Hi Everyone,
I was at a conference all last week and fell way behind on my
email. Since Mat's message below is the latest, I'll assume this
is the best version of quantize_xrpow so far?
(but it seems to be missing the .5,
> XRPOW_FTOI(x, rx);
should be
> XRPOW_FTOI(x-.5, rx);
?
I just committed the new quantize_xrpow with the -.5 fix, and
quantize_xrpow_ISO from Acy's last message, but also went back to the
old fft.c routines. Takehiro's code had the minor force_ms problem,
but also produced changes in the bitstream. Not unusual if you change
something like the fft, but it breaks my validation tests. In such
cases I want to make sure I understand exactly where the differences
are coming from. I hope to do this with the new fft in the next few
days.
Mark
(the force_ms problems were caused by the fact that the old
force_ms mode was removed one or two revisions ago. The new force_ms
mode just encodes as regular jstereo and forces all frames to be
mid/side. The old force_ms mode was a messy hack.
>
> > From: Mathew Hendry [mailto:[EMAIL PROTECTED]]
> >
>
> /*
> Timings for the MSVC asm version, before and after, testing "fools.wav"
>
> Func Func+Child Hit
> Time % Time % Count Function
> ---------------------------------------------------------
> 860.617 6.2 860.617 6.2 84439 _quantize_xrpow
> 789.424 5.7 789.424 5.7 84439 _quantize_xrpow
> */
>
> void quantize_xrpow(FLOAT8 xr[576], int ix[576], gr_info *cod_info) {
> /* quantize on xr^(3/4) instead of xr */
> const FLOAT8 quantizerStepSize = cod_info->quantizerStepSize;
> const FLOAT8 istep = pow(2.0, quantizerStepSize * -0.1875);
>
> #ifndef _MSC_VER
> {
> FLOAT8 x;
> int j, rx;
> for (j = 576 / 4; j > 0; --j) {
> x = *xr++ * istep;
> XRPOW_FTOI(x, rx);
> XRPOW_FTOI(x + QUANTFAC(rx), *ix++);
> x = *xr++ * istep;
> XRPOW_FTOI(x, rx);
> XRPOW_FTOI(x + QUANTFAC(rx), *ix++);
> x = *xr++ * istep;
> XRPOW_FTOI(x, rx);
> XRPOW_FTOI(x + QUANTFAC(rx), *ix++);
> x = *xr++ * istep;
> XRPOW_FTOI(x, rx);
> XRPOW_FTOI(x + QUANTFAC(rx), *ix++);
> }
> }
> #else
> /* def _MSC_VER */
> {
> /* asm from Acy Stapp <[EMAIL PROTECTED]> */
> int rx[4];
> _asm {
> fld qword ptr [istep]
> mov esi, dword ptr [xr]
> lea edi, dword ptr [adj43asm]
> mov edx, dword ptr [ix]
> mov ecx, 576/4
> } loop1: _asm {
> fld qword ptr [esi] // 0
> fld qword ptr [esi+8] // 1 0
> fld qword ptr [esi+16] // 2 1 0
> fld qword ptr [esi+24] // 3 2 1 0
>
> fxch st(3) // 0 2 1 3
> fmul st(0), st(4)
> fxch st(2) // 1 2 0 3
> fmul st(0), st(4)
> fxch st(1) // 2 1 0 3
> fmul st(0), st(4)
> fxch st(3) // 3 1 0 2
> fmul st(0), st(4)
>
> add esi, 32
> add edx, 16
>
> fxch st(2) // 0 1 3 2
> fist dword ptr [rx]
> fxch st(1) // 1 0 3 2
> fist dword ptr [rx+4]
> fxch st(3) // 2 0 3 1
> fist dword ptr [rx+8]
> fxch st(2) // 3 0 2 1
> fist dword ptr [rx+12]
>
> dec ecx
>
> mov eax, dword ptr [rx]
> mov ebx, dword ptr [rx+4]
> fxch st(1) // 0 3 2 1
> fadd qword ptr [edi+eax*8]
> fxch st(3) // 1 3 2 0
> fadd qword ptr [edi+ebx*8]
>
> mov eax, dword ptr [rx+8]
> mov ebx, dword ptr [rx+12]
> fxch st(2) // 2 3 1 0
> fadd qword ptr [edi+eax*8]
> fxch st(1) // 3 2 1 0
> fadd qword ptr [edi+ebx*8]
>
> fxch st(3) // 0 2 1 3
> fistp dword ptr [edx-16] // 2 1 3
> fxch st(1) // 1 2 3
> fistp dword ptr [edx-12] // 2 3
> fistp dword ptr [edx-8] // 3
> fistp dword ptr [edx-4]
>
> jnz loop1
>
> mov dword ptr [xr], esi
> mov dword ptr [ix], edx
> fstp st(0)
> }
> }
> #endif
> }
>
> -- Mat.
> --
> MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
>
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
