RPCS3/soundtouch
1
0
Fork 0
mirror of https://github.com/RPCS3/soundtouch.git synced 2024-09-20 00:11:38 +02:00
Code Issues Projects Releases Wiki Activity

Cleaned comments

Browse Source
This commit is contained in:
oparviai 2011-07-16 12:27:28 +00:00
parent 3b35f34ab8
commit 6a297adf2e
1 changed files with 0 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

82
source/SoundTouch/sse_optimized.cpp
View File

@ -423,88 +423,6 @@ uint FIRFilterSSE::evaluateFilterStereo(float *dest, const float *source, uint n
dest += 4;
}
*/
/* Similar routine in assembly, again obsoleted due to maintainability
_asm
{
// Very important note: data in 'src' _must_ be aligned to
// 16-byte boundary!
mov edx, count
mov ebx, dword ptr src
mov eax, dword ptr dest
shr edx, 1
loop1:
// "outer loop" : during each round 2*2 output samples are calculated
// give prefetch hints to CPU of what data are to be needed soonish
prefetcht0 [ebx]
prefetcht0 [filterCoeffsLocal]
mov esi, ebx
mov edi, filterCoeffsLocal
xorps xmm0, xmm0
xorps xmm1, xmm1
mov ecx, lengthLocal
loop2:
// "inner loop" : during each round eight FIR filter taps are evaluated for 2*2 samples
prefetcht0 [esi + 32] // give a prefetch hint to CPU what data are to be needed soonish
prefetcht0 [edi + 32] // give a prefetch hint to CPU what data are to be needed soonish
movups xmm2, [esi] // possibly unaligned load
movups xmm3, [esi + 8] // possibly unaligned load
mulps xmm2, [edi]
mulps xmm3, [edi]
addps xmm0, xmm2
addps xmm1, xmm3
movups xmm4, [esi + 16] // possibly unaligned load
movups xmm5, [esi + 24] // possibly unaligned load
mulps xmm4, [edi + 16]
mulps xmm5, [edi + 16]
addps xmm0, xmm4
addps xmm1, xmm5
prefetcht0 [esi + 64] // give a prefetch hint to CPU what data are to be needed soonish
prefetcht0 [edi + 64] // give a prefetch hint to CPU what data are to be needed soonish
movups xmm6, [esi + 32] // possibly unaligned load
movups xmm7, [esi + 40] // possibly unaligned load
mulps xmm6, [edi + 32]
mulps xmm7, [edi + 32]
addps xmm0, xmm6
addps xmm1, xmm7
movups xmm4, [esi + 48] // possibly unaligned load
movups xmm5, [esi + 56] // possibly unaligned load
mulps xmm4, [edi + 48]
mulps xmm5, [edi + 48]
addps xmm0, xmm4
addps xmm1, xmm5
add esi, 64
add edi, 64
dec ecx
jnz loop2
// Now xmm0 and xmm1 both have a filtered 2-channel sample each, but we still need
// to sum the two hi- and lo-floats of these registers together.
movhlps xmm2, xmm0 // xmm2 = xmm2_3 xmm2_2 xmm0_3 xmm0_2
movlhps xmm2, xmm1 // xmm2 = xmm1_1 xmm1_0 xmm0_3 xmm0_2
shufps xmm0, xmm1, 0xe4 // xmm0 = xmm1_3 xmm1_2 xmm0_1 xmm0_0
addps xmm0, xmm2
movaps [eax], xmm0
add ebx, 16
add eax, 16
dec edx
jnz loop1
}
*/
}
#endif // SOUNDTOUCH_ALLOW_SSE