FreeChainXenon/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

re PR target/85317 (missing constant propagation on _mm(256)_movemask_*)

Browse source 
PR target/85317
	* config/i386/i386.c (ix86_fold_builtin): Handle
	IX86_BUILTIN_{,P}MOVMSK{PS,PD,B}{,128,256}.

	* gcc.target/i386/pr85317.c: New test.
	* gcc.target/i386/avx2-vpmovmskb-2.c (avx2_test): Add asm volatile
	optimization barrier to avoid optimizing away the expected insn.

From-SVN: r260040
This commit is contained in:
Jakub Jelinek 2018-05-08 14:04:25 +02:00 • committed by Jakub Jelinek
parent 294c6f6637
commit ac68185918
5 changed files with 79 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
gcc/ChangeLog
View file

@ -1,5 +1,9 @@
2018-05-08 Jakub Jelinek <jakub@redhat.com>
PR target/85317
* config/i386/i386.c (ix86_fold_builtin): Handle
IX86_BUILTIN_{,P}MOVMSK{PS,PD,B}{,128,256}.
PR target/85480
* config/i386/sse.md (ssequaterinsnmode): New mode attribute.
(*<extract_type>_vinsert<shuffletype><extract_suf>_0): New pattern.

31
gcc/config/i386/i386.c
View file

@ -33477,6 +33477,37 @@ ix86_fold_builtin (tree fndecl, int n_args,
}
break;
case IX86_BUILTIN_MOVMSKPS:
case IX86_BUILTIN_PMOVMSKB:
case IX86_BUILTIN_MOVMSKPD:
case IX86_BUILTIN_PMOVMSKB128:
case IX86_BUILTIN_MOVMSKPD256:
case IX86_BUILTIN_MOVMSKPS256:
case IX86_BUILTIN_PMOVMSKB256:
gcc_assert (n_args == 1);
if (TREE_CODE (args[0]) == VECTOR_CST)
{
HOST_WIDE_INT res = 0;
for (unsigned i = 0; i < VECTOR_CST_NELTS (args[0]); ++i)
{
tree e = VECTOR_CST_ELT (args[0], i);
if (TREE_CODE (e) == INTEGER_CST && !TREE_OVERFLOW (e))
{
if (wi::neg_p (wi::to_wide (e)))
res |= HOST_WIDE_INT_1 << i;
}
else if (TREE_CODE (e) == REAL_CST && !TREE_OVERFLOW (e))
{
if (TREE_REAL_CST (e).sign)
res |= HOST_WIDE_INT_1 << i;
}
else
return NULL_TREE;
}
return build_int_cst (TREE_TYPE (TREE_TYPE (fndecl)), res);
}
break;
default:
break;
}

5
gcc/testsuite/ChangeLog
View file

@ -1,5 +1,10 @@
2018-05-08 Jakub Jelinek <jakub@redhat.com>
PR target/85317
* gcc.target/i386/pr85317.c: New test.
* gcc.target/i386/avx2-vpmovmskb-2.c (avx2_test): Add asm volatile
optimization barrier to avoid optimizing away the expected insn.
PR target/85480
* gcc.target/i386/avx512dq-pr85480-1.c: New test.
* gcc.target/i386/avx512dq-pr85480-2.c: New test.

1
gcc/testsuite/gcc.target/i386/avx2-vpmovmskb-2.c
View file

@ -14,6 +14,7 @@ avx2_test (void)
s.x = _mm256_set_epi8 (1, 2, 3, 4, 10, 20, 30, 90, -80, -40, -100,
15, 98, 25, 98, 7, 1, 2, 3, 4, 10, 20, 30, 90,
-80, -40, -100, -15, 98, 25, 98, 7);
__asm volatile ("" : "+m" (s) : : "memory");
res = _mm256_movemask_epi8 (s.x);

38
gcc/testsuite/gcc.target/i386/pr85317.c Normal file
View file

@ -0,0 +1,38 @@
/* PR target/85317 */
/* { dg-do compile } */
/* { dg-options "-O2 -mavx2 -fdump-tree-optimized" } */
/* { dg-final { scan-tree-dump-not "link_error" "optimized" } } */
#include <x86intrin.h>
extern void link_error (void);
int
main ()
{
int a = _mm_movemask_pd (_mm_set_pd (-2.0, 2.0));
if (a != 2) link_error ();
int b = _mm_movemask_pd (_mm_set_pd (0.0, __builtin_copysign (0.0, -4.0)));
if (b != 1) link_error ();
int c = _mm_movemask_ps (_mm_set_ps (__builtin_copysignf (0.0f, -4.0f), 0.0f,
-4.0f, 4.0f));
if (c != 10) link_error ();
int d = _mm_movemask_epi8 (_mm_set_epi8 (-4, 8, -8, -12, 12, 15, 0, -1, -3,
-128, 127, 126, 120, -120, 0, 5));
if (d != 0xb1c4) link_error ();
int e = _mm256_movemask_pd (_mm256_set_pd (-4.0, 0.0, 4.0,
__builtin_copysign (0.0, -4.0)));
if (e != 9) link_error ();
int f = _mm256_movemask_ps (_mm256_set_ps (-8.0f, -16.0f, 12.0f, 127.0f,
-4.0f, 0.0f, 4.0f,
__builtin_copysign (0.0f,
-4.0f)));
if (f != 0xc9) link_error ();
int g = _mm256_movemask_epi8 (_mm256_set_epi8 (-4, 8, -8, -12, 12, 15, 0, -1,
-3, -128, 127, 126, 120, -120,
0, 5, 12, 100, -20, -50, -70,
2, -65, 0, -1, 1, 2, -2, -9,
-9, 19, -64));
if (g != (int) 0xb1c43a9dU) link_error ();
return 0;
}