From 70bf42392d8a1b5e21e28cbb473d0f630c96d05b Mon Sep 17 00:00:00 2001 From: nmlgc Date: Sat, 26 Aug 2023 21:24:58 +0200 Subject: [PATCH] Fix SSE2 sample swapping in mono expansion. The SSE2 code paths for mono expansion introduced in Version 0.11.15 mixed up the parameters of `_mm_shuffle_ps()`, which in turn caused adjacent PCM frames to be swapped in the channel-expanded output. --- extras/miniaudio_split/miniaudio.c | 4 ++-- miniaudio.h | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/extras/miniaudio_split/miniaudio.c b/extras/miniaudio_split/miniaudio.c index e976cba6..a2042f65 100644 --- a/extras/miniaudio_split/miniaudio.c +++ b/extras/miniaudio_split/miniaudio.c @@ -41417,7 +41417,7 @@ static ma_result ma_channel_map_apply_mono_in_f32(float* MA_RESTRICT pFramesOut, for (iFrame = 0; iFrame < unrolledFrameCount; iFrame += 1) { __m128 in0 = _mm_set1_ps(pFramesIn[iFrame*2 + 0]); __m128 in1 = _mm_set1_ps(pFramesIn[iFrame*2 + 1]); - _mm_storeu_ps(&pFramesOut[iFrame*4 + 0], _mm_shuffle_ps(in1, in0, _MM_SHUFFLE(0, 0, 0, 0))); + _mm_storeu_ps(&pFramesOut[iFrame*4 + 0], _mm_shuffle_ps(in0, in1, _MM_SHUFFLE(0, 0, 0, 0))); } /* Tail. */ @@ -41443,7 +41443,7 @@ static ma_result ma_channel_map_apply_mono_in_f32(float* MA_RESTRICT pFramesOut, __m128 in1 = _mm_set1_ps(pFramesIn[iFrame*2 + 1]); _mm_storeu_ps(&pFramesOut[iFrame*12 + 0], in0); - _mm_storeu_ps(&pFramesOut[iFrame*12 + 4], _mm_shuffle_ps(in1, in0, _MM_SHUFFLE(0, 0, 0, 0))); + _mm_storeu_ps(&pFramesOut[iFrame*12 + 4], _mm_shuffle_ps(in0, in1, _MM_SHUFFLE(0, 0, 0, 0))); _mm_storeu_ps(&pFramesOut[iFrame*12 + 8], in1); } diff --git a/miniaudio.h b/miniaudio.h index c4d24567..376032be 100644 --- a/miniaudio.h +++ b/miniaudio.h @@ -52835,7 +52835,7 @@ static ma_result ma_channel_map_apply_mono_in_f32(float* MA_RESTRICT pFramesOut, for (iFrame = 0; iFrame < unrolledFrameCount; iFrame += 1) { __m128 in0 = _mm_set1_ps(pFramesIn[iFrame*2 + 0]); __m128 in1 = _mm_set1_ps(pFramesIn[iFrame*2 + 1]); - _mm_storeu_ps(&pFramesOut[iFrame*4 + 0], _mm_shuffle_ps(in1, in0, _MM_SHUFFLE(0, 0, 0, 0))); + _mm_storeu_ps(&pFramesOut[iFrame*4 + 0], _mm_shuffle_ps(in0, in1, _MM_SHUFFLE(0, 0, 0, 0))); } /* Tail. */ @@ -52861,7 +52861,7 @@ static ma_result ma_channel_map_apply_mono_in_f32(float* MA_RESTRICT pFramesOut, __m128 in1 = _mm_set1_ps(pFramesIn[iFrame*2 + 1]); _mm_storeu_ps(&pFramesOut[iFrame*12 + 0], in0); - _mm_storeu_ps(&pFramesOut[iFrame*12 + 4], _mm_shuffle_ps(in1, in0, _MM_SHUFFLE(0, 0, 0, 0))); + _mm_storeu_ps(&pFramesOut[iFrame*12 + 4], _mm_shuffle_ps(in0, in1, _MM_SHUFFLE(0, 0, 0, 0))); _mm_storeu_ps(&pFramesOut[iFrame*12 + 8], in1); }