Emulator-Archive/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2024-11-23 13:30:02 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add NEON code too for the atrac3 hotspot

Browse Source
This commit is contained in:
Henrik Rydgård 2024-04-13 10:35:09 +02:00
parent 7b6b7ebe3c
commit 857584532f
1 changed files with 20 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
ext/at3_standalone/atrac3plusdsp.cpp
View File

@ -663,6 +663,26 @@ void ff_atrac3p_ipqf(FFTContext *dct_ctx, Atrac3pIPQFChannelCtx *hist,
_mm_storeu_ps(outp + 12, _mm_add_ps(_mm_loadu_ps(outp + 12), _mm_add_ps( _mm_storeu_ps(outp + 12, _mm_add_ps(_mm_loadu_ps(outp + 12), _mm_add_ps(
_mm_mul_ps(_mm_reverse(_mm_loadu_ps(buf1)), _mm_loadu_ps(coeffs1 + 12)), _mm_mul_ps(_mm_reverse(_mm_loadu_ps(buf1)), _mm_loadu_ps(coeffs1 + 12)),
_mm_mul_ps(_mm_reverse(_mm_loadu_ps(buf2)), _mm_loadu_ps(coeffs2 + 12))))); _mm_mul_ps(_mm_reverse(_mm_loadu_ps(buf2)), _mm_loadu_ps(coeffs2 + 12)))));
#elif PPSSPP_ARCH(ARM_NEON)
auto vreverseq_f32 = [](float32x4_t x) -> float32x4_t {
float32x4_t rev = vrev64q_f32(x);
float32x2_t high = vget_high_f32(rev); //{4,3}
float32x2_t low = vget_low_f32(rev); //{1,2}
return vcombine_f32(high, low); //{4,3,2,1}
};
vst1q_f32(outp, vaddq_f32(vld1q_f32(outp), vaddq_f32(
vmulq_f32(vld1q_f32(buf1), vld1q_f32(coeffs1)),
vmulq_f32(vld1q_f32(buf2), vld1q_f32(coeffs2)))));
vst1q_f32(outp + 4, vaddq_f32(vld1q_f32(outp + 4), vaddq_f32(
vmulq_f32(vld1q_f32(buf1 + 4), vld1q_f32(coeffs1 + 4)),
vmulq_f32(vld1q_f32(buf2 + 4), vld1q_f32(coeffs2 + 4)))));
vst1q_f32(outp + 8, vaddq_f32(vld1q_f32(outp + 8), vaddq_f32(
vmulq_f32(vreverseq_f32(vld1q_f32(buf1 + 4)), vld1q_f32(coeffs1 + 8)),
vmulq_f32(vreverseq_f32(vld1q_f32(buf2 + 4)), vld1q_f32(coeffs2 + 8)))));
vst1q_f32(outp + 12, vaddq_f32(vld1q_f32(outp + 12), vaddq_f32(
vmulq_f32(vreverseq_f32(vld1q_f32(buf1)), vld1q_f32(coeffs1 + 12)),
vmulq_f32(vreverseq_f32(vld1q_f32(buf2)), vld1q_f32(coeffs2 + 12)))));
#else #else
for (i = 0; i < 8; i++) { for (i = 0; i < 8; i++) {
outp[i] += buf1[i] * coeffs1[i] + buf2[i] * coeffs2[i]; outp[i] += buf1[i] * coeffs1[i] + buf2[i] * coeffs2[i];