Refine maskloadmn pattern with UNSPEC_MASKLOAD.

If mem_addr points to a memory region with less than whole vector size
bytes of accessible memory and k is a mask that would prevent reading
the inaccessible bytes from mem_addr, add UNSPEC_MASKLOAD to prevent
it to be transformed to vpblendd.

gcc/ChangeLog:

	PR target/110309
	* config/i386/sse.md (maskload<mode><avx512fmaskmodelower>):
	Refine pattern with UNSPEC_MASKLOAD.
	(maskload<mode><avx512fmaskmodelower>): Ditto.
	(*<avx512>_load<mode>_mask): Extend mode iterator to
	VI12HFBF_AVX512VL.
	(*<avx512>_load<mode>): Ditto.

gcc/testsuite/ChangeLog:

	* gcc.target/i386/pr110309.c: New test.
This commit is contained in:
liuhongt 2023-06-20 15:41:00 +08:00
parent ec1f1d3f7a
commit c79476da46
2 changed files with 28 additions and 14 deletions

View file

@ -1465,12 +1465,12 @@
})
(define_insn "*<avx512>_load<mode>_mask"
[(set (match_operand:VI12_AVX512VL 0 "register_operand" "=v")
(vec_merge:VI12_AVX512VL
(unspec:VI12_AVX512VL
[(match_operand:VI12_AVX512VL 1 "memory_operand" "m")]
[(set (match_operand:VI12HFBF_AVX512VL 0 "register_operand" "=v")
(vec_merge:VI12HFBF_AVX512VL
(unspec:VI12HFBF_AVX512VL
[(match_operand:VI12HFBF_AVX512VL 1 "memory_operand" "m")]
UNSPEC_MASKLOAD)
(match_operand:VI12_AVX512VL 2 "nonimm_or_0_operand" "0C")
(match_operand:VI12HFBF_AVX512VL 2 "nonimm_or_0_operand" "0C")
(match_operand:<avx512fmaskmode> 3 "register_operand" "Yk")))]
"TARGET_AVX512BW"
"vmovdqu<ssescalarsize>\t{%1, %0%{%3%}%N2|%0%{%3%}%N2, %1}"
@ -1479,9 +1479,9 @@
(set_attr "mode" "<sseinsnmode>")])
(define_insn_and_split "*<avx512>_load<mode>"
[(set (match_operand:VI12_AVX512VL 0 "register_operand" "=v")
(unspec:VI12_AVX512VL
[(match_operand:VI12_AVX512VL 1 "memory_operand" "m")]
[(set (match_operand:VI12HFBF_AVX512VL 0 "register_operand" "=v")
(unspec:VI12HFBF_AVX512VL
[(match_operand:VI12HFBF_AVX512VL 1 "memory_operand" "m")]
UNSPEC_MASKLOAD))]
"TARGET_AVX512BW"
"#"
@ -26979,17 +26979,21 @@
"TARGET_AVX")
(define_expand "maskload<mode><avx512fmaskmodelower>"
[(set (match_operand:V48H_AVX512VL 0 "register_operand")
(vec_merge:V48H_AVX512VL
(match_operand:V48H_AVX512VL 1 "memory_operand")
[(set (match_operand:V48_AVX512VL 0 "register_operand")
(vec_merge:V48_AVX512VL
(unspec:V48_AVX512VL
[(match_operand:V48_AVX512VL 1 "memory_operand")]
UNSPEC_MASKLOAD)
(match_dup 0)
(match_operand:<avx512fmaskmode> 2 "register_operand")))]
"TARGET_AVX512F")
(define_expand "maskload<mode><avx512fmaskmodelower>"
[(set (match_operand:VI12_AVX512VL 0 "register_operand")
(vec_merge:VI12_AVX512VL
(match_operand:VI12_AVX512VL 1 "memory_operand")
[(set (match_operand:VI12HFBF_AVX512VL 0 "register_operand")
(vec_merge:VI12HFBF_AVX512VL
(unspec:VI12HFBF_AVX512VL
[(match_operand:VI12HFBF_AVX512VL 1 "memory_operand")]
UNSPEC_MASKLOAD)
(match_dup 0)
(match_operand:<avx512fmaskmode> 2 "register_operand")))]
"TARGET_AVX512BW")

View file

@ -0,0 +1,10 @@
/* { dg-do compile } */
/* { dg-options "-O3 --param vect-partial-vector-usage=1 -march=znver4 -mprefer-vector-width=256" } */
/* { dg-final { scan-assembler-not {(?n)vpblendd.*ymm} } } */
void foo (int * __restrict a, int *b)
{
for (int i = 0; i < 6; ++i)
a[i] = b[i] + 42;
}