AuroraMiddleware/AuroraOpenALSoft
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Revert "Apply HRTF coefficient stepping separately"

Browse Source 
This reverts commit 25b9c3d0c1.

Conflicts:
	Alc/mixer_neon.c

Unfortunately this also undoes the Neon-enhanced ApplyCoeffsStep method.
This commit is contained in:
Chris Robinson 2014-02-23 21:17:09 -08:00
parent c68ce288d0
commit 9a4ded2491
4 changed files with 74 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
Alc/mixer_c.c
View File

@ -46,13 +46,18 @@ DECL_TEMPLATE(cubic32)
#undef DECL_TEMPLATE #undef DECL_TEMPLATE
static inline void ApplyCoeffsStep(const ALuint IrSize, static inline void ApplyCoeffsStep(ALuint Offset, ALfloat (*restrict Values)[2],
const ALuint IrSize,
ALfloat (*restrict Coeffs)[2], ALfloat (*restrict Coeffs)[2],
const ALfloat (*restrict CoeffStep)[2]) const ALfloat (*restrict CoeffStep)[2],
ALfloat left, ALfloat right)
{ {
ALuint c; ALuint c;
for(c = 0;c < IrSize;c++) for(c = 0;c < IrSize;c++)
{ {
const ALuint off = (Offset+c)&HRIR_MASK;
Values[off][0] += Coeffs[c][0] * left;
Values[off][1] += Coeffs[c][1] * right;
Coeffs[c][0] += CoeffStep[c][0]; Coeffs[c][0] += CoeffStep[c][0];
Coeffs[c][1] += CoeffStep[c][1]; Coeffs[c][1] += CoeffStep[c][1];
} }

9
Alc/mixer_inc.c
View File

@ -19,9 +19,11 @@
#define MixDirect_Hrtf MERGE2(MixDirect_Hrtf_,SUFFIX) #define MixDirect_Hrtf MERGE2(MixDirect_Hrtf_,SUFFIX)
static inline void ApplyCoeffsStep(const ALuint irSize, static inline void ApplyCoeffsStep(ALuint Offset, ALfloat (*restrict Values)[2],
const ALuint irSize,
ALfloat (*restrict Coeffs)[2], ALfloat (*restrict Coeffs)[2],
const ALfloat (*restrict CoeffStep)[2]); const ALfloat (*restrict CoeffStep)[2],
ALfloat left, ALfloat right);
static inline void ApplyCoeffs(ALuint Offset, ALfloat (*restrict Values)[2], static inline void ApplyCoeffs(ALuint Offset, ALfloat (*restrict Values)[2],
const ALuint irSize, const ALuint irSize,
ALfloat (*restrict Coeffs)[2], ALfloat (*restrict Coeffs)[2],
@ -91,10 +93,9 @@ void MixDirect_Hrtf(const DirectParams *params, const ALfloat *restrict data, AL
Values[(Offset+IrSize)&HRIR_MASK][1] = 0.0f; Values[(Offset+IrSize)&HRIR_MASK][1] = 0.0f;
Offset++; Offset++;
ApplyCoeffs(Offset, Values, IrSize, Coeffs, left, right); ApplyCoeffsStep(Offset, Values, IrSize, Coeffs, CoeffStep, left, right);
DryBuffer[FrontLeft][OutPos] += Values[Offset&HRIR_MASK][0]; DryBuffer[FrontLeft][OutPos] += Values[Offset&HRIR_MASK][0];
DryBuffer[FrontRight][OutPos] += Values[Offset&HRIR_MASK][1]; DryBuffer[FrontRight][OutPos] += Values[Offset&HRIR_MASK][1];
ApplyCoeffsStep(IrSize, Coeffs, CoeffStep);
OutPos++; OutPos++;
Counter--; Counter--;

15
Alc/mixer_neon.c
View File

@ -10,19 +10,22 @@
#include "alu.h" #include "alu.h"
static inline void ApplyCoeffsStep(const ALuint IrSize, static inline void ApplyCoeffsStep(ALuint Offset, ALfloat (*restrict Values)[2],
const ALuint IrSize,
ALfloat (*restrict Coeffs)[2], ALfloat (*restrict Coeffs)[2],
const ALfloat (*restrict CoeffStep)[2]) const ALfloat (*restrict CoeffStep)[2],
ALfloat left, ALfloat right)
{ {
float32x4_t coeffs, deltas; float32x4_t coeffs, deltas;
ALuint c; ALuint c;
for(c = 0;c < IrSize;c += 2) for(c = 0;c < IrSize;c += 2)
{ {
coeffs = vld1q_f32(&Coeffs[c][0]); const ALuint off = (Offset+c)&HRIR_MASK;
deltas = vld1q_f32(&CoeffStep[c][0]); Values[off][0] += Coeffs[c][0] * left;
coeffs = vaddq_f32(coeffs, deltas); Values[off][1] += Coeffs[c][1] * right;
vst1q_f32(&Coeffs[c][0], coeffs); Coeffs[c][0] += CoeffStep[c][0];
Coeffs[c][1] += CoeffStep[c][1];
} }
} }

60
Alc/mixer_sse.c
View File

@ -21,19 +21,65 @@
#include "mixer_defs.h" #include "mixer_defs.h"
static inline void ApplyCoeffsStep(const ALuint IrSize, static inline void ApplyCoeffsStep(ALuint Offset, ALfloat (*restrict Values)[2],
const ALuint IrSize,
ALfloat (*restrict Coeffs)[2], ALfloat (*restrict Coeffs)[2],
const ALfloat (*restrict CoeffStep)[2]) const ALfloat (*restrict CoeffStep)[2],
ALfloat left, ALfloat right)
{ {
__m128 coeffs, deltas; const __m128 lrlr = { left, right, left, right };
__m128 coeffs, deltas, imp0, imp1;
__m128 vals = _mm_setzero_ps();
ALuint i; ALuint i;
for(i = 0;i < IrSize;i += 2) if((Offset&1))
{ {
coeffs = _mm_load_ps(&Coeffs[i][0]); const ALuint o0 = Offset&HRIR_MASK;
deltas = _mm_load_ps(&CoeffStep[i][0]); const ALuint o1 = (Offset+IrSize-1)&HRIR_MASK;
coeffs = _mm_load_ps(&Coeffs[0][0]);
deltas = _mm_load_ps(&CoeffStep[0][0]);
vals = _mm_loadl_pi(vals, (__m64*)&Values[o0][0]);
imp0 = _mm_mul_ps(lrlr, coeffs);
coeffs = _mm_add_ps(coeffs, deltas); coeffs = _mm_add_ps(coeffs, deltas);
_mm_store_ps(&Coeffs[i][0], coeffs); vals = _mm_add_ps(imp0, vals);
_mm_store_ps(&Coeffs[0][0], coeffs);
_mm_storel_pi((__m64*)&Values[o0][0], vals);
for(i = 1;i < IrSize-1;i += 2)
{
const ALuint o2 = (Offset+i)&HRIR_MASK;
coeffs = _mm_load_ps(&Coeffs[i+1][0]);
deltas = _mm_load_ps(&CoeffStep[i+1][0]);
vals = _mm_load_ps(&Values[o2][0]);
imp1 = _mm_mul_ps(lrlr, coeffs);
coeffs = _mm_add_ps(coeffs, deltas);
imp0 = _mm_shuffle_ps(imp0, imp1, _MM_SHUFFLE(1, 0, 3, 2));
vals = _mm_add_ps(imp0, vals);
_mm_store_ps(&Coeffs[i+1][0], coeffs);
_mm_store_ps(&Values[o2][0], vals);
imp0 = imp1;
}
vals = _mm_loadl_pi(vals, (__m64*)&Values[o1][0]);
imp0 = _mm_movehl_ps(imp0, imp0);
vals = _mm_add_ps(imp0, vals);
_mm_storel_pi((__m64*)&Values[o1][0], vals);
}
else
{
for(i = 0;i < IrSize;i += 2)
{
const ALuint o = (Offset + i)&HRIR_MASK;
coeffs = _mm_load_ps(&Coeffs[i][0]);
deltas = _mm_load_ps(&CoeffStep[i][0]);
vals = _mm_load_ps(&Values[o][0]);
imp0 = _mm_mul_ps(lrlr, coeffs);
coeffs = _mm_add_ps(coeffs, deltas);
vals = _mm_add_ps(imp0, vals);
_mm_store_ps(&Coeffs[i][0], coeffs);
_mm_store_ps(&Values[o][0], vals);
}
} }
} }