x86/intreadwrite: use intrinsics instead of inline asm for AV_ZERO128

When called inside a loop, the inline asm version results in one pxor
unnecessarely emitted per iteration, as the contents of the __asm__() block are
opaque to the compiler's instruction scheduler.
This is not the case with intrinsics, where pxor will be emitted once with any
half decent compiler.

This also has the benefit of removing any SSE -> AVX penalty that may happen
when the compiler emits VEX encoded instructions.

Signed-off-by: James Almer <jamrial@gmail.com>
release/7.1
James Almer 2 years ago
parent 34b4ca8696
commit 4a04cca69a
  1. 3
      configure
  2. 15
      libavutil/x86/intreadwrite.h

3
configure vendored

@ -2314,6 +2314,7 @@ HEADERS_LIST="
INTRINSICS_LIST="
intrinsics_neon
intrinsics_sse2
"
MATH_FUNCS="
@ -2743,6 +2744,7 @@ armv6t2_deps="arm"
armv8_deps="aarch64"
neon_deps_any="aarch64 arm"
intrinsics_neon_deps="neon"
intrinsics_sse2_deps="sse2"
vfp_deps="arm"
vfpv3_deps="vfp"
setend_deps="arm"
@ -6444,6 +6446,7 @@ elif enabled loongarch; then
fi
check_cc intrinsics_neon arm_neon.h "int16x8_t test = vdupq_n_s16(0)"
check_cc intrinsics_sse2 emmintrin.h "__m128i test = _mm_setzero_si128()"
check_ldflags -Wl,--as-needed
check_ldflags -Wl,-z,noexecstack

@ -22,6 +22,9 @@
#define AVUTIL_X86_INTREADWRITE_H
#include <stdint.h>
#if HAVE_INTRINSICS_SSE2
#include <emmintrin.h>
#endif
#include "config.h"
#include "libavutil/attributes.h"
@ -43,20 +46,16 @@ static av_always_inline void AV_COPY128(void *d, const void *s)
#endif /* __SSE__ */
#ifdef __SSE2__
#if HAVE_INTRINSICS_SSE2
#define AV_ZERO128 AV_ZERO128
static av_always_inline void AV_ZERO128(void *d)
{
struct v {uint64_t v[2];};
__asm__("pxor %%xmm0, %%xmm0 \n\t"
"movdqa %%xmm0, %0 \n\t"
: "=m"(*(struct v*)d)
:: "xmm0");
__m128i zero = _mm_setzero_si128();
_mm_store_si128(d, zero);
}
#endif /* __SSE2__ */
#endif /* HAVE_INTRINSICS_SSE2 */
#endif /* HAVE_MMX */

Loading…
Cancel
Save