mpegaudiodec: optimized iMDCT transform

Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>

4 files changed, 803 insertions, 1 deletions

diff --git a/libavcodec/mpegaudiodec.c b/libavcodec/mpegaudiodec.c
index 70c5f76..6a06afa 100644
--- a/libavcodec/mpegaudiodec.c
+++ b/libavcodec/mpegaudiodec.c
@@ -58,7 +58,7 @@ typedef struct GranuleDef {
     int preflag;
     int short_start, long_end; /* long/short band indexes */
     uint8_t scale_factors[40];
-    INTFLOAT sb_hybrid[SBLIMIT * 18]; /* 576 samples */
+    DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */
 } GranuleDef;
 
 typedef struct MPADecodeContext {
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index aa97942..2abe4fb 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -33,6 +33,7 @@ YASM-OBJS-$(CONFIG_AC3DSP)             += x86/ac3dsp.o
 MMX-OBJS-$(CONFIG_CAVS_DECODER)        += x86/cavsdsp_mmx.o
 MMX-OBJS-$(CONFIG_DNXHD_ENCODER)       += x86/dnxhd_mmx.o
 MMX-OBJS-$(CONFIG_MPEGAUDIODSP)        += x86/mpegaudiodec_mmx.o
+YASM-OBJS-$(CONFIG_MPEGAUDIODSP)       += x86/imdct36_sse.o
 MMX-OBJS-$(CONFIG_ENCODERS)            += x86/dsputilenc_mmx.o
 YASM-OBJS-$(CONFIG_ENCODERS)           += x86/dsputilenc_yasm.o
 MMX-OBJS-$(CONFIG_GPL)                 += x86/idct_mmx.o
diff --git a/libavcodec/x86/imdct36_sse.asm b/libavcodec/x86/imdct36_sse.asm
new file mode 100644
index 0000000..2908459
--- /dev/null
+++ b/libavcodec/x86/imdct36_sse.asm
@@ -0,0 +1,721 @@
+;******************************************************************************
+;* 36 point SSE-optimized IMDCT transform
+;* Copyright (c) 2011 Vitor Sessak
+;*
+;* This file is part of Libav.
+;*
+;* Libav is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* Libav is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with Libav; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86inc.asm"
+%include "libavutil/x86/x86util.asm"
+
+SECTION_RODATA
+
+align 16
+ps_mask:  dd 0, ~0, ~0, ~0
+ps_mask2: dd 0, ~0,  0, ~0
+ps_mask3: dd 0,  0,  0, ~0
+ps_mask4: dd 0, ~0,  0,  0
+
+ps_val1:  dd          -0.5,          -0.5, -0.8660254038, -0.8660254038
+ps_val2:  dd           1.0,           1.0,  0.8660254038,  0.8660254038
+ps_val3:  dd  0.1736481777,  0.1736481777,  0.3420201433,  0.3420201433
+ps_val4:  dd -0.7660444431, -0.7660444431,  0.8660254038,  0.8660254038
+ps_val5:  dd -0.9396926208, -0.9396926208, -0.9848077530, -0.9848077530
+ps_val6:  dd           0.5,           0.5, -0.6427876097, -0.6427876097
+ps_val7:  dd           1.0,           1.0, -0.6427876097, -0.6427876097
+
+ps_p1p1m1m1: dd 0,          0, 0x80000000, 0x80000000
+ps_p1m1p1m1: dd 0, 0x80000000,          0, 0x80000000
+
+ps_cosh:       dd 1.0, 0.50190991877167369479,  1.0,  5.73685662283492756461
+               dd 1.0, 0.51763809020504152469,  1.0,  1.93185165257813657349
+               dd 1.0, 0.55168895948124587824, -1.0, -1.18310079157624925896
+               dd 1.0, 0.61038729438072803416, -1.0, -0.87172339781054900991
+               dd 1.0, 0.70710678118654752439,  0.0,  0.0
+
+ps_cosh_sse3:  dd 1.0, -0.50190991877167369479,  1.0, -5.73685662283492756461
+               dd 1.0, -0.51763809020504152469,  1.0, -1.93185165257813657349
+               dd 1.0, -0.55168895948124587824, -1.0,  1.18310079157624925896
+               dd 1.0, -0.61038729438072803416, -1.0,  0.87172339781054900991
+               dd 1.0,  0.70710678118654752439,  0.0,  0.0
+
+costabs:  times 4 dd  0.98480773
+          times 4 dd  0.93969262
+          times 4 dd  0.86602539
+          times 4 dd -0.76604444
+          times 4 dd -0.64278764
+          times 4 dd  0.50000000
+          times 4 dd -0.50000000
+          times 4 dd -0.34202015
+          times 4 dd -0.17364818
+          times 4 dd  0.50190992
+          times 4 dd  0.51763808
+          times 4 dd  0.55168896
+          times 4 dd  0.61038726
+          times 4 dd  0.70710677
+          times 4 dd  0.87172341
+          times 4 dd  1.18310082
+          times 4 dd  1.93185163
+          times 4 dd  5.73685646
+
+%define SBLIMIT 32
+SECTION_TEXT
+
+%macro PSHUFD 3
+%if cpuflag(sse2) && notcpuflag(avx)
+    pshufd %1, %2, %3
+%else
+    shufps %1, %2, %2, %3
+%endif
+%endmacro
+
+; input  %2={x1,x2,x3,x4}, %3={y1,y2,y3,y4}
+; output %1={x3,x4,y1,y2}
+%macro BUILDINVHIGHLOW 3
+%if cpuflag(avx)
+    shufps %1, %2, %3, 0x4e
+%else
+    movlhps %1, %3
+    movhlps %1, %2
+%endif
+%endmacro
+
+; input  %2={x1,x2,x3,x4}, %3={y1,y2,y3,y4}
+; output %1={x4,y1,y2,y3}
+%macro ROTLEFT 3
+%if cpuflag(ssse3)
+    palignr  %1, %3, %2, 12
+%else
+    BUILDINVHIGHLOW %1, %2, %3
+    shufps  %1, %1, %3, 0x99
+%endif
+%endmacro
+
+%macro INVERTHL 2
+%if cpuflag(sse2)
+    PSHUFD  %1, %2, 0x4e
+%else
+    movhlps %1, %2
+    movlhps %1, %2
+%endif
+%endmacro
+
+%macro BUTTERF 3
+    INVERTHL %2, %1
+    xorps    %1, [ps_p1p1m1m1]
+    addps    %1, %2
+%if cpuflag(sse3)
+    mulps    %1, %1, [ps_cosh_sse3 + %3]
+    PSHUFD   %2, %1, 0xb1
+    addsubps %1, %1, %2
+%else
+    mulps    %1, [ps_cosh + %3]
+    PSHUFD   %2, %1, 0xb1
+    xorps    %1, [ps_p1m1p1m1]
+    addps    %1, %2
+%endif
+%endmacro
+
+%macro STORE 4
+    movhlps %2, %1
+    movss   [%3       ], %1
+    movss   [%3 + 2*%4], %2
+    shufps  %1, %1, 0xb1
+    movss   [%3 +   %4], %1
+    movhlps %2, %1
+    movss   [%3 + 3*%4], %2
+%endmacro
+
+%macro LOAD 4
+    movlps  %1, [%3       ]
+    movhps  %1, [%3 +   %4]
+    movlps  %2, [%3 + 2*%4]
+    movhps  %2, [%3 + 3*%4]
+    shufps  %1, %2, 0x88
+%endmacro
+
+%macro LOADA64 2
+%if cpuflag(avx)
+   movu     %1, [%2]
+%else
+   movlps   %1, [%2]
+   movhps   %1, [%2 + 8]
+%endif
+%endmacro
+
+%macro DEFINE_IMDCT 0
+cglobal imdct36_float, 4,4,9, out, buf, in, win
+
+    ; for(i=17;i>=1;i--) in[i] += in[i-1];
+    LOADA64 m0, inq
+    LOADA64 m1, inq + 16
+
+    ROTLEFT m5, m0, m1
+
+    PSHUFD  m6, m0, 0x93
+    andps   m6, m6, [ps_mask]
+    addps   m0, m0, m6
+
+    LOADA64 m2, inq + 32
+
+    ROTLEFT m7, m1, m2
+
+    addps   m1, m1, m5
+    LOADA64 m3, inq + 48
+
+    ROTLEFT m5, m2, m3
+
+    xorps   m4, m4, m4
+    movlps  m4, [inq+64]
+    BUILDINVHIGHLOW m6, m3, m4
+    shufps  m6, m6, m4, 0xa9
+
+    addps   m4, m4, m6
+    addps   m2, m2, m7
+    addps   m3, m3, m5
+
+    ; for(i=17;i>=3;i-=2) in[i] += in[i-2];
+    movlhps m5, m5, m0
+    andps   m5, m5, [ps_mask3]
+
+    BUILDINVHIGHLOW m7, m0, m1
+    andps   m7, m7, [ps_mask2]
+
+    addps   m0, m0, m5
+
+    BUILDINVHIGHLOW m6, m1, m2
+    andps   m6, m6, [ps_mask2]
+
+    addps  m1, m1, m7
+
+    BUILDINVHIGHLOW m7, m2, m3
+    andps   m7, m7, [ps_mask2]
+
+    addps   m2, m2, m6
+
+    movhlps m6, m6, m3
+    andps   m6, m6, [ps_mask4]
+
+    addps  m3, m3, m7
+    addps  m4, m4, m6
+
+    ; Populate tmp[]
+    movlhps m6, m1, m5    ; zero out high values
+    subps   m6, m6, m4
+
+    subps  m5, m0, m3
+
+%ifdef ARCH_X86_64
+    SWAP   m5, m8
+%endif
+
+    mulps  m7, m2, [ps_val1]
+
+%ifdef ARCH_X86_64
+    mulps  m5, m8, [ps_val2]
+%else
+    mulps  m5, m5, [ps_val2]
+%endif
+    addps  m7, m7, m5
+
+    mulps  m5, m6, [ps_val1]
+    subps  m7, m7, m5
+
+%ifdef ARCH_X86_64
+    SWAP   m5, m8
+%else
+    subps  m5, m0, m3
+%endif
+
+    subps  m5, m5, m6
+    addps  m5, m5, m2
+
+    shufps m6, m4, m3, 0xe4
+    subps  m6, m6, m2
+    mulps  m6, m6, [ps_val3]
+
+    addps  m4, m4, m1
+    mulps  m4, m4, [ps_val4]
+
+    shufps m1, m1, m0, 0xe4
+    addps  m1, m1, m2
+    mulps  m1, m1, [ps_val5]
+
+    mulps  m3, m3, [ps_val6]
+    mulps  m0, m0, [ps_val7]
+    addps  m0, m0, m3
+
+    xorps  m2, m1, [ps_p1p1m1m1]
+    subps  m2, m2, m4
+    addps  m2, m2, m0
+
+    addps  m3, m4, m0
+    subps  m3, m3, m6
+    xorps  m3, m3, [ps_p1p1m1m1]
+
+    shufps m0, m0, m4, 0xe4
+    subps  m0, m0, m1
+    addps  m0, m0, m6
+
+    BUILDINVHIGHLOW m4, m2, m3
+    shufps  m3, m3, m2, 0x4e
+
+    ; we have tmp = {SwAPLH(m0), SwAPLH(m7), m3, m4, m5}
+
+    BUTTERF  m0, m1, 0
+    BUTTERF  m7, m2, 16
+    BUTTERF  m3, m6, 32
+    BUTTERF  m4, m1, 48
+
+    mulps   m5, m5, [ps_cosh + 64]
+    PSHUFD  m1, m5, 0xe1
+    xorps   m5, m5, [ps_p1m1p1m1]
+    addps   m5, m5, m1
+
+    ; permutates:
+    ; m0    0  1  2  3     =>     2  6 10 14   m1
+    ; m7    4  5  6  7     =>     3  7 11 15   m2
+    ; m3    8  9 10 11     =>    17 13  9  5   m3
+    ; m4   12 13 14 15     =>    16 12  8  4   m5
+    ; m5   16 17 xx xx     =>     0  1 xx xx   m0
+
+    unpckhps m1, m0, m7
+    unpckhps m6, m3, m4
+    movhlps  m2, m6, m1
+    movlhps  m1, m1, m6
+
+    unpcklps m5, m5, m4
+    unpcklps m3, m3, m7
+    movhlps  m4, m3, m5
+    movlhps  m5, m5, m3
+    SWAP m4, m3
+    ; permutation done
+
+    PSHUFD  m6, m2, 0xb1
+    movss   m4, [bufq + 4*68]
+    movss   m7, [bufq + 4*64]
+    unpcklps  m7, m7, m4
+    mulps   m6, m6, [winq + 16*4]
+    addps   m6, m6, m7
+    movss   [outq + 64*SBLIMIT], m6
+    shufps  m6, m6, m6, 0xb1
+    movss   [outq + 68*SBLIMIT], m6
+
+    mulps   m6, m3, [winq + 4*4]
+    LOAD    m4, m7, bufq + 4*16, 16
+    addps   m6, m6, m4
+    STORE   m6, m7, outq + 16*SBLIMIT, 4*SBLIMIT
+
+    shufps  m4, m0, m3, 0xb5
+    mulps   m4, m4, [winq + 8*4]
+    LOAD    m7, m6, bufq + 4*32, 16
+    addps   m4, m4, m7
+    STORE   m4, m6, outq + 32*SBLIMIT, 4*SBLIMIT
+
+    shufps  m3, m3, m2, 0xb1
+    mulps   m3, m3, [winq + 12*4]
+    LOAD    m7, m6, bufq + 4*48, 16
+    addps   m3, m3, m7
+    STORE   m3, m7, outq + 48*SBLIMIT, 4*SBLIMIT
+
+    mulps   m2, m2, [winq]
+    LOAD    m6, m7, bufq, 16
+    addps   m2, m2, m6
+    STORE   m2, m7, outq, 4*SBLIMIT
+
+    mulps    m4, m1, [winq + 20*4]
+    STORE    m4, m7, bufq, 16
+
+    mulps    m3, m5, [winq + 24*4]
+    STORE    m3, m7, bufq + 4*16, 16
+
+    shufps   m0, m0, m5, 0xb0
+    mulps    m0, m0, [winq + 28*4]
+    STORE    m0, m7, bufq + 4*32, 16
+
+    shufps   m5, m5, m1, 0xb1
+    mulps    m5, m5, [winq + 32*4]
+    STORE    m5, m7, bufq + 4*48, 16
+
+    shufps   m1, m1, m1, 0xb1
+    mulps    m1, m1, [winq + 36*4]
+    movss    [bufq + 4*64], m1
+    shufps   m1, m1, 0xb1
+    movss    [bufq + 4*68], m1
+    RET
+%endmacro
+
+INIT_XMM sse
+DEFINE_IMDCT
+
+INIT_XMM sse2
+DEFINE_IMDCT
+
+INIT_XMM sse3
+DEFINE_IMDCT
+
+INIT_XMM ssse3
+DEFINE_IMDCT
+
+INIT_XMM avx
+DEFINE_IMDCT
+
+INIT_XMM sse
+
+%ifdef ARCH_X86_64
+%define SPILL SWAP
+%define UNSPILL SWAP
+%define SPILLED(x) m %+ x
+%else
+%define SPILLED(x) [tmpq+(x-8)*16 + 32*4]
+%macro SPILL 2 ; xmm#, mempos
+    movaps SPILLED(%2), m%1
+%endmacro
+%macro UNSPILL 2
+    movaps m%1, SPILLED(%2)
+%endmacro
+%endif
+
+%macro DEFINE_FOUR_IMDCT 0
+cglobal four_imdct36_float, 5,5,8, out, buf, in, win, tmp
+    movlps  m0, [inq+64]
+    movhps  m0, [inq+64 +   72]
+    movlps  m3, [inq+64 + 2*72]
+    movhps  m3, [inq+64 + 3*72]
+
+    shufps  m5, m0, m3, 0xdd
+    shufps  m0, m0, m3, 0x88
+
+    mova     m1, [inq+48]
+    movu     m6, [inq+48 +   72]
+    mova     m7, [inq+48 + 2*72]
+    movu     m3, [inq+48 + 3*72]
+
+    TRANSPOSE4x4PS 1, 6, 7, 3, 4
+
+    addps   m4, m6, m7
+    mova    [tmpq+4*28], m4
+
+    addps    m7, m3
+    addps    m6, m1
+    addps    m3, m0
+    addps    m0, m5
+    addps    m0, m7
+    addps    m7, m6
+    mova    [tmpq+4*12], m7
+    SPILL   3, 12
+
+    mova     m4, [inq+32]
+    movu     m5, [inq+32 +   72]
+    mova     m2, [inq+32 + 2*72]
+    movu     m7, [inq+32 + 3*72]
+
+    TRANSPOSE4x4PS 4, 5, 2, 7, 3
+
+    addps   m1, m7
+    SPILL   1, 11
+
+    addps   m3, m5, m2
+    SPILL   3, 13
+
+    addps    m7, m2
+    addps    m5, m4
+    addps    m6, m7
+    mova    [tmpq], m6
+    addps   m7, m5
+    mova    [tmpq+4*16], m7
+
+    mova    m2, [inq+16]
+    movu    m7, [inq+16 +   72]
+    mova    m1, [inq+16 + 2*72]
+    movu    m6, [inq+16 + 3*72]
+
+    TRANSPOSE4x4PS 2, 7, 1, 6, 3
+
+    addps   m4, m6
+    addps   m6, m1
+    addps   m1, m7
+    addps   m7, m2
+    addps   m5, m6
+    SPILL   5, 15
+    addps   m6, m7
+    mulps   m6, [costabs + 16*2]
+    mova    [tmpq+4*8], m6
+    SPILL   1, 10
+    SPILL   0, 14
+
+    mova    m1, [inq]
+    movu    m6, [inq +   72]
+    mova    m3, [inq + 2*72]
+    movu    m5, [inq + 3*72]
+
+    TRANSPOSE4x4PS 1, 6, 3, 5, 0
+
+    addps    m2, m5
+    addps    m5, m3
+    addps    m7, m5
+    addps    m3, m6
+    addps    m6, m1
+    SPILL    7, 8
+    addps    m5, m6
+    SPILL    6, 9
+    addps    m6, m4, SPILLED(12)
+    subps    m6, m2
+    UNSPILL  7, 11
+    SPILL    5, 11
+    subps    m5, m1, m7
+    mulps    m7, [costabs + 16*5]
+    addps    m7, m1
+    mulps    m0, m6, [costabs + 16*6]
+    addps    m0, m5
+    mova     [tmpq+4*24], m0
+    addps    m6, m5
+    mova     [tmpq+4*4], m6
+    addps    m6, m4, m2
+    mulps    m6, [costabs + 16*1]
+    subps    m4, SPILLED(12)
+    mulps    m4, [costabs + 16*8]
+    addps    m2, SPILLED(12)
+    mulps    m2, [costabs + 16*3]
+    subps    m5, m7, m6
+    subps    m5, m2
+    addps    m6, m7
+    addps    m6, m4
+    addps    m7, m2
+    subps    m7, m4
+    mova     [tmpq+4*20], m7
+    mova     m2, [tmpq+4*28]
+    mova     [tmpq+4*28], m5
+    UNSPILL  7, 13
+    subps    m5, m7, m2
+    mulps    m5, [costabs + 16*7]
+    UNSPILL  1, 10
+    mulps    m1, [costabs + 16*2]
+    addps    m4, m3, m2
+    mulps    m4, [costabs + 16*4]
+    addps    m2, m7
+    addps    m7, m3
+    mulps    m7, [costabs]
+    subps    m3, m2
+    mulps    m3, [costabs + 16*2]
+    addps    m2, m7, m5
+    addps    m2, m1
+    SPILL    2, 10
+    addps    m7, m4
+    subps    m7, m1
+    SPILL    7, 12
+    subps    m5, m4
+    subps    m5, m1
+    UNSPILL  0, 14
+    SPILL    5, 13
+    addps    m1, m0, SPILLED(15)
+    subps    m1, SPILLED(8)
+    mova     m4, [costabs + 16*5]
+    mulps    m4, [tmpq]
+    UNSPILL  2, 9
+    addps    m4, m2
+    subps    m2, [tmpq]
+    mulps    m5, m1, [costabs + 16*6]
+    addps    m5, m2
+    SPILL    5, 9
+    addps    m2, m1
+    SPILL    2, 14
+    UNSPILL  5, 15
+    subps    m7, m5, m0
+    addps    m5, SPILLED(8)
+    mulps    m5, [costabs + 16*1]
+    mulps    m7, [costabs + 16*8]
+    addps    m0, SPILLED(8)
+    mulps    m0, [costabs + 16*3]
+    subps    m2, m4, m5
+    subps    m2, m0
+    SPILL    2, 15
+    addps    m5, m4
+    addps    m5, m7
+    addps    m4, m0
+    subps    m4, m7
+    SPILL    4, 8
+    mova     m7, [tmpq+4*16]
+    mova     m2, [tmpq+4*12]
+    addps    m0, m7, m2
+    subps    m0, SPILLED(11)
+    mulps    m0, [costabs + 16*2]
+    addps    m4, m7, SPILLED(11)
+    mulps    m4, [costabs]
+    subps    m7, m2
+    mulps    m7, [costabs + 16*7]
+    addps    m2, SPILLED(11)
+    mulps    m2, [costabs + 16*4]
+    addps    m1, m7, [tmpq+4*8]
+    addps    m1, m4
+    addps    m4, m2
+    subps    m4, [tmpq+4*8]
+    SPILL    4, 11
+    subps    m7, m2
+    subps    m7, [tmpq+4*8]
+    addps    m4, m6, SPILLED(10)
+    subps    m6, SPILLED(10)
+    addps    m2, m5, m1
+    mulps    m2, [costabs + 16*9]
+    subps    m5, m1
+    mulps    m5, [costabs + 16*17]
+    subps    m1, m4, m2
+    addps    m4, m2
+    mulps    m2, m1, [winq+4*36]
+    addps    m2, [bufq+4*36]
+    mova     [outq+1152], m2
+    mulps    m1, [winq+4*32]
+    addps    m1, [bufq+4*32]
+    mova     [outq+1024], m1
+    mulps    m1, m4, [winq+4*116]
+    mova     [bufq+4*36], m1
+    mulps    m4, [winq+4*112]
+    mova     [bufq+4*32], m4
+    addps    m2, m6, m5
+    subps    m6, m5
+    mulps    m1, m6, [winq+4*68]
+    addps    m1, [bufq+4*68]
+    mova     [outq+2176], m1
+    mulps    m6, [winq]
+    addps    m6, [bufq]
+    mova     [outq], m6
+    mulps    m1, m2, [winq+4*148]
+    mova     [bufq+4*68], m1
+    mulps    m2, [winq+4*80]
+    mova     [bufq], m2
+    addps    m5, m3, [tmpq+4*24]
+    mova     m2, [tmpq+4*24]
+    subps    m2, m3
+    mova     m1, SPILLED(9)
+    subps    m1, m0
+    mulps    m1, [costabs + 16*10]
+    addps    m0, SPILLED(9)
+    mulps    m0, [costabs + 16*16]
+    addps    m6, m5, m1
+    subps    m5, m1
+    mulps    m3, m5, [winq+4*40]
+    addps    m3, [bufq+4*40]
+    mova     [outq+1280], m3
+    mulps    m5, [winq+4*28]
+    addps    m5, [bufq+4*28]
+    mova     [outq+896], m5
+    mulps    m1, m6, [winq+4*120]
+    mova     [bufq+4*40], m1
+    mulps    m6, [winq+4*108]
+    mova     [bufq+4*28], m6
+    addps    m1, m2, m0
+    subps    m2, m0
+    mulps    m5, m2, [winq+4*64]
+    addps    m5, [bufq+4*64]
+    mova     [outq+2048], m5
+    mulps    m2, [winq+4*4]
+    addps    m2, [bufq+4*4]
+    mova     [outq+128], m2
+    mulps    m0, m1, [winq+4*144]
+    mova     [bufq+4*64], m0
+    mulps    m1, [winq+4*84]
+    mova     [bufq+4*4], m1
+    mova     m1, [tmpq+4*28]
+    mova     m5, m1
+    addps    m1, SPILLED(13)
+    subps    m5, SPILLED(13)
+    UNSPILL  3, 15
+    addps    m2, m7, m3
+    mulps    m2, [costabs + 16*11]
+    subps    m3, m7
+    mulps    m3, [costabs + 16*15]
+    addps    m0, m2, m1
+    subps    m1, m2
+    SWAP     m0, m2
+    mulps    m6, m1, [winq+4*44]
+    addps    m6, [bufq+4*44]
+    mova     [outq+1408], m6
+    mulps    m1, [winq+4*24]
+    addps    m1, [bufq+4*24]
+    mova     [outq+768], m1
+    mulps    m0, m2, [winq+4*124]
+    mova     [bufq+4*44], m0
+    mulps    m2, [winq+4*104]
+    mova     [bufq+4*24], m2
+    addps    m0, m5, m3
+    subps    m5, m3
+    mulps    m1, m5, [winq+4*60]
+    addps    m1, [bufq+4*60]
+    mova     [outq+1920], m1
+    mulps    m5, [winq+4*8]
+    addps    m5, [bufq+4*8]
+    mova     [outq+256], m5
+    mulps    m1, m0, [winq+4*140]
+    mova     [bufq+4*60], m1
+    mulps    m0, [winq+4*88]
+    mova     [bufq+4*8], m0
+    mova     m1, [tmpq+4*20]
+    addps    m1, SPILLED(12)
+    mova     m2, [tmpq+4*20]
+    subps    m2, SPILLED(12)
+    UNSPILL  7, 8
+    subps    m0, m7, SPILLED(11)
+    addps    m7, SPILLED(11)
+    mulps    m4, m7, [costabs + 16*12]
+    mulps    m0, [costabs + 16*14]
+    addps    m5, m1, m4
+    subps    m1, m4
+    mulps    m7, m1, [winq+4*48]
+    addps    m7, [bufq+4*48]
+    mova     [outq+1536], m7
+    mulps    m1, [winq+4*20]
+    addps    m1, [bufq+4*20]
+    mova     [outq+640], m1
+    mulps    m1, m5, [winq+4*128]
+    mova     [bufq+4*48], m1
+    mulps    m5, [winq+4*100]
+    mova     [bufq+4*20], m5
+    addps    m6, m2, m0
+    subps    m2, m0
+    mulps    m1, m2, [winq+4*56]
+    addps    m1, [bufq+4*56]
+    mova     [outq+1792], m1
+    mulps    m2, [winq+4*12]
+    addps    m2, [bufq+4*12]
+    mova     [outq+384], m2
+    mulps    m0, m6, [winq+4*136]
+    mova    [bufq+4*56], m0
+    mulps    m6, [winq+4*92]
+    mova     [bufq+4*12], m6
+    UNSPILL  0, 14
+    mulps    m0, [costabs + 16*13]
+    mova     m3, [tmpq+4*4]
+    addps    m2, m0, m3
+    subps    m3, m0
+    mulps    m0, m3, [winq+4*52]
+    addps    m0, [bufq+4*52]
+    mova     [outq+1664], m0
+    mulps    m3, [winq+4*16]
+    addps    m3, [bufq+4*16]
+    mova     [outq+512], m3
+    mulps    m0, m2, [winq+4*132]
+    mova     [bufq+4*52], m0
+    mulps    m2, [winq+4*96]
+    mova     [bufq+4*16], m2
+    RET
+%endmacro
+
+INIT_XMM sse
+DEFINE_FOUR_IMDCT
+
+INIT_XMM avx
+DEFINE_FOUR_IMDCT
diff --git a/libavcodec/x86/mpegaudiodec_mmx.c b/libavcodec/x86/mpegaudiodec_mmx.c
index b644615..06ffbca 100644
--- a/libavcodec/x86/mpegaudiodec_mmx.c
+++ b/libavcodec/x86/mpegaudiodec_mmx.c
@@ -24,6 +24,18 @@
 #include "libavcodec/dsputil.h"
 #include "libavcodec/mpegaudiodsp.h"
 
+void ff_imdct36_float_sse(float *out, float *buf, float *in, float *win);
+void ff_imdct36_float_sse2(float *out, float *buf, float *in, float *win);
+void ff_imdct36_float_sse3(float *out, float *buf, float *in, float *win);
+void ff_imdct36_float_ssse3(float *out, float *buf, float *in, float *win);
+void ff_imdct36_float_avx(float *out, float *buf, float *in, float *win);
+void ff_four_imdct36_float_sse(float *out, float *buf, float *in, float *win,
+                               float *tmpbuf);
+void ff_four_imdct36_float_avx(float *out, float *buf, float *in, float *win,
+                               float *tmpbuf);
+
+DECLARE_ALIGNED(16, static float, mdct_win_sse)[2][4][4*40];
+
 #define MACS(rt, ra, rb) rt+=(ra)*(rb)
 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)
 
@@ -147,11 +159,79 @@ static void apply_window_mp3(float *in, float *win, int *unused, float *out,
     *out = sum;
 }
 
+
+#define DECL_IMDCT_BLOCKS(CPU1, CPU2)                                       \
+static void imdct36_blocks_ ## CPU1(float *out, float *buf, float *in,      \
+                               int count, int switch_point, int block_type) \
+{                                                                           \
+    int align_end = count - (count & 3);                                \
+    int j;                                                              \
+    for (j = 0; j < align_end; j+= 4) {                                 \
+        LOCAL_ALIGNED_16(float, tmpbuf, [1024]);                        \
+        float *win = mdct_win_sse[switch_point && j < 4][block_type];   \
+        /* apply window & overlap with previous buffer */               \
+                                                                        \
+        /* select window */                                             \
+        ff_four_imdct36_float_ ## CPU2(out, buf, in, win, tmpbuf);      \
+        in      += 4*18;                                                \
+        buf     += 4*18;                                                \
+        out     += 4;                                                   \
+    }                                                                   \
+    for (; j < count; j++) {                                            \
+        /* apply window & overlap with previous buffer */               \
+                                                                        \
+        /* select window */                                             \
+        int win_idx = (switch_point && j < 2) ? 0 : block_type;         \
+        float *win = ff_mdct_win_float[win_idx + (4 & -(j & 1))];       \
+                                                                        \
+        ff_imdct36_float_ ## CPU1(out, buf, in, win);                   \
+                                                                        \
+        in  += 18;                                                      \
+        buf++;                                                          \
+        out++;                                                          \
+    }                                                                   \
+}
+
+DECL_IMDCT_BLOCKS(sse,sse)
+DECL_IMDCT_BLOCKS(sse2,sse)
+DECL_IMDCT_BLOCKS(sse3,sse)
+DECL_IMDCT_BLOCKS(ssse3,sse)
+DECL_IMDCT_BLOCKS(avx,avx)
+
 void ff_mpadsp_init_mmx(MPADSPContext *s)
 {
     int mm_flags = av_get_cpu_flags();
 
+    int i, j;
+    for (j = 0; j < 4; j++) {
+        for (i = 0; i < 40; i ++) {
+            mdct_win_sse[0][j][4*i    ] = ff_mdct_win_float[j    ][i];
+            mdct_win_sse[0][j][4*i + 1] = ff_mdct_win_float[j + 4][i];
+            mdct_win_sse[0][j][4*i + 2] = ff_mdct_win_float[j    ][i];
+            mdct_win_sse[0][j][4*i + 3] = ff_mdct_win_float[j + 4][i];
+            mdct_win_sse[1][j][4*i    ] = ff_mdct_win_float[0    ][i];
+            mdct_win_sse[1][j][4*i + 1] = ff_mdct_win_float[4    ][i];
+            mdct_win_sse[1][j][4*i + 2] = ff_mdct_win_float[j    ][i];
+            mdct_win_sse[1][j][4*i + 3] = ff_mdct_win_float[j + 4][i];
+        }
+    }
+
     if (mm_flags & AV_CPU_FLAG_SSE2) {
         s->apply_window_float = apply_window_mp3;
     }
+#if HAVE_YASM
+    if (mm_flags & AV_CPU_FLAG_AVX && HAVE_AVX) {
+        s->imdct36_blocks_float = imdct36_blocks_avx;
+#if HAVE_SSE
+    } else if (mm_flags & AV_CPU_FLAG_SSSE3) {
+        s->imdct36_blocks_float = imdct36_blocks_ssse3;
+    } else if (mm_flags & AV_CPU_FLAG_SSE3) {
+        s->imdct36_blocks_float = imdct36_blocks_sse3;
+    } else if (mm_flags & AV_CPU_FLAG_SSE2) {
+        s->imdct36_blocks_float = imdct36_blocks_sse2;
+    } else if (mm_flags & AV_CPU_FLAG_SSE) {
+        s->imdct36_blocks_float = imdct36_blocks_sse;
+#endif /* HAVE_SSE */
+    }
+#endif /* HAVE_YASM */
 }