i386: Combine the FADD(A, FMA(B, C, 0)) to FMA(B, C, A) and combine FADD(A, FMUL(B, C)) to FMA(B, C, A).
Commit Message
Hi,
This patch is to support transform in fast-math something like _mm512_add_ph(x1, _mm512_fmadd_pch(a, b, _mm512_setzero_ph())) to _mm512_fmadd_pch(a, b, x1).
And support transform _mm512_add_ph(x1, _mm512_fmul_pch(a, b)) to _mm512_fmadd_pch(a, b, x1).
Ok for master?
gcc/ChangeLog:
* config/i386/sse.md (fma_<mode>_fadd_fmul): Add new
define_insn_and_split.
(fma_<mode>_fadd_fcmul):Likewise
(fma_<complexopname>_<mode>_fma_zero):Likewise
gcc/testsuite/ChangeLog:
* gcc.target/i386/avx512fp16-complex-fma.c: New test.
---
gcc/config/i386/sse.md | 52 +++++++++++++++++++
.../gcc.target/i386/avx512fp16-complex-fma.c | 18 +++++++
2 files changed, 70 insertions(+)
create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-complex-fma.c
+avx512f_test (void)
+{
+ res = _mm512_add_ph (x1, _mm512_fmadd_pch (a, b,
+_mm512_setzero_ph()));
+ res = _mm512_add_ph (x1, _mm512_fcmadd_pch (a, b,
+_mm512_setzero_ph()));
+
+ res = _mm512_add_ph (x1, _mm512_fmul_pch (a, b));
+ res = _mm512_add_ph (x1, _mm512_fcmul_pch (a, b)); }
--
2.18.1
Comments
On Fri, Oct 22, 2021 at 1:57 PM Kong, Lingling via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> Hi,
>
> This patch is to support transform in fast-math something like _mm512_add_ph(x1, _mm512_fmadd_pch(a, b, _mm512_setzero_ph())) to _mm512_fmadd_pch(a, b, x1).
>
> And support transform _mm512_add_ph(x1, _mm512_fmul_pch(a, b)) to _mm512_fmadd_pch(a, b, x1).
> Ok for master?
LGTM.
Also please add cfma_optab/conj_cfma_optab, so vectorizer can catch
some complex fma pattern match optimization.
>
> gcc/ChangeLog:
>
> * config/i386/sse.md (fma_<mode>_fadd_fmul): Add new
> define_insn_and_split.
> (fma_<mode>_fadd_fcmul):Likewise
> (fma_<complexopname>_<mode>_fma_zero):Likewise
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/i386/avx512fp16-complex-fma.c: New test.
> ---
> gcc/config/i386/sse.md | 52 +++++++++++++++++++
> .../gcc.target/i386/avx512fp16-complex-fma.c | 18 +++++++
> 2 files changed, 70 insertions(+)
> create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-complex-fma.c
>
> diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md index fbf056bf9e6..36407ca4a59 100644
> --- a/gcc/config/i386/sse.md
> +++ b/gcc/config/i386/sse.md
> @@ -5958,6 +5958,58 @@
> [(set_attr "type" "ssemuladd")
> (set_attr "mode" "<MODE>")])
>
> +(define_insn_and_split "fma_<mode>_fadd_fmul"
> + [(set (match_operand:VF_AVX512FP16VL 0 "register_operand")
> + (plus:VF_AVX512FP16VL
> + (unspec:VF_AVX512FP16VL
> + [(match_operand:VF_AVX512FP16VL 1 "vector_operand")
> + (match_operand:VF_AVX512FP16VL 2 "vector_operand")]
> + UNSPEC_COMPLEX_FMUL)
> + (match_operand:VF_AVX512FP16VL 3 "vector_operand")))]
> + "TARGET_AVX512FP16 && flag_unsafe_math_optimizations
> + && ix86_pre_reload_split()"
> + "#"
> + "&& 1"
> + [(set (match_dup 0)
> + (unspec:VF_AVX512FP16VL
> + [(match_dup 1) (match_dup 2) (match_dup 3)]
> + UNSPEC_COMPLEX_FMA))])
> +
> +(define_insn_and_split "fma_<mode>_fadd_fcmul"
> + [(set (match_operand:VF_AVX512FP16VL 0 "register_operand")
> + (plus:VF_AVX512FP16VL
> + (unspec:VF_AVX512FP16VL
> + [(match_operand:VF_AVX512FP16VL 1 "vector_operand")
> + (match_operand:VF_AVX512FP16VL 2 "vector_operand")]
> + UNSPEC_COMPLEX_FCMUL)
> + (match_operand:VF_AVX512FP16VL 3 "vector_operand")))]
> + "TARGET_AVX512FP16 && flag_unsafe_math_optimizations
> + && ix86_pre_reload_split()"
> + "#"
> + "&& 1"
> + [(set (match_dup 0)
> + (unspec:VF_AVX512FP16VL
> + [(match_dup 1) (match_dup 2) (match_dup 3)]
> + UNSPEC_COMPLEX_FCMA))])
> +
> +(define_insn_and_split "fma_<complexopname>_<mode>_fma_zero"
> + [(set (match_operand:VF_AVX512FP16VL 0 "register_operand")
> + (plus:VF_AVX512FP16VL
> + (unspec:VF_AVX512FP16VL
> + [(match_operand:VF_AVX512FP16VL 1 "vector_operand")
> + (match_operand:VF_AVX512FP16VL 2 "vector_operand")
> + (match_operand:VF_AVX512FP16VL 3 "const0_operand")]
> + UNSPEC_COMPLEX_F_C_MA)
> + (match_operand:VF_AVX512FP16VL 4 "vector_operand")))]
> + "TARGET_AVX512FP16 && flag_unsafe_math_optimizations
> + && ix86_pre_reload_split()"
> + "#"
> + "&& 1"
> + [(set (match_dup 0)
> + (unspec:VF_AVX512FP16VL
> + [(match_dup 1) (match_dup 2) (match_dup 4)]
> + UNSPEC_COMPLEX_F_C_MA))])
> +
> (define_insn "<avx512>_<complexopname>_<mode>_mask<round_name>"
> [(set (match_operand:VF_AVX512FP16VL 0 "register_operand" "=&v")
> (vec_merge:VF_AVX512FP16VL
> diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-complex-fma.c b/gcc/testsuite/gcc.target/i386/avx512fp16-complex-fma.c
> new file mode 100644
> index 00000000000..2dfd369e785
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-complex-fma.c
> @@ -0,0 +1,18 @@
> +/* { dg-do compile } */
> +/* { dg-options "-mavx512fp16 -O2 -Ofast" } */
> +/* { dg-final { scan-assembler-times "vfmaddcph\[
> +\\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(
> +?:\n|\[ \\t\]+#)" 2 } } */
> +/* { dg-final { scan-assembler-not "vaddph\[
> +\\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(
> +?:\n|\[ \\t\]+#)"} } */
> +/* { dg-final { scan-assembler-not "vfmulcph\[
> +\\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(
> +?:\n|\[ \\t\]+#)"} } */
> +/* { dg-final { scan-assembler-times "vfcmaddcph\[
> +\\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(
> +?:\n|\[ \\t\]+#)" 2 } } */
> +
> +#include <immintrin.h>
> +volatile __m512h x1, x2, res, a, b;
> +void extern
> +avx512f_test (void)
> +{
> + res = _mm512_add_ph (x1, _mm512_fmadd_pch (a, b,
> +_mm512_setzero_ph()));
> + res = _mm512_add_ph (x1, _mm512_fcmadd_pch (a, b,
> +_mm512_setzero_ph()));
> +
> + res = _mm512_add_ph (x1, _mm512_fmul_pch (a, b));
> + res = _mm512_add_ph (x1, _mm512_fcmul_pch (a, b)); }
> --
> 2.18.1
>
@@ -5958,6 +5958,58 @@
[(set_attr "type" "ssemuladd")
(set_attr "mode" "<MODE>")])
+(define_insn_and_split "fma_<mode>_fadd_fmul"
+ [(set (match_operand:VF_AVX512FP16VL 0 "register_operand")
+ (plus:VF_AVX512FP16VL
+ (unspec:VF_AVX512FP16VL
+ [(match_operand:VF_AVX512FP16VL 1 "vector_operand")
+ (match_operand:VF_AVX512FP16VL 2 "vector_operand")]
+ UNSPEC_COMPLEX_FMUL)
+ (match_operand:VF_AVX512FP16VL 3 "vector_operand")))]
+ "TARGET_AVX512FP16 && flag_unsafe_math_optimizations
+ && ix86_pre_reload_split()"
+ "#"
+ "&& 1"
+ [(set (match_dup 0)
+ (unspec:VF_AVX512FP16VL
+ [(match_dup 1) (match_dup 2) (match_dup 3)]
+ UNSPEC_COMPLEX_FMA))])
+
+(define_insn_and_split "fma_<mode>_fadd_fcmul"
+ [(set (match_operand:VF_AVX512FP16VL 0 "register_operand")
+ (plus:VF_AVX512FP16VL
+ (unspec:VF_AVX512FP16VL
+ [(match_operand:VF_AVX512FP16VL 1 "vector_operand")
+ (match_operand:VF_AVX512FP16VL 2 "vector_operand")]
+ UNSPEC_COMPLEX_FCMUL)
+ (match_operand:VF_AVX512FP16VL 3 "vector_operand")))]
+ "TARGET_AVX512FP16 && flag_unsafe_math_optimizations
+ && ix86_pre_reload_split()"
+ "#"
+ "&& 1"
+ [(set (match_dup 0)
+ (unspec:VF_AVX512FP16VL
+ [(match_dup 1) (match_dup 2) (match_dup 3)]
+ UNSPEC_COMPLEX_FCMA))])
+
+(define_insn_and_split "fma_<complexopname>_<mode>_fma_zero"
+ [(set (match_operand:VF_AVX512FP16VL 0 "register_operand")
+ (plus:VF_AVX512FP16VL
+ (unspec:VF_AVX512FP16VL
+ [(match_operand:VF_AVX512FP16VL 1 "vector_operand")
+ (match_operand:VF_AVX512FP16VL 2 "vector_operand")
+ (match_operand:VF_AVX512FP16VL 3 "const0_operand")]
+ UNSPEC_COMPLEX_F_C_MA)
+ (match_operand:VF_AVX512FP16VL 4 "vector_operand")))]
+ "TARGET_AVX512FP16 && flag_unsafe_math_optimizations
+ && ix86_pre_reload_split()"
+ "#"
+ "&& 1"
+ [(set (match_dup 0)
+ (unspec:VF_AVX512FP16VL
+ [(match_dup 1) (match_dup 2) (match_dup 4)]
+ UNSPEC_COMPLEX_F_C_MA))])
+
(define_insn "<avx512>_<complexopname>_<mode>_mask<round_name>"
[(set (match_operand:VF_AVX512FP16VL 0 "register_operand" "=&v")
(vec_merge:VF_AVX512FP16VL
new file mode 100644
@@ -0,0 +1,18 @@
+/* { dg-do compile } */
+/* { dg-options "-mavx512fp16 -O2 -Ofast" } */
+/* { dg-final { scan-assembler-times "vfmaddcph\[
+\\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(
+?:\n|\[ \\t\]+#)" 2 } } */
+/* { dg-final { scan-assembler-not "vaddph\[
+\\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(
+?:\n|\[ \\t\]+#)"} } */
+/* { dg-final { scan-assembler-not "vfmulcph\[
+\\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(
+?:\n|\[ \\t\]+#)"} } */
+/* { dg-final { scan-assembler-times "vfcmaddcph\[
+\\t\]+\[^\{\n\]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+\[^\n\r]*%zmm\[0-9\]+(
+?:\n|\[ \\t\]+#)" 2 } } */
+
+#include <immintrin.h>
+volatile __m512h x1, x2, res, a, b;
+void extern