,44 @@
> std::swap (operands[1], operands[2]);
> operands[0] = gen_lowpart (<ssebytemode>mode, operands[0]);
> operands[1] = gen_lowpart (<ssebytemode>mode, operands[1]);
> + operands[2] = force_reg (<ssebytemode>mode,
> + gen_lowpart (<ssebytemode>mode, operands[2]));
> +})
> +
> +(define_insn_and_split "*avx_cmp<mode>3_ltint_not"
> + [(set (match_operand:VI48_AVX 0 "register_operand")
> + (vec_merge:VI48_AVX
> + (match_operand:VI48_AVX 1 "vector_operand")
> + (match_operand:VI48_AVX 2 "vector_operand")
> + (unspec:<avx512fmaskmode>
> + [(subreg:VI48_AVX
> + (not:<ssebytemode>
> + (match_operand:<ssebytemode> 3 "vector_operand")) 0)
> + (match_operand:VI48_AVX 4 "const0_operand")
> + (match_operand:SI 5 "const_0_to_7_operand")]
> + UNSPEC_PCMP)))]
> + "TARGET_AVX512VL && ix86_pre_reload_split ()
> + /* not LT or GE 0 */
> + && ((INTVAL (operands[5]) == 1) || (INTVAL (operands[5]) == 5))"
> + "#"
> + "&& 1"
> + [(set (match_dup 0)
> + (unspec:<ssebytemode>
> + [(match_dup 1)
> + (match_dup 2)
> + (subreg:<ssebytemode>
> + (lt:VI48_AVX
> + (match_dup 3)
> + (match_dup 4)) 0)]
> + UNSPEC_BLENDV))]
> +{
> + if (INTVAL (operands[5]) == 5)
> + std::swap (operands[1], operands[2]);
> + operands[0] = gen_lowpart (<ssebytemode>mode, operands[0]);
> + operands[1] = force_reg (<ssebytemode>mode,
> + gen_lowpart (<ssebytemode>mode, operands[1]));
> operands[2] = gen_lowpart (<ssebytemode>mode, operands[2]);
> + operands[3] = lowpart_subreg (<MODE>mode, operands[3], <ssebytemode>mode);
> })
>
> (define_insn "avx_vmcmp<mode>3"
> diff --git a/gcc/testsuite/g++.target/i386/avx512vl-pr100738-1.C b/gcc/testsuite/g++.target/i386/avx512vl-pr100738-1.C
> new file mode 100755
> index 00000000000..ac4d62b94d1
> --- /dev/null
> +++ b/gcc/testsuite/g++.target/i386/avx512vl-pr100738-1.C
> @@ -0,0 +1,8 @@
> +/* { dg-do compile } */
> +/* { dg-options "-Ofast -march=cascadelake" } */
> +/* { dg-final {scan-assembler-times "vblendvps\[ \\t\]" 2 } } */
> +/* { dg-final {scan-assembler-not "vpcmpeqd\[ \\t\]" } } */
> +/* { dg-final {scan-assembler-not "vpxor\[ \\t\]" } } */
> +/* { dg-final {scan-assembler-not "vpternlogd\[ \\t\]" } } */
> +
> +#include "pr100738-1.C"
> --
> 2.18.1
>
--
BR,
Hongtao
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2021-12-17 5:45 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-12-15 1:25 [PATCH] [i386][avx512]Add combine splitter to transform vpternlogd/vpcmpeqd/vpxor/vblendvps to vblendvps for ~op0 Haochen Jiang
2021-12-17 5:45 ` Hongtao Liu
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).