Dima353/openal-soft
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Improve the leftover and realignment mixing loops

Browse Source 
Using a mask helps the compiler recognize that the leftover (any remaining non-
multiple-of-4) and realignment loops will only have 3 iterations at most, which
it can unroll or otherwise make more meaningful optimizations for. Previously
it would try to vectorize and partially unroll the loops, which is wasteful
when there would never be enough to vectorize.
This commit is contained in:
Chris Robinson
2020-05-02 16:42:24 -07:00
parent 3898b77dd6
commit 6823fa2de2
2 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
alc/mixer/mixer_neon.cpp
+4 -4
View File
@@ -219,7 +219,7 @@ void Mix_<NEONTag>(const al::span<const float> InSamples, const al::span<FloatBu
{
const float delta{(Counter > 0) ? 1.0f / static_cast<float>(Counter) : 0.0f};
const auto min_len = minz(Counter, InSamples.size());
const auto aligned_len = minz((min_len+3) & ~size_t{3}, InSamples.size());
const auto aligned_len = minz((min_len+3) & ~size_t{3}, InSamples.size()) - min_len;
for(FloatBufferLine &output : OutBuffer)
{
float *RESTRICT dst{al::assume_aligned<16>(output.data()+OutPos)};
@@ -258,7 +258,7 @@ void Mix_<NEONTag>(const al::span<const float> InSamples, const al::span<FloatBu
step_count = vgetq_lane_f32(step_count4, 0);
}
/* Mix with applying left over gain steps that aren't aligned multiples of 4. */
for(;pos != min_len;++pos)
for(uint_fast32_t leftover{min_len&3};leftover;++pos,--leftover)
{
dst[pos] += InSamples[pos] * (gain + step*step_count);
step_count += 1.0f;
@@ -269,7 +269,7 @@ void Mix_<NEONTag>(const al::span<const float> InSamples, const al::span<FloatBu
gain += step*step_count;
/* Mix until pos is aligned with 4 or the mix is done. */
for(;pos != aligned_len;++pos)
for(uint_fast32_t leftover{aligned_len&3};leftover;++pos,--leftover)
dst[pos] += InSamples[pos] * gain;
}
*CurrentGains = gain;
@@ -289,7 +289,7 @@ void Mix_<NEONTag>(const al::span<const float> InSamples, const al::span<FloatBu
pos += 4;
} while(--todo);
}
for(;pos != InSamples.size();++pos)
for(uint_fast32_t leftover{InSamples.size()&3};leftover;++pos,--leftover)
dst[pos] += InSamples[pos] * gain;
}
}
alc/mixer/mixer_sse.cpp
+4 -4
View File
@@ -192,7 +192,7 @@ void Mix_<SSETag>(const al::span<const float> InSamples, const al::span<FloatBuf
{
const float delta{(Counter > 0) ? 1.0f / static_cast<float>(Counter) : 0.0f};
const auto min_len = minz(Counter, InSamples.size());
const auto aligned_len = minz((min_len+3) & ~size_t{3}, InSamples.size());
const auto aligned_len = minz((min_len+3) & ~size_t{3}, InSamples.size()) - min_len;
for(FloatBufferLine &output : OutBuffer)
{
float *RESTRICT dst{al::assume_aligned<16>(output.data()+OutPos)};
@@ -230,7 +230,7 @@ void Mix_<SSETag>(const al::span<const float> InSamples, const al::span<FloatBuf
step_count = _mm_cvtss_f32(step_count4);
}
/* Mix with applying left over gain steps that aren't aligned multiples of 4. */
for(;pos != min_len;++pos)
for(uint_fast32_t leftover{min_len&3};leftover;++pos,--leftover)
{
dst[pos] += InSamples[pos] * (gain + step*step_count);
step_count += 1.0f;
@@ -241,7 +241,7 @@ void Mix_<SSETag>(const al::span<const float> InSamples, const al::span<FloatBuf
gain += step*step_count;
/* Mix until pos is aligned with 4 or the mix is done. */
for(;pos != aligned_len;++pos)
for(uint_fast32_t leftover{aligned_len&3};leftover;++pos,--leftover)
dst[pos] += InSamples[pos] * gain;
}
*CurrentGains = gain;
@@ -261,7 +261,7 @@ void Mix_<SSETag>(const al::span<const float> InSamples, const al::span<FloatBuf
pos += 4;
} while(--todo);
}
for(;pos != InSamples.size();++pos)
for(uint_fast32_t leftover{InSamples.size()&3};leftover;++pos,--leftover)
dst[pos] += InSamples[pos] * gain;
}
}