From 25b9c3d0c15e959d544f5d0ac7ea507ea5f6d69f Mon Sep 17 00:00:00 2001 From: Chris Robinson Date: Sun, 10 Nov 2013 05:52:22 -0800 Subject: Apply HRTF coefficient stepping separately --- Alc/mixer_sse.c | 60 +++++++-------------------------------------------------- 1 file changed, 7 insertions(+), 53 deletions(-) (limited to 'Alc/mixer_sse.c') diff --git a/Alc/mixer_sse.c b/Alc/mixer_sse.c index 56a4bdae..719ebd23 100644 --- a/Alc/mixer_sse.c +++ b/Alc/mixer_sse.c @@ -21,65 +21,19 @@ #include "mixer_defs.h" -static inline void ApplyCoeffsStep(ALuint Offset, ALfloat (*restrict Values)[2], - const ALuint IrSize, +static inline void ApplyCoeffsStep(const ALuint IrSize, ALfloat (*restrict Coeffs)[2], - const ALfloat (*restrict CoeffStep)[2], - ALfloat left, ALfloat right) + const ALfloat (*restrict CoeffStep)[2]) { - const __m128 lrlr = { left, right, left, right }; - __m128 coeffs, deltas, imp0, imp1; - __m128 vals = _mm_setzero_ps(); + __m128 coeffs, deltas; ALuint i; - if((Offset&1)) + for(i = 0;i < IrSize;i += 2) { - const ALuint o0 = Offset&HRIR_MASK; - 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_load_ps(&Coeffs[i][0]); + deltas = _mm_load_ps(&CoeffStep[i][0]); coeffs = _mm_add_ps(coeffs, deltas); - 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); - } + _mm_store_ps(&Coeffs[i][0], coeffs); } } -- cgit v1.2.3