diff options
author | Jakub Jelinek <jakub@redhat.com> | 2020-01-15 00:30:10 +0100 |
---|---|---|
committer | Jakub Jelinek <jakub@redhat.com> | 2020-01-15 00:30:10 +0100 |
commit | fbbc4c24fd7ba87e0c47cd965ae624afba6fa375 (patch) | |
tree | 8175b3ddc03ae580e8dde005c60d4b41c09c0aa3 | |
parent | f474fbd5e3cca37ebc886a4950827e93d1c665c8 (diff) | |
download | gcc-fbbc4c24fd7ba87e0c47cd965ae624afba6fa375.zip gcc-fbbc4c24fd7ba87e0c47cd965ae624afba6fa375.tar.gz gcc-fbbc4c24fd7ba87e0c47cd965ae624afba6fa375.tar.bz2 |
i386: Fix wrong-code x86 issue with avx512{f,vl} fma PR93009
As mentioned in the PR, the following testcase is miscompiled with avx512vl.
The reason is that the fma *_bcst_1 define_insns have two alternatives:
"=v,v" "0,v" "v,0" "m,m" and use the same
vfmadd213* %3<avx512bcst>, %2, %0<sd_mask_op4>
pattern. If the first alternative is chosen, everything is ok, but if the
second alternative is chosen, %2 and %0 are the same register, so instead
of doing dest=dest*another+membcst we do dest=dest*dest+membcst.
Now, to fix this, either we'd need separate:
"vfmadd213<ssemodesuffix>\t{%3<avx512bcst>, %2, %0<sd_mask_op4>|%0<sd_mask_op4>, %2, %3<avx512bcst>}
vfmadd213<ssemodesuffix>\t{%3<avx512bcst>, %1, %0<sd_mask_op4>|%0<sd_mask_op4>, %1, %3<avx512bcst>}"
where for the second alternative, we'd just use %1 instead of %2, but
what I think is actually cleaner is just use a single alternative and
make the two multiplication operands commutative, which they really are.
2020-01-15 Jakub Jelinek <jakub@redhat.com>
PR target/93009
* config/i386/sse.md
(*<sd_mask_codefor>fma_fmadd_<mode><sd_maskz_name>_bcst_1,
*<sd_mask_codefor>fma_fmsub_<mode><sd_maskz_name>_bcst_1,
*<sd_mask_codefor>fma_fnmadd_<mode><sd_maskz_name>_bcst_1,
*<sd_mask_codefor>fma_fnmsub_<mode><sd_maskz_name>_bcst_1): Use
just a single alternative instead of two, make operands 1 and 2
commutative.
* gcc.target/i386/avx512vl-pr93009.c: New test.
-rw-r--r-- | gcc/ChangeLog | 11 | ||||
-rw-r--r-- | gcc/config/i386/sse.md | 32 | ||||
-rw-r--r-- | gcc/testsuite/ChangeLog | 5 | ||||
-rw-r--r-- | gcc/testsuite/gcc.target/i386/avx512vl-pr93009.c | 38 |
4 files changed, 70 insertions, 16 deletions
diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 33ca91a..356bc63 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,14 @@ +2020-01-15 Jakub Jelinek <jakub@redhat.com> + + PR target/93009 + * config/i386/sse.md + (*<sd_mask_codefor>fma_fmadd_<mode><sd_maskz_name>_bcst_1, + *<sd_mask_codefor>fma_fmsub_<mode><sd_maskz_name>_bcst_1, + *<sd_mask_codefor>fma_fnmadd_<mode><sd_maskz_name>_bcst_1, + *<sd_mask_codefor>fma_fnmsub_<mode><sd_maskz_name>_bcst_1): Use + just a single alternative instead of two, make operands 1 and 2 + commutative. + 2020-01-14 Jan Hubicka <hubicka@ucw.cz> PR lto/91576 diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md index f4c8a39..b8d41b7 100644 --- a/gcc/config/i386/sse.md +++ b/gcc/config/i386/sse.md @@ -4172,12 +4172,12 @@ (set_attr "mode" "<MODE>")]) (define_insn "*<sd_mask_codefor>fma_fmadd_<mode><sd_maskz_name>_bcst_1" - [(set (match_operand:VF_AVX512 0 "register_operand" "=v,v") + [(set (match_operand:VF_AVX512 0 "register_operand" "=v") (fma:VF_AVX512 - (match_operand:VF_AVX512 1 "register_operand" "0,v") - (match_operand:VF_AVX512 2 "register_operand" "v,0") + (match_operand:VF_AVX512 1 "register_operand" "%0") + (match_operand:VF_AVX512 2 "register_operand" "v") (vec_duplicate:VF_AVX512 - (match_operand:<ssescalarmode> 3 "memory_operand" "m,m"))))] + (match_operand:<ssescalarmode> 3 "memory_operand" "m"))))] "TARGET_AVX512F && <sd_mask_mode512bit_condition>" "vfmadd213<ssemodesuffix>\t{%3<avx512bcst>, %2, %0<sd_mask_op4>|%0<sd_mask_op4>, %2, %3<avx512bcst>}" [(set_attr "type" "ssemuladd") @@ -4289,13 +4289,13 @@ (set_attr "mode" "<MODE>")]) (define_insn "*<sd_mask_codefor>fma_fmsub_<mode><sd_maskz_name>_bcst_1" - [(set (match_operand:VF_AVX512 0 "register_operand" "=v,v") + [(set (match_operand:VF_AVX512 0 "register_operand" "=v") (fma:VF_AVX512 - (match_operand:VF_AVX512 1 "register_operand" "0,v") - (match_operand:VF_AVX512 2 "register_operand" "v,0") + (match_operand:VF_AVX512 1 "register_operand" "%0") + (match_operand:VF_AVX512 2 "register_operand" "v") (neg:VF_AVX512 (vec_duplicate:VF_AVX512 - (match_operand:<ssescalarmode> 3 "memory_operand" "m,m")))))] + (match_operand:<ssescalarmode> 3 "memory_operand" "m")))))] "TARGET_AVX512F && <sd_mask_mode512bit_condition>" "vfmsub213<ssemodesuffix>\t{%3<avx512bcst>, %2, %0<sd_mask_op4>|%0<sd_mask_op4>, %2, %3<avx512bcst>}" [(set_attr "type" "ssemuladd") @@ -4411,13 +4411,13 @@ (set_attr "mode" "<MODE>")]) (define_insn "*<sd_mask_codefor>fma_fnmadd_<mode><sd_maskz_name>_bcst_1" - [(set (match_operand:VF_AVX512 0 "register_operand" "=v,v") + [(set (match_operand:VF_AVX512 0 "register_operand" "=v") (fma:VF_AVX512 (neg:VF_AVX512 - (match_operand:VF_AVX512 1 "register_operand" "0,v")) - (match_operand:VF_AVX512 2 "register_operand" "v,0") + (match_operand:VF_AVX512 1 "register_operand" "%0")) + (match_operand:VF_AVX512 2 "register_operand" "v") (vec_duplicate:VF_AVX512 - (match_operand:<ssescalarmode> 3 "memory_operand" "m,m"))))] + (match_operand:<ssescalarmode> 3 "memory_operand" "m"))))] "TARGET_AVX512F && <sd_mask_mode512bit_condition>" "vfnmadd213<ssemodesuffix>\t{%3<avx512bcst>, %2, %0<sd_mask_op4>|%0<sd_mask_op4>, %2, %3<avx512bcst>}" [(set_attr "type" "ssemuladd") @@ -4535,14 +4535,14 @@ (set_attr "mode" "<MODE>")]) (define_insn "*<sd_mask_codefor>fma_fnmsub_<mode><sd_maskz_name>_bcst_1" - [(set (match_operand:VF_AVX512 0 "register_operand" "=v,v") + [(set (match_operand:VF_AVX512 0 "register_operand" "=v") (fma:VF_AVX512 (neg:VF_AVX512 - (match_operand:VF_AVX512 1 "register_operand" "0,v")) - (match_operand:VF_AVX512 2 "register_operand" "v,0") + (match_operand:VF_AVX512 1 "register_operand" "%0")) + (match_operand:VF_AVX512 2 "register_operand" "v") (neg:VF_AVX512 (vec_duplicate:VF_AVX512 - (match_operand:<ssescalarmode> 3 "memory_operand" "m,m")))))] + (match_operand:<ssescalarmode> 3 "memory_operand" "m")))))] "TARGET_AVX512F && <sd_mask_mode512bit_condition>" "vfnmsub213<ssemodesuffix>\t{%3<avx512bcst>, %2, %0<sd_mask_op4>|%0<sd_mask_op4>, %2, %3<avx512bcst>}" [(set_attr "type" "ssemuladd") diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index dc42601..ea37d69 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,8 @@ +2020-01-15 Jakub Jelinek <jakub@redhat.com> + + PR target/93009 + * gcc.target/i386/avx512vl-pr93009.c: New test. + 2020-01-14 Jan Hubicka <hubicka@ucw.cz> PR lto/91576 diff --git a/gcc/testsuite/gcc.target/i386/avx512vl-pr93009.c b/gcc/testsuite/gcc.target/i386/avx512vl-pr93009.c new file mode 100644 index 0000000..4dfc4a9 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/avx512vl-pr93009.c @@ -0,0 +1,38 @@ +/* PR target/93009 */ +/* { dg-do run { target { avx512vl } } } */ +/* { dg-options "-O2 -mavx512vl" } */ + +#define AVX512VL +#define AVX512F_LEN 512 +#define AVX512F_LEN_HALF 256 + +#include "avx512f-check.h" + +typedef double v2df __attribute__((vector_size (16))); + +__attribute__((noipa)) v2df +foo (v2df x, v2df y, double *z) +{ + return x * y + (v2df) { z[0], z[0] }; +} + +__attribute__((noipa)) v2df +bar (v2df x, v2df y, double *z) +{ + return y * x + (v2df) { z[0], z[0] }; +} + +static void +test_256 (void) +{ +} + +static void +test_128 (void) +{ + double z = 5.0; + v2df x = foo ((v2df) { 1.0, 2.0 }, (v2df) { 3.0, 4.0 }, &z); + v2df y = bar ((v2df) { 6.0, 7.0 }, (v2df) { 8.0, 9.0 }, &z); + if (x[0] != 8.0 || x[1] != 13.0 || y[0] != 53.0 || y[1] != 68.0) + abort (); +} |