| Message ID | 20220903150518.169643-2-remi@remlab.net |
|---|---|
| State | Accepted |
| Commit | 620e6e14878de7392f9b5fd109cc8f5ed90dd835 |
| Headers | show |
| Series | [FFmpeg-devel,1/2] aarch64: relax byte-swap assembler constraints | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | success | Make fate finished |
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
On Sat, 3 Sep 2022, remi@remlab.net wrote: > From: Rémi Denis-Courmont <remi@remlab.net> > > There are no particular reasons to force the compiler to use the same > register as output and input operand. This forces an extra MOV > instruction if the input value needs to be reused after the swap. > > In most cases, this makes no differences, as the compiler will seleect > the same register for both operands either way. > --- > libavutil/arm/bswap.h | 17 ++++++++++------- > 1 file changed, 10 insertions(+), 7 deletions(-) These two patches LGTM - I can push them later after running them through some test builds (even though I don't expect there to be any issues). // Martin
diff --git a/libavutil/arm/bswap.h b/libavutil/arm/bswap.h index 611ff0ad5b..c3460e035d 100644 --- a/libavutil/arm/bswap.h +++ b/libavutil/arm/bswap.h @@ -39,8 +39,10 @@ static av_always_inline av_const uint32_t av_bswap32(uint32_t x) #define av_bswap16 av_bswap16 static av_always_inline av_const unsigned av_bswap16(unsigned x) { - __asm__("rev16 %0, %0" : "+r"(x)); - return x; + unsigned y; + + __asm__("rev16 %0, %1" : "=r"(y) : "r"(x)); + return y; } #endif @@ -48,17 +50,18 @@ static av_always_inline av_const unsigned av_bswap16(unsigned x) #define av_bswap32 av_bswap32 static av_always_inline av_const uint32_t av_bswap32(uint32_t x) { + uint32_t y; #if HAVE_ARMV6_INLINE - __asm__("rev %0, %0" : "+r"(x)); + __asm__("rev %0, %1" : "=r"(y) : "r"(x)); #else uint32_t t; - __asm__ ("eor %1, %0, %0, ror #16 \n\t" + __asm__ ("eor %1, %2, %2, ror #16 \n\t" "bic %1, %1, #0xFF0000 \n\t" - "mov %0, %0, ror #8 \n\t" + "mov %0, %2, ror #8 \n\t" "eor %0, %0, %1, lsr #8 \n\t" - : "+r"(x), "=&r"(t)); + : "=r"(y), "=&r"(t) : "r"(x)); #endif /* HAVE_ARMV6_INLINE */ - return x; + return y; } #endif /* AV_GCC_VERSION_AT_MOST(4,4) */
From: Rémi Denis-Courmont <remi@remlab.net> There are no particular reasons to force the compiler to use the same register as output and input operand. This forces an extra MOV instruction if the input value needs to be reused after the swap. In most cases, this makes no differences, as the compiler will seleect the same register for both operands either way. --- libavutil/arm/bswap.h | 17 ++++++++++------- 1 file changed, 10 insertions(+), 7 deletions(-)