aboutsummaryrefslogtreecommitdiffstats
path: root/core/mixer
diff options
context:
space:
mode:
Diffstat (limited to 'core/mixer')
-rw-r--r--core/mixer/defs.h9
-rw-r--r--core/mixer/hrtfbase.h12
-rw-r--r--core/mixer/hrtfdefs.h5
-rw-r--r--core/mixer/mixer_c.cpp18
-rw-r--r--core/mixer/mixer_neon.cpp18
-rw-r--r--core/mixer/mixer_sse.cpp18
-rw-r--r--core/mixer/mixer_sse2.cpp8
7 files changed, 46 insertions, 42 deletions
diff --git a/core/mixer/defs.h b/core/mixer/defs.h
index acf60350..ba304f22 100644
--- a/core/mixer/defs.h
+++ b/core/mixer/defs.h
@@ -6,6 +6,7 @@
#include "alspan.h"
#include "core/bufferline.h"
+#include "core/resampler_limits.h"
struct HrtfChannelState;
struct HrtfFilter;
@@ -19,12 +20,6 @@ constexpr int MixerFracBits{12};
constexpr int MixerFracOne{1 << MixerFracBits};
constexpr int MixerFracMask{MixerFracOne - 1};
-/* Maximum number of samples to pad on the ends of a buffer for resampling.
- * Note that the padding is symmetric (half at the beginning and half at the
- * end)!
- */
-constexpr int MaxResamplerPadding{48};
-
constexpr float GainSilenceThreshold{0.00001f}; /* -100dB */
@@ -80,7 +75,7 @@ template<typename InstTag>
void MixHrtfBlend_(const float *InSamples, float2 *AccumSamples, const uint IrSize,
const HrtfFilter *oldparams, const MixHrtfFilter *newparams, const size_t BufferSize);
template<typename InstTag>
-void MixDirectHrtf_(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
+void MixDirectHrtf_(const FloatBufferSpan LeftOut, const FloatBufferSpan RightOut,
const al::span<const FloatBufferLine> InSamples, float2 *AccumSamples,
float *TempBuf, HrtfChannelState *ChanState, const size_t IrSize, const size_t BufferSize);
diff --git a/core/mixer/hrtfbase.h b/core/mixer/hrtfbase.h
index 7419f960..79b09a3d 100644
--- a/core/mixer/hrtfbase.h
+++ b/core/mixer/hrtfbase.h
@@ -12,7 +12,7 @@
using uint = unsigned int;
using ApplyCoeffsT = void(&)(float2 *RESTRICT Values, const size_t irSize,
- const HrirArray &Coeffs, const float left, const float right);
+ const ConstHrirSpan Coeffs, const float left, const float right);
template<ApplyCoeffsT ApplyCoeffs>
inline void MixHrtfBase(const float *InSamples, float2 *RESTRICT AccumSamples, const size_t IrSize,
@@ -20,7 +20,7 @@ inline void MixHrtfBase(const float *InSamples, float2 *RESTRICT AccumSamples, c
{
ASSUME(BufferSize > 0);
- const HrirArray &Coeffs = *hrtfparams->Coeffs;
+ const ConstHrirSpan Coeffs{hrtfparams->Coeffs};
const float gainstep{hrtfparams->GainStep};
const float gain{hrtfparams->Gain};
@@ -45,9 +45,9 @@ inline void MixHrtfBlendBase(const float *InSamples, float2 *RESTRICT AccumSampl
{
ASSUME(BufferSize > 0);
- const auto &OldCoeffs = oldparams->Coeffs;
+ const ConstHrirSpan OldCoeffs{oldparams->Coeffs};
const float oldGainStep{oldparams->Gain / static_cast<float>(BufferSize)};
- const auto &NewCoeffs = *newparams->Coeffs;
+ const ConstHrirSpan NewCoeffs{newparams->Coeffs};
const float newGainStep{newparams->GainStep};
if LIKELY(oldparams->Gain > GainSilenceThreshold)
@@ -84,7 +84,7 @@ inline void MixHrtfBlendBase(const float *InSamples, float2 *RESTRICT AccumSampl
}
template<ApplyCoeffsT ApplyCoeffs>
-inline void MixDirectHrtfBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
+inline void MixDirectHrtfBase(const FloatBufferSpan LeftOut, const FloatBufferSpan RightOut,
const al::span<const FloatBufferLine> InSamples, float2 *RESTRICT AccumSamples,
float *TempBuf, HrtfChannelState *ChanState, const size_t IrSize, const size_t BufferSize)
{
@@ -133,7 +133,7 @@ inline void MixDirectHrtfBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOu
ChanState->mSplitter.processHfScale(tempbuf, ChanState->mHfScale);
/* Now apply the HRIR coefficients to this channel. */
- const auto &Coeffs = ChanState->mCoeffs;
+ const ConstHrirSpan Coeffs{ChanState->mCoeffs};
for(size_t i{0u};i < BufferSize;++i)
{
const float insample{tempbuf[i]};
diff --git a/core/mixer/hrtfdefs.h b/core/mixer/hrtfdefs.h
index 89a9bb8d..7046a31e 100644
--- a/core/mixer/hrtfdefs.h
+++ b/core/mixer/hrtfdefs.h
@@ -3,6 +3,7 @@
#include <array>
+#include "alspan.h"
#include "core/ambidefs.h"
#include "core/bufferline.h"
#include "core/filters/splitter.h"
@@ -28,9 +29,11 @@ constexpr uint MinIrLength{8};
constexpr uint HrtfDirectDelay{256};
using HrirArray = std::array<float2,HrirLength>;
+using HrirSpan = al::span<float2,HrirLength>;
+using ConstHrirSpan = al::span<const float2,HrirLength>;
struct MixHrtfFilter {
- const HrirArray *Coeffs;
+ const ConstHrirSpan Coeffs;
uint2 Delay;
float Gain;
float GainStep;
diff --git a/core/mixer/mixer_c.cpp b/core/mixer/mixer_c.cpp
index ff9538a4..f82f7dd1 100644
--- a/core/mixer/mixer_c.cpp
+++ b/core/mixer/mixer_c.cpp
@@ -32,15 +32,16 @@ inline float do_cubic(const InterpState&, const float *RESTRICT vals, const uint
inline float do_bsinc(const InterpState &istate, const float *RESTRICT vals, const uint frac)
{
const size_t m{istate.bsinc.m};
+ ASSUME(m > 0);
// Calculate the phase index and factor.
const uint pi{frac >> FracPhaseBitDiff};
const float pf{static_cast<float>(frac & (FracPhaseDiffOne-1)) * (1.0f/FracPhaseDiffOne)};
- const float *fil{istate.bsinc.filter + m*pi*4};
- const float *phd{fil + m};
- const float *scd{phd + m};
- const float *spd{scd + m};
+ const float *RESTRICT fil{istate.bsinc.filter + m*pi*2};
+ const float *RESTRICT phd{fil + m};
+ const float *RESTRICT scd{fil + BSincPhaseCount*2*m};
+ const float *RESTRICT spd{scd + m};
// Apply the scale and phase interpolated filter.
float r{0.0f};
@@ -51,13 +52,14 @@ inline float do_bsinc(const InterpState &istate, const float *RESTRICT vals, con
inline float do_fastbsinc(const InterpState &istate, const float *RESTRICT vals, const uint frac)
{
const size_t m{istate.bsinc.m};
+ ASSUME(m > 0);
// Calculate the phase index and factor.
const uint pi{frac >> FracPhaseBitDiff};
const float pf{static_cast<float>(frac & (FracPhaseDiffOne-1)) * (1.0f/FracPhaseDiffOne)};
- const float *fil{istate.bsinc.filter + m*pi*4};
- const float *phd{fil + m};
+ const float *RESTRICT fil{istate.bsinc.filter + m*pi*2};
+ const float *RESTRICT phd{fil + m};
// Apply the phase interpolated filter.
float r{0.0f};
@@ -83,7 +85,7 @@ float *DoResample(const InterpState *state, float *RESTRICT src, uint frac, uint
return dst.data();
}
-inline void ApplyCoeffs(float2 *RESTRICT Values, const size_t IrSize, const HrirArray &Coeffs,
+inline void ApplyCoeffs(float2 *RESTRICT Values, const size_t IrSize, const ConstHrirSpan Coeffs,
const float left, const float right)
{
ASSUME(IrSize >= MinIrLength);
@@ -149,7 +151,7 @@ void MixHrtfBlend_<CTag>(const float *InSamples, float2 *AccumSamples, const uin
}
template<>
-void MixDirectHrtf_<CTag>(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
+void MixDirectHrtf_<CTag>(const FloatBufferSpan LeftOut, const FloatBufferSpan RightOut,
const al::span<const FloatBufferLine> InSamples, float2 *AccumSamples,
float *TempBuf, HrtfChannelState *ChanState, const size_t IrSize, const size_t BufferSize)
{
diff --git a/core/mixer/mixer_neon.cpp b/core/mixer/mixer_neon.cpp
index f3e5f130..a3afdc6b 100644
--- a/core/mixer/mixer_neon.cpp
+++ b/core/mixer/mixer_neon.cpp
@@ -34,7 +34,7 @@ inline float32x4_t set_f4(float l0, float l1, float l2, float l3)
constexpr uint FracPhaseBitDiff{MixerFracBits - BSincPhaseBits};
constexpr uint FracPhaseDiffOne{1 << FracPhaseBitDiff};
-inline void ApplyCoeffs(float2 *RESTRICT Values, const size_t IrSize, const HrirArray &Coeffs,
+inline void ApplyCoeffs(float2 *RESTRICT Values, const size_t IrSize, const ConstHrirSpan Coeffs,
const float left, const float right)
{
float32x4_t leftright4;
@@ -118,6 +118,7 @@ float *Resample_<BSincTag,NEONTag>(const InterpState *state, float *RESTRICT src
const float *const filter{state->bsinc.filter};
const float32x4_t sf4{vdupq_n_f32(state->bsinc.sf)};
const size_t m{state->bsinc.m};
+ ASSUME(m > 0);
src -= state->bsinc.l;
for(float &out_sample : dst)
@@ -130,10 +131,10 @@ float *Resample_<BSincTag,NEONTag>(const InterpState *state, float *RESTRICT src
float32x4_t r4{vdupq_n_f32(0.0f)};
{
const float32x4_t pf4{vdupq_n_f32(pf)};
- const float *fil{filter + m*pi*4};
- const float *phd{fil + m};
- const float *scd{phd + m};
- const float *spd{scd + m};
+ const float *RESTRICT fil{filter + m*pi*2};
+ const float *RESTRICT phd{fil + m};
+ const float *RESTRICT scd{fil + BSincPhaseCount*2*m};
+ const float *RESTRICT spd{scd + m};
size_t td{m >> 2};
size_t j{0u};
@@ -163,6 +164,7 @@ float *Resample_<FastBSincTag,NEONTag>(const InterpState *state, float *RESTRICT
{
const float *const filter{state->bsinc.filter};
const size_t m{state->bsinc.m};
+ ASSUME(m > 0);
src -= state->bsinc.l;
for(float &out_sample : dst)
@@ -175,8 +177,8 @@ float *Resample_<FastBSincTag,NEONTag>(const InterpState *state, float *RESTRICT
float32x4_t r4{vdupq_n_f32(0.0f)};
{
const float32x4_t pf4{vdupq_n_f32(pf)};
- const float *fil{filter + m*pi*4};
- const float *phd{fil + m};
+ const float *RESTRICT fil{filter + m*pi*2};
+ const float *RESTRICT phd{fil + m};
size_t td{m >> 2};
size_t j{0u};
@@ -213,7 +215,7 @@ void MixHrtfBlend_<NEONTag>(const float *InSamples, float2 *AccumSamples, const
}
template<>
-void MixDirectHrtf_<NEONTag>(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
+void MixDirectHrtf_<NEONTag>(const FloatBufferSpan LeftOut, const FloatBufferSpan RightOut,
const al::span<const FloatBufferLine> InSamples, float2 *AccumSamples,
float *TempBuf, HrtfChannelState *ChanState, const size_t IrSize, const size_t BufferSize)
{
diff --git a/core/mixer/mixer_sse.cpp b/core/mixer/mixer_sse.cpp
index c0fd8fa1..3cfb00a5 100644
--- a/core/mixer/mixer_sse.cpp
+++ b/core/mixer/mixer_sse.cpp
@@ -26,7 +26,7 @@ constexpr uint FracPhaseDiffOne{1 << FracPhaseBitDiff};
#define MLA4(x, y, z) _mm_add_ps(x, _mm_mul_ps(y, z))
-inline void ApplyCoeffs(float2 *RESTRICT Values, const size_t IrSize, const HrirArray &Coeffs,
+inline void ApplyCoeffs(float2 *RESTRICT Values, const size_t IrSize, const ConstHrirSpan Coeffs,
const float left, const float right)
{
const __m128 lrlr{_mm_setr_ps(left, right, left, right)};
@@ -82,6 +82,7 @@ float *Resample_<BSincTag,SSETag>(const InterpState *state, float *RESTRICT src,
const float *const filter{state->bsinc.filter};
const __m128 sf4{_mm_set1_ps(state->bsinc.sf)};
const size_t m{state->bsinc.m};
+ ASSUME(m > 0);
src -= state->bsinc.l;
for(float &out_sample : dst)
@@ -94,10 +95,10 @@ float *Resample_<BSincTag,SSETag>(const InterpState *state, float *RESTRICT src,
__m128 r4{_mm_setzero_ps()};
{
const __m128 pf4{_mm_set1_ps(pf)};
- const float *fil{filter + m*pi*4};
- const float *phd{fil + m};
- const float *scd{phd + m};
- const float *spd{scd + m};
+ const float *RESTRICT fil{filter + m*pi*2};
+ const float *RESTRICT phd{fil + m};
+ const float *RESTRICT scd{fil + BSincPhaseCount*2*m};
+ const float *RESTRICT spd{scd + m};
size_t td{m >> 2};
size_t j{0u};
@@ -128,6 +129,7 @@ float *Resample_<FastBSincTag,SSETag>(const InterpState *state, float *RESTRICT
{
const float *const filter{state->bsinc.filter};
const size_t m{state->bsinc.m};
+ ASSUME(m > 0);
src -= state->bsinc.l;
for(float &out_sample : dst)
@@ -140,8 +142,8 @@ float *Resample_<FastBSincTag,SSETag>(const InterpState *state, float *RESTRICT
__m128 r4{_mm_setzero_ps()};
{
const __m128 pf4{_mm_set1_ps(pf)};
- const float *fil{filter + m*pi*4};
- const float *phd{fil + m};
+ const float *RESTRICT fil{filter + m*pi*2};
+ const float *RESTRICT phd{fil + m};
size_t td{m >> 2};
size_t j{0u};
@@ -179,7 +181,7 @@ void MixHrtfBlend_<SSETag>(const float *InSamples, float2 *AccumSamples, const u
}
template<>
-void MixDirectHrtf_<SSETag>(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
+void MixDirectHrtf_<SSETag>(const FloatBufferSpan LeftOut, const FloatBufferSpan RightOut,
const al::span<const FloatBufferLine> InSamples, float2 *AccumSamples,
float *TempBuf, HrtfChannelState *ChanState, const size_t IrSize, const size_t BufferSize)
{
diff --git a/core/mixer/mixer_sse2.cpp b/core/mixer/mixer_sse2.cpp
index f91d5dcd..99d04210 100644
--- a/core/mixer/mixer_sse2.cpp
+++ b/core/mixer/mixer_sse2.cpp
@@ -52,10 +52,10 @@ float *Resample_<LerpTag,SSE2Tag>(const InterpState*, float *RESTRICT src, uint
auto dst_iter = dst.begin();
for(size_t todo{dst.size()>>2};todo;--todo)
{
- const int pos0{_mm_cvtsi128_si32(_mm_shuffle_epi32(pos4, _MM_SHUFFLE(0, 0, 0, 0)))};
- const int pos1{_mm_cvtsi128_si32(_mm_shuffle_epi32(pos4, _MM_SHUFFLE(1, 1, 1, 1)))};
- const int pos2{_mm_cvtsi128_si32(_mm_shuffle_epi32(pos4, _MM_SHUFFLE(2, 2, 2, 2)))};
- const int pos3{_mm_cvtsi128_si32(_mm_shuffle_epi32(pos4, _MM_SHUFFLE(3, 3, 3, 3)))};
+ const int pos0{_mm_cvtsi128_si32(pos4)};
+ const int pos1{_mm_cvtsi128_si32(_mm_srli_si128(pos4, 4))};
+ const int pos2{_mm_cvtsi128_si32(_mm_srli_si128(pos4, 8))};
+ const int pos3{_mm_cvtsi128_si32(_mm_srli_si128(pos4, 12))};
const __m128 val1{_mm_setr_ps(src[pos0 ], src[pos1 ], src[pos2 ], src[pos3 ])};
const __m128 val2{_mm_setr_ps(src[pos0+1], src[pos1+1], src[pos2+1], src[pos3+1])};
generated by cgit v1.2.3 (git 2.39.1) at 2025-03-31 16:36:10 +0000