updating ffmpeg to 26-12-2001 CVS version

CVS patchset: 1310
CVS date: 2001/12/27 20:02:23

14 files changed, 4982 insertions, 0 deletions

diff --git a/src/libffmpeg/libavcodec/armv4l/Makefile.am b/src/libffmpeg/libavcodec/armv4l/Makefile.am
new file mode 100644
index 000000000..1fbda388f
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/Makefile.am
@@ -0,0 +1,38 @@
+##
+## Process this file with automake to produce Makefile.in
+##
+
+CFLAGS = @GLOBAL_CFLAGS@ @LIBFFMPEG_CFLAGS@ -DCONFIG_DECODERS -DHAVE_AV_CONFIG_H
+ASFLAGS =
+
+LIBTOOL = $(SHELL) $(top_builddir)/libtool-nofpic
+
+noinst_LTLIBRARIES = libavcodec_armv4l.la
+
+EXTRA_DIST =  jrevdct_arm.S dsputil_arm.c
+
+#if HAVE_ARMV4L
+#armv4l_modules = $(EXTRA_DIST)
+#endif
+armv4l_modules =
+
+libavcodec_armv4l_la_SOURCES = $(armv4l_modules)
+
+noinst_HEADERS = 
+
+.s.lo:
+	$(ASCOMPILE) -o $@ `test -f $< || echo '$(srcdir)/'`$<
+
+debug:
+	@$(MAKE) CFLAGS="@DEBUG_CFLAGS@ @LIBFFMPEG_CFLAGS@ -DCONFIG_DECODERS -DHAVE_AV_CONFIG_H"
+
+install-debug: debug
+	@$(MAKE) $(AM_MAKEFLAGS) install-exec-am install-data-am
+
+mostlyclean-generic:
+	-rm -f *~ \#* .*~ .\#*
+
+maintainer-clean-generic:
+	-@echo "This command is intended for maintainers to use;"
+	-@echo "it deletes files that may require special tools to rebuild."
+	-rm -f Makefile.in
diff --git a/src/libffmpeg/libavcodec/armv4l/dsputil_arm.c b/src/libffmpeg/libavcodec/armv4l/dsputil_arm.c
new file mode 100644
index 000000000..1cf7b4fba
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/dsputil_arm.c
@@ -0,0 +1,27 @@
+/*
+ * ARMv4L optimized DSP utils
+ * Copyright (c) 2001 Lionel Ulmer.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include "../dsputil.h"
+
+extern void j_rev_dct_ARM(DCTELEM *data);
+
+void dsputil_init_armv4l(void)
+{
+  ff_idct = j_rev_dct_ARM;
+}
diff --git a/src/libffmpeg/libavcodec/armv4l/jrevdct_arm.S b/src/libffmpeg/libavcodec/armv4l/jrevdct_arm.S
new file mode 100644
index 000000000..76eda57ea
--- /dev/null
+++ b/src/libffmpeg/libavcodec/armv4l/jrevdct_arm.S
@@ -0,0 +1,386 @@
+/* 
+   C-like prototype :
+	void j_rev_dct_ARM(DCTBLOCK data)
+
+   With DCTBLOCK being a pointer to an array of 64 'signed shorts'
+
+   Copyright (c) 2001 Lionel Ulmer (lionel.ulmer@free.fr / bbrox@bbrox.org)
+
+   Permission is hereby granted, free of charge, to any person obtaining a copy
+   of this software and associated documentation files (the "Software"), to deal
+   in the Software without restriction, including without limitation the rights
+   to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+   copies of the Software, and to permit persons to whom the Software is
+   furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+   COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
+   IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+   CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+	
+*/
+#define FIX_0_298631336 2446
+#define FIX_0_541196100 4433
+#define FIX_0_765366865 6270
+#define FIX_1_175875602 9633
+#define FIX_1_501321110 12299
+#define FIX_2_053119869 16819
+#define FIX_3_072711026 25172
+#define FIX_M_0_390180644 -3196
+#define FIX_M_0_899976223 -7373
+#define FIX_M_1_847759065 -15137
+#define FIX_M_1_961570560 -16069
+#define FIX_M_2_562915447 -20995
+#define FIX_0xFFFF 0xFFFF	
+		
+#define FIX_0_298631336_ID      0
+#define FIX_0_541196100_ID      4
+#define FIX_0_765366865_ID      8
+#define FIX_1_175875602_ID     12
+#define FIX_1_501321110_ID     16
+#define FIX_2_053119869_ID     20
+#define FIX_3_072711026_ID     24
+#define FIX_M_0_390180644_ID   28
+#define FIX_M_0_899976223_ID   32
+#define FIX_M_1_847759065_ID   36
+#define FIX_M_1_961570560_ID   40
+#define FIX_M_2_562915447_ID   44
+#define FIX_0xFFFF_ID          48
+	.text
+	.align
+	
+	.global j_rev_dct_ARM
+j_rev_dct_ARM:
+	stmdb   sp!, { r4 - r12, lr }   @ all callee saved regs
+
+	sub sp, sp, #4                  @ reserve some space on the stack
+	str r0, [ sp ]                  @ save the DCT pointer to the stack
+
+	mov lr, r0                      @ lr = pointer to the current row
+	mov r12, #8                     @ r12 = row-counter
+	add r11, pc, #(const_array-.-8) @ r11 = base pointer to the constants array	
+row_loop:
+	ldrsh r0, [lr, # 0]             @ r0 = 'd0'
+	ldrsh r1, [lr, # 8]             @ r1 = 'd1'
+
+	@ Optimization for row that have all items except the first set to 0
+	@ (this works as the DCTELEMS are always 4-byte aligned)
+	ldr r5, [lr, # 0]
+	ldr r2, [lr, # 4]
+	ldr r3, [lr, # 8]
+	ldr r4, [lr, #12]
+	orr r3, r3, r4
+	orr r3, r3, r2
+	orrs r5, r3, r5
+	beq end_of_row_loop             @ nothing to be done as ALL of them are '0'
+	orrs r2, r3, r1
+	beq empty_row
+	
+	ldrsh r2, [lr, # 2]             @ r2 = 'd2'
+	ldrsh r4, [lr, # 4]             @ r4 = 'd4'
+	ldrsh r6, [lr, # 6]             @ r6 = 'd6'
+	
+	ldr r3, [r11, #FIX_0_541196100_ID]
+	add r7, r2, r6
+	ldr r5, [r11, #FIX_M_1_847759065_ID]
+	mul r7, r3, r7                      @ r7 = z1
+	ldr r3, [r11, #FIX_0_765366865_ID]
+	mla r6, r5, r6, r7                  @ r6 = tmp2
+	add r5, r0, r4                      @ r5 = tmp0
+	mla r2, r3, r2, r7                  @ r2 = tmp3
+	sub r3, r0, r4                      @ r3 = tmp1
+
+	add r0, r2, r5, lsl #13             @ r0 = tmp10
+	rsb r2, r2, r5, lsl #13             @ r2 = tmp13
+	add r4, r6, r3, lsl #13             @ r4 = tmp11
+	rsb r3, r6, r3, lsl #13             @ r3 = tmp12
+
+	stmdb   sp!, { r0, r2, r3, r4 } @ save on the stack tmp10, tmp13, tmp12, tmp11
+	
+	ldrsh r3, [lr, #10]             @ r3 = 'd3'
+	ldrsh r5, [lr, #12]             @ r5 = 'd5'
+	ldrsh r7, [lr, #14]             @ r7 = 'd7'
+
+	add r0, r3, r5	                @ r0 = 'z2'
+	add r2, r1, r7                  @ r2 = 'z1'
+	add r4, r3, r7                  @ r4 = 'z3'
+	add r6, r1, r5                  @ r6 = 'z4'
+	ldr r9, [r11, #FIX_1_175875602_ID]
+	add r8, r4, r6                  @ r8 = z3 + z4
+	ldr r10, [r11, #FIX_M_0_899976223_ID]
+	mul r8, r9, r8                  @ r8 = 'z5'
+	ldr r9, [r11, #FIX_M_2_562915447_ID]
+	mul r2, r10, r2                 @ r2 = 'z1'
+	ldr r10, [r11, #FIX_M_1_961570560_ID]
+	mul r0, r9, r0                  @ r0 = 'z2'
+	ldr r9, [r11, #FIX_M_0_390180644_ID]
+	mla r4, r10, r4, r8             @ r4 = 'z3'
+	ldr r10, [r11, #FIX_0_298631336_ID]
+	mla r6, r9, r6, r8              @ r6 = 'z4'
+	ldr r9, [r11, #FIX_2_053119869_ID]
+	mla r7, r10, r7, r2             @ r7 = tmp0 + z1
+	ldr r10, [r11, #FIX_3_072711026_ID]
+	mla r5, r9, r5, r0              @ r5 = tmp1 + z2
+	ldr r9, [r11, #FIX_1_501321110_ID]
+	mla r3, r10, r3, r0             @ r3 = tmp2 + z2
+	add r7, r7, r4                  @ r7 = tmp0
+	mla r1, r9, r1, r2              @ r1 = tmp3 + z1
+	add r5,	r5, r6                  @ r5 = tmp1
+	add r3, r3, r4                  @ r3 = tmp2
+	add r1, r1, r6                  @ r1 = tmp3
+
+	ldmia sp!, { r0, r2, r4, r6 } @ r0 = tmp10 / r2 = tmp13 / r4 = tmp12 / r6 = tmp11
+	                              @ r1 = tmp3  / r3 = tmp2  / r5 = tmp1  / r7 = tmp0
+	
+	@ Compute DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS)
+	add r8, r0, r1
+	add r8, r8, #(1<<10)
+	mov r8, r8, asr #11
+	strh r8, [lr, # 0]
+	
+	@ Compute DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS)
+	sub r8, r0, r1
+	add r8, r8, #(1<<10)
+	mov r8, r8, asr #11
+	strh r8, [lr, #14]
+	
+	@ Compute DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS)
+	add r8, r6, r3
+	add r8, r8, #(1<<10)
+	mov r8, r8, asr #11
+	strh r8, [lr, # 2]
+	
+	@ Compute DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS)
+	sub r8, r6, r3
+	add r8, r8, #(1<<10)
+	mov r8, r8, asr #11
+	strh r8, [lr, #12]
+	
+	@ Compute DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS)
+	add r8, r4, r5
+	add r8, r8, #(1<<10)
+	mov r8, r8, asr #11
+	strh r8, [lr, # 4]
+	
+	@ Compute DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS)
+	sub r8, r4, r5
+	add r8, r8, #(1<<10)
+	mov r8, r8, asr #11
+	strh r8, [lr, #10]
+	
+	@ Compute DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS)
+	add r8, r2, r7
+	add r8, r8, #(1<<10)
+	mov r8, r8, asr #11
+	strh r8, [lr, # 6]
+	
+	@ Compute DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS)
+	sub r8, r2, r7
+	add r8, r8, #(1<<10)
+	mov r8, r8, asr #11
+	strh r8, [lr, # 8]
+
+	@ End of row loop
+	add lr, lr, #16
+	subs r12, r12, #1
+	bne row_loop
+	beq start_column_loop
+	
+empty_row:
+	ldr r1, [r11, #FIX_0xFFFF_ID]
+	mov r0, r0, lsl #2
+	and r0, r0, r1
+	add r0, r0, r0, lsl #16
+	str r0, [lr, # 0]
+	str r0, [lr, # 4]
+	str r0, [lr, # 8]
+	str r0, [lr, #12]
+
+end_of_row_loop:
+	@ End of loop
+	add lr, lr, #16
+	subs r12, r12, #1
+	bne row_loop
+
+start_column_loop:
+	@ Start of column loop
+	ldr lr, [ sp ]
+	mov r12, #8
+column_loop:
+	ldrsh r0, [lr, #( 0*8)]             @ r0 = 'd0'
+	ldrsh r2, [lr, #( 4*8)]             @ r2 = 'd2'
+	ldrsh r4, [lr, #( 8*8)]             @ r4 = 'd4'
+	ldrsh r6, [lr, #(12*8)]             @ r6 = 'd6'
+
+	ldr r3, [r11, #FIX_0_541196100_ID]
+	add r1, r2, r6
+	ldr r5, [r11, #FIX_M_1_847759065_ID]
+	mul r1, r3, r1                      @ r1 = z1
+	ldr r3, [r11, #FIX_0_765366865_ID]
+	mla r6, r5, r6, r1                  @ r6 = tmp2
+	add r5, r0, r4                      @ r5 = tmp0
+	mla r2, r3, r2, r1                  @ r2 = tmp3
+	sub r3, r0, r4                      @ r3 = tmp1
+
+	add r0, r2, r5, lsl #13             @ r0 = tmp10
+	rsb r2, r2, r5, lsl #13             @ r2 = tmp13
+	add r4, r6, r3, lsl #13             @ r4 = tmp11
+	rsb r6, r6, r3, lsl #13             @ r6 = tmp12
+
+	ldrsh r1, [lr, #( 2*8)]             @ r1 = 'd1'
+	ldrsh r3, [lr, #( 6*8)]             @ r3 = 'd3'
+	ldrsh r5, [lr, #(10*8)]             @ r5 = 'd5'
+	ldrsh r7, [lr, #(14*8)]             @ r7 = 'd7'
+
+	@ Check for empty odd column (happens about 20 to 25 % of the time according to my stats)
+	orr r9, r1, r3
+	orr r10, r5, r7
+	orrs r10, r9, r10
+	beq empty_odd_column
+
+	stmdb   sp!, { r0, r2, r4, r6 } @ save on the stack tmp10, tmp13, tmp12, tmp11
+		
+	add r0, r3, r5	                @ r0 = 'z2'
+	add r2, r1, r7                  @ r2 = 'z1'
+	add r4, r3, r7                  @ r4 = 'z3'
+	add r6, r1, r5                  @ r6 = 'z4'
+	ldr r9, [r11, #FIX_1_175875602_ID]
+	add r8, r4, r6
+	ldr r10, [r11, #FIX_M_0_899976223_ID]
+	mul r8, r9, r8                  @ r8 = 'z5'
+	ldr r9, [r11, #FIX_M_2_562915447_ID]
+	mul r2, r10, r2                 @ r2 = 'z1'
+	ldr r10, [r11, #FIX_M_1_961570560_ID]
+	mul r0, r9, r0                  @ r0 = 'z2'
+	ldr r9, [r11, #FIX_M_0_390180644_ID]
+	mla r4, r10, r4, r8             @ r4 = 'z3'
+	ldr r10, [r11, #FIX_0_298631336_ID]
+	mla r6, r9, r6, r8              @ r6 = 'z4'
+	ldr r9, [r11, #FIX_2_053119869_ID]
+	mla r7, r10, r7, r2             @ r7 = tmp0 + z1
+	ldr r10, [r11, #FIX_3_072711026_ID]
+	mla r5, r9, r5, r0              @ r5 = tmp1 + z2
+	ldr r9, [r11, #FIX_1_501321110_ID]
+	mla r3, r10, r3, r0             @ r3 = tmp2 + z2
+	add r7, r7, r4                  @ r7 = tmp0
+	mla r1, r9, r1, r2              @ r1 = tmp3 + z1
+	add r5,	r5, r6                  @ r5 = tmp1
+	add r3, r3, r4                  @ r3 = tmp2
+	add r1, r1, r6                  @ r1 = tmp3	
+	
+	ldmia sp!, { r0, r2, r4, r6 } @ r0 = tmp10 / r2 = tmp13 / r4 = tmp11 / r6 = tmp12
+	                              @ r1 = tmp3  / r3 = tmp2  / r5 = tmp1  / r7 = tmp0	
+
+	@ Compute DESCALE(tmp10 + tmp3, CONST_BITS+PASS1_BITS+3)
+	add r8, r0, r1
+	add r8, r8, #(1<<17)
+	mov r8, r8, asr #18
+	strh r8, [lr, #( 0*8)]
+	
+	@ Compute DESCALE(tmp10 - tmp3, CONST_BITS+PASS1_BITS+3)
+	sub r8, r0, r1
+	add r8, r8, #(1<<17)
+	mov r8, r8, asr #18
+	strh r8, [lr, #(14*8)]
+	
+	@ Compute DESCALE(tmp11 + tmp2, CONST_BITS+PASS1_BITS+3)
+	add r8, r4, r3
+	add r8, r8, #(1<<17)
+	mov r8, r8, asr #18
+	strh r8, [lr, #( 2*8)]
+	
+	@ Compute DESCALE(tmp11 - tmp2, CONST_BITS+PASS1_BITS+3)
+	sub r8, r4, r3
+	add r8, r8, #(1<<17)
+	mov r8, r8, asr #18
+	strh r8, [lr, #(12*8)]
+	
+	@ Compute DESCALE(tmp12 + tmp1, CONST_BITS+PASS1_BITS+3)
+	add r8, r6, r5
+	add r8, r8, #(1<<17)
+	mov r8, r8, asr #18
+	strh r8, [lr, #( 4*8)]
+	
+	@ Compute DESCALE(tmp12 - tmp1, CONST_BITS+PASS1_BITS+3)
+	sub r8, r6, r5
+	add r8, r8, #(1<<17)
+	mov r8, r8, asr #18
+	strh r8, [lr, #(10*8)]
+	
+	@ Compute DESCALE(tmp13 + tmp0, CONST_BITS+PASS1_BITS+3)
+	add r8, r2, r7
+	add r8, r8, #(1<<17)
+	mov r8, r8, asr #18
+	strh r8, [lr, #( 6*8)]
+	
+	@ Compute DESCALE(tmp13 - tmp0, CONST_BITS+PASS1_BITS+3)
+	sub r8, r2, r7
+	add r8, r8, #(1<<17)
+	mov r8, r8, asr #18
+	strh r8, [lr, #( 8*8)]
+
+	@ End of row loop
+	add lr, lr, #2
+	subs r12, r12, #1
+	bne column_loop
+	beq the_end
+	
+empty_odd_column:
+	@ Compute DESCALE(tmp10 + tmp3, CONST_BITS+PASS1_BITS+3)
+	@ Compute DESCALE(tmp10 - tmp3, CONST_BITS+PASS1_BITS+3)
+	add r0, r0, #(1<<17)
+	mov r0, r0, asr #18
+	strh r0, [lr, #( 0*8)]
+	strh r0, [lr, #(14*8)]
+	
+	@ Compute DESCALE(tmp11 + tmp2, CONST_BITS+PASS1_BITS+3)
+	@ Compute DESCALE(tmp11 - tmp2, CONST_BITS+PASS1_BITS+3)
+	add r4, r4, #(1<<17)
+	mov r4, r4, asr #18
+	strh r4, [lr, #( 2*8)]
+	strh r4, [lr, #(12*8)]
+	
+	@ Compute DESCALE(tmp12 + tmp1, CONST_BITS+PASS1_BITS+3)
+	@ Compute DESCALE(tmp12 - tmp1, CONST_BITS+PASS1_BITS+3)
+	add r6, r6, #(1<<17)
+	mov r6, r6, asr #18
+	strh r6, [lr, #( 4*8)]
+	strh r6, [lr, #(10*8)]
+	
+	@ Compute DESCALE(tmp13 + tmp0, CONST_BITS+PASS1_BITS+3)
+	@ Compute DESCALE(tmp13 - tmp0, CONST_BITS+PASS1_BITS+3)
+	add r2, r2, #(1<<17)
+	mov r2, r2, asr #18
+	strh r2, [lr, #( 6*8)]
+	strh r2, [lr, #( 8*8)]
+
+	@ End of row loop
+	add lr, lr, #2
+	subs r12, r12, #1
+	bne column_loop
+		
+the_end:	
+	@ The end....
+	add sp, sp, #4
+	ldmia   sp!, { r4 - r12, pc }   @ restore callee saved regs and return
+
+const_array:
+	.align
+	.word FIX_0_298631336
+	.word FIX_0_541196100
+	.word FIX_0_765366865
+	.word FIX_1_175875602
+	.word FIX_1_501321110
+	.word FIX_2_053119869
+	.word FIX_3_072711026
+	.word FIX_M_0_390180644
+	.word FIX_M_0_899976223
+	.word FIX_M_1_847759065
+	.word FIX_M_1_961570560
+	.word FIX_M_2_562915447
+	.word FIX_0xFFFF
diff --git a/src/libffmpeg/libavcodec/i386/Makefile.am b/src/libffmpeg/libavcodec/i386/Makefile.am
new file mode 100644
index 000000000..72cbcc92c
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/Makefile.am
@@ -0,0 +1,43 @@
+##
+## Process this file with automake to produce Makefile.in
+##
+
+#CFLAGS = -D_FILE_OFFSET_BITS=64 @GLOBAL_CFLAGS@ -DCONFIG_DECODERS -DHAVE_AV_CONFIG_H
+
+CFLAGS = @GLOBAL_CFLAGS@ @LIBFFMPEG_CFLAGS@ -DCONFIG_DECODERS -DHAVE_AV_CONFIG_H
+ASFLAGS =
+
+LIBTOOL = $(SHELL) $(top_builddir)/libtool-nofpic
+
+noinst_LTLIBRARIES = libavcodec_mmx.la
+
+EXTRA_DIST =  fdct_mmx.c \
+              dsputil_mmx.c mpegvideo_mmx.c \
+              idct_mmx.c motion_est_mmx.c \
+              simple_idct_mmx.c
+
+if HAVE_FFMMX
+mmx_modules = $(EXTRA_DIST)
+endif
+
+
+libavcodec_mmx_la_SOURCES = $(mmx_modules)
+
+noinst_HEADERS = dsputil_mmx_avg.h  mmx.h
+
+.s.lo:
+	$(ASCOMPILE) -o $@ `test -f $< || echo '$(srcdir)/'`$<
+
+debug:
+	@$(MAKE) CFLAGS="@DEBUG_CFLAGS@ @LIBFFMPEG_CFLAGS@ -DCONFIG_DECODERS -DHAVE_AV_CONFIG_H"
+
+install-debug: debug
+	@$(MAKE) $(AM_MAKEFLAGS) install-exec-am install-data-am
+
+mostlyclean-generic:
+	-rm -f *~ \#* .*~ .\#*
+
+maintainer-clean-generic:
+	-@echo "This command is intended for maintainers to use;"
+	-@echo "it deletes files that may require special tools to rebuild."
+	-rm -f Makefile.in
diff --git a/src/libffmpeg/libavcodec/i386/dsputil_mmx.c b/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
new file mode 100644
index 000000000..8647ed187
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
@@ -0,0 +1,1056 @@
+/*
+ * MMX optimized DSP utils
+ * Copyright (c) 2000, 2001 Gerard Lantau.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
+ */
+
+#include "../dsputil.h"
+#include "../simple_idct.h"
+
+int mm_flags; /* multimedia extension flags */
+
+int pix_abs16x16_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
+int pix_abs16x16_sse(UINT8 *blk1, UINT8 *blk2, int lx, int h);
+int pix_abs16x16_x2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
+int pix_abs16x16_y2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
+int pix_abs16x16_xy2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h);
+
+/* external functions, from idct_mmx.c */
+void ff_mmx_idct(DCTELEM *block);
+void ff_mmxext_idct(DCTELEM *block);
+
+/* pixel operations */
+static const unsigned long long int mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001;
+static const unsigned long long int mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002;
+//static const unsigned short mm_wone[4] __attribute__ ((aligned(8))) = { 0x1, 0x1, 0x1, 0x1 };
+//static const unsigned short mm_wtwo[4] __attribute__ ((aligned(8))) = { 0x2, 0x2, 0x2, 0x2 };
+
+/***********************************/
+/* 3Dnow specific */
+
+#define DEF(x) x ## _3dnow
+/* for Athlons PAVGUSB is prefered */
+#define PAVGB "pavgusb"
+
+#include "dsputil_mmx_avg.h"
+
+#undef DEF
+#undef PAVGB
+
+/***********************************/
+/* MMX2 specific */
+
+#define DEF(x) x ## _sse
+
+/* Introduced only in MMX2 set */
+#define PAVGB "pavgb"
+
+#include "dsputil_mmx_avg.h"
+
+#undef DEF
+#undef PAVGB
+
+/***********************************/
+/* standard MMX */
+
+static void get_pixels_mmx(DCTELEM *block, const UINT8 *pixels, int line_size)
+{
+    DCTELEM *p;
+    const UINT8 *pix;
+    int i;
+
+    /* read the pixels */
+    p = block;
+    pix = pixels;
+    __asm __volatile("pxor %%mm7, %%mm7":);
+    for(i=0;i<4;i++) {
+	__asm __volatile(
+		"movq	%1, %%mm0\n\t"
+		"movq	%2, %%mm1\n\t"
+		"movq	%%mm0, %%mm2\n\t"
+		"movq	%%mm1, %%mm3\n\t"
+		"punpcklbw %%mm7, %%mm0\n\t"
+		"punpckhbw %%mm7, %%mm2\n\t"
+		"punpcklbw %%mm7, %%mm1\n\t"
+		"punpckhbw %%mm7, %%mm3\n\t"
+		"movq	%%mm0, %0\n\t"
+		"movq	%%mm2, 8%0\n\t"
+		"movq	%%mm1, 16%0\n\t"
+		"movq	%%mm3, 24%0\n\t"
+		:"=m"(*p)
+		:"m"(*pix), "m"(*(pix+line_size))
+		:"memory");
+        pix += line_size*2;
+        p += 16;
+    }
+}
+
+static void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size)
+{
+    const DCTELEM *p;
+    UINT8 *pix;
+    int i;
+
+    /* read the pixels */
+    p = block;
+    pix = pixels;
+    for(i=0;i<2;i++) {
+	__asm __volatile(
+		"movq	%3, %%mm0\n\t"
+		"movq	8%3, %%mm1\n\t"
+		"movq	16%3, %%mm2\n\t"
+		"movq	24%3, %%mm3\n\t"
+		"movq	32%3, %%mm4\n\t"
+		"movq	40%3, %%mm5\n\t"
+		"movq	48%3, %%mm6\n\t"
+		"movq	56%3, %%mm7\n\t"
+		"packuswb %%mm1, %%mm0\n\t"
+		"packuswb %%mm3, %%mm2\n\t"
+		"packuswb %%mm5, %%mm4\n\t"
+		"packuswb %%mm7, %%mm6\n\t"
+		"movq	%%mm0, (%0)\n\t"
+		"movq	%%mm2, (%0, %1)\n\t"
+		"movq	%%mm4, (%0, %1, 2)\n\t"
+		"movq	%%mm6, (%0, %2)\n\t"
+		::"r" (pix), "r" (line_size), "r" (line_size*3), "m"(*p)
+		:"memory");
+        pix += line_size*4;
+        p += 32;
+    }
+}
+
+static void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size)
+{
+    const DCTELEM *p;
+    UINT8 *pix;
+    int i;
+
+    /* read the pixels */
+    p = block;
+    pix = pixels;
+	__asm __volatile("pxor	%%mm7, %%mm7":);
+    for(i=0;i<4;i++) {
+	__asm __volatile(
+		"movq	%2, %%mm0\n\t"
+		"movq	8%2, %%mm1\n\t"
+		"movq	16%2, %%mm2\n\t"
+		"movq	24%2, %%mm3\n\t"
+		"movq	%0, %%mm4\n\t"
+		"movq	%1, %%mm6\n\t"
+		"movq	%%mm4, %%mm5\n\t"
+		"punpcklbw %%mm7, %%mm4\n\t"
+		"punpckhbw %%mm7, %%mm5\n\t"
+		"paddsw	%%mm4, %%mm0\n\t"
+		"paddsw	%%mm5, %%mm1\n\t"
+		"movq	%%mm6, %%mm5\n\t"
+		"punpcklbw %%mm7, %%mm6\n\t"
+		"punpckhbw %%mm7, %%mm5\n\t"
+		"paddsw	%%mm6, %%mm2\n\t"
+		"paddsw	%%mm5, %%mm3\n\t"
+		"packuswb %%mm1, %%mm0\n\t"
+		"packuswb %%mm3, %%mm2\n\t"
+		"movq	%%mm0, %0\n\t"
+		"movq	%%mm2, %1\n\t"
+		:"+m"(*pix), "+m"(*(pix+line_size))
+		:"m"(*p)
+		:"memory");
+        pix += line_size*2;
+        p += 16;
+    }
+}
+
+static void put_pixels_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
+{
+    int dh, hh;
+    UINT8 *p;
+    const UINT8 *pix;
+    p   = block;
+    pix = pixels;
+    hh=h>>2;
+    dh=h&3;
+    while(hh--) {
+    __asm __volatile(
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%1, %2), %%mm1		\n\t"
+	"movq	(%1, %2, 2), %%mm2	\n\t"
+	"movq	(%1, %3), %%mm3		\n\t"
+	"movq	%%mm0, (%0)		\n\t"
+	"movq	%%mm1, (%0, %2)		\n\t"
+	"movq	%%mm2, (%0, %2, 2)	\n\t"
+	"movq	%%mm3, (%0, %3)		\n\t"
+	::"r"(p), "r"(pix), "r"(line_size), "r"(line_size*3)
+	:"memory");
+        pix = pix + line_size*4;
+        p =   p   + line_size*4;
+    }
+    while(dh--) {
+     __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"=m"(*p)
+	:"m"(*pix)
+	:"memory");
+        pix = pix + line_size;
+        p =   p   + line_size;
+    }
+}
+
+static void put_pixels_x2_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8 *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm4\n\t"
+	::"m"(mm_wone));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	1%1, %%mm1\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm4, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"=m"(*p)
+	:"m"(*pix)
+	:"memory");
+   pix += line_size; p += line_size;
+  } while (--h);
+}
+
+static void put_pixels_y2_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8 *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm4\n\t"
+	::"m"(mm_wone));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm4, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"=m"(*p)
+	:"m"(*pix),
+	 "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p += line_size;
+  } while (--h);
+}
+
+static void put_pixels_xy2_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8 *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm6\n\t"
+	::"m"(mm_wtwo));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	"movq	1%1, %%mm4\n\t"
+	"movq	1%2, %%mm5\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"movq	%%mm4, %%mm1\n\t"
+	"movq	%%mm5, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpcklbw %%mm7, %%mm5\n\t"
+	"punpckhbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm5, %%mm4\n\t"
+	"paddusw %%mm3, %%mm1\n\t"
+	"paddusw %%mm6, %%mm4\n\t"
+	"paddusw %%mm6, %%mm1\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm1, %%mm2\n\t"
+	"psrlw	$2, %%mm0\n\t"
+	"psrlw	$2, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"=m"(*p)
+	:"m"(*pix),
+	 "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p += line_size;
+  } while(--h);
+}
+
+static void   put_no_rnd_pixels_x2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile("pxor %%mm7, %%mm7\n\t":);
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	1%1, %%mm1\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"=m"(*p)
+	:"m"(*pix)
+	:"memory");
+   pix += line_size;
+   p +=   line_size;
+  } while (--h);
+}
+
+static void put_no_rnd_pixels_y2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile("pxor %%mm7, %%mm7\n\t":);
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"=m"(*p)
+	:"m"(*pix),
+	 "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p +=   line_size;
+  } while(--h);
+}
+
+static void   put_no_rnd_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm6\n\t"
+	::"m"(mm_wone));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	"movq	1%1, %%mm4\n\t"
+	"movq	1%2, %%mm5\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"movq	%%mm4, %%mm1\n\t"
+	"movq	%%mm5, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpcklbw %%mm7, %%mm5\n\t"
+	"punpckhbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm5, %%mm4\n\t"
+	"paddusw %%mm3, %%mm1\n\t"
+	"paddusw %%mm6, %%mm4\n\t"
+	"paddusw %%mm6, %%mm1\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm1, %%mm2\n\t"
+	"psrlw	$2, %%mm0\n\t"
+	"psrlw	$2, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"=m"(*p)
+	:"m"(*pix),
+	 "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p +=   line_size;
+  } while(--h);
+}
+
+static void avg_pixels_mmx(UINT8 *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm6\n\t"
+	::"m"(mm_wone));
+  do {
+    __asm __volatile(
+	"movq	%0, %%mm0\n\t"
+	"movq	%1, %%mm1\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"paddusw %%mm6, %%mm0\n\t"
+	"paddusw %%mm6, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix)
+	:"memory");
+   pix += line_size;
+   p +=   line_size;
+  }
+  while (--h);
+}
+
+static void   avg_pixels_x2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm6\n\t"
+	::"m"(mm_wone));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm1\n\t"
+	"movq	%0, %%mm0\n\t"
+	"movq	1%1, %%mm4\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"movq	%%mm4, %%mm5\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpckhbw %%mm7, %%mm5\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"paddusw %%mm4, %%mm1\n\t"
+	"paddusw %%mm5, %%mm3\n\t"
+	"paddusw %%mm6, %%mm1\n\t"
+	"paddusw %%mm6, %%mm3\n\t"
+	"psrlw	$1, %%mm1\n\t"
+	"psrlw	$1, %%mm3\n\t"
+	"paddusw %%mm6, %%mm0\n\t"
+	"paddusw %%mm6, %%mm2\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix)
+	:"memory");
+   pix += line_size;
+   p +=   line_size;
+  } while (--h);
+}
+
+static void   avg_pixels_y2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm6\n\t"
+	::"m"(mm_wone));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm1\n\t"
+	"movq	%0, %%mm0\n\t"
+	"movq	%2, %%mm4\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"movq	%%mm4, %%mm5\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpckhbw %%mm7, %%mm5\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"paddusw %%mm4, %%mm1\n\t"
+	"paddusw %%mm5, %%mm3\n\t"
+	"paddusw %%mm6, %%mm1\n\t"
+	"paddusw %%mm6, %%mm3\n\t"
+	"psrlw	$1, %%mm1\n\t"
+	"psrlw	$1, %%mm3\n\t"
+	"paddusw %%mm6, %%mm0\n\t"
+	"paddusw %%mm6, %%mm2\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix), "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p +=   line_size ;
+  } while(--h);
+}
+
+static void   avg_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm6\n\t"
+	::"m"(mm_wtwo));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	"movq	1%1, %%mm4\n\t"
+	"movq	1%2, %%mm5\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"movq	%%mm4, %%mm1\n\t"
+	"movq	%%mm5, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpcklbw %%mm7, %%mm5\n\t"
+	"punpckhbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm5, %%mm4\n\t"
+	"paddusw %%mm3, %%mm1\n\t"
+	"paddusw %%mm6, %%mm4\n\t"
+	"paddusw %%mm6, %%mm1\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm1, %%mm2\n\t"
+	"movq	%3, %%mm5\n\t"
+	"psrlw	$2, %%mm0\n\t"
+	"movq	%0, %%mm1\n\t"
+	"psrlw	$2, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"paddusw %%mm5, %%mm0\n\t"
+	"paddusw %%mm5, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix),
+	 "m"(*(pix+line_size)), "m"(mm_wone)
+	:"memory");
+   pix += line_size;
+   p +=   line_size ;
+  } while(--h);
+}
+
+static void avg_no_rnd_pixels_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile("pxor %%mm7, %%mm7\n\t":);
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%0, %%mm1\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix)
+	:"memory");
+   pix += line_size;
+   p +=   line_size ;
+  } while (--h);
+}
+
+static void   avg_no_rnd_pixels_x2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+      "pxor	%%mm7, %%mm7\n\t":);
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	1%1, %%mm1\n\t"
+	"movq	%0, %%mm4\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"movq	%%mm4, %%mm5\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpckhbw %%mm7, %%mm5\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm5, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix)
+	:"memory");
+   pix += line_size;
+   p +=   line_size;
+ } while (--h);
+}
+
+static void   avg_no_rnd_pixels_y2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+      "pxor	%%mm7, %%mm7\n\t":);
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	"movq	%0, %%mm4\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"movq	%%mm4, %%mm5\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpckhbw %%mm7, %%mm5\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm5, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix), "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p +=   line_size ;
+  } while(--h);
+}
+
+static void   avg_no_rnd_pixels_xy2_mmx( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm6\n\t"
+	::"m"(mm_wone));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	"movq	1%1, %%mm4\n\t"
+	"movq	1%2, %%mm5\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"movq	%%mm4, %%mm1\n\t"
+	"movq	%%mm5, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpcklbw %%mm7, %%mm5\n\t"
+	"punpckhbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm5, %%mm4\n\t"
+	"paddusw %%mm3, %%mm1\n\t"
+	"paddusw %%mm6, %%mm4\n\t"
+	"paddusw %%mm6, %%mm1\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm1, %%mm2\n\t"
+	"movq	%0, %%mm1\n\t"
+	"psrlw	$2, %%mm0\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"psrlw	$2, %%mm2\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"psrlw	$1, %%mm0\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix),
+	 "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p += line_size;
+  } while(--h);
+}
+
+static void sub_pixels_mmx( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)
+{
+  DCTELEM  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile("pxor %%mm7, %%mm7":);
+  do {
+    __asm __volatile(
+	"movq	%0, %%mm0\n\t"
+	"movq	%1, %%mm2\n\t"
+	"movq	8%0, %%mm1\n\t"
+	"movq	%%mm2, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"psubsw %%mm2, %%mm0\n\t"
+	"psubsw %%mm3, %%mm1\n\t"
+	"movq	%%mm0, %0\n\t"
+	"movq	%%mm1, 8%0\n\t"
+	:"+m"(*p)
+	:"m"(*pix)
+	:"memory");
+   pix += line_size;
+   p +=   8;
+  } while (--h);
+}
+
+static void sub_pixels_x2_mmx( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)
+{
+  DCTELEM  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+      "pxor	%%mm7, %%mm7\n\t"
+      "movq	%0, %%mm6"
+      ::"m"(mm_wone));
+  do {
+    __asm __volatile(
+	"movq	%0, %%mm0\n\t"
+	"movq	%1, %%mm2\n\t"
+	"movq	8%0, %%mm1\n\t"
+	"movq	1%1, %%mm4\n\t"
+	"movq	%%mm2, %%mm3\n\t"
+	"movq	%%mm4, %%mm5\n\t"
+	"punpcklbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpckhbw %%mm7, %%mm5\n\t"
+	"paddusw %%mm4, %%mm2\n\t"
+	"paddusw %%mm5, %%mm3\n\t"
+	"paddusw %%mm6, %%mm2\n\t"
+	"paddusw %%mm6, %%mm3\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"psrlw	$1, %%mm3\n\t"
+	"psubsw %%mm2, %%mm0\n\t"
+	"psubsw %%mm3, %%mm1\n\t"
+	"movq	%%mm0, %0\n\t"
+	"movq	%%mm1, 8%0\n\t"
+	:"+m"(*p)
+	:"m"(*pix)
+	:"memory");
+   pix += line_size;
+   p +=   8;
+ } while (--h);
+}
+
+static void sub_pixels_y2_mmx( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)
+{
+  DCTELEM  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+      "pxor	%%mm7, %%mm7\n\t"
+      "movq	%0, %%mm6"
+      ::"m"(mm_wone));
+  do {
+    __asm __volatile(
+	"movq	%0, %%mm0\n\t"
+	"movq	%1, %%mm2\n\t"
+	"movq	8%0, %%mm1\n\t"
+	"movq	%2, %%mm4\n\t"
+	"movq	%%mm2, %%mm3\n\t"
+	"movq	%%mm4, %%mm5\n\t"
+	"punpcklbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpckhbw %%mm7, %%mm5\n\t"
+	"paddusw %%mm4, %%mm2\n\t"
+	"paddusw %%mm5, %%mm3\n\t"
+	"paddusw %%mm6, %%mm2\n\t"
+	"paddusw %%mm6, %%mm3\n\t"
+	"psrlw	$1, %%mm2\n\t"
+	"psrlw	$1, %%mm3\n\t"
+	"psubsw %%mm2, %%mm0\n\t"
+	"psubsw %%mm3, %%mm1\n\t"
+	"movq	%%mm0, %0\n\t"
+	"movq	%%mm1, 8%0\n\t"
+	:"+m"(*p)
+	:"m"(*pix), "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p +=   8;
+ } while (--h);
+}
+
+static void   sub_pixels_xy2_mmx( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)
+{
+  DCTELEM  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm6\n\t"
+	::"m"(mm_wtwo));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	"movq	1%1, %%mm4\n\t"
+	"movq	1%2, %%mm5\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"movq	%%mm4, %%mm1\n\t"
+	"movq	%%mm5, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpcklbw %%mm7, %%mm5\n\t"
+	"punpckhbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm5, %%mm4\n\t"
+	"paddusw %%mm3, %%mm1\n\t"
+	"paddusw %%mm6, %%mm4\n\t"
+	"paddusw %%mm6, %%mm1\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm1, %%mm2\n\t"
+	"movq	%0, %%mm1\n\t"
+	"movq	8%0, %%mm3\n\t"
+	"psrlw	$2, %%mm0\n\t"
+	"psrlw	$2, %%mm2\n\t"
+	"psubsw %%mm0, %%mm1\n\t"
+	"psubsw %%mm2, %%mm3\n\t"
+	"movq	%%mm1, %0\n\t"
+	"movq	%%mm3, 8%0\n\t"
+	:"+m"(*p)
+	:"m"(*pix),
+	 "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p +=   8 ;
+  } while(--h);
+}
+
+void dsputil_init_mmx(void)
+{
+    mm_flags = mm_support();
+#if 0
+    printf("CPU flags:");
+    if (mm_flags & MM_MMX)
+        printf(" mmx");
+    if (mm_flags & MM_MMXEXT)
+        printf(" mmxext");
+    if (mm_flags & MM_3DNOW)
+        printf(" 3dnow");
+    if (mm_flags & MM_SSE)
+        printf(" sse");
+    if (mm_flags & MM_SSE2)
+        printf(" sse2");
+    printf("\n");
+#endif
+
+    if (mm_flags & MM_MMX) {
+        get_pixels = get_pixels_mmx;
+        put_pixels_clamped = put_pixels_clamped_mmx;
+        add_pixels_clamped = add_pixels_clamped_mmx;
+        
+        pix_abs16x16 = pix_abs16x16_mmx;
+        pix_abs16x16_x2 = pix_abs16x16_x2_mmx;
+        pix_abs16x16_y2 = pix_abs16x16_y2_mmx;
+        pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;
+        av_fdct = fdct_mmx;
+
+        put_pixels_tab[0] = put_pixels_mmx;
+        put_pixels_tab[1] = put_pixels_x2_mmx;
+        put_pixels_tab[2] = put_pixels_y2_mmx;
+        put_pixels_tab[3] = put_pixels_xy2_mmx;
+
+        put_no_rnd_pixels_tab[0] = put_pixels_mmx;
+        put_no_rnd_pixels_tab[1] = put_no_rnd_pixels_x2_mmx;
+        put_no_rnd_pixels_tab[2] = put_no_rnd_pixels_y2_mmx;
+        put_no_rnd_pixels_tab[3] = put_no_rnd_pixels_xy2_mmx;
+        
+        avg_pixels_tab[0] = avg_pixels_mmx;
+        avg_pixels_tab[1] = avg_pixels_x2_mmx;
+        avg_pixels_tab[2] = avg_pixels_y2_mmx;
+        avg_pixels_tab[3] = avg_pixels_xy2_mmx;
+
+        avg_no_rnd_pixels_tab[0] = avg_no_rnd_pixels_mmx;
+        avg_no_rnd_pixels_tab[1] = avg_no_rnd_pixels_x2_mmx;
+        avg_no_rnd_pixels_tab[2] = avg_no_rnd_pixels_y2_mmx;
+        avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels_xy2_mmx;
+        
+        sub_pixels_tab[0] = sub_pixels_mmx;
+        sub_pixels_tab[1] = sub_pixels_x2_mmx;
+        sub_pixels_tab[2] = sub_pixels_y2_mmx;
+        sub_pixels_tab[3] = sub_pixels_xy2_mmx;
+
+        if (mm_flags & MM_MMXEXT) {
+            pix_abs16x16 = pix_abs16x16_sse;
+        }
+
+        if (mm_flags & MM_SSE) {
+            put_pixels_tab[1] = put_pixels_x2_sse;
+            put_pixels_tab[2] = put_pixels_y2_sse;
+            
+            avg_pixels_tab[0] = avg_pixels_sse;
+            avg_pixels_tab[1] = avg_pixels_x2_sse;
+            avg_pixels_tab[2] = avg_pixels_y2_sse;
+            avg_pixels_tab[3] = avg_pixels_xy2_sse;
+
+            sub_pixels_tab[1] = sub_pixels_x2_sse;
+            sub_pixels_tab[2] = sub_pixels_y2_sse;
+        } else if (mm_flags & MM_3DNOW) {
+            put_pixels_tab[1] = put_pixels_x2_3dnow;
+            put_pixels_tab[2] = put_pixels_y2_3dnow;
+            
+            avg_pixels_tab[0] = avg_pixels_3dnow;
+            avg_pixels_tab[1] = avg_pixels_x2_3dnow;
+            avg_pixels_tab[2] = avg_pixels_y2_3dnow;
+            avg_pixels_tab[3] = avg_pixels_xy2_3dnow;
+
+            sub_pixels_tab[1] = sub_pixels_x2_3dnow;
+            sub_pixels_tab[2] = sub_pixels_y2_3dnow;
+        }
+
+        /* idct */
+        if (mm_flags & MM_MMXEXT) {
+            ff_idct = ff_mmxext_idct;
+        } else {
+            ff_idct = ff_mmx_idct;
+        }
+#ifdef SIMPLE_IDCT
+//	ff_idct = simple_idct;
+	ff_idct = simple_idct_mmx;
+#endif
+    }
+}
diff --git a/src/libffmpeg/libavcodec/i386/dsputil_mmx_avg.h b/src/libffmpeg/libavcodec/i386/dsputil_mmx_avg.h
new file mode 100644
index 000000000..830fe9f3b
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/dsputil_mmx_avg.h
@@ -0,0 +1,342 @@
+/*
+ * DSP utils : average functions are compiled twice for 3dnow/mmx2
+ * Copyright (c) 2000, 2001 Gerard Lantau.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
+ */
+
+static void DEF(put_pixels_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
+{
+  int dh, hh;
+  UINT8 *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  hh=h>>2;
+  dh=h&3;
+  while(hh--) {
+    __asm __volatile(
+	"movq	(%1), %%mm0\n\t"
+	"movq	1(%1), %%mm1\n\t"
+	"movq	(%1, %2), %%mm2\n\t"
+	"movq	1(%1, %2), %%mm3\n\t"
+	"movq	(%1, %2, 2), %%mm4\n\t"
+	"movq	1(%1, %2, 2), %%mm5\n\t"
+	"movq	(%1, %3), %%mm6\n\t"
+	"movq	1(%1, %3), %%mm7\n\t"
+	PAVGB"  %%mm1, %%mm0\n\t"
+	PAVGB"  %%mm3, %%mm2\n\t"
+	PAVGB"  %%mm5, %%mm4\n\t"
+	PAVGB"  %%mm7, %%mm6\n\t"
+	"movq	%%mm0, (%0)\n\t"
+	"movq	%%mm2, (%0, %2)\n\t"
+	"movq	%%mm4, (%0, %2, 2)\n\t"
+	"movq	%%mm6, (%0, %3)\n\t"
+	::"r"(p), "r"(pix), "r" (line_size), "r" (line_size*3)
+	:"memory");
+     pix += line_size*4; p += line_size*4;
+  }
+  while(dh--) {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	1%1, %%mm1\n\t"
+	PAVGB"  %%mm1, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"=m"(*p)
+	:"m"(*pix)
+	:"memory");
+     pix += line_size; p += line_size;
+  }
+}
+
+static void DEF(put_pixels_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
+{
+  int dh, hh;
+  UINT8 *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+
+  hh=h>>1;
+  dh=h&1;
+  while(hh--) {
+    __asm __volatile(
+	"movq	%2, %%mm0\n\t"
+	"movq	%3, %%mm1\n\t"
+	"movq	%4, %%mm2\n\t"
+	PAVGB"  %%mm1, %%mm0\n\t"
+	PAVGB"  %%mm2, %%mm1\n\t"
+	"movq	%%mm0, %0\n\t"
+	"movq	%%mm1, %1\n\t"
+	:"=m"(*p), "=m"(*(p+line_size))
+	:"m"(*pix), "m"(*(pix+line_size)),
+	 "m"(*(pix+line_size*2))
+	:"memory");
+     pix += line_size*2;
+     p += line_size*2;
+  }
+  if(dh) {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	PAVGB"  %%mm1, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"=m"(*p)
+	:"m"(*pix),
+	 "m"(*(pix+line_size))
+	:"memory");
+  }
+}
+
+static void DEF(avg_pixels)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
+{
+  int dh, hh;
+  UINT8 *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  hh=h>>2;
+  dh=h&3;
+  while(hh--) {
+    __asm __volatile(
+	"movq	(%0), %%mm0\n\t"
+	"movq	(%1), %%mm1\n\t"
+	"movq	(%0, %2), %%mm2\n\t"
+	"movq	(%1, %2), %%mm3\n\t"
+	"movq	(%0, %2, 2), %%mm4\n\t"
+	"movq	(%1, %2, 2), %%mm5\n\t"
+	"movq	(%0, %3), %%mm6\n\t"
+	"movq	(%1, %3), %%mm7\n\t"
+	PAVGB"  %%mm1, %%mm0\n\t"
+	PAVGB"  %%mm3, %%mm2\n\t"
+	PAVGB"  %%mm5, %%mm4\n\t"
+	PAVGB"  %%mm7, %%mm6\n\t"
+	"movq	%%mm0, (%0)\n\t"
+	"movq	%%mm2, (%0, %2)\n\t"
+	"movq	%%mm4, (%0, %2, 2)\n\t"
+	"movq	%%mm6, (%0, %3)\n\t"
+	::"r"(p), "r"(pix), "r" (line_size), "r" (line_size*3)
+	:"memory");
+     pix += line_size*4; p += line_size*4;
+  }
+  while(dh--) {
+    __asm __volatile(
+	"movq	%0, %%mm0\n\t"
+	"movq	%1, %%mm1\n\t"
+	PAVGB"  %%mm1, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix)
+	:"memory");
+     pix += line_size; p += line_size;
+  }
+}
+
+static void DEF(avg_pixels_x2)( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  int dh, hh;
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  hh=h>>1;
+  dh=h&1;
+  while(hh--) {
+    __asm __volatile(
+	"movq	%2, %%mm2\n\t"
+	"movq	1%2, %%mm3\n\t"
+	"movq	%3, %%mm4\n\t"
+	"movq	1%3, %%mm5\n\t"
+	"movq	%0, %%mm0\n\t"
+	"movq	%1, %%mm1\n\t"
+	PAVGB"	%%mm3, %%mm2\n\t"
+	PAVGB"	%%mm2, %%mm0\n\t"
+	PAVGB"	%%mm5, %%mm4\n\t"
+	PAVGB"	%%mm4, %%mm1\n\t"
+	"movq	%%mm0, %0\n\t"
+	"movq	%%mm1, %1\n\t"
+	:"+m"(*p), "+m"(*(p+line_size))
+	:"m"(*pix), "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size*2;
+   p +=   line_size*2;
+  }
+  if(dh) {
+    __asm __volatile(
+	"movq	%1, %%mm1\n\t"
+	"movq	1%1, %%mm2\n\t"
+	"movq	%0, %%mm0\n\t"
+	PAVGB"	%%mm2, %%mm1\n\t"
+	PAVGB"	%%mm1, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix)
+	:"memory");
+  }
+}
+
+static void  DEF(avg_pixels_y2)( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  int dh, hh;
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  hh=h>>1;
+  dh=h&1;
+  while(hh--) {
+    __asm __volatile(
+	"movq	%2, %%mm2\n\t"
+	"movq	%3, %%mm3\n\t"
+	"movq	%3, %%mm4\n\t"
+	"movq	%4, %%mm5\n\t"
+	"movq	%0, %%mm0\n\t"
+	"movq	%1, %%mm1\n\t"
+	PAVGB"	%%mm3, %%mm2\n\t"
+	PAVGB"	%%mm2, %%mm0\n\t"
+	PAVGB"	%%mm5, %%mm4\n\t"
+	PAVGB"	%%mm4, %%mm1\n\t"
+	"movq	%%mm0, %0\n\t"
+	"movq	%%mm1, %1\n\t"
+	:"+m"(*p), "+m"(*(p+line_size))
+	:"m"(*pix), "m"(*(pix+line_size)), "m"(*(pix+line_size*2))
+	:"memory");
+   pix += line_size*2;
+   p +=   line_size*2;
+  }
+  if(dh) {
+    __asm __volatile(
+	"movq	%1, %%mm1\n\t"
+	"movq	%2, %%mm2\n\t"
+	"movq	%0, %%mm0\n\t"
+	PAVGB"	%%mm2, %%mm1\n\t"
+	PAVGB"	%%mm1, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix), "m"(*(pix+line_size))
+	:"memory");
+  }
+}
+
+static void DEF(avg_pixels_xy2)( UINT8  *block, const UINT8 *pixels, int line_size, int h)
+{
+  UINT8  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+	"pxor	%%mm7, %%mm7\n\t"
+	"movq	%0, %%mm6\n\t"
+	::"m"(mm_wtwo));
+  do {
+    __asm __volatile(
+	"movq	%1, %%mm0\n\t"
+	"movq	%2, %%mm1\n\t"
+	"movq	1%1, %%mm4\n\t"
+	"movq	1%2, %%mm5\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm0\n\t"
+	"punpcklbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm1, %%mm0\n\t"
+	"paddusw %%mm3, %%mm2\n\t"
+	"movq	%%mm4, %%mm1\n\t"
+	"movq	%%mm5, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm4\n\t"
+	"punpcklbw %%mm7, %%mm5\n\t"
+	"punpckhbw %%mm7, %%mm1\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"paddusw %%mm5, %%mm4\n\t"
+	"paddusw %%mm3, %%mm1\n\t"
+	"paddusw %%mm6, %%mm4\n\t"
+	"paddusw %%mm6, %%mm1\n\t"
+	"paddusw %%mm4, %%mm0\n\t"
+	"paddusw %%mm1, %%mm2\n\t"
+	"psrlw	$2, %%mm0\n\t"
+	"psrlw	$2, %%mm2\n\t"
+	"packuswb  %%mm2, %%mm0\n\t"
+	PAVGB"	%0, %%mm0\n\t"
+	"movq	%%mm0, %0\n\t"
+	:"+m"(*p)
+	:"m"(*pix),
+	 "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p +=   line_size ;
+  } while(--h);
+}
+
+static void DEF(sub_pixels_x2)( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)
+{
+  DCTELEM  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+      "pxor	%%mm7, %%mm7":);
+  do {
+    __asm __volatile(
+	"movq	1%1, %%mm2\n\t"
+	"movq	%0, %%mm0\n\t"
+	PAVGB"	%1, %%mm2\n\t"
+	"movq	8%0, %%mm1\n\t"
+	"movq	%%mm2, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"psubsw %%mm2, %%mm0\n\t"
+	"psubsw %%mm3, %%mm1\n\t"
+	"movq	%%mm0, %0\n\t"
+	"movq	%%mm1, 8%0\n\t"
+	:"+m"(*p)
+	:"m"(*pix)
+	:"memory");
+   pix += line_size;
+   p +=   8;
+ } while (--h);
+}
+
+static void DEF(sub_pixels_y2)( DCTELEM  *block, const UINT8 *pixels, int line_size, int h)
+{
+  DCTELEM  *p;
+  const UINT8 *pix;
+  p = block;
+  pix = pixels;
+  __asm __volatile(
+      "pxor	%%mm7, %%mm7":);
+  do {
+    __asm __volatile(
+	"movq	%2, %%mm2\n\t"
+	"movq	%0, %%mm0\n\t"
+	PAVGB"	%1, %%mm2\n\t"
+	"movq	8%0, %%mm1\n\t"
+	"movq	%%mm2, %%mm3\n\t"
+	"punpcklbw %%mm7, %%mm2\n\t"
+	"punpckhbw %%mm7, %%mm3\n\t"
+	"psubsw %%mm2, %%mm0\n\t"
+	"psubsw %%mm3, %%mm1\n\t"
+	"movq	%%mm0, %0\n\t"
+	"movq	%%mm1, 8%0\n\t"
+	:"+m"(*p)
+	:"m"(*pix), "m"(*(pix+line_size))
+	:"memory");
+   pix += line_size;
+   p +=   8;
+ } while (--h);
+}
+
diff --git a/src/libffmpeg/libavcodec/i386/fdct_mmx.c b/src/libffmpeg/libavcodec/i386/fdct_mmx.c
new file mode 100644
index 000000000..a71b89bb1
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/fdct_mmx.c
@@ -0,0 +1,291 @@
+/*
+ * MMX optimized forward DCT
+ * The gcc porting is Copyright (c) 2001 Gerard Lantau.
+ *
+ * from  fdctam32.c - AP922 MMX(3D-Now) forward-DCT
+ * 
+ *  Intel Application Note AP-922 - fast, precise implementation of DCT
+ *        http://developer.intel.com/vtune/cbts/appnotes.htm
+ */
+#include "../common.h"
+#include "mmx.h"
+
+#define ATTR_ALIGN(align) __attribute__ ((__aligned__ (align)))
+
+//////////////////////////////////////////////////////////////////////
+//
+// constants for the forward DCT
+// -----------------------------
+//
+// Be sure to check that your compiler is aligning all constants to QWORD
+// (8-byte) memory boundaries!  Otherwise the unaligned memory access will
+// severely stall MMX execution.
+//
+//////////////////////////////////////////////////////////////////////
+
+#define BITS_FRW_ACC	3 //; 2 or 3 for accuracy
+#define SHIFT_FRW_COL	BITS_FRW_ACC
+#define SHIFT_FRW_ROW	(BITS_FRW_ACC + 17)
+//#define RND_FRW_ROW		(262144 * (BITS_FRW_ACC - 1)) //; 1 << (SHIFT_FRW_ROW-1)
+#define RND_FRW_ROW		(1 << (SHIFT_FRW_ROW-1))
+//#define RND_FRW_COL		(2 * (BITS_FRW_ACC - 1)) //; 1 << (SHIFT_FRW_COL-1)
+#define RND_FRW_COL		(1 << (SHIFT_FRW_COL-1))
+
+//concatenated table, for forward DCT transformation
+const int16_t fdct_tg_all_16[] ATTR_ALIGN(8) = {
+    13036, 13036, 13036, 13036,		// tg * (2<<16) + 0.5
+    27146, 27146, 27146, 27146,		// tg * (2<<16) + 0.5
+    -21746, -21746, -21746, -21746,	// tg * (2<<16) + 0.5
+};
+const int16_t cos_4_16[4] = {
+    -19195, -19195, -19195, -19195,	//cos * (2<<16) + 0.5
+};
+
+const int16_t ocos_4_16[4] = {
+    23170, 23170, 23170, 23170,	//cos * (2<<15) + 0.5
+};
+
+const mmx_t fdct_one_corr = {0x0001000100010001LL};
+const mmx_t fdct_r_row = {d:{RND_FRW_ROW, RND_FRW_ROW} };
+
+const int16_t tab_frw_01234567[] ATTR_ALIGN(8) = {  // forward_dct coeff table
+    //row0
+    16384, 16384, 21407, -8867,     //    w09 w01 w08 w00
+    16384, 16384, 8867, -21407,     //    w13 w05 w12 w04
+    16384, -16384, 8867, 21407,     //    w11 w03 w10 w02
+    -16384, 16384, -21407, -8867,   //    w15 w07 w14 w06
+    22725, 12873, 19266, -22725,    //    w22 w20 w18 w16
+    19266, 4520, -4520, -12873,     //    w23 w21 w19 w17
+    12873, 4520, 4520, 19266,       //    w30 w28 w26 w24
+    -22725, 19266, -12873, -22725,  //    w31 w29 w27 w25
+
+    //row1
+    22725, 22725, 29692, -12299,    //    w09 w01 w08 w00
+    22725, 22725, 12299, -29692,    //    w13 w05 w12 w04
+    22725, -22725, 12299, 29692,    //    w11 w03 w10 w02
+    -22725, 22725, -29692, -12299,  //    w15 w07 w14 w06
+    31521, 17855, 26722, -31521,    //    w22 w20 w18 w16
+    26722, 6270, -6270, -17855,     //    w23 w21 w19 w17
+    17855, 6270, 6270, 26722,       //    w30 w28 w26 w24
+    -31521, 26722, -17855, -31521,  //    w31 w29 w27 w25
+
+    //row2
+    21407, 21407, 27969, -11585,    //    w09 w01 w08 w00
+    21407, 21407, 11585, -27969,    //    w13 w05 w12 w04
+    21407, -21407, 11585, 27969,    //    w11 w03 w10 w02
+    -21407, 21407, -27969, -11585,  //    w15 w07 w14 w06
+    29692, 16819, 25172, -29692,    //    w22 w20 w18 w16
+    25172, 5906, -5906, -16819,     //    w23 w21 w19 w17
+    16819, 5906, 5906, 25172,       //    w30 w28 w26 w24
+    -29692, 25172, -16819, -29692,  //    w31 w29 w27 w25
+
+    //row3
+    19266, 19266, 25172, -10426,    //    w09 w01 w08 w00
+    19266, 19266, 10426, -25172,    //    w13 w05 w12 w04
+    19266, -19266, 10426, 25172,    //    w11 w03 w10 w02
+    -19266, 19266, -25172, -10426,  //    w15 w07 w14 w06, 
+    26722, 15137, 22654, -26722,    //    w22 w20 w18 w16
+    22654, 5315, -5315, -15137,     //    w23 w21 w19 w17
+    15137, 5315, 5315, 22654,       //    w30 w28 w26 w24
+    -26722, 22654, -15137, -26722,  //    w31 w29 w27 w25, 
+
+    //row4
+    16384, 16384, 21407, -8867,     //    w09 w01 w08 w00
+    16384, 16384, 8867, -21407,     //    w13 w05 w12 w04
+    16384, -16384, 8867, 21407,     //    w11 w03 w10 w02
+    -16384, 16384, -21407, -8867,   //    w15 w07 w14 w06
+    22725, 12873, 19266, -22725,    //    w22 w20 w18 w16
+    19266, 4520, -4520, -12873,     //    w23 w21 w19 w17
+    12873, 4520, 4520, 19266,       //    w30 w28 w26 w24
+    -22725, 19266, -12873, -22725,  //    w31 w29 w27 w25 
+
+    //row5
+    19266, 19266, 25172, -10426,    //    w09 w01 w08 w00
+    19266, 19266, 10426, -25172,    //    w13 w05 w12 w04
+    19266, -19266, 10426, 25172,    //    w11 w03 w10 w02
+    -19266, 19266, -25172, -10426,  //    w15 w07 w14 w06
+    26722, 15137, 22654, -26722,    //    w22 w20 w18 w16
+    22654, 5315, -5315, -15137,     //    w23 w21 w19 w17
+    15137, 5315, 5315, 22654,       //    w30 w28 w26 w24
+    -26722, 22654, -15137, -26722,  //    w31 w29 w27 w25
+
+    //row6
+    21407, 21407, 27969, -11585,    //    w09 w01 w08 w00
+    21407, 21407, 11585, -27969,    //    w13 w05 w12 w04
+    21407, -21407, 11585, 27969,    //    w11 w03 w10 w02
+    -21407, 21407, -27969, -11585,  //    w15 w07 w14 w06, 
+    29692, 16819, 25172, -29692,    //    w22 w20 w18 w16
+    25172, 5906, -5906, -16819,     //    w23 w21 w19 w17
+    16819, 5906, 5906, 25172,       //    w30 w28 w26 w24
+    -29692, 25172, -16819, -29692,  //    w31 w29 w27 w25, 
+
+    //row7
+    22725, 22725, 29692, -12299,    //    w09 w01 w08 w00
+    22725, 22725, 12299, -29692,    //    w13 w05 w12 w04
+    22725, -22725, 12299, 29692,    //    w11 w03 w10 w02
+    -22725, 22725, -29692, -12299,  //    w15 w07 w14 w06, 
+    31521, 17855, 26722, -31521,    //    w22 w20 w18 w16
+    26722, 6270, -6270, -17855,     //    w23 w21 w19 w17
+    17855, 6270, 6270, 26722,       //    w30 w28 w26 w24
+    -31521, 26722, -17855, -31521   //    w31 w29 w27 w25
+};
+
+
+static inline void fdct_col(const int16_t *in, int16_t *out, int offset)
+{
+    movq_m2r(*(in + offset + 1 * 8), mm0);
+    movq_m2r(*(in + offset + 6 * 8), mm1);
+    movq_r2r(mm0, mm2);
+    movq_m2r(*(in + offset + 2 * 8), mm3);
+    paddsw_r2r(mm1, mm0);
+    movq_m2r(*(in + offset + 5 * 8), mm4);
+    psllw_i2r(SHIFT_FRW_COL, mm0);
+    movq_m2r(*(in + offset + 0 * 8), mm5);
+    paddsw_r2r(mm3, mm4);
+    paddsw_m2r(*(in + offset + 7 * 8), mm5);
+    psllw_i2r(SHIFT_FRW_COL, mm4);
+    movq_r2r(mm0, mm6);
+    psubsw_r2r(mm1, mm2);
+    movq_m2r(*(fdct_tg_all_16 + 4), mm1);
+    psubsw_r2r(mm4, mm0);
+    movq_m2r(*(in + offset + 3 * 8), mm7);
+    pmulhw_r2r(mm0, mm1);
+    paddsw_m2r(*(in + offset + 4 * 8), mm7);
+    psllw_i2r(SHIFT_FRW_COL, mm5);
+    paddsw_r2r(mm4, mm6);
+    psllw_i2r(SHIFT_FRW_COL, mm7);
+    movq_r2r(mm5, mm4);
+    psubsw_r2r(mm7, mm5);
+    paddsw_r2r(mm5, mm1);
+    paddsw_r2r(mm7, mm4);
+    por_m2r(fdct_one_corr, mm1);
+    psllw_i2r(SHIFT_FRW_COL + 1, mm2);
+    pmulhw_m2r(*(fdct_tg_all_16 + 4), mm5);
+    movq_r2r(mm4, mm7);
+    psubsw_m2r(*(in + offset + 5 * 8), mm3);
+    psubsw_r2r(mm6, mm4);
+    movq_r2m(mm1, *(out + offset + 2 * 8));
+    paddsw_r2r(mm6, mm7);
+    movq_m2r(*(in + offset + 3 * 8), mm1);
+    psllw_i2r(SHIFT_FRW_COL + 1, mm3);
+    psubsw_m2r(*(in + offset + 4 * 8), mm1);
+    movq_r2r(mm2, mm6);
+    movq_r2m(mm4, *(out + offset + 4 * 8));
+    paddsw_r2r(mm3, mm2);
+    pmulhw_m2r(*ocos_4_16, mm2);
+    psubsw_r2r(mm3, mm6);
+    pmulhw_m2r(*ocos_4_16, mm6);
+    psubsw_r2r(mm0, mm5);
+    por_m2r(fdct_one_corr, mm5);
+    psllw_i2r(SHIFT_FRW_COL, mm1);
+    por_m2r(fdct_one_corr, mm2);
+    movq_r2r(mm1, mm4);
+    movq_m2r(*(in + offset + 0 * 8), mm3);
+    paddsw_r2r(mm6, mm1);
+    psubsw_m2r(*(in + offset + 7 * 8), mm3);
+    psubsw_r2r(mm6, mm4);
+    movq_m2r(*(fdct_tg_all_16 + 0), mm0);
+    psllw_i2r(SHIFT_FRW_COL, mm3);
+    movq_m2r(*(fdct_tg_all_16 + 8), mm6);
+    pmulhw_r2r(mm1, mm0);
+    movq_r2m(mm7, *(out + offset + 0 * 8));
+    pmulhw_r2r(mm4, mm6);
+    movq_r2m(mm5, *(out + offset + 6 * 8));
+    movq_r2r(mm3, mm7);
+    movq_m2r(*(fdct_tg_all_16 + 8), mm5);
+    psubsw_r2r(mm2, mm7);
+    paddsw_r2r(mm2, mm3);
+    pmulhw_r2r(mm7, mm5);
+    paddsw_r2r(mm3, mm0);
+    paddsw_r2r(mm4, mm6);
+    pmulhw_m2r(*(fdct_tg_all_16 + 0), mm3);
+    por_m2r(fdct_one_corr, mm0);
+    paddsw_r2r(mm7, mm5);
+    psubsw_r2r(mm6, mm7);
+    movq_r2m(mm0, *(out + offset + 1 * 8));
+    paddsw_r2r(mm4, mm5);
+    movq_r2m(mm7, *(out + offset + 3 * 8));
+    psubsw_r2r(mm1, mm3);
+    movq_r2m(mm5, *(out + offset + 5 * 8));
+    movq_r2m(mm3, *(out + offset + 7 * 8));
+}
+
+static inline void fdct_row(const int16_t *in, int16_t *out, const int16_t *table)
+{
+    movd_m2r(*(in + 6), mm5);
+    punpcklwd_m2r(*(in + 4), mm5);
+    movq_r2r(mm5, mm2);
+    psrlq_i2r(0x20, mm5);
+    movq_m2r(*(in + 0), mm0);
+    punpcklwd_r2r(mm2, mm5);
+    movq_r2r(mm0, mm1);
+    paddsw_r2r(mm5, mm0);
+    psubsw_r2r(mm5, mm1);
+    movq_r2r(mm0, mm2);
+    punpcklwd_r2r(mm1, mm0);
+    punpckhwd_r2r(mm1, mm2);
+    movq_r2r(mm2, mm1);
+    movq_r2r(mm0, mm2);
+    movq_m2r(*(table + 0), mm3);
+    punpcklwd_r2r(mm1, mm0);
+    movq_r2r(mm0, mm5);
+    punpckldq_r2r(mm0, mm0);
+    movq_m2r(*(table + 4), mm4);
+    punpckhwd_r2r(mm1, mm2);
+    pmaddwd_r2r(mm0, mm3);
+    movq_r2r(mm2, mm6);
+    movq_m2r(*(table + 16), mm1);
+    punpckldq_r2r(mm2, mm2);
+    pmaddwd_r2r(mm2, mm4);
+    punpckhdq_r2r(mm5, mm5);
+    pmaddwd_m2r(*(table + 8), mm0);
+    punpckhdq_r2r(mm6, mm6);
+    movq_m2r(*(table + 20), mm7);
+    pmaddwd_r2r(mm5, mm1);
+    paddd_m2r(fdct_r_row, mm3);
+    pmaddwd_r2r(mm6, mm7);
+    pmaddwd_m2r(*(table + 12), mm2);
+    paddd_r2r(mm4, mm3);
+    pmaddwd_m2r(*(table + 24), mm5);
+    pmaddwd_m2r(*(table + 28), mm6);
+    paddd_r2r(mm7, mm1);
+    paddd_m2r(fdct_r_row, mm0);
+    psrad_i2r(SHIFT_FRW_ROW, mm3);
+    paddd_m2r(fdct_r_row, mm1);
+    paddd_r2r(mm2, mm0);
+    paddd_m2r(fdct_r_row, mm5);
+    psrad_i2r(SHIFT_FRW_ROW, mm1);
+    paddd_r2r(mm6, mm5);
+    psrad_i2r(SHIFT_FRW_ROW, mm0);
+    psrad_i2r(SHIFT_FRW_ROW, mm5);
+    packssdw_r2r(mm0, mm3);
+    packssdw_r2r(mm5, mm1);
+    movq_r2r(mm3, mm6);
+    punpcklwd_r2r(mm1, mm3);
+    punpckhwd_r2r(mm1, mm6);
+    movq_r2m(mm3, *(out + 0));
+    movq_r2m(mm6, *(out + 4));
+}
+
+void fdct_mmx(int16_t *block)
+{
+    /* XXX: not thread safe */
+    static int16_t block_tmp[64] ATTR_ALIGN(8);
+    int16_t *block1, *out;
+    const int16_t *table;
+    int i;
+
+    block1 = block_tmp;
+    fdct_col(block, block1, 0);
+    fdct_col(block, block1, 4);
+
+    block1 = block_tmp;
+    table = tab_frw_01234567;
+    out = block;
+    for(i=8;i>0;i--) {
+        fdct_row(block1, out, table);
+        block1 += 8;
+        table += 32;
+        out += 8;
+    }
+}
diff --git a/src/libffmpeg/libavcodec/i386/idct_mmx.c b/src/libffmpeg/libavcodec/i386/idct_mmx.c
new file mode 100644
index 000000000..618c1cfde
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/idct_mmx.c
@@ -0,0 +1,590 @@
+/*
+ * Note: For libavcodec, this code can also be used under the LGPL license
+ */
+/*
+ * idct_mmx.c
+ * Copyright (C) 1999-2001 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
+ *
+ * This file is part of mpeg2dec, a free MPEG-2 video stream decoder.
+ *
+ * mpeg2dec is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * mpeg2dec 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "../common.h"
+
+#include "mmx.h"
+
+#define ATTR_ALIGN(align) __attribute__ ((__aligned__ (align)))
+
+#define ROW_SHIFT 11
+#define COL_SHIFT 6
+
+#define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))
+#define rounder(bias) {round (bias), round (bias)}
+
+#if 0
+/* C row IDCT - its just here to document the MMXEXT and MMX versions */
+static inline void idct_row (int16_t * row, int offset,
+			     int16_t * table, int32_t * rounder)
+{
+    int C1, C2, C3, C4, C5, C6, C7;
+    int a0, a1, a2, a3, b0, b1, b2, b3;
+
+    row += offset;
+
+    C1 = table[1];
+    C2 = table[2];
+    C3 = table[3];
+    C4 = table[4];
+    C5 = table[5];
+    C6 = table[6];
+    C7 = table[7];
+
+    a0 = C4*row[0] + C2*row[2] + C4*row[4] + C6*row[6] + *rounder;
+    a1 = C4*row[0] + C6*row[2] - C4*row[4] - C2*row[6] + *rounder;
+    a2 = C4*row[0] - C6*row[2] - C4*row[4] + C2*row[6] + *rounder;
+    a3 = C4*row[0] - C2*row[2] + C4*row[4] - C6*row[6] + *rounder;
+
+    b0 = C1*row[1] + C3*row[3] + C5*row[5] + C7*row[7];
+    b1 = C3*row[1] - C7*row[3] - C1*row[5] - C5*row[7];
+    b2 = C5*row[1] - C1*row[3] + C7*row[5] + C3*row[7];
+    b3 = C7*row[1] - C5*row[3] + C3*row[5] - C1*row[7];
+
+    row[0] = (a0 + b0) >> ROW_SHIFT;
+    row[1] = (a1 + b1) >> ROW_SHIFT;
+    row[2] = (a2 + b2) >> ROW_SHIFT;
+    row[3] = (a3 + b3) >> ROW_SHIFT;
+    row[4] = (a3 - b3) >> ROW_SHIFT;
+    row[5] = (a2 - b2) >> ROW_SHIFT;
+    row[6] = (a1 - b1) >> ROW_SHIFT;
+    row[7] = (a0 - b0) >> ROW_SHIFT;
+}
+#endif
+
+
+/* MMXEXT row IDCT */
+
+#define mmxext_table(c1,c2,c3,c4,c5,c6,c7)	{  c4,  c2, -c4, -c2,	\
+						   c4,  c6,  c4,  c6,	\
+						   c1,  c3, -c1, -c5,	\
+						   c5,  c7,  c3, -c7,	\
+						   c4, -c6,  c4, -c6,	\
+						  -c4,  c2,  c4, -c2,	\
+						   c5, -c1,  c3, -c1,	\
+						   c7,  c3,  c7, -c5 }
+
+static inline void mmxext_row_head (int16_t * row, int offset, int16_t * table)
+{
+    movq_m2r (*(row+offset), mm2);	// mm2 = x6 x4 x2 x0
+
+    movq_m2r (*(row+offset+4), mm5);	// mm5 = x7 x5 x3 x1
+    movq_r2r (mm2, mm0);		// mm0 = x6 x4 x2 x0
+
+    movq_m2r (*table, mm3);		// mm3 = -C2 -C4 C2 C4
+    movq_r2r (mm5, mm6);		// mm6 = x7 x5 x3 x1
+
+    movq_m2r (*(table+4), mm4);		// mm4 = C6 C4 C6 C4
+    pmaddwd_r2r (mm0, mm3);		// mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
+
+    pshufw_r2r (mm2, mm2, 0x4e);	// mm2 = x2 x0 x6 x4
+}
+
+static inline void mmxext_row (int16_t * table, int32_t * rounder)
+{
+    movq_m2r (*(table+8), mm1);		// mm1 = -C5 -C1 C3 C1
+    pmaddwd_r2r (mm2, mm4);		// mm4 = C4*x0+C6*x2 C4*x4+C6*x6
+
+    pmaddwd_m2r (*(table+16), mm0);	// mm0 = C4*x4-C6*x6 C4*x0-C6*x2
+    pshufw_r2r (mm6, mm6, 0x4e);	// mm6 = x3 x1 x7 x5
+
+    movq_m2r (*(table+12), mm7);	// mm7 = -C7 C3 C7 C5
+    pmaddwd_r2r (mm5, mm1);		// mm1 = -C1*x5-C5*x7 C1*x1+C3*x3
+
+    paddd_m2r (*rounder, mm3);		// mm3 += rounder
+    pmaddwd_r2r (mm6, mm7);		// mm7 = C3*x1-C7*x3 C5*x5+C7*x7
+
+    pmaddwd_m2r (*(table+20), mm2);	// mm2 = C4*x0-C2*x2 -C4*x4+C2*x6
+    paddd_r2r (mm4, mm3);		// mm3 = a1 a0 + rounder
+
+    pmaddwd_m2r (*(table+24), mm5);	// mm5 = C3*x5-C1*x7 C5*x1-C1*x3
+    movq_r2r (mm3, mm4);		// mm4 = a1 a0 + rounder
+
+    pmaddwd_m2r (*(table+28), mm6);	// mm6 = C7*x1-C5*x3 C7*x5+C3*x7
+    paddd_r2r (mm7, mm1);		// mm1 = b1 b0
+
+    paddd_m2r (*rounder, mm0);		// mm0 += rounder
+    psubd_r2r (mm1, mm3);		// mm3 = a1-b1 a0-b0 + rounder
+
+    psrad_i2r (ROW_SHIFT, mm3);		// mm3 = y6 y7
+    paddd_r2r (mm4, mm1);		// mm1 = a1+b1 a0+b0 + rounder
+
+    paddd_r2r (mm2, mm0);		// mm0 = a3 a2 + rounder
+    psrad_i2r (ROW_SHIFT, mm1);		// mm1 = y1 y0
+
+    paddd_r2r (mm6, mm5);		// mm5 = b3 b2
+    movq_r2r (mm0, mm4);		// mm4 = a3 a2 + rounder
+
+    paddd_r2r (mm5, mm0);		// mm0 = a3+b3 a2+b2 + rounder
+    psubd_r2r (mm5, mm4);		// mm4 = a3-b3 a2-b2 + rounder
+}
+
+static inline void mmxext_row_tail (int16_t * row, int store)
+{
+    psrad_i2r (ROW_SHIFT, mm0);		// mm0 = y3 y2
+
+    psrad_i2r (ROW_SHIFT, mm4);		// mm4 = y4 y5
+
+    packssdw_r2r (mm0, mm1);		// mm1 = y3 y2 y1 y0
+
+    packssdw_r2r (mm3, mm4);		// mm4 = y6 y7 y4 y5
+
+    movq_r2m (mm1, *(row+store));	// save y3 y2 y1 y0
+    pshufw_r2r (mm4, mm4, 0xb1);	// mm4 = y7 y6 y5 y4
+
+    /* slot */
+
+    movq_r2m (mm4, *(row+store+4));	// save y7 y6 y5 y4
+}
+
+static inline void mmxext_row_mid (int16_t * row, int store,
+				   int offset, int16_t * table)
+{
+    movq_m2r (*(row+offset), mm2);	// mm2 = x6 x4 x2 x0
+    psrad_i2r (ROW_SHIFT, mm0);		// mm0 = y3 y2
+
+    movq_m2r (*(row+offset+4), mm5);	// mm5 = x7 x5 x3 x1
+    psrad_i2r (ROW_SHIFT, mm4);		// mm4 = y4 y5
+
+    packssdw_r2r (mm0, mm1);		// mm1 = y3 y2 y1 y0
+    movq_r2r (mm5, mm6);		// mm6 = x7 x5 x3 x1
+
+    packssdw_r2r (mm3, mm4);		// mm4 = y6 y7 y4 y5
+    movq_r2r (mm2, mm0);		// mm0 = x6 x4 x2 x0
+
+    movq_r2m (mm1, *(row+store));	// save y3 y2 y1 y0
+    pshufw_r2r (mm4, mm4, 0xb1);	// mm4 = y7 y6 y5 y4
+
+    movq_m2r (*table, mm3);		// mm3 = -C2 -C4 C2 C4
+    movq_r2m (mm4, *(row+store+4));	// save y7 y6 y5 y4
+
+    pmaddwd_r2r (mm0, mm3);		// mm3 = -C4*x4-C2*x6 C4*x0+C2*x2
+
+    movq_m2r (*(table+4), mm4);		// mm4 = C6 C4 C6 C4
+    pshufw_r2r (mm2, mm2, 0x4e);	// mm2 = x2 x0 x6 x4
+}
+
+
+/* MMX row IDCT */
+
+#define mmx_table(c1,c2,c3,c4,c5,c6,c7)	{  c4,  c2,  c4,  c6,	\
+					   c4,  c6, -c4, -c2,	\
+					   c1,  c3,  c3, -c7,	\
+					   c5,  c7, -c1, -c5,	\
+					   c4, -c6,  c4, -c2,	\
+					  -c4,  c2,  c4, -c6,	\
+					   c5, -c1,  c7, -c5,	\
+					   c7,  c3,  c3, -c1 }
+
+static inline void mmx_row_head (int16_t * row, int offset, int16_t * table)
+{
+    movq_m2r (*(row+offset), mm2);	// mm2 = x6 x4 x2 x0
+
+    movq_m2r (*(row+offset+4), mm5);	// mm5 = x7 x5 x3 x1
+    movq_r2r (mm2, mm0);		// mm0 = x6 x4 x2 x0
+
+    movq_m2r (*table, mm3);		// mm3 = C6 C4 C2 C4
+    movq_r2r (mm5, mm6);		// mm6 = x7 x5 x3 x1
+
+    punpckldq_r2r (mm0, mm0);		// mm0 = x2 x0 x2 x0
+
+    movq_m2r (*(table+4), mm4);		// mm4 = -C2 -C4 C6 C4
+    pmaddwd_r2r (mm0, mm3);		// mm3 = C4*x0+C6*x2 C4*x0+C2*x2
+
+    movq_m2r (*(table+8), mm1);		// mm1 = -C7 C3 C3 C1
+    punpckhdq_r2r (mm2, mm2);		// mm2 = x6 x4 x6 x4
+}
+
+static inline void mmx_row (int16_t * table, int32_t * rounder)
+{
+    pmaddwd_r2r (mm2, mm4);		// mm4 = -C4*x4-C2*x6 C4*x4+C6*x6
+    punpckldq_r2r (mm5, mm5);		// mm5 = x3 x1 x3 x1
+
+    pmaddwd_m2r (*(table+16), mm0);	// mm0 = C4*x0-C2*x2 C4*x0-C6*x2
+    punpckhdq_r2r (mm6, mm6);		// mm6 = x7 x5 x7 x5
+
+    movq_m2r (*(table+12), mm7);	// mm7 = -C5 -C1 C7 C5
+    pmaddwd_r2r (mm5, mm1);		// mm1 = C3*x1-C7*x3 C1*x1+C3*x3
+
+    paddd_m2r (*rounder, mm3);		// mm3 += rounder
+    pmaddwd_r2r (mm6, mm7);		// mm7 = -C1*x5-C5*x7 C5*x5+C7*x7
+
+    pmaddwd_m2r (*(table+20), mm2);	// mm2 = C4*x4-C6*x6 -C4*x4+C2*x6
+    paddd_r2r (mm4, mm3);		// mm3 = a1 a0 + rounder
+
+    pmaddwd_m2r (*(table+24), mm5);	// mm5 = C7*x1-C5*x3 C5*x1-C1*x3
+    movq_r2r (mm3, mm4);		// mm4 = a1 a0 + rounder
+
+    pmaddwd_m2r (*(table+28), mm6);	// mm6 = C3*x5-C1*x7 C7*x5+C3*x7
+    paddd_r2r (mm7, mm1);		// mm1 = b1 b0
+
+    paddd_m2r (*rounder, mm0);		// mm0 += rounder
+    psubd_r2r (mm1, mm3);		// mm3 = a1-b1 a0-b0 + rounder
+
+    psrad_i2r (ROW_SHIFT, mm3);		// mm3 = y6 y7
+    paddd_r2r (mm4, mm1);		// mm1 = a1+b1 a0+b0 + rounder
+
+    paddd_r2r (mm2, mm0);		// mm0 = a3 a2 + rounder
+    psrad_i2r (ROW_SHIFT, mm1);		// mm1 = y1 y0
+
+    paddd_r2r (mm6, mm5);		// mm5 = b3 b2
+    movq_r2r (mm0, mm7);		// mm7 = a3 a2 + rounder
+
+    paddd_r2r (mm5, mm0);		// mm0 = a3+b3 a2+b2 + rounder
+    psubd_r2r (mm5, mm7);		// mm7 = a3-b3 a2-b2 + rounder
+}
+
+static inline void mmx_row_tail (int16_t * row, int store)
+{
+    psrad_i2r (ROW_SHIFT, mm0);		// mm0 = y3 y2
+
+    psrad_i2r (ROW_SHIFT, mm7);		// mm7 = y4 y5
+
+    packssdw_r2r (mm0, mm1);		// mm1 = y3 y2 y1 y0
+
+    packssdw_r2r (mm3, mm7);		// mm7 = y6 y7 y4 y5
+
+    movq_r2m (mm1, *(row+store));	// save y3 y2 y1 y0
+    movq_r2r (mm7, mm4);		// mm4 = y6 y7 y4 y5
+
+    pslld_i2r (16, mm7);		// mm7 = y7 0 y5 0
+
+    psrld_i2r (16, mm4);		// mm4 = 0 y6 0 y4
+
+    por_r2r (mm4, mm7);			// mm7 = y7 y6 y5 y4
+
+    /* slot */
+
+    movq_r2m (mm7, *(row+store+4));	// save y7 y6 y5 y4
+}
+
+static inline void mmx_row_mid (int16_t * row, int store,
+				int offset, int16_t * table)
+{
+    movq_m2r (*(row+offset), mm2);	// mm2 = x6 x4 x2 x0
+    psrad_i2r (ROW_SHIFT, mm0);		// mm0 = y3 y2
+
+    movq_m2r (*(row+offset+4), mm5);	// mm5 = x7 x5 x3 x1
+    psrad_i2r (ROW_SHIFT, mm7);		// mm7 = y4 y5
+
+    packssdw_r2r (mm0, mm1);		// mm1 = y3 y2 y1 y0
+    movq_r2r (mm5, mm6);		// mm6 = x7 x5 x3 x1
+
+    packssdw_r2r (mm3, mm7);		// mm7 = y6 y7 y4 y5
+    movq_r2r (mm2, mm0);		// mm0 = x6 x4 x2 x0
+
+    movq_r2m (mm1, *(row+store));	// save y3 y2 y1 y0
+    movq_r2r (mm7, mm1);		// mm1 = y6 y7 y4 y5
+
+    punpckldq_r2r (mm0, mm0);		// mm0 = x2 x0 x2 x0
+    psrld_i2r (16, mm7);		// mm7 = 0 y6 0 y4
+
+    movq_m2r (*table, mm3);		// mm3 = C6 C4 C2 C4
+    pslld_i2r (16, mm1);		// mm1 = y7 0 y5 0
+
+    movq_m2r (*(table+4), mm4);		// mm4 = -C2 -C4 C6 C4
+    por_r2r (mm1, mm7);			// mm7 = y7 y6 y5 y4
+
+    movq_m2r (*(table+8), mm1);		// mm1 = -C7 C3 C3 C1
+    punpckhdq_r2r (mm2, mm2);		// mm2 = x6 x4 x6 x4
+
+    movq_r2m (mm7, *(row+store+4));	// save y7 y6 y5 y4
+    pmaddwd_r2r (mm0, mm3);		// mm3 = C4*x0+C6*x2 C4*x0+C2*x2
+}
+
+
+#if 0
+// C column IDCT - its just here to document the MMXEXT and MMX versions
+static inline void idct_col (int16_t * col, int offset)
+{
+/* multiplication - as implemented on mmx */
+#define F(c,x) (((c) * (x)) >> 16)
+
+/* saturation - it helps us handle torture test cases */
+#define S(x) (((x)>32767) ? 32767 : ((x)<-32768) ? -32768 : (x))
+
+    int16_t x0, x1, x2, x3, x4, x5, x6, x7;
+    int16_t y0, y1, y2, y3, y4, y5, y6, y7;
+    int16_t a0, a1, a2, a3, b0, b1, b2, b3;
+    int16_t u04, v04, u26, v26, u17, v17, u35, v35, u12, v12;
+
+    col += offset;
+
+    x0 = col[0*8];
+    x1 = col[1*8];
+    x2 = col[2*8];
+    x3 = col[3*8];
+    x4 = col[4*8];
+    x5 = col[5*8];
+    x6 = col[6*8];
+    x7 = col[7*8];
+
+    u04 = S (x0 + x4);
+    v04 = S (x0 - x4);
+    u26 = S (F (T2, x6) + x2);
+    v26 = S (F (T2, x2) - x6);
+
+    a0 = S (u04 + u26);
+    a1 = S (v04 + v26);
+    a2 = S (v04 - v26);
+    a3 = S (u04 - u26);
+
+    u17 = S (F (T1, x7) + x1);
+    v17 = S (F (T1, x1) - x7);
+    u35 = S (F (T3, x5) + x3);
+    v35 = S (F (T3, x3) - x5);
+
+    b0 = S (u17 + u35);
+    b3 = S (v17 - v35);
+    u12 = S (u17 - u35);
+    v12 = S (v17 + v35);
+    u12 = S (2 * F (C4, u12));
+    v12 = S (2 * F (C4, v12));
+    b1 = S (u12 + v12);
+    b2 = S (u12 - v12);
+
+    y0 = S (a0 + b0) >> COL_SHIFT;
+    y1 = S (a1 + b1) >> COL_SHIFT;
+    y2 = S (a2 + b2) >> COL_SHIFT;
+    y3 = S (a3 + b3) >> COL_SHIFT;
+
+    y4 = S (a3 - b3) >> COL_SHIFT;
+    y5 = S (a2 - b2) >> COL_SHIFT;
+    y6 = S (a1 - b1) >> COL_SHIFT;
+    y7 = S (a0 - b0) >> COL_SHIFT;
+
+    col[0*8] = y0;
+    col[1*8] = y1;
+    col[2*8] = y2;
+    col[3*8] = y3;
+    col[4*8] = y4;
+    col[5*8] = y5;
+    col[6*8] = y6;
+    col[7*8] = y7;
+}
+#endif
+
+
+// MMX column IDCT
+static inline void idct_col (int16_t * col, int offset)
+{
+#define T1 13036
+#define T2 27146
+#define T3 43790
+#define C4 23170
+
+    static short _T1[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
+    static short _T2[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
+    static short _T3[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
+    static short _C4[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
+
+    /* column code adapted from peter gubanov */
+    /* http://www.elecard.com/peter/idct.shtml */
+
+    movq_m2r (*_T1, mm0);		// mm0 = T1
+
+    movq_m2r (*(col+offset+1*8), mm1);	// mm1 = x1
+    movq_r2r (mm0, mm2);		// mm2 = T1
+
+    movq_m2r (*(col+offset+7*8), mm4);	// mm4 = x7
+    pmulhw_r2r (mm1, mm0);		// mm0 = T1*x1
+
+    movq_m2r (*_T3, mm5);		// mm5 = T3
+    pmulhw_r2r (mm4, mm2);		// mm2 = T1*x7
+
+    movq_m2r (*(col+offset+5*8), mm6);	// mm6 = x5
+    movq_r2r (mm5, mm7);		// mm7 = T3-1
+
+    movq_m2r (*(col+offset+3*8), mm3);	// mm3 = x3
+    psubsw_r2r (mm4, mm0);		// mm0 = v17
+
+    movq_m2r (*_T2, mm4);		// mm4 = T2
+    pmulhw_r2r (mm3, mm5);		// mm5 = (T3-1)*x3
+
+    paddsw_r2r (mm2, mm1);		// mm1 = u17
+    pmulhw_r2r (mm6, mm7);		// mm7 = (T3-1)*x5
+
+    /* slot */
+
+    movq_r2r (mm4, mm2);		// mm2 = T2
+    paddsw_r2r (mm3, mm5);		// mm5 = T3*x3
+
+    pmulhw_m2r (*(col+offset+2*8), mm4);// mm4 = T2*x2
+    paddsw_r2r (mm6, mm7);		// mm7 = T3*x5
+
+    psubsw_r2r (mm6, mm5);		// mm5 = v35
+    paddsw_r2r (mm3, mm7);		// mm7 = u35
+
+    movq_m2r (*(col+offset+6*8), mm3);	// mm3 = x6
+    movq_r2r (mm0, mm6);		// mm6 = v17
+
+    pmulhw_r2r (mm3, mm2);		// mm2 = T2*x6
+    psubsw_r2r (mm5, mm0);		// mm0 = b3
+
+    psubsw_r2r (mm3, mm4);		// mm4 = v26
+    paddsw_r2r (mm6, mm5);		// mm5 = v12
+
+    movq_r2m (mm0, *(col+offset+3*8));	// save b3 in scratch0
+    movq_r2r (mm1, mm6);		// mm6 = u17
+
+    paddsw_m2r (*(col+offset+2*8), mm2);// mm2 = u26
+    paddsw_r2r (mm7, mm6);		// mm6 = b0
+
+    psubsw_r2r (mm7, mm1);		// mm1 = u12
+    movq_r2r (mm1, mm7);		// mm7 = u12
+
+    movq_m2r (*(col+offset+0*8), mm3);	// mm3 = x0
+    paddsw_r2r (mm5, mm1);		// mm1 = u12+v12
+
+    movq_m2r (*_C4, mm0);		// mm0 = C4/2
+    psubsw_r2r (mm5, mm7);		// mm7 = u12-v12
+
+    movq_r2m (mm6, *(col+offset+5*8));	// save b0 in scratch1
+    pmulhw_r2r (mm0, mm1);		// mm1 = b1/2
+
+    movq_r2r (mm4, mm6);		// mm6 = v26
+    pmulhw_r2r (mm0, mm7);		// mm7 = b2/2
+
+    movq_m2r (*(col+offset+4*8), mm5);	// mm5 = x4
+    movq_r2r (mm3, mm0);		// mm0 = x0
+
+    psubsw_r2r (mm5, mm3);		// mm3 = v04
+    paddsw_r2r (mm5, mm0);		// mm0 = u04
+
+    paddsw_r2r (mm3, mm4);		// mm4 = a1
+    movq_r2r (mm0, mm5);		// mm5 = u04
+
+    psubsw_r2r (mm6, mm3);		// mm3 = a2
+    paddsw_r2r (mm2, mm5);		// mm5 = a0
+
+    paddsw_r2r (mm1, mm1);		// mm1 = b1
+    psubsw_r2r (mm2, mm0);		// mm0 = a3
+
+    paddsw_r2r (mm7, mm7);		// mm7 = b2
+    movq_r2r (mm3, mm2);		// mm2 = a2
+
+    movq_r2r (mm4, mm6);		// mm6 = a1
+    paddsw_r2r (mm7, mm3);		// mm3 = a2+b2
+
+    psraw_i2r (COL_SHIFT, mm3);		// mm3 = y2
+    paddsw_r2r (mm1, mm4);		// mm4 = a1+b1
+
+    psraw_i2r (COL_SHIFT, mm4);		// mm4 = y1
+    psubsw_r2r (mm1, mm6);		// mm6 = a1-b1
+
+    movq_m2r (*(col+offset+5*8), mm1);	// mm1 = b0
+    psubsw_r2r (mm7, mm2);		// mm2 = a2-b2
+
+    psraw_i2r (COL_SHIFT, mm6);		// mm6 = y6
+    movq_r2r (mm5, mm7);		// mm7 = a0
+
+    movq_r2m (mm4, *(col+offset+1*8));	// save y1
+    psraw_i2r (COL_SHIFT, mm2);		// mm2 = y5
+
+    movq_r2m (mm3, *(col+offset+2*8));	// save y2
+    paddsw_r2r (mm1, mm5);		// mm5 = a0+b0
+
+    movq_m2r (*(col+offset+3*8), mm4);	// mm4 = b3
+    psubsw_r2r (mm1, mm7);		// mm7 = a0-b0
+
+    psraw_i2r (COL_SHIFT, mm5);		// mm5 = y0
+    movq_r2r (mm0, mm3);		// mm3 = a3
+
+    movq_r2m (mm2, *(col+offset+5*8));	// save y5
+    psubsw_r2r (mm4, mm3);		// mm3 = a3-b3
+
+    psraw_i2r (COL_SHIFT, mm7);		// mm7 = y7
+    paddsw_r2r (mm0, mm4);		// mm4 = a3+b3
+
+    movq_r2m (mm5, *(col+offset+0*8));	// save y0
+    psraw_i2r (COL_SHIFT, mm3);		// mm3 = y4
+
+    movq_r2m (mm6, *(col+offset+6*8));	// save y6
+    psraw_i2r (COL_SHIFT, mm4);		// mm4 = y3
+
+    movq_r2m (mm7, *(col+offset+7*8));	// save y7
+
+    movq_r2m (mm3, *(col+offset+4*8));	// save y4
+
+    movq_r2m (mm4, *(col+offset+3*8));	// save y3
+}
+
+
+static int32_t rounder0[] ATTR_ALIGN(8) =
+    rounder ((1 << (COL_SHIFT - 1)) - 0.5);
+static int32_t rounder4[] ATTR_ALIGN(8) = rounder (0);
+static int32_t rounder1[] ATTR_ALIGN(8) =
+    rounder (1.25683487303);	/* C1*(C1/C4+C1+C7)/2 */
+static int32_t rounder7[] ATTR_ALIGN(8) =
+    rounder (-0.25);		/* C1*(C7/C4+C7-C1)/2 */
+static int32_t rounder2[] ATTR_ALIGN(8) =
+    rounder (0.60355339059);	/* C2 * (C6+C2)/2 */
+static int32_t rounder6[] ATTR_ALIGN(8) =
+    rounder (-0.25);		/* C2 * (C6-C2)/2 */
+static int32_t rounder3[] ATTR_ALIGN(8) =
+    rounder (0.087788325588);	/* C3*(-C3/C4+C3+C5)/2 */
+static int32_t rounder5[] ATTR_ALIGN(8) =
+    rounder (-0.441341716183);	/* C3*(-C5/C4+C5-C3)/2 */
+
+
+#define declare_idct(idct,table,idct_row_head,idct_row,idct_row_tail,idct_row_mid)	\
+void idct (int16_t * block)					\
+{									\
+    static int16_t table04[] ATTR_ALIGN(16) =				\
+	table (22725, 21407, 19266, 16384, 12873,  8867, 4520);		\
+    static int16_t table17[] ATTR_ALIGN(16) =				\
+	table (31521, 29692, 26722, 22725, 17855, 12299, 6270);		\
+    static int16_t table26[] ATTR_ALIGN(16) =				\
+	table (29692, 27969, 25172, 21407, 16819, 11585, 5906);		\
+    static int16_t table35[] ATTR_ALIGN(16) =				\
+	table (26722, 25172, 22654, 19266, 15137, 10426, 5315);		\
+									\
+    idct_row_head (block, 0*8, table04);				\
+    idct_row (table04, rounder0);					\
+    idct_row_mid (block, 0*8, 4*8, table04);				\
+    idct_row (table04, rounder4);					\
+    idct_row_mid (block, 4*8, 1*8, table17);				\
+    idct_row (table17, rounder1);					\
+    idct_row_mid (block, 1*8, 7*8, table17);				\
+    idct_row (table17, rounder7);					\
+    idct_row_mid (block, 7*8, 2*8, table26);				\
+    idct_row (table26, rounder2);					\
+    idct_row_mid (block, 2*8, 6*8, table26);				\
+    idct_row (table26, rounder6);					\
+    idct_row_mid (block, 6*8, 3*8, table35);				\
+    idct_row (table35, rounder3);					\
+    idct_row_mid (block, 3*8, 5*8, table35);				\
+    idct_row (table35, rounder5);					\
+    idct_row_tail (block, 5*8);						\
+									\
+    idct_col (block, 0);						\
+    idct_col (block, 4);						\
+}
+
+
+declare_idct (ff_mmxext_idct, mmxext_table,
+	      mmxext_row_head, mmxext_row, mmxext_row_tail, mmxext_row_mid)
+
+declare_idct (ff_mmx_idct, mmx_table,
+	      mmx_row_head, mmx_row, mmx_row_tail, mmx_row_mid)
diff --git a/src/libffmpeg/libavcodec/i386/mmx.h b/src/libffmpeg/libavcodec/i386/mmx.h
new file mode 100644
index 000000000..2ba28898d
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/mmx.h
@@ -0,0 +1 @@
+#include "xineutils.h"
diff --git a/src/libffmpeg/libavcodec/i386/motion_est_mmx.c b/src/libffmpeg/libavcodec/i386/motion_est_mmx.c
new file mode 100644
index 000000000..35b16b711
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/motion_est_mmx.c
@@ -0,0 +1,244 @@
+/*
+ * MMX optimized motion estimation
+ * Copyright (c) 2001 Gerard Lantau.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ */
+#include "../dsputil.h"
+#include "mmx.h"
+
+static const unsigned long long int mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001;
+static const unsigned long long int mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002;
+
+/* mm7 is accumulator, mm6 is zero */
+static inline void sad_add(const UINT8 *p1, const UINT8 *p2)
+{
+    movq_m2r(*p1, mm0);
+    movq_m2r(*p2, mm1);
+    movq_r2r(mm0, mm2);
+    psubusb_r2r(mm1, mm0);
+    psubusb_r2r(mm2, mm1);
+    por_r2r(mm1, mm0); /* mm0 is absolute value */
+
+    movq_r2r(mm0, mm1);
+    punpcklbw_r2r(mm6, mm0);
+    punpckhbw_r2r(mm6, mm1);
+    paddusw_r2r(mm0, mm7);
+    paddusw_r2r(mm1, mm7);
+}
+
+/* convert mm7 to value */
+static inline int sad_end(void)
+{
+    int res;
+
+    movq_r2r(mm7, mm0);
+    psrlq_i2r(32, mm7);
+    paddusw_r2r(mm0, mm7);
+
+    movq_r2r(mm7, mm0);
+    psrlq_i2r(16, mm7);
+    paddusw_r2r(mm0, mm7);
+    __asm __volatile ("movd %%mm7, %0" : "=a" (res));
+    return res & 0xffff;
+}
+
+int pix_abs16x16_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h)
+{
+    const UINT8 *p1, *p2;
+
+    h >>= 1;
+    p1 = blk1;
+    p2 = blk2;
+    pxor_r2r(mm7, mm7); /* mm7 is accumulator */
+    pxor_r2r(mm6, mm6); /* mm7 is zero constant */
+    do {
+        sad_add(p1, p2);
+        sad_add(p1 + 8, p2 + 8);
+        p1 += lx;
+        p2 += lx;
+        sad_add(p1, p2);
+        sad_add(p1 + 8, p2 + 8);
+        p1 += lx;
+        p2 += lx;
+    } while (--h);
+    return sad_end();
+}
+
+/* please test it ! */
+static inline void sad_add_sse(const UINT8 *p1, const UINT8 *p2)
+{
+    movq_m2r(*(p1 + 0), mm0);
+    movq_m2r(*(p1 + 8), mm1);
+    psadbw_m2r(*(p2 + 0), mm0);
+    psadbw_m2r(*(p2 + 8), mm1);
+    paddusw_r2r(mm0, mm7);
+    paddusw_r2r(mm1, mm7);
+}
+
+int pix_abs16x16_sse(UINT8 *blk1, UINT8 *blk2, int lx, int h)
+{
+    const UINT8 *p1, *p2;
+
+    h >>= 1;
+    p1 = blk1;
+    p2 = blk2;
+    pxor_r2r(mm7, mm7); /* mm7 is accumulator */
+    do {
+        sad_add_sse(p1, p2);
+        p1 += lx;
+        p2 += lx;
+        sad_add_sse(p1, p2);
+        p1 += lx;
+        p2 += lx;
+    } while (--h);
+    return sad_end();
+}
+
+#define DUMP(reg) { mmx_t tmp; movq_r2m(reg, tmp); printf(#reg "=%016Lx\n", tmp.uq); }
+
+/* mm7 is accumulator, mm6 is zero */
+static inline void sad_add_x2(const UINT8 *p1, const UINT8 *p2, const UINT8 *p3)
+{
+    movq_m2r(*(p2 + 0), mm0);
+    movq_m2r(*(p3 + 0), mm1);
+    movq_r2r(mm0, mm2);
+    movq_r2r(mm1, mm3);
+    punpcklbw_r2r(mm6, mm0); /* extract 4 bytes low */
+    punpcklbw_r2r(mm6, mm1);
+    punpckhbw_r2r(mm6, mm2); /* high */
+    punpckhbw_r2r(mm6, mm3); 
+    paddusw_r2r(mm1, mm0);
+    paddusw_r2r(mm3, mm2);
+    movq_m2r(*(p1 + 0), mm1); /* mm1 : other value */
+    paddusw_r2r(mm5, mm0); /* + 1 */
+    paddusw_r2r(mm5, mm2); /* + 1 */
+    psrlw_i2r(1, mm0);
+    psrlw_i2r(1, mm2);
+    packuswb_r2r(mm2, mm0); /* average is in mm0 */
+
+    movq_r2r(mm1, mm2); 
+    psubusb_r2r(mm0, mm1);
+    psubusb_r2r(mm2, mm0);
+    por_r2r(mm1, mm0); /* mm0 is absolute value */
+
+    movq_r2r(mm0, mm1);
+    punpcklbw_r2r(mm6, mm0);
+    punpckhbw_r2r(mm6, mm1);
+    paddusw_r2r(mm0, mm7);
+    paddusw_r2r(mm1, mm7);
+}
+
+int pix_abs16x16_x2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h)
+{
+    const UINT8 *p1, *p2;
+
+    p1 = blk1;
+    p2 = blk2;
+    pxor_r2r(mm7, mm7); /* mm7 is accumulator */
+    pxor_r2r(mm6, mm6); /* mm7 is zero constant */
+    movq_m2r(mm_wone, mm5); /* one constant */
+    do {
+        sad_add_x2(p1, p2, p2 + 1);
+        sad_add_x2(p1 + 8, p2 + 8, p2 + 9);
+        p1 += lx;
+        p2 += lx;
+    } while (--h);
+    return sad_end();
+}
+
+int pix_abs16x16_y2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h)
+{
+    const UINT8 *p1, *p2;
+
+    p1 = blk1;
+    p2 = blk2;
+    pxor_r2r(mm7, mm7); /* mm7 is accumulator */
+    pxor_r2r(mm6, mm6); /* mm7 is zero constant */
+    movq_m2r(mm_wone, mm5); /* one constant */
+    do {
+        sad_add_x2(p1, p2, p2 + lx);
+        sad_add_x2(p1 + 8, p2 + 8, p2 + 8 + lx);
+        p1 += lx;
+        p2 += lx;
+    } while (--h);
+    return sad_end();
+}
+
+/* mm7 is accumulator, mm6 is zero */
+static inline void sad_add_xy2(const UINT8 *p1, const UINT8 *p2, const UINT8 *p3)
+{
+    movq_m2r(*(p2 + 0), mm0);
+    movq_m2r(*(p3 + 0), mm1);
+    movq_r2r(mm0, mm2);
+    movq_r2r(mm1, mm3);
+    punpcklbw_r2r(mm6, mm0); /* extract 4 bytes low */
+    punpcklbw_r2r(mm6, mm1);
+    punpckhbw_r2r(mm6, mm2); /* high */
+    punpckhbw_r2r(mm6, mm3); 
+    paddusw_r2r(mm1, mm0);
+    paddusw_r2r(mm3, mm2);
+
+    movq_m2r(*(p2 + 1), mm1);
+    movq_m2r(*(p3 + 1), mm3);
+    movq_r2r(mm1, mm4);
+    punpcklbw_r2r(mm6, mm1); /* low */
+    punpckhbw_r2r(mm6, mm4); /* high */
+    paddusw_r2r(mm1, mm0);
+    paddusw_r2r(mm4, mm2);
+    movq_r2r(mm3, mm4);
+    punpcklbw_r2r(mm6, mm3); /* low */
+    punpckhbw_r2r(mm6, mm4); /* high */
+    paddusw_r2r(mm3, mm0);
+    paddusw_r2r(mm4, mm2);
+    
+    movq_m2r(*(p1 + 0), mm1); /* mm1 : other value */
+    paddusw_r2r(mm5, mm0); /* + 2 */
+    paddusw_r2r(mm5, mm2); /* + 2 */
+    psrlw_i2r(2, mm0);
+    psrlw_i2r(2, mm2);
+    packuswb_r2r(mm2, mm0); /* average is in mm0 */
+
+    movq_r2r(mm1, mm2); 
+    psubusb_r2r(mm0, mm1);
+    psubusb_r2r(mm2, mm0);
+    por_r2r(mm1, mm0); /* mm0 is absolute value */
+
+    movq_r2r(mm0, mm1);
+    punpcklbw_r2r(mm6, mm0);
+    punpckhbw_r2r(mm6, mm1);
+    paddusw_r2r(mm0, mm7);
+    paddusw_r2r(mm1, mm7);
+}
+
+int pix_abs16x16_xy2_mmx(UINT8 *blk1, UINT8 *blk2, int lx, int h)
+{
+    const UINT8 *p1, *p2, *p3;
+
+    p1 = blk1;
+    p2 = blk2;
+    p3 = blk2 + lx;
+    pxor_r2r(mm7, mm7); /* mm7 is accumulator */
+    pxor_r2r(mm6, mm6); /* mm7 is zero constant */
+    movq_m2r(mm_wtwo, mm5); /* one constant */
+    do {
+        sad_add_xy2(p1, p2, p2 + lx);
+        sad_add_xy2(p1 + 8, p2 + 8, p2 + 8 + lx);
+        p1 += lx;
+        p2 += lx;
+    } while (--h);
+    return sad_end();
+}
diff --git a/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c b/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c
new file mode 100644
index 000000000..b34d4eb03
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c
@@ -0,0 +1,326 @@
+/*
+ * The simplest mpeg encoder (well, it was the simplest!)
+ * Copyright (c) 2000,2001 Gerard Lantau.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ * Optimized for ia32 cpus by Nick Kurshev <nickols_k@mail.ru>
+ */
+
+#include "../dsputil.h"
+#include "../mpegvideo.h"
+
+#if 0
+
+/* XXX: GL: I don't understand why this function needs optimization
+   (it is called only once per frame!), so I disabled it */
+
+void MPV_frame_start(MpegEncContext *s)
+{
+    if (s->pict_type == B_TYPE) {
+	__asm __volatile(
+	    "movl	(%1), %%eax\n\t"
+	    "movl	4(%1), %%edx\n\t"
+	    "movl	8(%1), %%ecx\n\t"
+	    "movl	%%eax, (%0)\n\t"
+	    "movl	%%edx, 4(%0)\n\t"
+	    "movl	%%ecx, 8(%0)\n\t"
+	    :
+	    :"r"(s->current_picture), "r"(s->aux_picture)
+	    :"eax","edx","ecx","memory");
+    } else {
+            /* swap next and last */
+	__asm __volatile(
+	    "movl	(%1), %%eax\n\t"
+	    "movl	4(%1), %%edx\n\t"
+	    "movl	8(%1), %%ecx\n\t"
+	    "xchgl	(%0), %%eax\n\t"
+	    "xchgl	4(%0), %%edx\n\t"
+	    "xchgl	8(%0), %%ecx\n\t"
+	    "movl	%%eax, (%1)\n\t"
+	    "movl	%%edx, 4(%1)\n\t"
+	    "movl	%%ecx, 8(%1)\n\t"
+	    "movl	%%eax, (%2)\n\t"
+	    "movl	%%edx, 4(%2)\n\t"
+	    "movl	%%ecx, 8(%2)\n\t"
+	    :
+	    :"r"(s->last_picture), "r"(s->next_picture), "r"(s->current_picture)
+	    :"eax","edx","ecx","memory");
+    }
+}
+#endif
+
+static const unsigned long long int mm_wabs __attribute__ ((aligned(8))) = 0xffffffffffffffffULL;
+static const unsigned long long int mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL;
+
+
+static void dct_unquantize_h263_mmx(MpegEncContext *s,
+                                  DCTELEM *block, int n, int qscale)
+{
+    int i, level, qmul, qadd;
+
+    qmul = s->qscale << 1;
+    qadd = (s->qscale - 1) | 1;
+
+    if (s->mb_intra) {
+        if (n < 4)
+            block[0] = block[0] * s->y_dc_scale;
+        else
+            block[0] = block[0] * s->c_dc_scale;
+
+	for(i=1; i<8; i++) {
+		level = block[i];
+		if (level) {
+			if (level < 0) {
+				level = level * qmul - qadd;
+			} else {
+				level = level * qmul + qadd;
+			}
+			block[i] = level;
+		}
+	}
+    } else {
+        i = 0;
+    }
+
+asm volatile(
+		"movd %1, %%mm6			\n\t" //qmul
+		"packssdw %%mm6, %%mm6		\n\t"
+		"packssdw %%mm6, %%mm6		\n\t"
+		"movd %2, %%mm5			\n\t" //qadd
+		"pxor %%mm7, %%mm7		\n\t"
+		"packssdw %%mm5, %%mm5		\n\t"
+		"packssdw %%mm5, %%mm5		\n\t"
+		"psubw %%mm5, %%mm7		\n\t"
+		"pxor %%mm4, %%mm4		\n\t"
+		".balign 16\n\t"
+		"1:				\n\t"
+		"movq (%0, %3), %%mm0		\n\t"
+		"movq 8(%0, %3), %%mm1		\n\t"
+
+		"pmullw %%mm6, %%mm0		\n\t"
+		"pmullw %%mm6, %%mm1		\n\t"
+
+		"movq (%0, %3), %%mm2		\n\t"
+		"movq 8(%0, %3), %%mm3		\n\t"
+
+		"pcmpgtw %%mm4, %%mm2		\n\t" // block[i] < 0 ? -1 : 0
+		"pcmpgtw %%mm4, %%mm3		\n\t" // block[i] < 0 ? -1 : 0
+
+		"pxor %%mm2, %%mm0		\n\t"
+		"pxor %%mm3, %%mm1		\n\t"
+
+		"paddw %%mm7, %%mm0		\n\t"
+		"paddw %%mm7, %%mm1		\n\t"
+
+		"pxor %%mm0, %%mm2		\n\t"
+		"pxor %%mm1, %%mm3		\n\t"
+
+		"pcmpeqw %%mm7, %%mm0		\n\t" // block[i] == 0 ? -1 : 0
+		"pcmpeqw %%mm7, %%mm1		\n\t" // block[i] == 0 ? -1 : 0
+
+		"pandn %%mm2, %%mm0		\n\t"
+		"pandn %%mm3, %%mm1		\n\t"
+
+		"movq %%mm0, (%0, %3)		\n\t"
+		"movq %%mm1, 8(%0, %3)		\n\t"
+
+		"addl $16, %3			\n\t"
+		"cmpl $128, %3			\n\t"
+		"jb 1b				\n\t"
+		::"r" (block), "g"(qmul), "g" (qadd), "r" (2*i)
+		: "memory"
+	);
+}
+
+
+/*
+  NK:
+  Note: looking at PARANOID:
+  "enable all paranoid tests for rounding, overflows, etc..."
+
+#ifdef PARANOID
+                if (level < -2048 || level > 2047)
+                    fprintf(stderr, "unquant error %d %d\n", i, level);
+#endif
+  We can suppose that result of two multiplications can't be greate of 0xFFFF
+  i.e. is 16-bit, so we use here only PMULLW instruction and can avoid
+  a complex multiplication.
+=====================================================
+ Full formula for multiplication of 2 integer numbers
+ which are represent as high:low words:
+ input: value1 = high1:low1
+        value2 = high2:low2
+ output: value3 = value1*value2
+ value3=high3:low3 (on overflow: modulus 2^32 wrap-around)
+ this mean that for 0x123456 * 0x123456 correct result is 0x766cb0ce4
+ but this algorithm will compute only 0x66cb0ce4
+ this limited by 16-bit size of operands
+ ---------------------------------
+ tlow1 = high1*low2
+ tlow2 = high2*low1
+ tlow1 = tlow1 + tlow2
+ high3:low3 = low1*low2
+ high3 += tlow1
+*/
+static void dct_unquantize_mpeg1_mmx(MpegEncContext *s,
+                                     DCTELEM *block, int n, int qscale)
+{
+    int i, level;
+    const UINT16 *quant_matrix;
+    if (s->mb_intra) {
+        if (n < 4) 
+            block[0] = block[0] * s->y_dc_scale;
+        else
+            block[0] = block[0] * s->c_dc_scale;
+        if (s->out_format == FMT_H263) {
+            i = 1;
+            goto unquant_even;
+        }
+        /* XXX: only mpeg1 */
+        quant_matrix = s->intra_matrix;
+	i=1;
+	/* Align on 4 elements boundary */
+	while(i&3)
+	{
+            level = block[i];
+            if (level) {
+                if (level < 0) level = -level;
+                    level = (int)(level * qscale * quant_matrix[i]) >> 3;
+                    level = (level - 1) | 1;
+                if (block[i] < 0) level = -level;
+                block[i] = level;
+            }
+	    i++;
+	}
+	__asm __volatile(
+	"movd	%0, %%mm6\n\t"       /* mm6 = qscale | 0  */
+	"punpckldq %%mm6, %%mm6\n\t" /* mm6 = qscale | qscale */
+	"movq	%2, %%mm4\n\t"
+	"movq	%%mm6, %%mm7\n\t"
+	"movq	%1, %%mm5\n\t"
+	"packssdw %%mm6, %%mm7\n\t" /* mm7 = qscale | qscale | qscale | qscale */
+	"pxor	%%mm6, %%mm6\n\t"
+	::"g"(qscale),"m"(mm_wone),"m"(mm_wabs):"memory");
+        for(;i<64;i+=4) {
+		__asm __volatile(
+			"movq	%1, %%mm0\n\t"
+			"movq	%%mm7, %%mm1\n\t"
+			"movq	%%mm0, %%mm2\n\t"
+			"movq	%%mm0, %%mm3\n\t"
+			"pcmpgtw %%mm6, %%mm2\n\t"
+			"pmullw	%2, %%mm1\n\t"
+			"pandn	%%mm4, %%mm2\n\t"
+			"por	%%mm5, %%mm2\n\t"
+			"pmullw	%%mm2, %%mm0\n\t" /* mm0 = abs(block[i]). */
+
+			"pcmpeqw %%mm6, %%mm3\n\t"
+			"pmullw	%%mm0, %%mm1\n\t"
+			"psraw	$3, %%mm1\n\t"
+			"psubw	%%mm5, %%mm1\n\t"   /* block[i] --; */
+			"pandn	%%mm4, %%mm3\n\t"  /* fake of pcmpneqw : mm0 != 0 then mm1 = -1 */
+			"por	%%mm5, %%mm1\n\t"   /* block[i] |= 1 */
+			"pmullw %%mm2, %%mm1\n\t"   /* change signs again */
+
+			"pand	%%mm3, %%mm1\n\t" /* nullify if was zero */
+			"movq	%%mm1, %0"
+			:"=m"(block[i])
+			:"m"(block[i]), "m"(quant_matrix[i])
+			:"memory");
+        }
+    } else {
+        i = 0;
+    unquant_even:
+        quant_matrix = s->non_intra_matrix;
+	/* Align on 4 elements boundary */
+	while(i&7)
+	{
+	    level = block[i];
+            if (level) {
+                if (level < 0) level = -level;
+                    level = (((level << 1) + 1) * qscale *
+                             ((int) quant_matrix[i])) >> 4;
+                    level = (level - 1) | 1;
+                if(block[i] < 0) level = -level;
+                block[i] = level;
+	    }
+	    i++;
+	}
+
+asm volatile(
+		"pcmpeqw %%mm7, %%mm7		\n\t"
+		"psrlw $15, %%mm7		\n\t"
+		"movd %2, %%mm6			\n\t"
+		"packssdw %%mm6, %%mm6		\n\t"
+		"packssdw %%mm6, %%mm6		\n\t"
+		".balign 16\n\t"
+		"1:				\n\t"
+		"movq (%0, %3), %%mm0		\n\t"
+		"movq 8(%0, %3), %%mm1		\n\t"
+		"movq (%1, %3), %%mm4		\n\t"
+		"movq 8(%1, %3), %%mm5		\n\t"
+		"pmullw %%mm6, %%mm4		\n\t" // q=qscale*quant_matrix[i]
+		"pmullw %%mm6, %%mm5		\n\t" // q=qscale*quant_matrix[i]
+		"pxor %%mm2, %%mm2		\n\t"
+		"pxor %%mm3, %%mm3		\n\t"
+		"pcmpgtw %%mm0, %%mm2		\n\t" // block[i] < 0 ? -1 : 0
+		"pcmpgtw %%mm1, %%mm3		\n\t" // block[i] < 0 ? -1 : 0
+		"pxor %%mm2, %%mm0		\n\t"
+		"pxor %%mm3, %%mm1		\n\t"
+		"psubw %%mm2, %%mm0		\n\t" // abs(block[i])
+		"psubw %%mm3, %%mm1		\n\t" // abs(block[i])
+		"paddw %%mm0, %%mm0		\n\t" // abs(block[i])*2
+		"paddw %%mm1, %%mm1		\n\t" // abs(block[i])*2
+		"paddw %%mm7, %%mm0		\n\t" // abs(block[i])*2 + 1
+		"paddw %%mm7, %%mm1		\n\t" // abs(block[i])*2 + 1
+		"pmullw %%mm4, %%mm0		\n\t" // (abs(block[i])*2 + 1)*q
+		"pmullw %%mm5, %%mm1		\n\t" // (abs(block[i])*2 + 1)*q
+		"pxor %%mm4, %%mm4		\n\t"
+		"pxor %%mm5, %%mm5		\n\t" // FIXME slow
+		"pcmpeqw (%0, %3), %%mm4	\n\t" // block[i] == 0 ? -1 : 0
+		"pcmpeqw 8(%0, %3), %%mm5	\n\t" // block[i] == 0 ? -1 : 0
+		"psraw $4, %%mm0		\n\t"
+		"psraw $4, %%mm1		\n\t"
+		"psubw %%mm7, %%mm0		\n\t"
+		"psubw %%mm7, %%mm1		\n\t"
+		"por %%mm7, %%mm0		\n\t"
+		"por %%mm7, %%mm1		\n\t"
+		"pxor %%mm2, %%mm0		\n\t"
+		"pxor %%mm3, %%mm1		\n\t"
+		"psubw %%mm2, %%mm0		\n\t"
+		"psubw %%mm3, %%mm1		\n\t"
+		"pandn %%mm0, %%mm4		\n\t"
+		"pandn %%mm1, %%mm5		\n\t"
+		"movq %%mm4, (%0, %3)		\n\t"
+		"movq %%mm5, 8(%0, %3)		\n\t"
+
+		"addl $16, %3			\n\t"
+		"cmpl $128, %3			\n\t"
+		"jb 1b				\n\t"
+		::"r" (block), "r"(quant_matrix), "g" (qscale), "r" (2*i)
+		: "memory"
+	);
+    }
+}
+
+void MPV_common_init_mmx(MpegEncContext *s)
+{
+    if (mm_flags & MM_MMX) {
+        if (s->out_format == FMT_H263)
+        	s->dct_unquantize = dct_unquantize_h263_mmx;
+	else
+        	s->dct_unquantize = dct_unquantize_mpeg1_mmx;
+    }
+}
diff --git a/src/libffmpeg/libavcodec/i386/simple_idct_mmx.c b/src/libffmpeg/libavcodec/i386/simple_idct_mmx.c
new file mode 100644
index 000000000..297f23724
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/simple_idct_mmx.c
@@ -0,0 +1,1455 @@
+/*
+    Copyright (C) 2001 Michael Niedermayer (michaelni@gmx.at)
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program 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 General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program; if not, write to the Free Software
+    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#include <inttypes.h>
+#include "../dsputil.h"
+
+#define C0 23170 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+#define C1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+#define C2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+#define C3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+#define C4 16384 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+#define C5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+#define C6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+#define C7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+
+#define ROW_SHIFT 11
+#define COL_SHIFT 20 // 6
+
+static uint64_t __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000ULL;
+static uint64_t __attribute__((aligned(8))) d40000= 0x0000000000040000ULL;
+static int16_t __attribute__((aligned(8))) temp[64];
+static int16_t __attribute__((aligned(8))) coeffs[]= {
+	1<<(ROW_SHIFT-1), 0, 1<<(ROW_SHIFT-1), 0,
+//	1<<(COL_SHIFT-1), 0, 1<<(COL_SHIFT-1), 0,
+//	0, 1<<(COL_SHIFT-1-16), 0, 1<<(COL_SHIFT-1-16),
+	1<<(ROW_SHIFT-1), 1, 1<<(ROW_SHIFT-1), 0,
+	// the 1 = ((1<<(COL_SHIFT-1))/C4)<<ROW_SHIFT :)
+//	0, 0, 0, 0,
+//	0, 0, 0, 0,
+
+	 C4,  C2,  C4,  C2,
+	 C4,  C6,  C4,  C6,
+	 C1,  C3,  C1,  C3,
+	 C5,  C7,  C5,  C7,
+
+	 C4,  C6,  C4,  C6,
+	-C4, -C2, -C4, -C2,
+	 C3, -C7,  C3, -C7,
+	-C1, -C5, -C1, -C5,
+
+	 C4, -C6,  C4, -C6,
+	-C4,  C2, -C4,  C2,
+	 C5, -C1,  C5, -C1,
+	 C7,  C3,  C7,  C3,
+
+	 C4, -C2,  C4, -C2,
+	 C4, -C6,  C4, -C6,
+	 C7, -C5,  C7, -C5,
+	 C3, -C1,  C3, -C1
+	};
+#if 0
+static void inline idctCol (int16_t * col, int16_t *input)
+{
+#undef C0
+#undef C1
+#undef C2
+#undef C3
+#undef C4
+#undef C5
+#undef C6
+#undef C7
+	int a0, a1, a2, a3, b0, b1, b2, b3;
+	const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C4 = 16384; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C6 = 8867; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C7 = 4520; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+/*
+	if( !(col[8*1] | col[8*2] |col[8*3] |col[8*4] |col[8*5] |col[8*6] | col[8*7])) {
+		col[8*0] = col[8*1] = col[8*2] = col[8*3] = col[8*4] =
+			col[8*5] = col[8*6] = col[8*7] = col[8*0]<<3;
+		return;
+	}*/
+
+col[8*0] = input[8*0 + 0];
+col[8*1] = input[8*2 + 0];
+col[8*2] = input[8*0 + 1];
+col[8*3] = input[8*2 + 1];
+col[8*4] = input[8*4 + 0];
+col[8*5] = input[8*6 + 0];
+col[8*6] = input[8*4 + 1];
+col[8*7] = input[8*6 + 1];
+
+	a0 = C4*col[8*0] + C2*col[8*2] + C4*col[8*4] + C6*col[8*6] + (1<<(COL_SHIFT-1));
+	a1 = C4*col[8*0] + C6*col[8*2] - C4*col[8*4] - C2*col[8*6] + (1<<(COL_SHIFT-1));
+	a2 = C4*col[8*0] - C6*col[8*2] - C4*col[8*4] + C2*col[8*6] + (1<<(COL_SHIFT-1));
+	a3 = C4*col[8*0] - C2*col[8*2] + C4*col[8*4] - C6*col[8*6] + (1<<(COL_SHIFT-1));
+
+	b0 = C1*col[8*1] + C3*col[8*3] + C5*col[8*5] + C7*col[8*7];
+	b1 = C3*col[8*1] - C7*col[8*3] - C1*col[8*5] - C5*col[8*7];
+	b2 = C5*col[8*1] - C1*col[8*3] + C7*col[8*5] + C3*col[8*7];
+	b3 = C7*col[8*1] - C5*col[8*3] + C3*col[8*5] - C1*col[8*7];
+
+	col[8*0] = (a0 + b0) >> COL_SHIFT;
+	col[8*1] = (a1 + b1) >> COL_SHIFT;
+	col[8*2] = (a2 + b2) >> COL_SHIFT;
+	col[8*3] = (a3 + b3) >> COL_SHIFT;
+	col[8*4] = (a3 - b3) >> COL_SHIFT;
+	col[8*5] = (a2 - b2) >> COL_SHIFT;
+	col[8*6] = (a1 - b1) >> COL_SHIFT;
+	col[8*7] = (a0 - b0) >> COL_SHIFT;
+}
+
+static void inline idctRow (int16_t * output, int16_t * input)
+{
+	int16_t row[8];
+
+	int a0, a1, a2, a3, b0, b1, b2, b3;
+	const int C0 = 23170; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C1 = 22725; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C2 = 21407; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C3 = 19266; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C4 = 16384; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C5 = 12873; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C6 = 8867; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+	const int C7 = 4520; //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
+
+row[0] = input[0];
+row[2] = input[1];
+row[4] = input[4];
+row[6] = input[5];
+row[1] = input[8];
+row[3] = input[9];
+row[5] = input[12];
+row[7] = input[13];
+
+	if( !(row[1] | row[2] |row[3] |row[4] |row[5] |row[6] | row[7]) ) {
+		row[0] = row[1] = row[2] = row[3] = row[4] =
+			row[5] = row[6] = row[7] = row[0]<<3;
+	output[0] = row[0];
+	output[2] = row[1];
+	output[4] = row[2];
+	output[6] = row[3];
+	output[8] = row[4];
+	output[10] = row[5];
+	output[12] = row[6];
+	output[14] = row[7];
+		return;
+	}
+
+	a0 = C4*row[0] + C2*row[2] + C4*row[4] + C6*row[6] + (1<<(ROW_SHIFT-1));
+	a1 = C4*row[0] + C6*row[2] - C4*row[4] - C2*row[6] + (1<<(ROW_SHIFT-1));
+	a2 = C4*row[0] - C6*row[2] - C4*row[4] + C2*row[6] + (1<<(ROW_SHIFT-1));
+	a3 = C4*row[0] - C2*row[2] + C4*row[4] - C6*row[6] + (1<<(ROW_SHIFT-1));
+
+	b0 = C1*row[1] + C3*row[3] + C5*row[5] + C7*row[7];
+	b1 = C3*row[1] - C7*row[3] - C1*row[5] - C5*row[7];
+	b2 = C5*row[1] - C1*row[3] + C7*row[5] + C3*row[7];
+	b3 = C7*row[1] - C5*row[3] + C3*row[5] - C1*row[7];
+
+	row[0] = (a0 + b0) >> ROW_SHIFT;
+	row[1] = (a1 + b1) >> ROW_SHIFT;
+	row[2] = (a2 + b2) >> ROW_SHIFT;
+	row[3] = (a3 + b3) >> ROW_SHIFT;
+	row[4] = (a3 - b3) >> ROW_SHIFT;
+	row[5] = (a2 - b2) >> ROW_SHIFT;
+	row[6] = (a1 - b1) >> ROW_SHIFT;
+	row[7] = (a0 - b0) >> ROW_SHIFT;
+
+	output[0] = row[0];
+	output[2] = row[1];
+	output[4] = row[2];
+	output[6] = row[3];
+	output[8] = row[4];
+	output[10] = row[5];
+	output[12] = row[6];
+	output[14] = row[7];
+}
+#endif
+
+static inline void idct(int16_t *block)
+{
+	int i;
+//for(i=0; i<64; i++) temp[i]= block[ block_permute_op(i) ];
+//for(i=0; i<64; i++) temp[block_permute_op(i)]= block[ i ];
+//for(i=0; i<64; i++) block[i]= temp[i];
+//block_permute(block);
+/*
+idctRow(temp, block);
+idctRow(temp+16, block+16);
+idctRow(temp+1, block+2);
+idctRow(temp+17, block+18);
+idctRow(temp+32, block+32);
+idctRow(temp+48, block+48);
+idctRow(temp+33, block+34);
+idctRow(temp+49, block+50);
+*/
+
+	asm volatile(
+//		"lea 64(%0), %%eax		\n\t"
+//r0,r2,R0,R2	r4,r6,R4,R6	r1,r3,R1,R3	r5,r7,R5,R7
+//src0		src4		src1		src5
+//r0,R0,r7,R7	r1,R1,r6,R6	r2,R2,r5,R5	r3,R3,r4,R4
+//dst0		dst1		dst2		dst3
+#if 0 //Alternative, simpler variant
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src4 ", %%mm1			\n\t" /* R6	R4	r6	r4 */\
+	"movq " #src1 ", %%mm2			\n\t" /* R3	R1	r3	r1 */\
+	"movq " #src5 ", %%mm3			\n\t" /* R7	R5	r7	r5 */\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 24(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R6+C4R4	C6r6+C4r4 */\
+	"movq 32(%2), %%mm6			\n\t" /* C3	C1	C3	C1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* C3R3+C1R1	C3r3+C1r1 */\
+	"movq 40(%2), %%mm7			\n\t" /* C7	C5	C7	C5 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C7R7+C5R5	C7r7+C5r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A0		a0 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B0		b0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0(%%mm6, %%mm4, dst) \
+\
+	"movq 56(%2), %%mm4			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* -C2R6-C4R4	-C2r6-C4r4 */\
+	"movq 64(%2), %%mm6			\n\t" /* -C7	C3	-C7	C3 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C7R3+C3R1	-C7r3+C3r1 */\
+	"movq 72(%2), %%mm7			\n\t" /* -C5	-C1	-C5	-C1 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* -C5R7-C1R5	-C5r7-C1r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A1		a1 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 80(%2), %%mm5			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE1(%%mm6, %%mm4, dst, %%mm7) \
+\
+	"movq 88(%2), %%mm4			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* C2R6-C4R4	C2r6-C4r4 */\
+	"movq 96(%2), %%mm6			\n\t" /* -C1	C5	-C1	C5 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C1R3+C5R1	-C1r3+C5r1 */\
+	"movq 104(%2), %%mm7			\n\t" /* C3	C7	C3	C7 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C3R7+C7R5	C3r7+C7r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A2		a2 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"pmaddwd 120(%2), %%mm1			\n\t" /* -C6R6+C4R4	-C6r6+C4r4 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 128(%2), %%mm2			\n\t" /* -C5R3+C7R1	-C5r3+C7r1 */\
+	"pmaddwd 136(%2), %%mm3			\n\t" /* -C1R7+C3R5	-C1r7+C3r5 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm1, %%mm0			\n\t" /* A3		a3 */\
+	#rounder ", %%mm0			\n\t"\
+	"paddd %%mm3, %%mm2			\n\t" /* B3		b3 */\
+	"paddd %%mm0, %%mm2			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm2, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2(%%mm6, %%mm4, %%mm2, %%mm0, dst)
+
+#define DC_COND_IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src4 ", %%mm1			\n\t" /* R6	R4	r6	r4 */\
+	"movq " #src1 ", %%mm2			\n\t" /* R3	R1	r3	r1 */\
+	"movq " #src5 ", %%mm3			\n\t" /* R7	R5	r7	r5 */\
+	"movq wm1010, %%mm4			\n\t"\
+	"pand %%mm0, %%mm4			\n\t"\
+	"por %%mm1, %%mm4			\n\t"\
+	"por %%mm2, %%mm4			\n\t"\
+	"por %%mm3, %%mm4			\n\t"\
+	"packssdw %%mm4,%%mm4			\n\t"\
+	"movd %%mm4, %%eax			\n\t"\
+	"orl %%eax, %%eax			\n\t"\
+	"jz 1f					\n\t"\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 24(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R6+C4R4	C6r6+C4r4 */\
+	"movq 32(%2), %%mm6			\n\t" /* C3	C1	C3	C1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* C3R3+C1R1	C3r3+C1r1 */\
+	"movq 40(%2), %%mm7			\n\t" /* C7	C5	C7	C5 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C7R7+C5R5	C7r7+C5r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A0		a0 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B0		b0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0(%%mm6, %%mm4, dst) \
+\
+	"movq 56(%2), %%mm4			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* -C2R6-C4R4	-C2r6-C4r4 */\
+	"movq 64(%2), %%mm6			\n\t" /* -C7	C3	-C7	C3 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C7R3+C3R1	-C7r3+C3r1 */\
+	"movq 72(%2), %%mm7			\n\t" /* -C5	-C1	-C5	-C1 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* -C5R7-C1R5	-C5r7-C1r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A1		a1 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 80(%2), %%mm5			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE1(%%mm6, %%mm4, dst, %%mm7) \
+\
+	"movq 88(%2), %%mm4			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* C2R6-C4R4	C2r6-C4r4 */\
+	"movq 96(%2), %%mm6			\n\t" /* -C1	C5	-C1	C5 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C1R3+C5R1	-C1r3+C5r1 */\
+	"movq 104(%2), %%mm7			\n\t" /* C3	C7	C3	C7 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C3R7+C7R5	C3r7+C7r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A2		a2 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"pmaddwd 120(%2), %%mm1			\n\t" /* -C6R6+C4R4	-C6r6+C4r4 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 128(%2), %%mm2			\n\t" /* -C5R3+C7R1	-C5r3+C7r1 */\
+	"pmaddwd 136(%2), %%mm3			\n\t" /* -C1R7+C3R5	-C1r7+C3r5 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm1, %%mm0			\n\t" /* A3		a3 */\
+	#rounder ", %%mm0			\n\t"\
+	"paddd %%mm3, %%mm2			\n\t" /* B3		b3 */\
+	"paddd %%mm0, %%mm2			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm2, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2(%%mm6, %%mm4, %%mm2, %%mm0, dst)\
+	"jmp 2f					\n\t"\
+	"1:					\n\t"\
+	WRITE3(%%mm0, dst)\
+	"2:					\n\t"\
+
+
+#define WRITE0(s0, s7, dst)\
+	"movq " #s0 ", " #dst "			\n\t" /* R0		r0 */\
+	"movq " #s7 ", 24+" #dst "		\n\t" /* R7		r7 */
+
+#define WRITE1(s1, s6, dst, tmp)\
+	"movq " #dst ", " #tmp "		\n\t" /* R0		r0 */\
+	"packssdw " #s1 ", " #tmp "		\n\t" /* R1	r1	R0	r0*/\
+	"movq " #tmp ", " #dst "		\n\t"\
+	"movq 24+" #dst ", " #tmp "		\n\t" /* R7		r7 */\
+	"packssdw " #tmp ", " #s6 "		\n\t" /* R7	r7	R6	r6*/\
+	"movq " #s6 ", 24+" #dst "		\n\t"
+
+#define WRITE2(s2, s5, s3, s4, dst)\
+	"packssdw " #s3 ", " #s2 "		\n\t" /* R3	r3	R2	r2*/\
+	"packssdw " #s5 ", " #s4 "		\n\t" /* R5	r5	R4	r4*/\
+	"movq " #s2 ", 8+" #dst "		\n\t"\
+	"movq " #s4 ", 16+" #dst "		\n\t"
+
+#define WRITE3(a, dst)\
+	"pslld $16, " #a "			\n\t"\
+	"psrad $13, " #a "			\n\t"\
+	"packssdw " #a ", " #a "		\n\t"\
+	"movq " #a ", " #dst "			\n\t"\
+	"movq " #a ", 8+" #dst "		\n\t"\
+	"movq " #a ", 16+" #dst "		\n\t"\
+	"movq " #a ", 24+" #dst "		\n\t"\
+
+//IDCT_CORE(          src0,   src4,   src1,   src5,    dst,   rounder, shift)
+IDCT_CORE(            (%0),  8(%0), 16(%0), 24(%0),  0(%1),paddd 8(%2), 11)
+/*
+DC_COND_IDCT_CORE(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1),paddd (%2), 11)
+DC_COND_IDCT_CORE(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11)
+DC_COND_IDCT_CORE(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11)
+*/
+IDCT_CORE(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1),paddd (%2), 11)
+IDCT_CORE(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11)
+IDCT_CORE(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11)
+
+#undef WRITE0
+#undef WRITE1
+#undef WRITE2
+
+#define WRITE0(s0, s7, dst)\
+	"packssdw " #s0 ", " #s0 "		\n\t" /* C0, c0, C0, c0 */\
+	"packssdw " #s7 ", " #s7 "		\n\t" /* C7, c7, C7, c7 */\
+	"movd " #s0 ", " #dst "			\n\t" /* C0, c0 */\
+	"movd " #s7 ", 112+" #dst "		\n\t" /* C7, c7 */
+
+#define WRITE1(s1, s6, dst, tmp)\
+	"packssdw " #s1 ", " #s1 "		\n\t" /* C1, c1, C1, c1 */\
+	"packssdw " #s6 ", " #s6 "		\n\t" /* C6, c6, C6, c6 */\
+	"movd " #s1 ", 16+" #dst "		\n\t" /* C1, c1 */\
+	"movd " #s6 ", 96+" #dst "		\n\t" /* C6, c6 */
+
+#define WRITE2(s2, s5, s3, s4, dst)\
+	"packssdw " #s2 ", " #s2 "		\n\t" /* C2, c2, C2, c2 */\
+	"packssdw " #s3 ", " #s3 "		\n\t" /* C3, c3, C3, c3 */\
+	"movd " #s2 ", 32+" #dst "		\n\t" /* C2, c2 */\
+	"movd " #s3 ", 48+" #dst "		\n\t" /* C3, c3 */\
+	"packssdw " #s4 ", " #s4 "		\n\t" /* C4, c4, C4, c4 */\
+	"packssdw " #s5 ", " #s5 "		\n\t" /* C5, c5, C5, c5 */\
+	"movd " #s4 ", 64+" #dst "		\n\t" /* C4, c4 */\
+	"movd " #s5 ", 80+" #dst "		\n\t" /* C5, c5 */\
+
+//IDCT_CORE(  src0,   src4,   src1,    src5,    dst, rounder, shift)
+IDCT_CORE(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
+IDCT_CORE(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
+IDCT_CORE(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
+IDCT_CORE(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
+
+#else
+
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src4 ", %%mm1			\n\t" /* R6	R4	r6	r4 */\
+	"movq " #src1 ", %%mm2			\n\t" /* R3	R1	r3	r1 */\
+	"movq " #src5 ", %%mm3			\n\t" /* R7	R5	r7	r5 */\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 24(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R6+C4R4	C6r6+C4r4 */\
+	"movq 32(%2), %%mm6			\n\t" /* C3	C1	C3	C1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* C3R3+C1R1	C3r3+C1r1 */\
+	"movq 40(%2), %%mm7			\n\t" /* C7	C5	C7	C5 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C7R7+C5R5	C7r7+C5r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A0		a0 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B0		b0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0(%%mm6, %%mm4, dst) \
+\
+	"movq 56(%2), %%mm4			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* -C2R6-C4R4	-C2r6-C4r4 */\
+	"movq 64(%2), %%mm6			\n\t" /* -C7	C3	-C7	C3 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C7R3+C3R1	-C7r3+C3r1 */\
+	"movq 72(%2), %%mm7			\n\t" /* -C5	-C1	-C5	-C1 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* -C5R7-C1R5	-C5r7-C1r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A1		a1 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 80(%2), %%mm5			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE1(%%mm6, %%mm4, dst, %%mm7) \
+\
+	"movq 88(%2), %%mm4			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* C2R6-C4R4	C2r6-C4r4 */\
+	"movq 96(%2), %%mm6			\n\t" /* -C1	C5	-C1	C5 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C1R3+C5R1	-C1r3+C5r1 */\
+	"movq 104(%2), %%mm7			\n\t" /* C3	C7	C3	C7 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C3R7+C7R5	C3r7+C7r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A2		a2 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"pmaddwd 120(%2), %%mm1			\n\t" /* -C6R6+C4R4	-C6r6+C4r4 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 128(%2), %%mm2			\n\t" /* -C5R3+C7R1	-C5r3+C7r1 */\
+	"pmaddwd 136(%2), %%mm3			\n\t" /* -C1R7+C3R5	-C1r7+C3r5 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm1, %%mm0			\n\t" /* A3		a3 */\
+	#rounder ", %%mm0			\n\t"\
+	"paddd %%mm3, %%mm2			\n\t" /* B3		b3 */\
+	"paddd %%mm0, %%mm2			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm2, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2(%%mm6, %%mm4, %%mm2, %%mm0, dst)
+
+#define DC_COND_IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src4 ", %%mm1			\n\t" /* R6	R4	r6	r4 */\
+	"movq " #src1 ", %%mm2			\n\t" /* R3	R1	r3	r1 */\
+	"movq " #src5 ", %%mm3			\n\t" /* R7	R5	r7	r5 */\
+	"movq wm1010, %%mm4			\n\t"\
+	"pand %%mm0, %%mm4			\n\t"\
+	"por %%mm1, %%mm4			\n\t"\
+	"por %%mm2, %%mm4			\n\t"\
+	"por %%mm3, %%mm4			\n\t"\
+	"packssdw %%mm4,%%mm4			\n\t"\
+	"movd %%mm4, %%eax			\n\t"\
+	"orl %%eax, %%eax			\n\t"\
+	"jz 1f					\n\t"\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 24(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R6+C4R4	C6r6+C4r4 */\
+	"movq 32(%2), %%mm6			\n\t" /* C3	C1	C3	C1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* C3R3+C1R1	C3r3+C1r1 */\
+	"movq 40(%2), %%mm7			\n\t" /* C7	C5	C7	C5 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C7R7+C5R5	C7r7+C5r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A0		a0 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B0		b0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0(%%mm6, %%mm4, dst) \
+\
+	"movq 56(%2), %%mm4			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* -C2R6-C4R4	-C2r6-C4r4 */\
+	"movq 64(%2), %%mm6			\n\t" /* -C7	C3	-C7	C3 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C7R3+C3R1	-C7r3+C3r1 */\
+	"movq 72(%2), %%mm7			\n\t" /* -C5	-C1	-C5	-C1 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* -C5R7-C1R5	-C5r7-C1r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A1		a1 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 80(%2), %%mm5			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE1(%%mm6, %%mm4, dst, %%mm7) \
+\
+	"movq 88(%2), %%mm4			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* C2R6-C4R4	C2r6-C4r4 */\
+	"movq 96(%2), %%mm6			\n\t" /* -C1	C5	-C1	C5 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C1R3+C5R1	-C1r3+C5r1 */\
+	"movq 104(%2), %%mm7			\n\t" /* C3	C7	C3	C7 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C3R7+C7R5	C3r7+C7r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A2		a2 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"pmaddwd 120(%2), %%mm1			\n\t" /* -C6R6+C4R4	-C6r6+C4r4 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 128(%2), %%mm2			\n\t" /* -C5R3+C7R1	-C5r3+C7r1 */\
+	"pmaddwd 136(%2), %%mm3			\n\t" /* -C1R7+C3R5	-C1r7+C3r5 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm1, %%mm0			\n\t" /* A3		a3 */\
+	#rounder ", %%mm0			\n\t"\
+	"paddd %%mm3, %%mm2			\n\t" /* B3		b3 */\
+	"paddd %%mm0, %%mm2			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm2, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2(%%mm6, %%mm4, %%mm2, %%mm0, dst)\
+	"jmp 2f					\n\t"\
+	"#.balign 16				\n\t"\
+	"1:					\n\t"\
+	WRITE3(%%mm0, dst)\
+	"2:					\n\t"\
+
+#define Z_COND_IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift, bt) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src4 ", %%mm1			\n\t" /* R6	R4	r6	r4 */\
+	"movq " #src1 ", %%mm2			\n\t" /* R3	R1	r3	r1 */\
+	"movq " #src5 ", %%mm3			\n\t" /* R7	R5	r7	r5 */\
+	"movq %%mm0, %%mm4			\n\t"\
+	"por %%mm1, %%mm4			\n\t"\
+	"por %%mm2, %%mm4			\n\t"\
+	"por %%mm3, %%mm4			\n\t"\
+	"packssdw %%mm4, %%mm4			\n\t"\
+	"movd %%mm4, %%eax			\n\t"\
+	"orl %%eax, %%eax			\n\t"\
+	"jz " #bt "				\n\t"\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 24(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R6+C4R4	C6r6+C4r4 */\
+	"movq 32(%2), %%mm6			\n\t" /* C3	C1	C3	C1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* C3R3+C1R1	C3r3+C1r1 */\
+	"movq 40(%2), %%mm7			\n\t" /* C7	C5	C7	C5 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C7R7+C5R5	C7r7+C5r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A0		a0 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B0		b0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0(%%mm6, %%mm4, dst) \
+\
+	"movq 56(%2), %%mm4			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* -C2R6-C4R4	-C2r6-C4r4 */\
+	"movq 64(%2), %%mm6			\n\t" /* -C7	C3	-C7	C3 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C7R3+C3R1	-C7r3+C3r1 */\
+	"movq 72(%2), %%mm7			\n\t" /* -C5	-C1	-C5	-C1 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* -C5R7-C1R5	-C5r7-C1r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A1		a1 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"movq 80(%2), %%mm5			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE1(%%mm6, %%mm4, dst, %%mm7) \
+\
+	"movq 88(%2), %%mm4			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* C2R6-C4R4	C2r6-C4r4 */\
+	"movq 96(%2), %%mm6			\n\t" /* -C1	C5	-C1	C5 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C1R3+C5R1	-C1r3+C5r1 */\
+	"movq 104(%2), %%mm7			\n\t" /* C3	C7	C3	C7 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C3R7+C7R5	C3r7+C7r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A2		a2 */\
+	#rounder ", %%mm4			\n\t"\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"pmaddwd 120(%2), %%mm1			\n\t" /* -C6R6+C4R4	-C6r6+C4r4 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 128(%2), %%mm2			\n\t" /* -C5R3+C7R1	-C5r3+C7r1 */\
+	"pmaddwd 136(%2), %%mm3			\n\t" /* -C1R7+C3R5	-C1r7+C3r5 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm1, %%mm0			\n\t" /* A3		a3 */\
+	#rounder ", %%mm0			\n\t"\
+	"paddd %%mm3, %%mm2			\n\t" /* B3		b3 */\
+	"paddd %%mm0, %%mm2			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm2, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2(%%mm6, %%mm4, %%mm2, %%mm0, dst)\
+
+
+#define WRITE0(s0, s7, dst)\
+	"movq " #s0 ", " #dst "			\n\t" /* R0		r0 */\
+	"movq " #s7 ", 24+" #dst "		\n\t" /* R7		r7 */
+
+#define WRITE1(s1, s6, dst, tmp)\
+	"movq " #dst ", " #tmp "		\n\t" /* R0		r0 */\
+	"packssdw " #s1 ", " #tmp "		\n\t" /* R1	r1	R0	r0*/\
+	"movq " #tmp ", " #dst "		\n\t"\
+	"movq 24+" #dst ", " #tmp "		\n\t" /* R7		r7 */\
+	"packssdw " #tmp ", " #s6 "		\n\t" /* R7	r7	R6	r6*/\
+	"movq " #s6 ", 24+" #dst "		\n\t"
+
+#define WRITE2(s2, s5, s3, s4, dst)\
+	"packssdw " #s3 ", " #s2 "		\n\t" /* R3	r3	R2	r2*/\
+	"packssdw " #s5 ", " #s4 "		\n\t" /* R5	r5	R4	r4*/\
+	"movq " #s2 ", 8+" #dst "		\n\t"\
+	"movq " #s4 ", 16+" #dst "		\n\t"
+
+#define WRITE3(a, dst)\
+	"pslld $16, " #a "			\n\t"\
+	"paddd d40000, " #a "			\n\t"\
+	"psrad $13, " #a "			\n\t"\
+	"packssdw " #a ", " #a "		\n\t"\
+	"movq " #a ", " #dst "			\n\t"\
+	"movq " #a ", 8+" #dst "		\n\t"\
+	"movq " #a ", 16+" #dst "		\n\t"\
+	"movq " #a ", 24+" #dst "		\n\t"\
+
+#define WRITE0b(s0, s7, dst)\
+	"packssdw " #s0 ", " #s0 "		\n\t" /* C0, c0, C0, c0 */\
+	"packssdw " #s7 ", " #s7 "		\n\t" /* C7, c7, C7, c7 */\
+	"movd " #s0 ", " #dst "			\n\t" /* C0, c0 */\
+	"movd " #s7 ", 112+" #dst "		\n\t" /* C7, c7 */
+
+#define WRITE1b(s1, s6, dst, tmp)\
+	"packssdw " #s1 ", " #s1 "		\n\t" /* C1, c1, C1, c1 */\
+	"packssdw " #s6 ", " #s6 "		\n\t" /* C6, c6, C6, c6 */\
+	"movd " #s1 ", 16+" #dst "		\n\t" /* C1, c1 */\
+	"movd " #s6 ", 96+" #dst "		\n\t" /* C6, c6 */
+
+#define WRITE2b(s2, s5, s3, s4, dst)\
+	"packssdw " #s2 ", " #s2 "		\n\t" /* C2, c2, C2, c2 */\
+	"packssdw " #s3 ", " #s3 "		\n\t" /* C3, c3, C3, c3 */\
+	"movd " #s2 ", 32+" #dst "		\n\t" /* C2, c2 */\
+	"movd " #s3 ", 48+" #dst "		\n\t" /* C3, c3 */\
+	"packssdw " #s4 ", " #s4 "		\n\t" /* C4, c4, C4, c4 */\
+	"packssdw " #s5 ", " #s5 "		\n\t" /* C5, c5, C5, c5 */\
+	"movd " #s4 ", 64+" #dst "		\n\t" /* C4, c4 */\
+	"movd " #s5 ", 80+" #dst "		\n\t" /* C5, c5 */\
+
+
+//IDCT_CORE(         src0,   src4,   src1,   src5,    dst,   rounder, shift)
+DC_COND_IDCT_CORE(  0(%0),  8(%0), 16(%0), 24(%0),  0(%1),paddd 8(%2), 11)
+Z_COND_IDCT_CORE(  32(%0), 40(%0), 48(%0), 56(%0), 32(%1),paddd (%2), 11, 4f)
+Z_COND_IDCT_CORE(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11, 2f)
+Z_COND_IDCT_CORE(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 1f)
+
+#undef IDCT_CORE
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src4 ", %%mm1			\n\t" /* R6	R4	r6	r4 */\
+	"movq " #src1 ", %%mm2			\n\t" /* R3	R1	r3	r1 */\
+	"movq " #src5 ", %%mm3			\n\t" /* R7	R5	r7	r5 */\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 24(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R6+C4R4	C6r6+C4r4 */\
+	"movq 32(%2), %%mm6			\n\t" /* C3	C1	C3	C1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* C3R3+C1R1	C3r3+C1r1 */\
+	"movq 40(%2), %%mm7			\n\t" /* C7	C5	C7	C5 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C7R7+C5R5	C7r7+C5r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A0		a0 */\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B0		b0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0b(%%mm6, %%mm4, dst) \
+\
+	"movq 56(%2), %%mm4			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* -C2R6-C4R4	-C2r6-C4r4 */\
+	"movq 64(%2), %%mm6			\n\t" /* -C7	C3	-C7	C3 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C7R3+C3R1	-C7r3+C3r1 */\
+	"movq 72(%2), %%mm7			\n\t" /* -C5	-C1	-C5	-C1 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* -C5R7-C1R5	-C5r7-C1r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A1		a1 */\
+\
+	"movq 80(%2), %%mm5			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE1b(%%mm6, %%mm4, dst, %%mm7) \
+\
+	"movq 88(%2), %%mm4			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* C2R6-C4R4	C2r6-C4r4 */\
+	"movq 96(%2), %%mm6			\n\t" /* -C1	C5	-C1	C5 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C1R3+C5R1	-C1r3+C5r1 */\
+	"movq 104(%2), %%mm7			\n\t" /* C3	C7	C3	C7 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C3R7+C7R5	C3r7+C7r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A2		a2 */\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"pmaddwd 120(%2), %%mm1			\n\t" /* -C6R6+C4R4	-C6r6+C4r4 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 128(%2), %%mm2			\n\t" /* -C5R3+C7R1	-C5r3+C7r1 */\
+	"pmaddwd 136(%2), %%mm3			\n\t" /* -C1R7+C3R5	-C1r7+C3r5 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm1, %%mm0			\n\t" /* A3		a3 */\
+	"paddd %%mm3, %%mm2			\n\t" /* B3		b3 */\
+	"paddd %%mm0, %%mm2			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm2, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2b(%%mm6, %%mm4, %%mm2, %%mm0, dst)
+
+//IDCT_CORE(  src0,   src4,   src1,    src5,    dst, rounder, shift)
+IDCT_CORE(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
+IDCT_CORE(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
+IDCT_CORE(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
+IDCT_CORE(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
+	"jmp 9f					\n\t"
+
+	"#.balign 16				\n\t"\
+	"4:					\n\t"
+Z_COND_IDCT_CORE(  64(%0), 72(%0), 80(%0), 88(%0), 64(%1),paddd (%2), 11, 6f)
+Z_COND_IDCT_CORE(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 5f)
+
+#undef IDCT_CORE
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src4 ", %%mm1			\n\t" /* R6	R4	r6	r4 */\
+	"movq " #src5 ", %%mm3			\n\t" /* R7	R5	r7	r5 */\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 24(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R6+C4R4	C6r6+C4r4 */\
+	"movq 40(%2), %%mm7			\n\t" /* C7	C5	C7	C5 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C7R7+C5R5	C7r7+C5r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A0		a0 */\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm4, %%mm7			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm7, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm7		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0b(%%mm7, %%mm4, dst) \
+\
+	"movq 56(%2), %%mm4			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* -C2R6-C4R4	-C2r6-C4r4 */\
+	"movq 72(%2), %%mm7			\n\t" /* -C5	-C1	-C5	-C1 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* -C5R7-C1R5	-C5r7-C1r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A1		a1 */\
+\
+	"movq 80(%2), %%mm5			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm4, %%mm7			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm7, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm7		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE1b(%%mm7, %%mm4, dst, %%mm6) \
+\
+	"movq 88(%2), %%mm4			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* C2R6-C4R4	C2r6-C4r4 */\
+	"movq 104(%2), %%mm7			\n\t" /* C3	C7	C3	C7 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C3R7+C7R5	C3r7+C7r5 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A2		a2 */\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm4, %%mm7			\n\t" /* A1+B1		a1+b1 */\
+	"pmaddwd 120(%2), %%mm1			\n\t" /* -C6R6+C4R4	-C6r6+C4r4 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm7, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 136(%2), %%mm3			\n\t" /* -C1R7+C3R5	-C1r7+C3r5 */\
+	"psrad $" #shift ", %%mm7		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm1, %%mm0			\n\t" /* A3		a3 */\
+	"paddd %%mm0, %%mm3			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm3, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm3		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2b(%%mm7, %%mm4, %%mm3, %%mm0, dst)
+
+//IDCT_CORE(  src0,   src4,   src1,    src5,    dst, rounder, shift)
+IDCT_CORE(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
+IDCT_CORE(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
+IDCT_CORE(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
+IDCT_CORE(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
+	"jmp 9f					\n\t"
+
+	"#.balign 16				\n\t"\
+	"6:					\n\t"
+Z_COND_IDCT_CORE(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 7f)
+
+#undef IDCT_CORE
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src5 ", %%mm3			\n\t" /* R7	R5	r7	r5 */\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 40(%2), %%mm7			\n\t" /* C7	C5	C7	C5 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C7R7+C5R5	C7r7+C5r5 */\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm4, %%mm7			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm7, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm7		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0b(%%mm7, %%mm4, dst) \
+\
+	"movq 72(%2), %%mm7			\n\t" /* -C5	-C1	-C5	-C1 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* -C5R7-C1R5	-C5r7-C1r5 */\
+\
+	"movq 80(%2), %%mm4			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm5, %%mm7			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm5, %%mm5			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm7, %%mm5			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm7		\n\t"\
+	"psrad $" #shift ", %%mm5		\n\t"\
+	WRITE1b(%%mm7, %%mm5, dst, %%mm6) \
+\
+	"movq 104(%2), %%mm7			\n\t" /* C3	C7	C3	C7 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C3R7+C7R5	C3r7+C7r5 */\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm4, %%mm7			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm7, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 136(%2), %%mm3			\n\t" /* -C1R7+C3R5	-C1r7+C3r5 */\
+	"psrad $" #shift ", %%mm7		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm0, %%mm3			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm3, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm3		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2b(%%mm7, %%mm4, %%mm3, %%mm0, dst)
+
+//IDCT_CORE(  src0,   src4,   src1,    src5,    dst, rounder, shift)
+IDCT_CORE(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
+IDCT_CORE(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
+IDCT_CORE(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
+IDCT_CORE(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
+	"jmp 9f					\n\t"
+
+	"#.balign 16				\n\t"\
+	"2:					\n\t"
+Z_COND_IDCT_CORE(  96(%0),104(%0),112(%0),120(%0), 96(%1),paddd (%2), 11, 3f)
+
+#undef IDCT_CORE
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src1 ", %%mm2			\n\t" /* R3	R1	r3	r1 */\
+	"movq " #src5 ", %%mm3			\n\t" /* R7	R5	r7	r5 */\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 32(%2), %%mm6			\n\t" /* C3	C1	C3	C1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* C3R3+C1R1	C3r3+C1r1 */\
+	"movq 40(%2), %%mm7			\n\t" /* C7	C5	C7	C5 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C7R7+C5R5	C7r7+C5r5 */\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B0		b0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0b(%%mm6, %%mm4, dst) \
+\
+	"movq 64(%2), %%mm6			\n\t" /* -C7	C3	-C7	C3 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C7R3+C3R1	-C7r3+C3r1 */\
+	"movq 72(%2), %%mm7			\n\t" /* -C5	-C1	-C5	-C1 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* -C5R7-C1R5	-C5r7-C1r5 */\
+\
+	"movq 80(%2), %%mm4			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm5, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm5, %%mm5			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm5			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm5		\n\t"\
+	WRITE1b(%%mm6, %%mm5, dst, %%mm7) \
+\
+	"movq 96(%2), %%mm6			\n\t" /* -C1	C5	-C1	C5 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C1R3+C5R1	-C1r3+C5r1 */\
+	"movq 104(%2), %%mm7			\n\t" /* C3	C7	C3	C7 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /* C3R7+C7R5	C3r7+C7r5 */\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm7, %%mm6			\n\t" /* B1		b1 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 128(%2), %%mm2			\n\t" /* -C5R3+C7R1	-C5r3+C7r1 */\
+	"pmaddwd 136(%2), %%mm3			\n\t" /* -C1R7+C3R5	-C1r7+C3r5 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm3, %%mm2			\n\t" /* B3		b3 */\
+	"paddd %%mm0, %%mm2			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm2, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2b(%%mm6, %%mm4, %%mm2, %%mm0, dst)
+
+//IDCT_CORE(  src0,   src4,   src1,    src5,    dst, rounder, shift)
+IDCT_CORE(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
+IDCT_CORE(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
+IDCT_CORE(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
+IDCT_CORE(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
+	"jmp 9f					\n\t"
+
+	"#.balign 16				\n\t"\
+	"3:					\n\t"
+#undef IDCT_CORE
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src1 ", %%mm2			\n\t" /* R3	R1	r3	r1 */\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 32(%2), %%mm6			\n\t" /* C3	C1	C3	C1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* C3R3+C1R1	C3r3+C1r1 */\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0b(%%mm6, %%mm4, dst) \
+\
+	"movq 64(%2), %%mm6			\n\t" /* -C7	C3	-C7	C3 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C7R3+C3R1	-C7r3+C3r1 */\
+\
+	"movq 80(%2), %%mm4			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm5, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm5, %%mm5			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm5			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm5		\n\t"\
+	WRITE1b(%%mm6, %%mm5, dst, %%mm7) \
+\
+	"movq 96(%2), %%mm6			\n\t" /* -C1	C5	-C1	C5 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C1R3+C5R1	-C1r3+C5r1 */\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 128(%2), %%mm2			\n\t" /* -C5R3+C7R1	-C5r3+C7r1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm0, %%mm2			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm2, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2b(%%mm6, %%mm4, %%mm2, %%mm0, dst)
+
+//IDCT_CORE(  src0,   src4,   src1,    src5,    dst, rounder, shift)
+IDCT_CORE(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
+IDCT_CORE(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
+IDCT_CORE(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
+IDCT_CORE(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
+	"jmp 9f					\n\t"
+
+	"#.balign 16				\n\t"\
+	"5:					\n\t"
+#undef IDCT_CORE
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"movq %%mm4, %%mm6\n\t"\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq " #src4 ", %%mm1			\n\t" /* R6	R4	r6	r4 */\
+	"movq 24(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"movq %%mm5, %%mm7\n\t"\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R6+C4R4	C6r6+C4r4 */\
+	"movq 8+" #src0 ", %%mm2		\n\t" /*2R2	R0	r2	r0 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /*2C2R2+C4R0	C2r2+C4r0 */\
+	"movq 8+" #src4 ", %%mm3		\n\t" /*2R6	R4	r6	r4 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /*2C6R6+C4R4	C6r6+C4r4 */\
+\
+	"paddd %%mm5, %%mm4			\n\t" /* A0		a0 */\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+\
+	"paddd %%mm7, %%mm6			\n\t" /*2A0		a0 */\
+	"movq 56(%2), %%mm7			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"pmaddwd %%mm1, %%mm7			\n\t" /* -C2R6-C4R4	-C2r6-C4r4 */\
+\
+	"packssdw %%mm6, %%mm4			\n\t" /* C0, c0, C0, c0 */\
+	"movq 48(%2), %%mm6			\n\t" /* C6	C4	C6	C4 */\
+	"movq %%mm4, " #dst "			\n\t" /* C0, c0 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /*2C6R2+C4R0	C6r2+C4r0 */\
+\
+	"movq %%mm4, 112+" #dst "		\n\t" /* C0, c0 */\
+	"movq 56(%2), %%mm4			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"pmaddwd %%mm3, %%mm4			\n\t" /*2-C2R6-C4R4	-C2r6-C4r4 */\
+\
+	"paddd %%mm5, %%mm7			\n\t" /* A1		a1 */\
+	"movq 80(%2), %%mm5			\n\t" /* -C6	C4	-C6	C4 */\
+	"psrad $" #shift ", %%mm7		\n\t"\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+\
+	"paddd %%mm4, %%mm6			\n\t" /*2A1		a1 */\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"movq 88(%2), %%mm4			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* C2R6-C4R4	C2r6-C4r4 */\
+\
+	"pmaddwd 120(%2), %%mm1			\n\t" /* -C6R6+C4R4	-C6r6+C4r4 */\
+	"packssdw %%mm6, %%mm7			\n\t" /* C1, c1, C1, c1 */\
+\
+	"movq 80(%2), %%mm6			\n\t" /* -C6	C4	-C6	C4 */\
+	"movq %%mm7, 16+" #dst "		\n\t" /* C1, c1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /*2-C6R2+C4R0	-C6r2+C4r0 */\
+\
+	"movq %%mm7, 96+" #dst "		\n\t" /* C1, c1 */\
+	"movq 88(%2), %%mm7			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm3, %%mm7			\n\t" /*2C2R6-C4R4	C2r6-C4r4 */\
+\
+	"pmaddwd 112(%2), %%mm2			\n\t" /*2-C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A2		a2 */\
+\
+	"pmaddwd 120(%2), %%mm3			\n\t" /*2-C6R6+C4R4	-C6r6+C4r4 */\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm7, %%mm6			\n\t" /*2A2		a2 */\
+	"paddd %%mm1, %%mm0			\n\t" /* A3		a3 */\
+\
+	"psrad $" #shift ", %%mm6		\n\t"\
+\
+	"packssdw %%mm6, %%mm4			\n\t" /* C2, c2, C2, c2 */\
+	"movq %%mm4, 32+" #dst "		\n\t" /* C2, c2 */\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	"paddd %%mm3, %%mm2			\n\t" /*2A3		a3 */\
+\
+	"movq %%mm4, 80+" #dst "		\n\t" /* C2, c2 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+\
+	"packssdw %%mm2, %%mm0			\n\t" /* C3, c3, C3, c3 */\
+	"movq %%mm0, 48+" #dst "		\n\t" /* C3, c3 */\
+	"movq %%mm0, 64+" #dst "		\n\t" /* C3, c3 */\
+
+//IDCT_CORE(  src0,   src4,   src1,    src5,    dst, rounder, shift)
+IDCT_CORE(    0(%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
+//IDCT_CORE(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
+IDCT_CORE(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
+//IDCT_CORE(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
+	"jmp 9f					\n\t"
+
+
+	"#.balign 16				\n\t"\
+	"1:					\n\t"
+#undef IDCT_CORE
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq " #src4 ", %%mm1			\n\t" /* R6	R4	r6	r4 */\
+	"movq " #src1 ", %%mm2			\n\t" /* R3	R1	r3	r1 */\
+	"movq 16(%2), %%mm4			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 24(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R6+C4R4	C6r6+C4r4 */\
+	"movq 32(%2), %%mm6			\n\t" /* C3	C1	C3	C1 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* C3R3+C1R1	C3r3+C1r1 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A0		a0 */\
+\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A0+B0		a0+b0 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A0		2a0 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A0-B0		a0-b0 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE0b(%%mm6, %%mm4, dst) \
+\
+	"movq 56(%2), %%mm4			\n\t" /* -C2	-C4	-C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* -C2R6-C4R4	-C2r6-C4r4 */\
+	"movq 64(%2), %%mm6			\n\t" /* -C7	C3	-C7	C3 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C7R3+C3R1	-C7r3+C3r1 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A1		a1 */\
+\
+	"movq 80(%2), %%mm5			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm5			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	WRITE1b(%%mm6, %%mm4, dst, %%mm7) \
+\
+	"movq 88(%2), %%mm4			\n\t" /* C2	-C4	C2	-C4 */\
+	"pmaddwd %%mm1, %%mm4			\n\t" /* C2R6-C4R4	C2r6-C4r4 */\
+	"movq 96(%2), %%mm6			\n\t" /* -C1	C5	-C1	C5 */\
+	"pmaddwd %%mm2, %%mm6			\n\t" /* -C1R3+C5R1	-C1r3+C5r1 */\
+	"paddd %%mm5, %%mm4			\n\t" /* A2		a2 */\
+\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"paddd %%mm4, %%mm6			\n\t" /* A1+B1		a1+b1 */\
+	"pmaddwd 120(%2), %%mm1			\n\t" /* -C6R6+C4R4	-C6r6+C4r4 */\
+	"paddd %%mm4, %%mm4			\n\t" /* 2A1		2a1 */\
+	"psubd %%mm6, %%mm4			\n\t" /* A1-B1		a1-b1 */\
+	"pmaddwd 128(%2), %%mm2			\n\t" /* -C5R3+C7R1	-C5r3+C7r1 */\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"psrad $" #shift ", %%mm4		\n\t"\
+\
+	"paddd %%mm1, %%mm0			\n\t" /* A3		a3 */\
+	"paddd %%mm0, %%mm2			\n\t" /* A3+B3		a3+b3 */\
+	"paddd %%mm0, %%mm0			\n\t" /* 2A3		2a3 */\
+	"psubd %%mm2, %%mm0			\n\t" /* A3-B3		a3-b3 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	WRITE2b(%%mm6, %%mm4, %%mm2, %%mm0, dst)
+
+//IDCT_CORE(  src0,   src4,   src1,    src5,    dst, rounder, shift)
+IDCT_CORE(    (%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
+IDCT_CORE(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
+IDCT_CORE(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
+IDCT_CORE(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
+	"jmp 9f					\n\t"
+
+
+	"#.balign 16				\n\t"
+	"7:					\n\t"
+#undef IDCT_CORE
+#define IDCT_CORE(src0, src4, src1, src5, dst, rounder, shift) \
+	"movq " #src0 ", %%mm0			\n\t" /* R2	R0	r2	r0 */\
+	"movq 16(%2), %%mm2			\n\t" /* C2	C4	C2	C4 */\
+	"movq 8+" #src0 ", %%mm1		\n\t" /* R2	R0	r2	r0 */\
+	"pmaddwd %%mm0, %%mm2			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+	"movq 16(%2), %%mm3			\n\t" /* C2	C4	C2	C4 */\
+	"pmaddwd %%mm1, %%mm3			\n\t" /* C2R2+C4R0	C2r2+C4r0 */\
+\
+	"movq 48(%2), %%mm4			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm0, %%mm4			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"movq 48(%2), %%mm5			\n\t" /* C6	C4	C6	C4 */\
+	"pmaddwd %%mm1, %%mm5			\n\t" /* C6R2+C4R0	C6r2+C4r0 */\
+	"movq 80(%2), %%mm6			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm0, %%mm6			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"movq 80(%2), %%mm7			\n\t" /* -C6	C4	-C6	C4 */\
+	"pmaddwd %%mm1, %%mm7			\n\t" /* -C6R2+C4R0	-C6r2+C4r0 */\
+	"pmaddwd 112(%2), %%mm0			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"psrad $" #shift ", %%mm2		\n\t"\
+	"psrad $" #shift ", %%mm3		\n\t"\
+	"pmaddwd 112(%2), %%mm1			\n\t" /* -C2R2+C4R0	-C2r2+C4r0 */\
+	"packssdw %%mm3, %%mm2			\n\t" /* C0, c0, C0, c0 */\
+	"movq %%mm2, " #dst "			\n\t" /* C0, c0 */\
+	"psrad $" #shift ", %%mm4		\n\t"\
+	"psrad $" #shift ", %%mm5		\n\t"\
+	"movq %%mm2, 112+" #dst "		\n\t" /* C0, c0 */\
+	"packssdw %%mm5, %%mm4			\n\t" /* C1, c1, C1, c1 */\
+	"movq %%mm4, 16+" #dst "		\n\t" /* C0, c0 */\
+	"psrad $" #shift ", %%mm7		\n\t"\
+	"psrad $" #shift ", %%mm6		\n\t"\
+	"movq %%mm4, 96+" #dst "		\n\t" /* C0, c0 */\
+	"packssdw %%mm7, %%mm6			\n\t" /* C2, c2, C2, c2 */\
+	"movq %%mm6, 32+" #dst "		\n\t" /* C0, c0 */\
+	"psrad $" #shift ", %%mm0		\n\t"\
+	"movq %%mm6, 80+" #dst "		\n\t" /* C0, c0 */\
+	"psrad $" #shift ", %%mm1		\n\t"\
+	"packssdw %%mm1, %%mm0			\n\t" /* C3, c3, C3, c3 */\
+	"movq %%mm0, 48+" #dst "		\n\t" /* C0, c0 */\
+	"movq %%mm0, 64+" #dst "		\n\t" /* C0, c0 */\
+
+//IDCT_CORE(  src0,   src4,   src1,    src5,    dst, rounder, shift)
+IDCT_CORE(   0(%1), 64(%1), 32(%1),  96(%1),  0(%0),/nop, 20)
+//IDCT_CORE(   8(%1), 72(%1), 40(%1), 104(%1),  4(%0),/nop, 20)
+IDCT_CORE(  16(%1), 80(%1), 48(%1), 112(%1),  8(%0),/nop, 20)
+//IDCT_CORE(  24(%1), 88(%1), 56(%1), 120(%1), 12(%0),/nop, 20)
+
+
+#endif
+
+/*
+Input
+ 00 20 02 22 40 60 42 62
+ 10 30 12 32 50 70 52 72
+ 01 21 03 23 41 61 43 63
+ 11 31 13 33 51 71 53 73
+ 04 24 06 26 44 64 46 66
+ 14 34 16 36 54 74 56 76
+...
+*/
+/*
+Temp
+ 00 02 10 12 20 22 30 32
+ 40 42 50 52 60 62 70 72
+ 01 03 11 13 21 23 31 33
+ 41 43 51 53 61 63 71 73
+ 04 06 14 16 24 26 34 36
+ 44 46 54 56 64 66 74 76
+ 05 07 15 17 25 27 35 37
+ 45 47 55 57 65 67 75 77
+*/
+
+/*
+Output
+ 00 10 20 30 40 50 60 70
+ 01 11 21 31 41 51 61 71
+...
+*/
+
+"9: \n\t"
+		:: "r" (block), "r" (temp), "r" (coeffs)
+		: "%eax"
+	);
+/*
+idctCol(block, temp);
+idctCol(block+1, temp+2);
+idctCol(block+2, temp+4);
+idctCol(block+3, temp+6);
+idctCol(block+4, temp+8);
+idctCol(block+5, temp+10);
+idctCol(block+6, temp+12);
+idctCol(block+7, temp+14);
+*/
+}
+
+void simple_idct_mmx(int16_t *block)
+{
+	static int imax=0, imin=0;
+	static int omax=0, omin=0;
+	int i, j;
+/*
+	for(i=0; i<64; i++)
+	{
+		if(block[i] > imax)
+		{
+			imax= block[i];
+			printf("Input-Max: %d\n", imax);
+			printf("Input-Min: %d\n", imin);
+			printf("Output-Max: %d\n", omax);
+			printf("Output-Min: %d\n", omin);
+		}
+		if(block[i] < imin)
+		{
+			imin= block[i];
+			printf("Input-Max: %d\n", imax);
+			printf("Input-Min: %d\n", imin);
+			printf("Output-Max: %d\n", omax);
+			printf("Output-Min: %d\n", omin);
+		}
+	}*/
+/*	static int stat[64];
+	for(j=0; j<4; j++)
+	{
+		static int line[8]={0,2,1,3,4,6,5,7};
+		for(i=0; i<16; i++)
+		{
+			if(block[j*16+i])
+			{
+				stat[j*16+1]++;
+				break;
+			}
+		}
+		for(i=0; i<16; i++)
+		{
+			if(block[j*16+i] && i!=0 && i!=2)
+			{
+				stat[j*16+2]++;
+				break;
+			}
+		}
+	}
+	stat[0]++;*/
+/*	for(i=1; i<8; i++)
+	{
+		if(block[i] != 0)
+		{
+			stat[1]++;
+			break;
+		}
+	}
+	for(i=32; i<64; i++)
+	{
+		if(block[i] != 0)
+		{
+			stat[2]++;
+			break;
+		}
+	}
+	stat[0]++;
+*/
+//	return;
+	idct(block);
+//	memset(block, 0, 128);
+/*
+	if(stat[0] > 100000)
+		for(i=0; i<64; i++)
+		{
+			if((i&7) == 0) printf("\n");
+			printf("%06d ", stat[i]);
+		}
+*/
+/*
+	for(i=0; i<4; i++) printf("%d", stat[1+i*16]);
+	printf("  ");
+	for(i=0; i<4; i++) printf("%d", stat[2+i*16]);
+	printf("\n");
+*/
+//	printf("%d", stat[2]);
+
+//	memset(stat, 0, 256);
+
+/*
+	for(i=0; i<64; i++)
+	{
+		if(block[i] > omax)
+		{
+			omax= block[i];
+			printf("Input-Max: %d\n", imax);
+			printf("Input-Min: %d\n", imin);
+			printf("Output-Max: %d\n", omax);
+			printf("Output-Min: %d\n", omin);
+		}
+		if(block[i] < omin)
+		{
+			omin= block[i];
+			printf("Input-Max: %d\n", imax);
+			printf("Input-Min: %d\n", imin);
+			printf("Output-Max: %d\n", omax);
+			printf("Output-Min: %d\n", omin);
+		}
+	}*/
+}
diff --git a/src/libffmpeg/libavcodec/mlib/Makefile.am b/src/libffmpeg/libavcodec/mlib/Makefile.am
new file mode 100644
index 000000000..92dc8cc30
--- /dev/null
+++ b/src/libffmpeg/libavcodec/mlib/Makefile.am
@@ -0,0 +1,39 @@
+##
+## Process this file with automake to produce Makefile.in
+##
+
+#CFLAGS = -D_FILE_OFFSET_BITS=64 @GLOBAL_CFLAGS@ -DCONFIG_DECODERS -DHAVE_AV_CONFIG_H
+
+CFLAGS = @GLOBAL_CFLAGS@ @LIBFFMPEG_CFLAGS@ -DCONFIG_DECODERS -DHAVE_AV_CONFIG_H
+ASFLAGS =
+
+LIBTOOL = $(SHELL) $(top_builddir)/libtool-nofpic
+
+noinst_LTLIBRARIES = libavcodec_mlib.la
+
+EXTRA_DIST = dsputil_mlib.c
+
+if HAVE_MLIB
+mlib_modules = $(EXTRA_DIST)
+endif
+
+libavcodec_mlib_la_SOURCES = $(mlib_modules)
+
+noinst_HEADERS = 
+
+.s.lo:
+	$(ASCOMPILE) -o $@ `test -f $< || echo '$(srcdir)/'`$<
+
+debug:
+	@$(MAKE) CFLAGS="@DEBUG_CFLAGS@ @LIBFFMPEG_CFLAGS@ -DCONFIG_DECODERS -DHAVE_AV_CONFIG_H"
+
+install-debug: debug
+	@$(MAKE) $(AM_MAKEFLAGS) install-exec-am install-data-am
+
+mostlyclean-generic:
+	-rm -f *~ \#* .*~ .\#*
+
+maintainer-clean-generic:
+	-@echo "This command is intended for maintainers to use;"
+	-@echo "it deletes files that may require special tools to rebuild."
+	-rm -f Makefile.in
diff --git a/src/libffmpeg/libavcodec/mlib/dsputil_mlib.c b/src/libffmpeg/libavcodec/mlib/dsputil_mlib.c
new file mode 100644
index 000000000..236c9206a
--- /dev/null
+++ b/src/libffmpeg/libavcodec/mlib/dsputil_mlib.c
@@ -0,0 +1,144 @@
+/*
+ * Sun mediaLib optimized DSP utils
+ * Copyright (c) 2001 Gerard Lantau.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include "../dsputil.h"
+
+#include <mlib_types.h>
+#include <mlib_status.h>
+#include <mlib_sys.h>
+#include <mlib_video.h>
+
+
+static void put_pixels_mlib (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    assert(height == 16 || height == 8);
+    if (height == 16)
+	mlib_VideoCopyRef_U8_U8_8x16(dest, (uint8_t *)ref, stride);
+    else
+	mlib_VideoCopyRef_U8_U8_8x8 (dest, (uint8_t *)ref, stride);
+}
+
+static void put_pixels_x2_mlib (uint8_t * dest, const uint8_t * ref,
+				int stride, int height)
+{
+    assert(height == 16 || height == 8);
+    if (height == 16)
+	mlib_VideoInterpX_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
+    else
+	mlib_VideoInterpX_U8_U8_8x8 (dest, (uint8_t *)ref, stride, stride);
+}
+
+static void put_pixels_y2_mlib (uint8_t * dest, const uint8_t * ref,
+				int stride, int height)
+{
+    assert(height == 16 || height == 8);
+    if (height == 16)
+	mlib_VideoInterpY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
+    else
+	mlib_VideoInterpY_U8_U8_8x8 (dest, (uint8_t *)ref, stride, stride);
+}
+
+static void put_pixels_xy2_mlib(uint8_t * dest, const uint8_t * ref,
+				int stride, int height)
+{
+    assert(height == 16 || height == 8);
+    if (height == 16) 
+	mlib_VideoInterpXY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
+    else
+	mlib_VideoInterpXY_U8_U8_8x8 (dest, (uint8_t *)ref, stride, stride);
+}
+
+static void avg_pixels_mlib (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    assert(height == 16 || height == 8);
+    if (height == 16)
+	mlib_VideoCopyRefAve_U8_U8_8x16(dest, (uint8_t *)ref, stride);
+    else
+	mlib_VideoCopyRefAve_U8_U8_8x8 (dest, (uint8_t *)ref, stride);
+}
+
+static void avg_pixels_x2_mlib (uint8_t * dest, const uint8_t * ref,
+				int stride, int height)
+{
+    assert(height == 16 || height == 8);
+    if (height == 16)
+	mlib_VideoInterpAveX_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
+    else
+	mlib_VideoInterpAveX_U8_U8_8x8 (dest, (uint8_t *)ref, stride, stride);
+}
+
+static void avg_pixels_y2_mlib (uint8_t * dest, const uint8_t * ref,
+				int stride, int height)
+{
+    assert(height == 16 || height == 8);
+    if (height == 16)
+	mlib_VideoInterpAveY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
+    else
+	mlib_VideoInterpAveY_U8_U8_8x8 (dest, (uint8_t *)ref, stride, stride);
+}
+
+static void avg_pixels_xy2_mlib (uint8_t * dest, const uint8_t * ref,
+				 int stride, int height)
+{
+    assert(height == 16 || height == 8);
+    if (height == 16)
+	mlib_VideoInterpAveXY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
+    else
+	mlib_VideoInterpAveXY_U8_U8_8x8 (dest, (uint8_t *)ref, stride, stride);
+}
+
+
+static void add_pixels_clamped_mlib(const DCTELEM *block, UINT8 *pixels, int line_size)
+{
+    mlib_VideoAddBlock_U8_S16(pixels, (mlib_s16 *)block, line_size);
+}
+
+
+void ff_idct_mlib(DCTELEM *data)
+{
+    mlib_VideoIDCT8x8_S16_S16 (data, data);
+}
+
+
+void ff_fdct_mlib(DCTELEM *data)
+{
+    mlib_VideoDCT8x8_S16_S16 (data, data);
+}
+
+void dsputil_init_mlib(void)
+{
+    av_fdct = ff_fdct_mlib;
+    ff_idct = ff_idct_mlib;
+
+    put_pixels_tab[0] = put_pixels_mlib;
+    put_pixels_tab[1] = put_pixels_x2_mlib;
+    put_pixels_tab[2] = put_pixels_y2_mlib;
+    put_pixels_tab[3] = put_pixels_xy2_mlib;
+
+    avg_pixels_tab[0] = avg_pixels_mlib;
+    avg_pixels_tab[1] = avg_pixels_x2_mlib;
+    avg_pixels_tab[2] = avg_pixels_y2_mlib;
+    avg_pixels_tab[3] = avg_pixels_xy2_mlib;
+    
+    put_no_rnd_pixels_tab[0] = put_pixels_mlib;
+    
+    add_pixels_clamped = add_pixels_clamped_mlib;
+}