Emulator-Archive/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2024-10-06 18:33:34 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Simplify the inner loop of ff_atrac3p_ipqf (prepare for SIMD)

Browse Source
This commit is contained in:
Henrik Rydgård 2024-04-13 09:40:16 +02:00
parent effae82208
commit 52111103b8
3 changed files with 19 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
.editorconfig
View File

@ -29,6 +29,10 @@ charset = utf-8-bom
[Windows/{aboutbox.rc,version.rc}]
charset = utf-8
[ext/at3_standalone/**.{cpp,h}]
indent_style = space
indent_size = 4
[libretro/**.{cpp,h}]
indent_style = space
indent_size = 3

5
ext/at3_standalone/at3_decoders.h
View File

@ -2,6 +2,11 @@
#include <cstdint>
// Notes
//
// Performance-wise, these are OK.
// For Atrac3+, the bottleneck is two functions: decode_qu_spectra and ff_atrac3p_ipqf. At least the latter is quite SIMD-able.
// The full external API for the standalone Atrac3/3+ decoder.
struct ATRAC3Context;

14
ext/at3_standalone/atrac3plusdsp.cpp
View File

@ -624,12 +624,18 @@ void ff_atrac3p_ipqf(FFTContext *dct_ctx, Atrac3pIPQFChannelCtx *hist,
pos_next = mod23_lut[pos_now + 2]; // pos_next = (pos_now + 1) % 23;
for (t = 0; t < ATRAC3P_PQF_FIR_LEN; t++) {
const float *buf1 = hist->buf1[pos_now];
const float *buf2 = hist->buf2[pos_next];
const float *coeffs1 = ipqf_coeffs1[t];
const float *coeffs2 = ipqf_coeffs2[t];
float *outp = out + s * 16;
for (i = 0; i < 8; i++) {
out[s * 16 + i + 0] += hist->buf1[pos_now][i] * ipqf_coeffs1[t][i] +
hist->buf2[pos_next][i] * ipqf_coeffs2[t][i];
out[s * 16 + i + 8] += hist->buf1[pos_now][7 - i] * ipqf_coeffs1[t][i + 8] +
hist->buf2[pos_next][7 - i] * ipqf_coeffs2[t][i + 8];
outp[i] += buf1[i] * coeffs1[i] + buf2[i] * coeffs2[i];
}
for (i = 0; i < 8; i++) {
outp[i + 8] += buf1[7 - i] * coeffs1[i + 8] + buf2[7 - i] * coeffs2[i + 8];
}
pos_now = mod23_lut[pos_next + 2]; // pos_now = (pos_now + 2) % 23;
pos_next = mod23_lut[pos_now + 2]; // pos_next = (pos_next + 2) % 23;