kode54
/
cog
2
1
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Add SIMD optimized volume scaling

CQTexperiment
Christopher Snowhill 2022-01-16 00:41:54 -08:00
parent 669c52e63d
commit c640481c25
1 changed files with 45 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
Audio/Chain/ConverterNode.m
View File

@ -18,6 +18,14 @@
#import "lpc.h"
#import <TargetConditionals.h>
#if TARGET_CPU_X86 || TARGET_CPU_X86_64
#include <emmintrin.h>
#elif TARGET_CPU_ARM || TARGET_CPU_ARM64
#include <arm_neon.h>
#endif
void PrintStreamDesc (AudioStreamBasicDescription *inDesc)
{
if (!inDesc) {
@ -262,8 +270,45 @@ static void upmix(float * buffer, int inchannels, int outchannels, size_t count)
void scale_by_volume(float * buffer, size_t count, float volume)
{
if ( volume != 1.0 )
{
#if TARGET_CPU_X86 || TARGET_CPU_X86_64
if ( count >= 8 )
{
__m128 vgf = _mm_set1_ps(volume);
while ( count >= 8 )
{
__m128 input = _mm_loadu_ps(buffer);
__m128 input2 = _mm_loadu_ps(buffer + 4);
__m128 output = _mm_mul_ps(input, vgf);
__m128 output2 = _mm_mul_ps(input2, vgf);
_mm_storeu_ps(buffer + 0, output);
_mm_storeu_ps(buffer + 4, output2);
buffer += 8;
count -= 8;
}
}
#elif TARGET_CPU_ARM || TARGET_CPU_ARM64
if ( count >= 8 )
{
float32x4_t vgf = vdupq_n_f32(volume);
while ( count >= 8 )
{
float32x4x2_t oreg;
float32x4x2_t inreg = vld1q_f32_x2(buffer);
oreg.val[0] = vmulq_f32(inreg.val[0], vgf);
oreg.val[1] = vmulq_f32(inreg.val[1], vgf);
vst1q_f32_x2(buffer, oreg);
buffer += 8;
count -= 8;
}
}
#endif
for (size_t i = 0; i < count; ++i)
buffer[i] *= volume;
}
}
/**