mirror of
https://github.com/love2d/megasource.git
synced 2026-08-20 04:30:45 +02:00
update OpenAL-Soft to 1.24.3.
This commit is contained in:
Sasha Szpakowski
@@ -14,10 +14,12 @@
|
||||
#include "core/bufferline.h"
|
||||
#include "core/cubic_defs.h"
|
||||
#include "core/mixer/hrtfdefs.h"
|
||||
#include "core/resampler_limits.h"
|
||||
#include "defs.h"
|
||||
#include "hrtfbase.h"
|
||||
#include "opthelpers.h"
|
||||
|
||||
struct CTag;
|
||||
struct NEONTag;
|
||||
struct LerpTag;
|
||||
struct CubicTag;
|
||||
@@ -63,60 +65,62 @@ inline float32x4_t set_f4(float l0, float l1, float l2, float l3)
|
||||
return ret;
|
||||
}
|
||||
|
||||
inline void ApplyCoeffs(float2 *RESTRICT Values, const size_t IrSize, const ConstHrirSpan Coeffs,
|
||||
const float left, const float right)
|
||||
inline void ApplyCoeffs(const al::span<float2> Values, const size_t IrSize,
|
||||
const ConstHrirSpan Coeffs, const float left, const float right)
|
||||
{
|
||||
auto dup_samples = [left,right]
|
||||
ASSUME(IrSize >= MinIrLength);
|
||||
ASSUME(IrSize <= HrirLength);
|
||||
|
||||
auto dup_samples = [left,right]() -> float32x4_t
|
||||
{
|
||||
float32x2_t leftright2{vset_lane_f32(right, vmov_n_f32(left), 1)};
|
||||
return vcombine_f32(leftright2, leftright2);
|
||||
};
|
||||
const float32x4_t leftright4{dup_samples()};
|
||||
const auto leftright4 = dup_samples();
|
||||
|
||||
ASSUME(IrSize >= MinIrLength);
|
||||
/* Using a loop here instead of std::transform since some builds seem to
|
||||
* have an issue with accessing an array/span of float32x4_t.
|
||||
*/
|
||||
for(size_t c{0};c < IrSize;c += 2)
|
||||
{
|
||||
float32x4_t vals = vld1q_f32(&Values[c][0]);
|
||||
float32x4_t coefs = vld1q_f32(&Coeffs[c][0]);
|
||||
|
||||
vals = vmlaq_f32(vals, coefs, leftright4);
|
||||
|
||||
auto vals = vld1q_f32(&Values[c][0]);
|
||||
vals = vmlaq_f32(vals, vld1q_f32(&Coeffs[c][0]), leftright4);
|
||||
vst1q_f32(&Values[c][0], vals);
|
||||
}
|
||||
}
|
||||
|
||||
force_inline void MixLine(const al::span<const float> InSamples, float *RESTRICT dst,
|
||||
float &CurrentGain, const float TargetGain, const float delta, const size_t min_len,
|
||||
const size_t aligned_len, size_t Counter)
|
||||
force_inline void MixLine(const al::span<const float> InSamples, const al::span<float> dst,
|
||||
float &CurrentGain, const float TargetGain, const float delta, const size_t fade_len,
|
||||
const size_t realign_len, size_t Counter)
|
||||
{
|
||||
float gain{CurrentGain};
|
||||
const float step{(TargetGain-gain) * delta};
|
||||
const auto step = float{(TargetGain-CurrentGain) * delta};
|
||||
|
||||
size_t pos{0};
|
||||
if(!(std::abs(step) > std::numeric_limits<float>::epsilon()))
|
||||
gain = TargetGain;
|
||||
else
|
||||
auto pos = size_t{0};
|
||||
if(std::abs(step) > std::numeric_limits<float>::epsilon())
|
||||
{
|
||||
float step_count{0.0f};
|
||||
const auto gain = float{CurrentGain};
|
||||
auto step_count = float{0.0f};
|
||||
/* Mix with applying gain steps in aligned multiples of 4. */
|
||||
if(size_t todo{min_len >> 2})
|
||||
if(const size_t todo{fade_len >> 2})
|
||||
{
|
||||
const float32x4_t four4{vdupq_n_f32(4.0f)};
|
||||
const float32x4_t step4{vdupq_n_f32(step)};
|
||||
const float32x4_t gain4{vdupq_n_f32(gain)};
|
||||
float32x4_t step_count4{vdupq_n_f32(0.0f)};
|
||||
step_count4 = vsetq_lane_f32(1.0f, step_count4, 1);
|
||||
step_count4 = vsetq_lane_f32(2.0f, step_count4, 2);
|
||||
step_count4 = vsetq_lane_f32(3.0f, step_count4, 3);
|
||||
const auto four4 = vdupq_n_f32(4.0f);
|
||||
const auto step4 = vdupq_n_f32(step);
|
||||
const auto gain4 = vdupq_n_f32(gain);
|
||||
auto step_count4 = set_f4(0.0f, 1.0f, 2.0f, 3.0f);
|
||||
|
||||
const auto in4 = al::span{reinterpret_cast<const float32x4_t*>(InSamples.data()),
|
||||
InSamples.size()/4}.first(todo);
|
||||
const auto out4 = al::span{reinterpret_cast<float32x4_t*>(dst.data()), dst.size()/4};
|
||||
std::transform(in4.begin(), in4.end(), out4.begin(), out4.begin(),
|
||||
[gain4,step4,four4,&step_count4](const float32x4_t val4, float32x4_t dry4)
|
||||
{
|
||||
/* dry += val * (gain + step*step_count) */
|
||||
dry4 = vmlaq_f32(dry4, val4, vmlaq_f32(gain4, step4, step_count4));
|
||||
step_count4 = vaddq_f32(step_count4, four4);
|
||||
return dry4;
|
||||
});
|
||||
pos += in4.size()*4;
|
||||
|
||||
do {
|
||||
const float32x4_t val4 = vld1q_f32(&InSamples[pos]);
|
||||
float32x4_t dry4 = vld1q_f32(&dst[pos]);
|
||||
dry4 = vmlaq_f32(dry4, val4, vmlaq_f32(gain4, step4, step_count4));
|
||||
step_count4 = vaddq_f32(step_count4, four4);
|
||||
vst1q_f32(&dst[pos], dry4);
|
||||
pos += 4;
|
||||
} while(--todo);
|
||||
/* NOTE: step_count4 now represents the next four counts after the
|
||||
* last four mixed samples, so the lowest element represents the
|
||||
* next step count to apply.
|
||||
@@ -124,43 +128,70 @@ force_inline void MixLine(const al::span<const float> InSamples, float *RESTRICT
|
||||
step_count = vgetq_lane_f32(step_count4, 0);
|
||||
}
|
||||
/* Mix with applying left over gain steps that aren't aligned multiples of 4. */
|
||||
for(size_t leftover{min_len&3};leftover;++pos,--leftover)
|
||||
if(const size_t leftover{fade_len&3})
|
||||
{
|
||||
dst[pos] += InSamples[pos] * (gain + step*step_count);
|
||||
step_count += 1.0f;
|
||||
const auto in = InSamples.subspan(pos, leftover);
|
||||
const auto out = dst.subspan(pos);
|
||||
|
||||
std::transform(in.begin(), in.end(), out.begin(), out.begin(),
|
||||
[gain,step,&step_count](const float val, float dry) noexcept -> float
|
||||
{
|
||||
dry += val * (gain + step*step_count);
|
||||
step_count += 1.0f;
|
||||
return dry;
|
||||
});
|
||||
pos += leftover;
|
||||
}
|
||||
if(pos < Counter)
|
||||
{
|
||||
CurrentGain = gain + step*step_count;
|
||||
return;
|
||||
}
|
||||
if(pos == Counter)
|
||||
gain = TargetGain;
|
||||
else
|
||||
gain += step*step_count;
|
||||
|
||||
/* Mix until pos is aligned with 4 or the mix is done. */
|
||||
for(size_t leftover{aligned_len&3};leftover;++pos,--leftover)
|
||||
dst[pos] += InSamples[pos] * gain;
|
||||
}
|
||||
CurrentGain = gain;
|
||||
if(const size_t leftover{realign_len&3})
|
||||
{
|
||||
const auto in = InSamples.subspan(pos, leftover);
|
||||
const auto out = dst.subspan(pos);
|
||||
|
||||
if(!(std::abs(gain) > GainSilenceThreshold))
|
||||
return;
|
||||
if(size_t todo{(InSamples.size()-pos) >> 2})
|
||||
{
|
||||
const float32x4_t gain4 = vdupq_n_f32(gain);
|
||||
do {
|
||||
const float32x4_t val4 = vld1q_f32(&InSamples[pos]);
|
||||
float32x4_t dry4 = vld1q_f32(&dst[pos]);
|
||||
dry4 = vmlaq_f32(dry4, val4, gain4);
|
||||
vst1q_f32(&dst[pos], dry4);
|
||||
pos += 4;
|
||||
} while(--todo);
|
||||
std::transform(in.begin(), in.end(), out.begin(), out.begin(),
|
||||
[TargetGain](const float val, const float dry) noexcept -> float
|
||||
{ return dry + val*TargetGain; });
|
||||
pos += leftover;
|
||||
}
|
||||
}
|
||||
CurrentGain = TargetGain;
|
||||
|
||||
if(!(std::abs(TargetGain) > GainSilenceThreshold))
|
||||
return;
|
||||
if(const size_t todo{(InSamples.size()-pos) >> 2})
|
||||
{
|
||||
const auto in4 = al::span{reinterpret_cast<const float32x4_t*>(InSamples.data()),
|
||||
InSamples.size()/4}.last(todo);
|
||||
const auto out = dst.subspan(pos);
|
||||
const auto out4 = al::span{reinterpret_cast<float32x4_t*>(out.data()), out.size()/4};
|
||||
|
||||
const auto gain4 = vdupq_n_f32(TargetGain);
|
||||
std::transform(in4.begin(), in4.end(), out4.begin(), out4.begin(),
|
||||
[gain4](const float32x4_t val4, const float32x4_t dry4) -> float32x4_t
|
||||
{ return vmlaq_f32(dry4, val4, gain4); });
|
||||
pos += in4.size()*4;
|
||||
}
|
||||
if(const size_t leftover{(InSamples.size()-pos)&3})
|
||||
{
|
||||
const auto in = InSamples.last(leftover);
|
||||
const auto out = dst.subspan(pos);
|
||||
|
||||
std::transform(in.begin(), in.end(), out.begin(), out.begin(),
|
||||
[TargetGain](const float val, const float dry) noexcept -> float
|
||||
{ return dry + val*TargetGain; });
|
||||
}
|
||||
for(size_t leftover{(InSamples.size()-pos)&3};leftover;++pos,--leftover)
|
||||
dst[pos] += InSamples[pos] * gain;
|
||||
}
|
||||
|
||||
} // namespace
|
||||
|
||||
template<>
|
||||
void Resample_<LerpTag,NEONTag>(const InterpState*, const float *src, uint frac,
|
||||
void Resample_<LerpTag,NEONTag>(const InterpState*, const al::span<const float> src, uint frac,
|
||||
const uint increment, const al::span<float> dst)
|
||||
{
|
||||
ASSUME(frac < MixerFracOne);
|
||||
@@ -169,18 +200,19 @@ void Resample_<LerpTag,NEONTag>(const InterpState*, const float *src, uint frac,
|
||||
const float32x4_t fracOne4 = vdupq_n_f32(1.0f/MixerFracOne);
|
||||
const uint32x4_t fracMask4 = vdupq_n_u32(MixerFracMask);
|
||||
|
||||
alignas(16) std::array<uint,4> pos_, frac_;
|
||||
InitPosArrays(frac, increment, al::span{frac_}, al::span{pos_});
|
||||
alignas(16) std::array<uint,4> pos_{}, frac_{};
|
||||
InitPosArrays(MaxResamplerEdge, frac, increment, al::span{frac_}, al::span{pos_});
|
||||
uint32x4_t frac4 = vld1q_u32(frac_.data());
|
||||
uint32x4_t pos4 = vld1q_u32(pos_.data());
|
||||
|
||||
auto vecout = al::span<float32x4_t>{reinterpret_cast<float32x4_t*>(dst.data()), dst.size()/4};
|
||||
auto vecout = al::span{reinterpret_cast<float32x4_t*>(dst.data()), dst.size()/4};
|
||||
std::generate(vecout.begin(), vecout.end(), [=,&pos4,&frac4]() -> float32x4_t
|
||||
{
|
||||
const uint pos0{vgetq_lane_u32(pos4, 0)};
|
||||
const uint pos1{vgetq_lane_u32(pos4, 1)};
|
||||
const uint pos2{vgetq_lane_u32(pos4, 2)};
|
||||
const uint pos3{vgetq_lane_u32(pos4, 3)};
|
||||
ASSUME(pos0 <= pos1); ASSUME(pos1 <= pos2); ASSUME(pos2 <= pos3);
|
||||
const float32x4_t val1{set_f4(src[pos0], src[pos1], src[pos2], src[pos3])};
|
||||
const float32x4_t val2{set_f4(src[pos0+1_uz], src[pos1+1_uz], src[pos2+1_uz], src[pos3+1_uz])};
|
||||
|
||||
@@ -197,24 +229,26 @@ void Resample_<LerpTag,NEONTag>(const InterpState*, const float *src, uint frac,
|
||||
|
||||
if(size_t todo{dst.size()&3})
|
||||
{
|
||||
src += vgetq_lane_u32(pos4, 0);
|
||||
auto pos = size_t{vgetq_lane_u32(pos4, 0)};
|
||||
frac = vgetq_lane_u32(frac4, 0);
|
||||
|
||||
std::generate(dst.end()-ptrdiff_t(todo), dst.end(), [&src,&frac,increment]
|
||||
const auto out = dst.last(todo);
|
||||
std::generate(out.begin(), out.end(), [&pos,&frac,src,increment]
|
||||
{
|
||||
const float out{lerpf(src[0], src[1], static_cast<float>(frac) * (1.0f/MixerFracOne))};
|
||||
const float output{lerpf(src[pos+0], src[pos+1],
|
||||
static_cast<float>(frac) * (1.0f/MixerFracOne))};
|
||||
|
||||
frac += increment;
|
||||
src += frac>>MixerFracBits;
|
||||
pos += frac>>MixerFracBits;
|
||||
frac &= MixerFracMask;
|
||||
return out;
|
||||
return output;
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
template<>
|
||||
void Resample_<CubicTag,NEONTag>(const InterpState *state, const float *src, uint frac,
|
||||
const uint increment, const al::span<float> dst)
|
||||
void Resample_<CubicTag,NEONTag>(const InterpState *state, const al::span<const float> src,
|
||||
uint frac, const uint increment, const al::span<float> dst)
|
||||
{
|
||||
ASSUME(frac < MixerFracOne);
|
||||
|
||||
@@ -225,23 +259,23 @@ void Resample_<CubicTag,NEONTag>(const InterpState *state, const float *src, uin
|
||||
const float32x4_t fracDiffOne4{vdupq_n_f32(1.0f/CubicPhaseDiffOne)};
|
||||
const uint32x4_t fracDiffMask4{vdupq_n_u32(CubicPhaseDiffMask)};
|
||||
|
||||
alignas(16) std::array<uint,4> pos_, frac_;
|
||||
InitPosArrays(frac, increment, al::span{frac_}, al::span{pos_});
|
||||
alignas(16) std::array<uint,4> pos_{}, frac_{};
|
||||
InitPosArrays(MaxResamplerEdge-1, frac, increment, al::span{frac_}, al::span{pos_});
|
||||
uint32x4_t frac4{vld1q_u32(frac_.data())};
|
||||
uint32x4_t pos4{vld1q_u32(pos_.data())};
|
||||
|
||||
src -= 1;
|
||||
auto vecout = al::span<float32x4_t>{reinterpret_cast<float32x4_t*>(dst.data()), dst.size()/4};
|
||||
std::generate(vecout.begin(), vecout.end(), [=,&pos4,&frac4]() -> float32x4_t
|
||||
auto vecout = al::span{reinterpret_cast<float32x4_t*>(dst.data()), dst.size()/4};
|
||||
std::generate(vecout.begin(), vecout.end(), [=,&pos4,&frac4]
|
||||
{
|
||||
const uint pos0{vgetq_lane_u32(pos4, 0)};
|
||||
const uint pos1{vgetq_lane_u32(pos4, 1)};
|
||||
const uint pos2{vgetq_lane_u32(pos4, 2)};
|
||||
const uint pos3{vgetq_lane_u32(pos4, 3)};
|
||||
const float32x4_t val0{vld1q_f32(src+pos0)};
|
||||
const float32x4_t val1{vld1q_f32(src+pos1)};
|
||||
const float32x4_t val2{vld1q_f32(src+pos2)};
|
||||
const float32x4_t val3{vld1q_f32(src+pos3)};
|
||||
ASSUME(pos0 <= pos1); ASSUME(pos1 <= pos2); ASSUME(pos2 <= pos3);
|
||||
const float32x4_t val0{vld1q_f32(&src[pos0])};
|
||||
const float32x4_t val1{vld1q_f32(&src[pos1])};
|
||||
const float32x4_t val2{vld1q_f32(&src[pos2])};
|
||||
const float32x4_t val3{vld1q_f32(&src[pos3])};
|
||||
|
||||
const uint32x4_t pi4{vshrq_n_u32(frac4, CubicPhaseDiffBits)};
|
||||
const uint pi0{vgetq_lane_u32(pi4, 0)}; ASSUME(pi0 < CubicPhaseCount);
|
||||
@@ -276,10 +310,11 @@ void Resample_<CubicTag,NEONTag>(const InterpState *state, const float *src, uin
|
||||
|
||||
if(const size_t todo{dst.size()&3})
|
||||
{
|
||||
src += vgetq_lane_u32(pos4, 0);
|
||||
auto pos = size_t{vgetq_lane_u32(pos4, 0)};
|
||||
frac = vgetq_lane_u32(frac4, 0);
|
||||
|
||||
std::generate(dst.end()-ptrdiff_t(todo), dst.end(), [&src,&frac,increment,filter]
|
||||
auto out = dst.last(todo);
|
||||
std::generate(out.begin(), out.end(), [&pos,&frac,src,increment,filter]
|
||||
{
|
||||
const uint pi{frac >> CubicPhaseDiffBits}; ASSUME(pi < CubicPhaseCount);
|
||||
const float pf{static_cast<float>(frac&CubicPhaseDiffMask) * (1.0f/CubicPhaseDiffOne)};
|
||||
@@ -287,13 +322,13 @@ void Resample_<CubicTag,NEONTag>(const InterpState *state, const float *src, uin
|
||||
|
||||
const float32x4_t f4{vmlaq_f32(vld1q_f32(filter[pi].mCoeffs.data()), pf4,
|
||||
vld1q_f32(filter[pi].mDeltas.data()))};
|
||||
float32x4_t r4{vmulq_f32(f4, vld1q_f32(src))};
|
||||
float32x4_t r4{vmulq_f32(f4, vld1q_f32(&src[pos]))};
|
||||
|
||||
r4 = vaddq_f32(r4, vrev64q_f32(r4));
|
||||
const float output{vget_lane_f32(vadd_f32(vget_low_f32(r4), vget_high_f32(r4)), 0)};
|
||||
|
||||
frac += increment;
|
||||
src += frac>>MixerFracBits;
|
||||
pos += frac>>MixerFracBits;
|
||||
frac &= MixerFracMask;
|
||||
return output;
|
||||
});
|
||||
@@ -301,31 +336,34 @@ void Resample_<CubicTag,NEONTag>(const InterpState *state, const float *src, uin
|
||||
}
|
||||
|
||||
template<>
|
||||
void Resample_<BSincTag,NEONTag>(const InterpState *state, const float *src, uint frac,
|
||||
const uint increment, const al::span<float> dst)
|
||||
void Resample_<BSincTag,NEONTag>(const InterpState *state, const al::span<const float> src,
|
||||
uint frac, const uint increment, const al::span<float> dst)
|
||||
{
|
||||
const auto &bsinc = std::get<BsincState>(*state);
|
||||
const float *const filter{bsinc.filter};
|
||||
const float32x4_t sf4{vdupq_n_f32(bsinc.sf)};
|
||||
const size_t m{bsinc.m};
|
||||
const auto sf4 = vdupq_n_f32(bsinc.sf);
|
||||
const auto m = size_t{bsinc.m};
|
||||
ASSUME(m > 0);
|
||||
ASSUME(m <= MaxResamplerPadding);
|
||||
ASSUME(frac < MixerFracOne);
|
||||
|
||||
src -= bsinc.l;
|
||||
std::generate(dst.begin(), dst.end(), [&src,&frac,increment,filter,sf4,m]() -> float
|
||||
const auto filter = bsinc.filter.first(4_uz*BSincPhaseCount*m);
|
||||
|
||||
ASSUME(bsinc.l <= MaxResamplerEdge);
|
||||
auto pos = size_t{MaxResamplerEdge-bsinc.l};
|
||||
std::generate(dst.begin(), dst.end(), [&pos,&frac,src,increment,sf4,m,filter]() -> float
|
||||
{
|
||||
// Calculate the phase index and factor.
|
||||
const uint pi{frac >> BSincPhaseDiffBits};
|
||||
const uint pi{frac >> BSincPhaseDiffBits}; ASSUME(pi < BSincPhaseCount);
|
||||
const float pf{static_cast<float>(frac&BSincPhaseDiffMask) * (1.0f/BSincPhaseDiffOne)};
|
||||
|
||||
// Apply the scale and phase interpolated filter.
|
||||
float32x4_t r4{vdupq_n_f32(0.0f)};
|
||||
{
|
||||
const float32x4_t pf4{vdupq_n_f32(pf)};
|
||||
const float *fil{filter + m*pi*2_uz};
|
||||
const float *phd{fil + m};
|
||||
const float *scd{fil + BSincPhaseCount*2_uz*m};
|
||||
const float *spd{scd + m};
|
||||
const auto fil = filter.subspan(2_uz*pi*m);
|
||||
const auto phd = fil.subspan(m);
|
||||
const auto scd = fil.subspan(2_uz*BSincPhaseCount*m);
|
||||
const auto spd = scd.subspan(m);
|
||||
size_t td{m >> 2};
|
||||
size_t j{0u};
|
||||
|
||||
@@ -335,7 +373,7 @@ void Resample_<BSincTag,NEONTag>(const InterpState *state, const float *src, uin
|
||||
vmlaq_f32(vld1q_f32(&fil[j]), sf4, vld1q_f32(&scd[j])),
|
||||
pf4, vmlaq_f32(vld1q_f32(&phd[j]), sf4, vld1q_f32(&spd[j])));
|
||||
/* r += f*src */
|
||||
r4 = vmlaq_f32(r4, f4, vld1q_f32(&src[j]));
|
||||
r4 = vmlaq_f32(r4, f4, vld1q_f32(&src[pos+j]));
|
||||
j += 4;
|
||||
} while(--td);
|
||||
}
|
||||
@@ -343,35 +381,38 @@ void Resample_<BSincTag,NEONTag>(const InterpState *state, const float *src, uin
|
||||
const float output{vget_lane_f32(vadd_f32(vget_low_f32(r4), vget_high_f32(r4)), 0)};
|
||||
|
||||
frac += increment;
|
||||
src += frac>>MixerFracBits;
|
||||
pos += frac>>MixerFracBits;
|
||||
frac &= MixerFracMask;
|
||||
return output;
|
||||
});
|
||||
}
|
||||
|
||||
template<>
|
||||
void Resample_<FastBSincTag,NEONTag>(const InterpState *state, const float *src,
|
||||
void Resample_<FastBSincTag,NEONTag>(const InterpState *state, const al::span<const float> src,
|
||||
uint frac, const uint increment, const al::span<float> dst)
|
||||
{
|
||||
const auto &bsinc = std::get<BsincState>(*state);
|
||||
const float *const filter{bsinc.filter};
|
||||
const size_t m{bsinc.m};
|
||||
const auto m = size_t{bsinc.m};
|
||||
ASSUME(m > 0);
|
||||
ASSUME(m <= MaxResamplerPadding);
|
||||
ASSUME(frac < MixerFracOne);
|
||||
|
||||
src -= bsinc.l;
|
||||
std::generate(dst.begin(), dst.end(), [&src,&frac,increment,filter,m]() -> float
|
||||
const auto filter = bsinc.filter.first(2_uz*BSincPhaseCount*m);
|
||||
|
||||
ASSUME(bsinc.l <= MaxResamplerEdge);
|
||||
auto pos = size_t{MaxResamplerEdge-bsinc.l};
|
||||
std::generate(dst.begin(), dst.end(), [&pos,&frac,src,increment,m,filter]() -> float
|
||||
{
|
||||
// Calculate the phase index and factor.
|
||||
const uint pi{frac >> BSincPhaseDiffBits};
|
||||
const uint pi{frac >> BSincPhaseDiffBits}; ASSUME(pi < BSincPhaseCount);
|
||||
const float pf{static_cast<float>(frac&BSincPhaseDiffMask) * (1.0f/BSincPhaseDiffOne)};
|
||||
|
||||
// Apply the phase interpolated filter.
|
||||
float32x4_t r4{vdupq_n_f32(0.0f)};
|
||||
{
|
||||
const float32x4_t pf4{vdupq_n_f32(pf)};
|
||||
const float *fil{filter + m*pi*2_uz};
|
||||
const float *phd{fil + m};
|
||||
const auto fil = filter.subspan(2_uz*pi*m);
|
||||
const auto phd = fil.subspan(m);
|
||||
size_t td{m >> 2};
|
||||
size_t j{0u};
|
||||
|
||||
@@ -379,7 +420,7 @@ void Resample_<FastBSincTag,NEONTag>(const InterpState *state, const float *src,
|
||||
/* f = fil + pf*phd */
|
||||
const float32x4_t f4 = vmlaq_f32(vld1q_f32(&fil[j]), pf4, vld1q_f32(&phd[j]));
|
||||
/* r += f*src */
|
||||
r4 = vmlaq_f32(r4, f4, vld1q_f32(&src[j]));
|
||||
r4 = vmlaq_f32(r4, f4, vld1q_f32(&src[pos+j]));
|
||||
j += 4;
|
||||
} while(--td);
|
||||
}
|
||||
@@ -387,7 +428,7 @@ void Resample_<FastBSincTag,NEONTag>(const InterpState *state, const float *src,
|
||||
const float output{vget_lane_f32(vadd_f32(vget_low_f32(r4), vget_high_f32(r4)), 0)};
|
||||
|
||||
frac += increment;
|
||||
src += frac>>MixerFracBits;
|
||||
pos += frac>>MixerFracBits;
|
||||
frac &= MixerFracMask;
|
||||
return output;
|
||||
});
|
||||
@@ -395,50 +436,59 @@ void Resample_<FastBSincTag,NEONTag>(const InterpState *state, const float *src,
|
||||
|
||||
|
||||
template<>
|
||||
void MixHrtf_<NEONTag>(const float *InSamples, float2 *AccumSamples, const uint IrSize,
|
||||
const MixHrtfFilter *hrtfparams, const size_t BufferSize)
|
||||
{ MixHrtfBase<ApplyCoeffs>(InSamples, AccumSamples, IrSize, hrtfparams, BufferSize); }
|
||||
void MixHrtf_<NEONTag>(const al::span<const float> InSamples, const al::span<float2> AccumSamples,
|
||||
const uint IrSize, const MixHrtfFilter *hrtfparams, const size_t SamplesToDo)
|
||||
{ MixHrtfBase<ApplyCoeffs>(InSamples, AccumSamples, IrSize, hrtfparams, SamplesToDo); }
|
||||
|
||||
template<>
|
||||
void MixHrtfBlend_<NEONTag>(const float *InSamples, float2 *AccumSamples, const uint IrSize,
|
||||
const HrtfFilter *oldparams, const MixHrtfFilter *newparams, const size_t BufferSize)
|
||||
void MixHrtfBlend_<NEONTag>(const al::span<const float> InSamples,
|
||||
const al::span<float2> AccumSamples, const uint IrSize, const HrtfFilter *oldparams,
|
||||
const MixHrtfFilter *newparams, const size_t SamplesToDo)
|
||||
{
|
||||
MixHrtfBlendBase<ApplyCoeffs>(InSamples, AccumSamples, IrSize, oldparams, newparams,
|
||||
BufferSize);
|
||||
SamplesToDo);
|
||||
}
|
||||
|
||||
template<>
|
||||
void MixDirectHrtf_<NEONTag>(const FloatBufferSpan LeftOut, const FloatBufferSpan RightOut,
|
||||
const al::span<const FloatBufferLine> InSamples, float2 *AccumSamples,
|
||||
const al::span<float,BufferLineSize> TempBuf, HrtfChannelState *ChanState, const size_t IrSize,
|
||||
const size_t BufferSize)
|
||||
const al::span<const FloatBufferLine> InSamples, const al::span<float2> AccumSamples,
|
||||
const al::span<float,BufferLineSize> TempBuf, const al::span<HrtfChannelState> ChanState,
|
||||
const size_t IrSize, const size_t SamplesToDo)
|
||||
{
|
||||
MixDirectHrtfBase<ApplyCoeffs>(LeftOut, RightOut, InSamples, AccumSamples, TempBuf, ChanState,
|
||||
IrSize, BufferSize);
|
||||
IrSize, SamplesToDo);
|
||||
}
|
||||
|
||||
|
||||
template<>
|
||||
void Mix_<NEONTag>(const al::span<const float> InSamples, const al::span<FloatBufferLine> OutBuffer,
|
||||
float *CurrentGains, const float *TargetGains, const size_t Counter, const size_t OutPos)
|
||||
void Mix_<NEONTag>(const al::span<const float> InSamples,const al::span<FloatBufferLine> OutBuffer,
|
||||
const al::span<float> CurrentGains, const al::span<const float> TargetGains,
|
||||
const size_t Counter, const size_t OutPos)
|
||||
{
|
||||
const float delta{(Counter > 0) ? 1.0f / static_cast<float>(Counter) : 0.0f};
|
||||
const auto min_len = std::min(Counter, InSamples.size());
|
||||
const auto aligned_len = std::min((min_len+3_uz) & ~3_uz, InSamples.size()) - min_len;
|
||||
if((OutPos&3) != 0) UNLIKELY
|
||||
return Mix_<CTag>(InSamples, OutBuffer, CurrentGains, TargetGains, Counter, OutPos);
|
||||
|
||||
const float delta{(Counter > 0) ? 1.0f / static_cast<float>(Counter) : 0.0f};
|
||||
const auto fade_len = std::min(Counter, InSamples.size());
|
||||
const auto realign_len = std::min((fade_len+3_uz) & ~3_uz, InSamples.size()) - fade_len;
|
||||
|
||||
auto curgains = CurrentGains.begin();
|
||||
auto targetgains = TargetGains.cbegin();
|
||||
for(FloatBufferLine &output : OutBuffer)
|
||||
MixLine(InSamples, al::assume_aligned<16>(output.data()+OutPos), *CurrentGains++,
|
||||
*TargetGains++, delta, min_len, aligned_len, Counter);
|
||||
MixLine(InSamples, al::span{output}.subspan(OutPos), *curgains++, *targetgains++, delta,
|
||||
fade_len, realign_len, Counter);
|
||||
}
|
||||
|
||||
template<>
|
||||
void Mix_<NEONTag>(const al::span<const float> InSamples, float *OutBuffer, float &CurrentGain,
|
||||
const float TargetGain, const size_t Counter)
|
||||
void Mix_<NEONTag>(const al::span<const float> InSamples, const al::span<float> OutBuffer,
|
||||
float &CurrentGain, const float TargetGain, const size_t Counter)
|
||||
{
|
||||
const float delta{(Counter > 0) ? 1.0f / static_cast<float>(Counter) : 0.0f};
|
||||
const auto min_len = std::min(Counter, InSamples.size());
|
||||
const auto aligned_len = std::min((min_len+3_uz) & ~3_uz, InSamples.size()) - min_len;
|
||||
if((reinterpret_cast<uintptr_t>(OutBuffer.data())&15) != 0) UNLIKELY
|
||||
return Mix_<CTag>(InSamples, OutBuffer, CurrentGain, TargetGain, Counter);
|
||||
|
||||
MixLine(InSamples, al::assume_aligned<16>(OutBuffer), CurrentGain, TargetGain, delta, min_len,
|
||||
aligned_len, Counter);
|
||||
const float delta{(Counter > 0) ? 1.0f / static_cast<float>(Counter) : 0.0f};
|
||||
const auto fade_len = std::min(Counter, InSamples.size());
|
||||
const auto realign_len = std::min((fade_len+3_uz) & ~3_uz, InSamples.size()) - fade_len;
|
||||
|
||||
MixLine(InSamples, OutBuffer, CurrentGain, TargetGain, delta, fade_len, realign_len, Counter);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user