https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120708
Bug ID: 120708
Summary: ix86_expand_set_or_cpymem ignores MOVE_MAX
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: hjl.tools at gmail dot com
CC: liuhongt at gcc dot gnu.org
Target Milestone: ---
Target: x86-64
i386 defines
/* Max number of bytes we can move from memory to memory in one
reasonably fast instruction, as opposed to MOVE_MAX_PIECES which
is the number of bytes at a time which we can move efficiently.
MOVE_MAX_PIECES defaults to MOVE_MAX. */
#define MOVE_MAX \
((TARGET_AVX512F \
&& (ix86_move_max == PVW_AVX512 \
|| ix86_store_max == PVW_AVX512)) \
? 64 \
: ((TARGET_AVX \
&& (ix86_move_max >= PVW_AVX256 \
|| ix86_store_max >= PVW_AVX256)) \
? 32 \
: ((TARGET_SSE2 \
&& TARGET_SSE_UNALIGNED_LOAD_OPTIMAL \
&& TARGET_SSE_UNALIGNED_STORE_OPTIMAL) \
? 16 : UNITS_PER_WORD)))
If TARGET_SSE_UNALIGNED_LOAD_OPTIMAL or TARGET_SSE_UNALIGNED_STORE_OPTIMAL are
false, MOVE_MAX is defined UNITS_PER_WORD. For -march=atom, both are false.
But ix86_expand_set_or_cpymem ignores it. As the result,
memcpy-vector_loop-1.c
and memset-vector_loop-2.c, which are compiled with -march=atom, are compiled
with
SSE instructions:
movdqa %xmm3, a(%rax)
movdqa %xmm2, a+16(%rax)
movdqa %xmm1, a+32(%rax)
movdqa %xmm0, a+48(%rax)
