Simplify counting for SIMD MixRow functions

This commit is contained in:
Chris Robinson
2018-05-15 23:14:23 -07:00
parent 72e39ba1c9
commit 588a6bcb4f
2 changed files with 20 additions and 16 deletions
+10 -9
View File
@@ -250,13 +250,10 @@ void Mix_Neon(const ALfloat *data, ALsizei OutChans, ALfloat (*restrict OutBuffe
void MixRow_Neon(ALfloat *OutBuffer, const ALfloat *Gains, const ALfloat (*restrict data)[BUFFERSIZE], ALsizei InChans, ALsizei InPos, ALsizei BufferSize)
{
float32x4_t gain4;
ALsizei c;
ASSUME(InChans > 0);
ASSUME(BufferSize > 0);
data = ASSUME_ALIGNED(data, 16);
OutBuffer = ASSUME_ALIGNED(OutBuffer, 16);
for(c = 0;c < InChans;c++)
{
@@ -265,13 +262,17 @@ void MixRow_Neon(ALfloat *OutBuffer, const ALfloat *Gains, const ALfloat (*restr
if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
continue;
gain4 = vdupq_n_f32(gain);
for(;BufferSize-pos > 3;pos += 4)
if(LIKELY(BufferSize > 3))
{
const float32x4_t val4 = vld1q_f32(&data[c][InPos+pos]);
float32x4_t dry4 = vld1q_f32(&OutBuffer[pos]);
dry4 = vmlaq_f32(dry4, val4, gain4);
vst1q_f32(&OutBuffer[pos], dry4);
ALsizei todo = BufferSize >> 2;
float32x4_t gain4 = vdupq_n_f32(gain);
do {
const float32x4_t val4 = vld1q_f32(&data[c][InPos+pos]);
float32x4_t dry4 = vld1q_f32(&OutBuffer[pos]);
dry4 = vmlaq_f32(dry4, val4, gain4);
vst1q_f32(&OutBuffer[pos], dry4);
pos += 4;
} while(--todo);
}
for(;pos < BufferSize;pos++)
OutBuffer[pos] += data[c][InPos+pos]*gain;
+10 -7
View File
@@ -215,7 +215,6 @@ void Mix_SSE(const ALfloat *data, ALsizei OutChans, ALfloat (*restrict OutBuffer
void MixRow_SSE(ALfloat *OutBuffer, const ALfloat *Gains, const ALfloat (*restrict data)[BUFFERSIZE], ALsizei InChans, ALsizei InPos, ALsizei BufferSize)
{
__m128 gain4;
ALsizei c;
ASSUME(InChans > 0);
@@ -228,13 +227,17 @@ void MixRow_SSE(ALfloat *OutBuffer, const ALfloat *Gains, const ALfloat (*restri
if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
continue;
gain4 = _mm_set1_ps(gain);
for(;BufferSize-pos > 3;pos += 4)
if(LIKELY(BufferSize > 3))
{
const __m128 val4 = _mm_load_ps(&data[c][InPos+pos]);
__m128 dry4 = _mm_load_ps(&OutBuffer[pos]);
dry4 = _mm_add_ps(dry4, _mm_mul_ps(val4, gain4));
_mm_store_ps(&OutBuffer[pos], dry4);
ALsizei todo = BufferSize >> 2;
const __m128 gain4 = _mm_set1_ps(gain);
do {
const __m128 val4 = _mm_load_ps(&data[c][InPos+pos]);
__m128 dry4 = _mm_load_ps(&OutBuffer[pos]);
dry4 = _mm_add_ps(dry4, _mm_mul_ps(val4, gain4));
_mm_store_ps(&OutBuffer[pos], dry4);
pos += 4;
} while(--todo);
}
for(;pos < BufferSize;pos++)
OutBuffer[pos] += data[c][InPos+pos]*gain;