x86: Fix *vec_dupv4hi constraints [PR94942]

This insn and split splits into HI->V?HImode broadcast for avx2 and later,
but either the operands need to be %xmm0-%xmm15 (i.e. VEX encoded insn), or
the insn needs both AVX512BW and AVX512VL.
Now, Yv constraint is v for AVX512VL and x otherwise, so for -mavx512vl -mno-avx512bw
we ICE if we end up with a %xmm16+ register from RA.
Yw constraint is v for AVX512VL and AVX512BW and nothing otherwise, so
in this pattern we actually need xYw.

2020-05-05  Jakub Jelinek  <jakub@redhat.com>

	PR target/94942
	* config/i386/mmx.md (*vec_dupv4hi): Use xYw constraints instead of Yv.

	* gcc.target/i386/pr94942.c: New test.
This commit is contained in:
Jakub Jelinek 2020-05-05 11:33:47 +02:00
parent 6d938a5d77
commit 7f916201ac
4 changed files with 32 additions and 2 deletions

View File

@ -1,5 +1,8 @@
2020-05-05 Jakub Jelinek <jakub@redhat.com>
PR target/94942
* config/i386/mmx.md (*vec_dupv4hi): Use xYw constraints instead of Yv.
PR tree-optimization/94914
* match.pd ((((type)A * B) >> prec) != 0 to .MUL_OVERFLOW(A, B) != 0):
New simplification.

View File

@ -1613,10 +1613,10 @@
(set_attr "mode" "DI")])
(define_insn_and_split "*vec_dupv4hi"
[(set (match_operand:V4HI 0 "register_operand" "=y,Yv,Yw")
[(set (match_operand:V4HI 0 "register_operand" "=y,xYw,Yw")
(vec_duplicate:V4HI
(truncate:HI
(match_operand:SI 1 "register_operand" "0,Yv,r"))))]
(match_operand:SI 1 "register_operand" "0,xYw,r"))))]
"(TARGET_MMX || TARGET_MMX_WITH_SSE)
&& (TARGET_SSE || TARGET_3DNOW_A)"
"@

View File

@ -1,5 +1,8 @@
2020-05-05 Jakub Jelinek <jakub@redhat.com>
PR target/94942
* gcc.target/i386/pr94942.c: New test.
PR tree-optimization/94914
* gcc.target/i386/pr94914.c: New test.

View File

@ -0,0 +1,24 @@
/* PR target/94942 */
/* { dg-do compile } */
/* { dg-options "-O -flive-range-shrinkage -ftree-vrp -mavx512vl -mno-avx512bw -Wno-div-by-zero" } */
typedef unsigned __attribute__((__vector_size__(8))) U;
typedef short __attribute__((__vector_size__(8))) V;
typedef char __attribute__((__vector_size__(16))) W;
typedef int __attribute__((__vector_size__(16))) Z;
int i, j, n, o;
W k;
Z l;
char m;
U
foo (U q, long long r, V s)
{
Z t = (i & i - (Z){10} & 4) - (0 != j);
Z u = o * (j * l);
s -= (char)__builtin_clrsbll (n);
W v = (k | k >> m + (W){4}) % 0;
W w = v + (W)t + (W)u;
U x = ((union { W a; U b; })w).b + q + (U)s + (U)r;
return x;
}