Relintai/sdl2_frt
1
0
Fork 0
mirror of https://github.com/Relintai/sdl2_frt.git synced 2024-11-21 20:47:19 +01:00
Code Issues Packages Projects Releases Wiki Activity

audiocvt: stereo-to-mono SSE3 now uses unaligned accesses.

Browse Source 
On modern CPUs, there's no penalty for using the unaligned instruction on
aligned memory, but now it can vectorize unaligned data too, which even if
it's not optimal, is still going to be faster than the scalar fallback.

Fixes #4532.
This commit is contained in:
Ryan C. Gordon 2021-07-27 12:23:46 -04:00
parent a894ce4c14
commit 8d790b10f8
No known key found for this signature in database
GPG Key ID: FA148B892AB48044
1 changed files with 7 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
src/audio/SDL_audiocvt.c
View File

@ -52,6 +52,7 @@
static void SDLCALL static void SDLCALL
SDL_ConvertStereoToMono_SSE3(SDL_AudioCVT * cvt, SDL_AudioFormat format) SDL_ConvertStereoToMono_SSE3(SDL_AudioCVT * cvt, SDL_AudioFormat format)
{ {
const __m128 divby2 = _mm_set1_ps(0.5f);
float *dst = (float *) cvt->buf; float *dst = (float *) cvt->buf;
const float *src = dst; const float *src = dst;
int i = cvt->len_cvt / 8; int i = cvt->len_cvt / 8;
@ -59,15 +60,12 @@ SDL_ConvertStereoToMono_SSE3(SDL_AudioCVT * cvt, SDL_AudioFormat format)
LOG_DEBUG_CONVERT("stereo", "mono (using SSE3)"); LOG_DEBUG_CONVERT("stereo", "mono (using SSE3)");
SDL_assert(format == AUDIO_F32SYS); SDL_assert(format == AUDIO_F32SYS);
/* We can only do this if dst is aligned to 16 bytes; since src is the /* Do SSE blocks as long as we have 16 bytes available.
same pointer and it moves by 2, it can't be forcibly aligned. */ Just use unaligned load/stores, if the memory at runtime is
if ((((size_t) dst) & 15) == 0) { aligned it'll be just as fast on modern processors */
/* Aligned! Do SSE blocks as long as we have 16 bytes available. */ while (i >= 4) { /* 4 * float32 */
const __m128 divby2 = _mm_set1_ps(0.5f); _mm_storeu_ps(dst, _mm_mul_ps(_mm_hadd_ps(_mm_load_ps(src), _mm_loadu_ps(src+4)), divby2));
while (i >= 4) { /* 4 * float32 */ i -= 4; src += 8; dst += 4;
_mm_store_ps(dst, _mm_mul_ps(_mm_hadd_ps(_mm_load_ps(src), _mm_load_ps(src+4)), divby2));
i -= 4; src += 8; dst += 4;
}
} }
/* Finish off any leftovers with scalar operations. */ /* Finish off any leftovers with scalar operations. */