justinruggles

SECTION_RODATA 32

pb_shuf_unpack_even: db 11, 10, -1, -1,  9,  8, -1, -1, 3, 2, -1, -1, 1, 0, -1, -1
pb_shuf_unpack_odd:  db 15, 14, -1, -1, 13, 12, -1, -1, 7, 6, -1, -1, 5, 4, -1, -1

SECTION_TEXT

;------------------------------------------------------------------------------
; void ff_conv_s16p_to_flt_6ch(float *dst, int16_t *const *src, int len,
;                              int channels);

justinruggles

pb_perm_unpackw_0: db  0,  1,  2,  3, 16, 17, 18, 19,  8,  9, 10, 11, 24, 25, 26, 27
pb_perm_unpackw_1: db  8,  9, 10, 11, 24, 25, 26, 27, 12, 13, 14, 15, 28, 29, 30, 31

[...]

    punpcklwd  m0, m1            ; m0 =  0,  1,  6,  7, 12, 13, 18, 19
    punpcklwd  m2, m3            ; m2 =  4,  5, 10, 11, 16, 17, 22, 23
    punpcklwd  m4, m5            ; m4 =  2,  3,  8,  9, 14, 15, 20, 21
%if cpuflag(xop)
    vpperm     m1, m0, m4, [pb_perm_unpackw_0]  ; m1 =  0,  1,  2,  3, 12, 13, 14, 15

justinruggles

From ac7c4b9b628f14f9bd8f457a9d15b0fee31e44bb Mon Sep 17 00:00:00 2001
From: Justin Ruggles <[email protected]>
Date: Fri, 20 Jul 2012 15:24:40 -0400
Subject: [PATCH 1/4] dsputil: x86: convert PMINSD, PMAXSD, and CLIPD macros to use cpuflags

---
 libavcodec/x86/dsputil_mmx.c    |    6 ++--
 libavcodec/x86/dsputil_yasm.asm |   67 +++++++++++++++++++--------------------
 libavutil/x86/x86util.asm       |   34 ++++++++++++-------

justinruggles

make tests/tiny_psnr

float mono test:
avconv -cpuflags 0    -i <input> -filter:a aformat=sample_formats\=flt -channel_layout mono -y out_flt_mono_c.wav
avconv -cpuflags fma4 -i <input> -filter:a aformat=sample_formats\=flt -channel_layout mono -y out_flt_mono_fma4.wav
./tests/tiny_psnr out_flt_mono_c.wav out_flt_mono_fma4.wav

float stereo test:
avconv -cpuflags 0    -i <input> -filter:a aformat=sample_formats\=flt -channel_layout stereo -y out_flt_stereo_c.wav
avconv -cpuflags fma4 -i <input> -filter:a aformat=sample_formats\=flt -channel_layout stereo -y out_flt_stereo_fma4.wav

justinruggles

time of s->imdct_512.imdct_half() in ac3dec
     C: 38800
 3dnow:  7500
3dnow2:  6700
   SSE:  3700

time of mdct in ac3enc
     C: 41941
 3dnow: 25469
3dnow2: 24576

justinruggles

TEST    fft-4
make: *** [fate-fft-4] Error 1
TEST    ifft-4
make: *** [fate-ifft-4] Error 1
TEST    mdct-4
TEST    imdct-4
TEST    rdft-4
TEST    irdft-4
TEST    dct1d-4
TEST    idct1d-4

justinruggles

From 926c0121c9fd3b7234f175648db87bdb6e3c15e5 Mon Sep 17 00:00:00 2001
From: Justin Ruggles <[email protected]>
Date: Fri, 20 Jul 2012 19:53:40 -0400
Subject: [PATCH] dsputil: x86: use cpuflags for apply_window_int16()

---
 libavcodec/x86/dsputil_mmx.c    |   12 +++---
 libavcodec/x86/dsputil_yasm.asm |   79 +++++++++++++++++---------------------
 2 files changed, 41 insertions(+), 50 deletions(-)

justinruggles

Program received signal SIGSEGV, Segmentation fault.
ff_imdct_half_3dnow.pre ()
    at /home/justin/src/libav2/libavcodec/x86/fft_mmx.asm:1071
1071    DECL_IMDCT POSROTATESHUF_3DNOW
(gdb) disp/i $pc
1: x/i $pc
=> 0x85dfa1a <ff_imdct_half_3dnow.pre+67>:  movq   %mm0,0x0(%ebp)
(gdb) info all-registers
eax            0x8ee2940    149825856
ecx            0x8ee8d00    149851392

justinruggles

Dump of assembler code for function ff_imdct_half_3dnow.pre:
   0x085df9d7 <+0>:	mov    0x4(%esp),%ebp
   0x085df9db <+4>:	movd   -0x4(%edx,%esi,4),%mm0
   0x085df9e0 <+9>:	movd   (%edi,%ebx,2),%mm1
   0x085df9e4 <+13>:	punpckldq (%edx,%ebx,4),%mm0
   0x085df9e8 <+17>:	punpckldq 0x0(%ebp,%ebx,2),%mm1
   0x085df9ed <+22>:	movq   %mm0,%mm2
   0x085df9f0 <+25>:	movq   %mm1,%mm3
   0x085df9f3 <+28>:	psrlq  $0x20,%mm3
   0x085df9f7 <+32>:	punpckldq %mm1,%mm3

justinruggles

Dump of assembler code for function ff_imdct_half_3dnow.pre:
   0x085df9e7 <+0>:	mov    0x4(%esp),%ebp
   0x085df9eb <+4>:	movd   (%edx,%ebx,4),%mm0
   0x085df9ef <+8>:	movd   -0x4(%edi,%esi,4),%mm1
   0x085df9f4 <+13>:	punpckldq -0x4(%edx,%esi,4),%mm0
   0x085df9f9 <+18>:	punpckldq -0x4(%ebp,%esi,4),%mm1
   0x085df9fe <+23>:	movq   %mm0,%mm2
   0x085dfa01 <+26>:	movq   %mm1,%mm3
   0x085dfa04 <+29>:	psrlq  $0x20,%mm3
   0x085dfa08 <+33>:	punpckldq %mm1,%mm3