This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Fix 4 avx512* mask reg constraints
- From: Jakub Jelinek <jakub at redhat dot com>
- To: Kirill Yukhin <kirill dot yukhin at gmail dot com>, Uros Bizjak <ubizjak at gmail dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Fri, 12 Aug 2016 12:51:19 +0200
- Subject: [PATCH] Fix 4 avx512* mask reg constraints
- Authentication-results: sourceware.org; auth=none
- Reply-to: Jakub Jelinek <jakub at redhat dot com>
Hi!
For the VEXTRACT* insns, I can't find anything that would make them
special wrt. masking, so I believe they can't accept {%k0}, just
no masking or {%k1} through {%k7}.
I've tried to reproduce it with:
#include <x86intrin.h>
__m128i
f1 (__m128i a, __m512i b, __mmask8 c)
{
register __mmask8 d __asm ("k0");
d = c;
asm ("" : "+k" (d));
return _mm512_mask_extracti64x2_epi64 (a, d, b, 3);
}
but for some reason it still chose k1, still I believe k constraint
is wrong in these cases.
Ok for trunk?
2016-08-12 Jakub Jelinek <jakub@redhat.com>
* config/i386/sse.md (avx512dq_vextract<shuffletype>64x2_1_maskm,
vec_extract_hi_<mode>_mask, vec_extract_lo_<mode>_mask,
vec_extract_hi_<mode>_mask): Use Yk constraint instead of k.
--- gcc/config/i386/sse.md.jj 2016-07-14 20:28:33.000000000 +0200
+++ gcc/config/i386/sse.md 2016-08-12 12:14:19.618486123 +0200
@@ -6957,7 +6957,7 @@ (define_insn "avx512dq_vextract<shufflet
(parallel [(match_operand 2 "const_0_to_7_operand")
(match_operand 3 "const_0_to_7_operand")]))
(match_operand:<ssequartermode> 4 "memory_operand" "0")
- (match_operand:QI 5 "register_operand" "k")))]
+ (match_operand:QI 5 "register_operand" "Yk")))]
"TARGET_AVX512DQ
&& (INTVAL (operands[2]) % 2 == 0)
&& (INTVAL (operands[2]) == INTVAL (operands[3]) - 1)
@@ -7171,7 +7171,7 @@ (define_insn "vec_extract_hi_<mode>_mask
(const_int 12) (const_int 13)
(const_int 14) (const_int 15)]))
(match_operand:<ssehalfvecmode> 2 "memory_operand" "0")
- (match_operand:QI 3 "register_operand" "k")))]
+ (match_operand:QI 3 "register_operand" "Yk")))]
"TARGET_AVX512DQ
&& rtx_equal_p (operands[2], operands[0])"
"vextract<shuffletype>32x8\t{$0x1, %1, %0%{%3%}|%0%{%3%}, %1, 0x1}"
@@ -7377,7 +7377,7 @@ (define_insn "vec_extract_lo_<mode>_mask
(parallel [(const_int 0) (const_int 1)
(const_int 2) (const_int 3)]))
(match_operand:<ssehalfvecmode> 2 "memory_operand" "0")
- (match_operand:QI 3 "register_operand" "k")))]
+ (match_operand:QI 3 "register_operand" "Yk")))]
"TARGET_AVX512VL && TARGET_AVX512F
&& rtx_equal_p (operands[2], operands[0])"
"vextract<shuffletype>32x4\t{$0x0, %1, %0%{%3%}|%0%{%3%}, %1, 0x0}"
@@ -7395,7 +7395,7 @@ (define_insn "vec_extract_hi_<mode>_mask
(parallel [(const_int 4) (const_int 5)
(const_int 6) (const_int 7)]))
(match_operand:<ssehalfvecmode> 2 "memory_operand" "0")
- (match_operand:<ssehalfvecmode> 3 "register_operand" "k")))]
+ (match_operand:<ssehalfvecmode> 3 "register_operand" "Yk")))]
"TARGET_AVX512F && TARGET_AVX512VL
&& rtx_equal_p (operands[2], operands[0])"
"vextract<shuffletype>32x4\t{$0x1, %1, %0%{%3%}|%0%{%3%}, %1, 0x1}"
Jakub