On Tue, May 28, 2024 at 4:00 PM Hu, Lin1 <lin1...@intel.com> wrote:
>
> Hi all,
>
> This patch aims to acheive EQ/NE comparison between avx512 kmask and -1
> by using kxortest with checking CF.
>
> Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,-m64}. Ok for trunk?
Ok.
>
> BRs,
> Lin
>
> gcc/ChangeLog:
>
> PR target/113609
> * config/i386/sse.md
> (*kortest_cmp<mode>_setcc): New define_insn_and_split.
> (*kortest_cmp<mode>_jcc): Ditto.
>
> gcc/testsuite/ChangeLog:
>
> PR target/113609
> * gcc.target/i386/pr113609-1.c: New test.
> * gcc.target/i386/pr113609-2.c: Ditto.
> ---
> gcc/config/i386/sse.md | 67 +++++++
> gcc/testsuite/gcc.target/i386/pr113609-1.c | 194 +++++++++++++++++++++
> gcc/testsuite/gcc.target/i386/pr113609-2.c | 161 +++++++++++++++++
> 3 files changed, 422 insertions(+)
> create mode 100644 gcc/testsuite/gcc.target/i386/pr113609-1.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr113609-2.c
>
> diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> index b59c988fc31..34fd2e4afac 100644
> --- a/gcc/config/i386/sse.md
> +++ b/gcc/config/i386/sse.md
> @@ -2201,6 +2201,73 @@ (define_expand "kortest<mode>"
> UNSPEC_KORTEST))]
> "TARGET_AVX512F")
>
> +;; Optimize cmp + setcc with mask register by kortest + setcc.
> +(define_insn_and_split "*kortest_cmp<mode>_setcc"
> + [(set (match_operand:QI 0 "nonimmediate_operand" "=qm, qm")
> + (match_operator:QI 1 "bt_comparison_operator"
> + [(match_operand:SWI1248_AVX512BWDQ_64 2 "register_operand" "?k,
> <r>")
> + (const_int -1)]))
> + (clobber (reg:CC FLAGS_REG))]
> + "TARGET_AVX512BW"
> + "#"
> + "&& reload_completed"
> + [(const_int 0)]
> +{
> + if (MASK_REGNO_P (REGNO (operands[2])))
> + {
> + emit_insn (gen_kortest<mode>_ccc (operands[2], operands[2]));
> + operands[4] = gen_rtx_REG (CCCmode, FLAGS_REG);
> + }
> + else
> + {
> + operands[4] = gen_rtx_REG (CCZmode, FLAGS_REG);
> + emit_insn (gen_rtx_SET (operands[4],
> + gen_rtx_COMPARE (CCZmode,
> + operands[2],
> + constm1_rtx)));
> + }
> + ix86_expand_setcc (operands[0],
> + GET_CODE (operands[1]),
> + operands[4],
> + const0_rtx);
> + DONE;
> +})
> +
> +;; Optimize cmp + jcc with mask register by kortest + jcc.
> +(define_insn_and_split "*kortest_cmp<mode>_jcc"
> + [(set (pc)
> + (if_then_else
> + (match_operator 0 "bt_comparison_operator"
> + [(match_operand:SWI1248_AVX512BWDQ_64 1 "register_operand" "?k,
> <r>")
> + (const_int -1)])
> + (label_ref (match_operand 2))
> + (pc)))
> + (clobber (reg:CC FLAGS_REG))]
> + "TARGET_AVX512BW"
> + "#"
> + "&& reload_completed"
> + [(const_int 0)]
> +{
> + if (MASK_REGNO_P (REGNO (operands[1])))
> + {
> + emit_insn (gen_kortest<mode>_ccc (operands[1], operands[1]));
> + operands[4] = gen_rtx_REG (CCCmode, FLAGS_REG);
> + }
> + else
> + {
> + operands[4] = gen_rtx_REG (CCZmode, FLAGS_REG);
> + emit_insn (gen_rtx_SET (operands[4],
> + gen_rtx_COMPARE (CCZmode,
> + operands[1],
> + constm1_rtx)));
> + }
> + ix86_expand_branch (GET_CODE (operands[0]),
> + operands[4],
> + const0_rtx,
> + operands[2]);
> + DONE;
> +})
> +
> (define_insn "kunpckhi"
> [(set (match_operand:HI 0 "register_operand" "=k")
> (ior:HI
> diff --git a/gcc/testsuite/gcc.target/i386/pr113609-1.c
> b/gcc/testsuite/gcc.target/i386/pr113609-1.c
> new file mode 100644
> index 00000000000..f0639b8500a
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr113609-1.c
> @@ -0,0 +1,194 @@
> +/* PR target/113609 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -march=x86-64-v4" } */
> +/* { dg-final { scan-assembler-not "^cmp" } } */
> +/* { dg-final { scan-assembler-not "\[ \\t\]+sete" { target { ! ia32 } } } }
> */
> +/* { dg-final { scan-assembler-not "\[ \\t\]+setne" { target { ! ia32 } } }
> } */
> +/* { dg-final { scan-assembler-not "\[ \\t\]+je" { target { ! ia32 } } } } */
> +/* { dg-final { scan-assembler-not "\[ \\t\]+jne" { target { ! ia32 } } } }
> */
> +/* { dg-final { scan-assembler-times "\[ \\t\]+sete" 1 { target { ia32 } } }
> } */
> +/* { dg-final { scan-assembler-times "\[ \\t\]+setne" 1 { target { ia32 } }
> } } */
> +/* { dg-final { scan-assembler-times "\[ \\t\]+je" 1 { target { ia32 } } } }
> */
> +/* { dg-final { scan-assembler-times "\[ \\t\]+jne" 2 { target { ia32 } } }
> } */
> +/* { dg-final { scan-assembler-times "kortest" 12 { target { ia32 } } } } */
> +/* { dg-final { scan-assembler-times "kortest" 17 { target { ! ia32 } } } }
> */
> +
> +#include <immintrin.h>
> +
> +unsigned int
> +cmp_vector_sete_mask8(__m128i a, __m128i b)
> +{
> + __mmask8 k = _mm_cmpeq_epi16_mask (a, b);
> + if (k == (__mmask8) -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_vector_sete_mask16(__m128i a, __m128i b)
> +{
> + __mmask16 k = _mm_cmpeq_epi8_mask (a, b);
> + if (k == (__mmask16) -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_vector_sete_mask32(__m256i a, __m256i b)
> +{
> + __mmask32 k = _mm256_cmpeq_epi8_mask (a, b);
> + if (k == (__mmask32) -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_vector_sete_mask64(__m512i a, __m512i b)
> +{
> + __mmask64 k = _mm512_cmpeq_epi8_mask (a, b);
> + if (k == (__mmask64) -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_vector_setne_mask8(__m128i a, __m128i b)
> +{
> + __mmask8 k = _mm_cmpeq_epi16_mask (a, b);
> + if (k != (__mmask8) -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_vector_setne_mask16(__m128i a, __m128i b)
> +{
> + __mmask16 k = _mm_cmpeq_epi8_mask (a, b);
> + if (k != (__mmask16) -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_vector_setne_mask32(__m256i a, __m256i b)
> +{
> + __mmask32 k = _mm256_cmpeq_epi8_mask (a, b);
> + if (k != (__mmask32) -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_vector_setne_mask64(__m512i a, __m512i b)
> +{
> + __mmask64 k = _mm512_cmpeq_epi8_mask (a, b);
> + if (k != (__mmask64) -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +__m128i
> +cmp_vector_je_mask8(__m128i a, __m128i b) {
> + __mmask8 k = _mm_cmpeq_epi16_mask (a, b);
> + if (k == (__mmask8) -1) {
> + a[0] = a[0] + 1;
> + }
> + else {
> + a[0] = a[0] - 1;
> + }
> + return a;
> +}
> +
> +__m128i
> +cmp_vector_je_mask16(__m128i a, __m128i b) {
> + __mmask16 k = _mm_cmpeq_epi8_mask (a, b);
> + if (k == (__mmask16) -1) {
> + a[0] = a[0] + 1;
> + }
> + else {
> + a[0] = a[0] - 1;
> + }
> + return a;
> +}
> +
> +__m256i
> +cmp_vector_je_mask32(__m256i a, __m256i b) {
> + __mmask32 k = _mm256_cmpeq_epi8_mask (a, b);
> + if (k == (__mmask32) -1) {
> + a[0] = a[0] + 1;
> + }
> + else {
> + a[0] = a[0] - 1;
> + }
> + return a;
> +}
> +
> +__m512i
> +cmp_vector_je_mask64(__m512i a, __m512i b) {
> + __mmask64 k = _mm512_cmpeq_epi8_mask (a, b);
> + if (k == (__mmask64) -1) {
> + a[0] = a[0] + 1;
> + }
> + else {
> + a[0] = a[0] - 5;
> + }
> + return a;
> +}
> +
> +__m128i
> +cmp_vector_jne_mask8(__m128i a, __m128i b) {
> + __mmask8 k = _mm_cmpeq_epi16_mask (a, b);
> + if (k == (__mmask8) -1) {
> + a[0] = a[0] + 1;
> + }
> + a[0] = a[0] - 4;
> + return a;
> +}
> +
> +__m128i
> +cmp_vector_jne_mask16(__m128i a, __m128i b) {
> + __mmask16 k = _mm_cmpeq_epi8_mask (a, b);
> + if (k == (__mmask16) -1) {
> + a[0] = a[0] + 1;
> + }
> + a[0] = a[0] - 4;
> + return a;
> +}
> +
> +__m256i
> +cmp_vector_jne_mask32(__m256i a, __m256i b) {
> + __mmask32 k = _mm256_cmpeq_epi8_mask (a, b);
> + if (k == (__mmask32) -1) {
> + a[0] = a[0] + 1;
> + }
> + a[0] = a[0] - 4;
> + return a;
> +}
> +
> +__m512i
> +cmp_vector_jne_mask64(__m512i a, __m512i b) {
> + __mmask64 k = _mm512_cmpeq_epi8_mask (a, b);
> + if (k == (__mmask64) -1) {
> + a[0] = a[0] + 1;
> + }
> + a[0] = a[0] - 4;
> + return a;
> +}
> +
> +__m512i
> +mask_cmp_vector_jne_mask64(__m512i a, __m512i b) {
> + __mmask64 k = _mm512_mask_cmpeq_epi8_mask
> ((__mmask64)0xffffffefffffffff, a, b);
> + if (k == (__mmask64) -1) {
> + a[0] = a[0] + 1;
> + }
> + a[0] = a[0] - 4;
> + return a;
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr113609-2.c
> b/gcc/testsuite/gcc.target/i386/pr113609-2.c
> new file mode 100644
> index 00000000000..e9503f51538
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr113609-2.c
> @@ -0,0 +1,161 @@
> +/* PR target/113609 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -march=x86-64-v4" } */
> +/* { dg-final { scan-assembler-times "\[ \\t\]+sete" 4 } } */
> +/* { dg-final { scan-assembler-times "\[ \\t\]+setne" 4 } } */
> +/* { dg-final { scan-assembler-times "\[ \\t\]+je" 4 } } */
> +/* { dg-final { scan-assembler-times "\[ \\t\]+jne" 4 } } */
> +
> +#include <immintrin.h>
> +
> +unsigned int
> +cmp_pi8_setcc(char a)
> +{
> + if (a == -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_pi16_setcc(short a)
> +{
> + if (a == -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_pi32_setcc(int a)
> +{
> + if (a == -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_pi64_setcc(long long a)
> +{
> + if (a == -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_pi8_setne(char a)
> +{
> + if (a != -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_pi16_setne(short a)
> +{
> + if (a != -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_pi32_setne(int a)
> +{
> + if (a != -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +unsigned int
> +cmp_pi64_setne(long long a)
> +{
> + if (a != -1)
> + return 1;
> + else
> + return 0;
> +}
> +
> +__m128i
> +cmp_pi8_je(__m128i a, char b) {
> + if (b == -1) {
> + a[0] = a[0] + 1;
> + }
> + else {
> + a[0] = a[0] - 1;
> + }
> + return a;
> +}
> +
> +__m128i
> +cmp_pi16_je(__m128i a, short b) {
> + if (b == -1) {
> + a[0] = a[0] + 1;
> + }
> + else {
> + a[0] = a[0] - 1;
> + }
> + return a;
> +}
> +
> +__m128i
> +cmp_pi32_je(__m128i a, int b) {
> + if (b == -1) {
> + a[0] = a[0] + 1;
> + }
> + else {
> + a[0] = a[0] - 1;
> + }
> + return a;
> +}
> +
> +__m128i
> +cmp_pi64_je(__m128i a, long long b) {
> + if (b == -1) {
> + a[0] = a[0] + 1;
> + }
> + else {
> + a[0] = a[0] - 1;
> + }
> + return a;
> +}
> +
> +__m128i
> +cmp_pi8_jne(__m128i a, char b) {
> + if (b == -1) {
> + a[0] = a[0] + 1;
> + }
> + a[0] = a[0] - 4;
> + return a;
> +}
> +
> +__m128i
> +cmp_pi16_jne(__m128i a, short b) {
> + if (b == -1) {
> + a[0] = a[0] + 1;
> + }
> + a[0] = a[0] - 4;
> + return a;
> +}
> +
> +__m128i
> +cmp_pi32_jne(__m128i a, int b) {
> + if (b == -1) {
> + a[0] = a[0] + 1;
> + }
> + a[0] = a[0] - 4;
> + return a;
> +}
> +
> +__m128i
> +cmp_pi64_jne(__m128i a, long long b) {
> + if (b == -1) {
> + a[0] = a[0] + 1;
> + }
> + a[0] = a[0] - 4;
> + return a;
> +}
> --
> 2.31.1
>
--
BR,
Hongtao
