Better ARM support for mplayer/ffmpeg, ported from atty fork

while playing with some new hardware, I found it's running a forked mplayer
 -- and it looks like they're following the GPL.

 The maintainer's page is here: http://atty.jp/?Zaurus/mplayer
 Unfortunately it's mostly in Japanese, so it's hard to figure out any
  details.

  Their code looks quite interesting (at least to those of us w/ ARM CPUs).

  The patches I've attached are the patches from atty.jp with a couple of
  modifications by myself:
  - ported to current CVS
  - reverted their change of removing SNOW support from ffmpeg
  - cleaned up their bswap mess
  - removed DOS-style linebreaks from various files

patch by (Bernhard Rosenkraenzer: bero, arklinux org)

Originally committed as revision 4311 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/Makefile

@@ -316,8 +316,11 @@ endif
 
 # armv4l specific stuff
 ifeq ($(TARGET_ARCH_ARMV4L),yes)
-ASM_OBJS += armv4l/jrevdct_arm.o armv4l/simple_idct_arm.o
+ASM_OBJS += armv4l/jrevdct_arm.o armv4l/simple_idct_arm.o armv4l/dsputil_arm_s.o
 OBJS += armv4l/dsputil_arm.o armv4l/mpegvideo_arm.o
+ifeq ($(TARGET_IWMMXT),yes)
+OBJS += armv4l/dsputil_iwmmxt.o armv4l/mpegvideo_iwmmxt.o
+endif
 endif
 
 # sun mediaLib specific stuff
@@ -327,6 +330,12 @@ OBJS += mlib/dsputil_mlib.o
 CFLAGS += $(MLIB_INC)
 endif
 
+# Intel IPP specific stuff
+# currently only works when libavcodec is used in mplayer
+ifeq ($(HAVE_IPP),yes)
+CFLAGS += $(IPP_INC)
+endif
+
 # alpha specific stuff
 ifeq ($(TARGET_ARCH_ALPHA),yes)
 OBJS += alpha/dsputil_alpha.o alpha/mpegvideo_alpha.o \

libavcodec/armv4l/dsputil_arm.c

@@ -18,6 +18,13 @@
  */
 
 #include "../dsputil.h"
+#ifdef HAVE_IPP
+#include "ipp.h"
+#endif
+
+#ifdef HAVE_IWMMXT
+extern void dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx);
+#endif
 
 extern void j_rev_dct_ARM(DCTELEM *data);
 extern void simple_idct_ARM(DCTELEM *data);
@@ -26,6 +33,146 @@ extern void simple_idct_ARM(DCTELEM *data);
 static void (*ff_put_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
 static void (*ff_add_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
 
+void put_pixels8_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+void put_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+void put_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+void put_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+
+void put_no_rnd_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+void put_no_rnd_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+void put_no_rnd_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+
+void put_pixels16_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
+static void put_pixels16_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+    put_pixels8_x2_arm(block, pixels, line_size, h);
+    put_pixels8_x2_arm(block + 8, pixels + 8, line_size, h);
+}
+
+static void put_pixels16_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+    put_pixels8_y2_arm(block, pixels, line_size, h);
+    put_pixels8_y2_arm(block + 8, pixels + 8, line_size, h);
+}
+
+static void put_pixels16_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+    put_pixels8_xy2_arm(block, pixels, line_size, h);
+    put_pixels8_xy2_arm(block + 8, pixels + 8, line_size, h);
+}
+
+static void put_no_rnd_pixels16_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+    put_no_rnd_pixels8_x2_arm(block, pixels, line_size, h);
+    put_no_rnd_pixels8_x2_arm(block + 8, pixels + 8, line_size, h);
+}
+
+static void put_no_rnd_pixels16_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+    put_no_rnd_pixels8_y2_arm(block, pixels, line_size, h);
+    put_no_rnd_pixels8_y2_arm(block + 8, pixels + 8, line_size, h);
+}
+
+static void put_no_rnd_pixels16_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+    put_no_rnd_pixels8_xy2_arm(block, pixels, line_size, h);
+    put_no_rnd_pixels8_xy2_arm(block + 8, pixels + 8, line_size, h);
+}
+
+static void add_pixels_clamped_ARM(short *block, unsigned char *dest, int line_size)
+{
+    asm volatile (
+                  "mov r10, #8 \n\t"
+
+                  "1: \n\t"
+
+                  /* load dest */
+                  "ldr r4, [%1] \n\t"
+                  /* block[0] and block[1]*/
+                  "ldrsh r5, [%0] \n\t"
+                  "ldrsh r7, [%0, #2] \n\t"
+                  "and r6, r4, #0xFF \n\t"
+                  "and r8, r4, #0xFF00 \n\t"
+                  "add r6, r5, r6 \n\t"
+                  "add r8, r7, r8, lsr #8 \n\t"
+                  "mvn r5, r5 \n\t"
+                  "mvn r7, r7 \n\t"
+                  "tst r6, #0x100 \n\t"
+                  "movne r6, r5, lsr #24 \n\t"
+                  "tst r8, #0x100 \n\t"
+                  "movne r8, r7, lsr #24 \n\t"
+                  "mov r9, r6 \n\t"
+                  "ldrsh r5, [%0, #4] \n\t" /* moved form [A] */
+                  "orr r9, r9, r8, lsl #8 \n\t"
+                  /* block[2] and block[3] */
+                  /* [A] */
+                  "ldrsh r7, [%0, #6] \n\t"
+                  "and r6, r4, #0xFF0000 \n\t"
+                  "and r8, r4, #0xFF000000 \n\t"
+                  "add r6, r5, r6, lsr #16 \n\t"
+                  "add r8, r7, r8, lsr #24 \n\t"
+                  "mvn r5, r5 \n\t"
+                  "mvn r7, r7 \n\t"
+                  "tst r6, #0x100 \n\t"
+                  "movne r6, r5, lsr #24 \n\t"
+                  "tst r8, #0x100 \n\t"
+                  "movne r8, r7, lsr #24 \n\t"
+                  "orr r9, r9, r6, lsl #16 \n\t"
+                  "ldr r4, [%1, #4] \n\t"       /* moved form [B] */
+                  "orr r9, r9, r8, lsl #24 \n\t"
+                  /* store dest */
+                  "ldrsh r5, [%0, #8] \n\t" /* moved form [C] */
+                  "str r9, [%1] \n\t"
+
+                  /* load dest */
+                  /* [B] */
+                  /* block[4] and block[5] */
+                  /* [C] */
+                  "ldrsh r7, [%0, #10] \n\t"
+                  "and r6, r4, #0xFF \n\t"
+                  "and r8, r4, #0xFF00 \n\t"
+                  "add r6, r5, r6 \n\t"
+                  "add r8, r7, r8, lsr #8 \n\t"
+                  "mvn r5, r5 \n\t"
+                  "mvn r7, r7 \n\t"
+                  "tst r6, #0x100 \n\t"
+                  "movne r6, r5, lsr #24 \n\t"
+                  "tst r8, #0x100 \n\t"
+                  "movne r8, r7, lsr #24 \n\t"
+                  "mov r9, r6 \n\t"
+                  "ldrsh r5, [%0, #12] \n\t" /* moved from [D] */
+                  "orr r9, r9, r8, lsl #8 \n\t"
+                  /* block[6] and block[7] */
+                  /* [D] */
+                  "ldrsh r7, [%0, #14] \n\t"
+                  "and r6, r4, #0xFF0000 \n\t"
+                  "and r8, r4, #0xFF000000 \n\t"
+                  "add r6, r5, r6, lsr #16 \n\t"
+                  "add r8, r7, r8, lsr #24 \n\t"
+                  "mvn r5, r5 \n\t"
+                  "mvn r7, r7 \n\t"
+                  "tst r6, #0x100 \n\t"
+                  "movne r6, r5, lsr #24 \n\t"
+                  "tst r8, #0x100 \n\t"
+                  "movne r8, r7, lsr #24 \n\t"
+                  "orr r9, r9, r6, lsl #16 \n\t"
+                  "add %0, %0, #16 \n\t" /* moved from [E] */
+                  "orr r9, r9, r8, lsl #24 \n\t"
+                  "subs r10, r10, #1 \n\t" /* moved from [F] */
+                  /* store dest */
+                  "str r9, [%1, #4] \n\t"
+
+                  /* [E] */
+                  /* [F] */
+                  "add %1, %1, %2 \n\t"
+                  "bne 1b \n\t"
+                  :
+                  : "r"(block),
+                    "r"(dest),
+                    "r"(line_size)
+                  : "r4", "r5", "r6", "r7", "r8", "r9", "r10", "cc", "memory" );
+}
+
 /* XXX: those functions should be suppressed ASAP when all IDCTs are
    converted */
 static void j_rev_dct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
@@ -48,6 +195,34 @@ static void simple_idct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
     simple_idct_ARM (block);
     ff_add_pixels_clamped(block, dest, line_size);
 }
+static void simple_idct_ipp(DCTELEM *block)
+{
+#ifdef HAVE_IPP
+    ippiDCT8x8Inv_Video_16s_C1I(block);
+#endif
+}
+static void simple_idct_ipp_put(uint8_t *dest, int line_size, DCTELEM *block)
+{
+#ifdef HAVE_IPP
+    ippiDCT8x8Inv_Video_16s8u_C1R(block, dest, line_size);
+#endif
+}
+
+#ifdef HAVE_IWMMXT
+void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size);
+#endif
+
+static void simple_idct_ipp_add(uint8_t *dest, int line_size, DCTELEM *block)
+{
+#ifdef HAVE_IPP
+    ippiDCT8x8Inv_Video_16s_C1I(block);
+#ifdef HAVE_IWMMXT
+    add_pixels_clamped_iwmmxt(block, dest, line_size);
+#else
+    add_pixels_clamped_ARM(block, dest, line_size);
+#endif
+#endif
+}
 
 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx)
 {
@@ -56,7 +231,11 @@ void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx)
     ff_put_pixels_clamped = c->put_pixels_clamped;
     ff_add_pixels_clamped = c->add_pixels_clamped;
 
+#ifdef HAVE_IPP
+    if(idct_algo==FF_IDCT_ARM){
+#else
     if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_ARM){
+#endif
         c->idct_put= j_rev_dct_ARM_put;
         c->idct_add= j_rev_dct_ARM_add;
 	c->idct    = j_rev_dct_ARM;
@@ -66,5 +245,37 @@ void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx)
 	c->idct_add= simple_idct_ARM_add;
 	c->idct    = simple_idct_ARM;
 	c->idct_permutation_type= FF_NO_IDCT_PERM;
+#ifdef HAVE_IPP
+    } else if (idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_IPP){
+#else
+    } else if (idct_algo==FF_IDCT_IPP){
+#endif
+        c->idct_put= simple_idct_ipp_put;
+        c->idct_add= simple_idct_ipp_add;
+        c->idct    = simple_idct_ipp;
+        c->idct_permutation_type= FF_NO_IDCT_PERM;
     }
+
+/*     c->put_pixels_tab[0][0] = put_pixels16_arm; */ // NG!
+    c->put_pixels_tab[0][1] = put_pixels16_x2_arm; //OK!
+    c->put_pixels_tab[0][2] = put_pixels16_y2_arm; //OK!
+/*     c->put_pixels_tab[0][3] = put_pixels16_xy2_arm; /\* NG *\/ */
+/*     c->put_no_rnd_pixels_tab[0][0] = put_pixels16_arm; // ?(»È¤ï¤ì¤Ê¤¤) */
+    c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_arm; // OK
+    c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_arm; //OK
+/*     c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_arm; //NG */
+    c->put_pixels_tab[1][0] = put_pixels8_arm; //OK
+    c->put_pixels_tab[1][1] = put_pixels8_x2_arm; //OK
+/*     c->put_pixels_tab[1][2] = put_pixels8_y2_arm; //NG */
+/*     c->put_pixels_tab[1][3] = put_pixels8_xy2_arm; //NG */
+    c->put_no_rnd_pixels_tab[1][0] = put_pixels8_arm;//OK
+    c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_arm; //OK
+    c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_arm; //OK
+/*     c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_arm;//NG */
+
+#if 1
+#ifdef HAVE_IWMMXT
+    dsputil_init_iwmmxt(c, avctx);
+#endif
+#endif
 }

libavcodec/armv4l/dsputil_arm_s.S

@@ -0,0 +1,694 @@
+@
+@ ARMv4L optimized DSP utils
+@ Copyright (c) 2004 AGAWA Koji <i (AT) atty (DOT) jp>
+@
+@ This library is free software; you can redistribute it and/or
+@ modify it under the terms of the GNU Lesser General Public
+@ License as published by the Free Software Foundation; either
+@ version 2 of the License, or (at your option) any later version.
+@
+@ This library is distributed in the hope that it will be useful,
+@ but WITHOUT ANY WARRANTY; without even the implied warranty of
+@ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+@ Lesser General Public License for more details.
+@
+@ You should have received a copy of the GNU Lesser General Public
+@ License along with this library; if not, write to the Free Software
+@ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+@
+
+.macro  ADJ_ALIGN_QUADWORD_D shift, Rd0, Rd1, Rd2, Rd3, Rn0, Rn1, Rn2, Rn3, Rn4
+        mov \Rd0, \Rn0, lsr #(\shift * 8)
+        mov \Rd1, \Rn1, lsr #(\shift * 8)
+        mov \Rd2, \Rn2, lsr #(\shift * 8)
+        mov \Rd3, \Rn3, lsr #(\shift * 8)
+        orr \Rd0, \Rd0, \Rn1, lsl #(32 - \shift * 8)
+        orr \Rd1, \Rd1, \Rn2, lsl #(32 - \shift * 8)
+        orr \Rd2, \Rd2, \Rn3, lsl #(32 - \shift * 8)
+        orr \Rd3, \Rd3, \Rn4, lsl #(32 - \shift * 8)
+.endm
+.macro  ADJ_ALIGN_DOUBLEWORD shift, R0, R1, R2
+        mov \R0, \R0, lsr #(\shift * 8)
+        orr \R0, \R0, \R1, lsl #(32 - \shift * 8)
+        mov \R1, \R1, lsr #(\shift * 8)
+        orr \R1, \R1, \R2, lsl #(32 - \shift * 8)
+.endm
+.macro  ADJ_ALIGN_DOUBLEWORD_D shift, Rdst0, Rdst1, Rsrc0, Rsrc1, Rsrc2
+        mov \Rdst0, \Rsrc0, lsr #(\shift * 8)
+        mov \Rdst1, \Rsrc1, lsr #(\shift * 8)
+        orr \Rdst0, \Rdst0, \Rsrc1, lsl #(32 - (\shift * 8))
+        orr \Rdst1, \Rdst1, \Rsrc2, lsl #(32 - (\shift * 8))
+.endm
+
+.macro  RND_AVG32 Rd0, Rd1, Rn0, Rn1, Rm0, Rm1, Rmask
+        @ Rd = (Rn | Rm) - (((Rn ^ Rm) & ~0x01010101) >> 1)
+        @ Rmask = 0xFEFEFEFE
+        @ Rn = destroy
+        eor \Rd0, \Rn0, \Rm0
+        eor \Rd1, \Rn1, \Rm1
+        orr \Rn0, \Rn0, \Rm0
+        orr \Rn1, \Rn1, \Rm1
+        and \Rd0, \Rd0, \Rmask
+        and \Rd1, \Rd1, \Rmask
+        sub \Rd0, \Rn0, \Rd0, lsr #1
+        sub \Rd1, \Rn1, \Rd1, lsr #1
+.endm
+
+.macro  NO_RND_AVG32 Rd0, Rd1, Rn0, Rn1, Rm0, Rm1, Rmask
+        @ Rd = (Rn & Rm) - (((Rn ^ Rm) & ~0x01010101) >> 1)
+        @ Rmask = 0xFEFEFEFE
+        @ Rn = destroy
+        eor \Rd0, \Rn0, \Rm0
+        eor \Rd1, \Rn1, \Rm1
+        and \Rn0, \Rn0, \Rm0
+        and \Rn1, \Rn1, \Rm1
+        and \Rd0, \Rd0, \Rmask
+        and \Rd1, \Rd1, \Rmask
+        add \Rd0, \Rn0, \Rd0, lsr #1
+        add \Rd1, \Rn1, \Rd1, lsr #1
+.endm
+
+@ ----------------------------------------------------------------
+        .align 8
+        .global put_pixels16_arm
+put_pixels16_arm:
+        @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+        @ block = word aligned, pixles = unaligned
+        pld [r1]
+        stmfd sp!, {r4-r11, lr} @ R14 is also called LR
+        adr r5, 5f
+        ands r4, r1, #3
+        bic r1, r1, #3
+        add r5, r5, r4, lsl #2
+        ldrne pc, [r5]
+1:
+        ldmia r1, {r4-r7}
+        add r1, r1, r2
+        stmia r0, {r4-r7}
+        pld [r1]
+        subs r3, r3, #1
+        add r0, r0, r2
+        bne 1b
+        ldmfd sp!, {r4-r11, pc}
+        .align 8
+2:
+        ldmia r1, {r4-r8}
+        add r1, r1, r2
+        ADJ_ALIGN_QUADWORD_D 1, r9, r10, r11, r12, r4, r5, r6, r7, r8
+        pld [r1]
+        subs r3, r3, #1
+        stmia r0, {r9-r12}
+        add r0, r0, r2
+        bne 2b
+        ldmfd sp!, {r4-r11, pc}
+        .align 8
+3:
+        ldmia r1, {r4-r8}
+        add r1, r1, r2
+        ADJ_ALIGN_QUADWORD_D 2, r9, r10, r11, r12, r4, r5, r6, r7, r8
+        pld [r1]
+        subs r3, r3, #1
+        stmia r0, {r9-r12}
+        add r0, r0, r2
+        bne 3b
+        ldmfd sp!, {r4-r11, pc}
+        .align 8
+4:
+        ldmia r1, {r4-r8}
+        add r1, r1, r2
+        ADJ_ALIGN_QUADWORD_D 3, r9, r10, r11, r12, r4, r5, r6, r7, r8
+        pld [r1]
+        subs r3, r3, #1
+        stmia r0, {r9-r12}
+        add r0, r0, r2
+        bne 4b
+        ldmfd sp!, {r4-r11,pc}
+        .align 8
+5:
+        .word 1b
+        .word 2b
+        .word 3b
+        .word 4b
+
+@ ----------------------------------------------------------------
+        .align 8
+        .global put_pixels8_arm
+put_pixels8_arm:
+        @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+        @ block = word aligned, pixles = unaligned
+        pld [r1]
+        stmfd sp!, {r4-r5,lr} @ R14 is also called LR
+        adr r5, 5f
+        ands r4, r1, #3
+        bic r1, r1, #3
+        add r5, r5, r4, lsl #2
+        ldrne pc, [r5]
+1:
+        ldmia r1, {r4-r5}
+        add r1, r1, r2
+        subs r3, r3, #1
+        pld [r1]
+        stmia r0, {r4-r5}
+        add r0, r0, r2
+        bne 1b
+        ldmfd sp!, {r4-r5,pc}
+        .align 8
+2:
+        ldmia r1, {r4-r5, r12}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r12
+        pld [r1]
+        subs r3, r3, #1
+        stmia r0, {r4-r5}
+        add r0, r0, r2
+        bne 2b
+        ldmfd sp!, {r4-r5,pc}
+        .align 8
+3:
+        ldmia r1, {r4-r5, r12}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r12
+        pld [r1]
+        subs r3, r3, #1
+        stmia r0, {r4-r5}
+        add r0, r0, r2
+        bne 3b
+        ldmfd sp!, {r4-r5,pc}
+        .align 8
+4:
+        ldmia r1, {r4-r5, r12}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r12
+        pld [r1]
+        subs r3, r3, #1
+        stmia r0, {r4-r5}
+        add r0, r0, r2
+        bne 4b
+        ldmfd sp!, {r4-r5,pc}
+        .align 8
+5:
+        .word 1b
+        .word 2b
+        .word 3b
+        .word 4b
+
+@ ----------------------------------------------------------------
+        .align 8
+        .global put_pixels8_x2_arm
+put_pixels8_x2_arm:
+        @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+        @ block = word aligned, pixles = unaligned
+        pld [r1]
+        stmfd sp!, {r4-r10,lr} @ R14 is also called LR
+        adr r5, 5f
+        ands r4, r1, #3
+        ldr r12, [r5]
+        add r5, r5, r4, lsl #2
+        bic r1, r1, #3
+        ldrne pc, [r5]
+1:
+        ldmia r1, {r4-r5, r10}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD_D 1, r6, r7, r4, r5, r10
+        pld [r1]
+        RND_AVG32 r8, r9, r4, r5, r6, r7, r12
+        subs r3, r3, #1
+        stmia r0, {r8-r9}
+        add r0, r0, r2
+        bne 1b
+        ldmfd sp!, {r4-r10,pc}
+        .align 8
+2:
+        ldmia r1, {r4-r5, r10}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD_D 1, r6, r7, r4, r5, r10
+        ADJ_ALIGN_DOUBLEWORD_D 2, r8, r9, r4, r5, r10
+        pld [r1]
+        RND_AVG32 r4, r5, r6, r7, r8, r9, r12
+        subs r3, r3, #1
+        stmia r0, {r4-r5}
+        add r0, r0, r2
+        bne 2b
+        ldmfd sp!, {r4-r10,pc}
+        .align 8
+3:
+        ldmia r1, {r4-r5, r10}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD_D 2, r6, r7, r4, r5, r10
+        ADJ_ALIGN_DOUBLEWORD_D 3, r8, r9, r4, r5, r10
+        pld [r1]
+        RND_AVG32 r4, r5, r6, r7, r8, r9, r12
+        subs r3, r3, #1
+        stmia r0, {r4-r5}
+        add r0, r0, r2
+        bne 3b
+        ldmfd sp!, {r4-r10,pc}
+        .align 8
+4:
+        ldmia r1, {r4-r5, r10}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD_D 3, r6, r7, r4, r5, r10
+        pld [r1]
+        RND_AVG32 r8, r9, r6, r7, r5, r10, r12
+        subs r3, r3, #1
+        stmia r0, {r8-r9}
+        add r0, r0, r2
+        bne 4b
+        ldmfd sp!, {r4-r10,pc} @@ update PC with LR content.
+        .align 8
+5:
+        .word 0xFEFEFEFE
+        .word 2b
+        .word 3b
+        .word 4b
+
+        .align 8
+        .global put_no_rnd_pixels8_x2_arm
+put_no_rnd_pixels8_x2_arm:
+        @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+        @ block = word aligned, pixles = unaligned
+        pld [r1]
+        stmfd sp!, {r4-r10,lr} @ R14 is also called LR
+        adr r5, 5f
+        ands r4, r1, #3
+        ldr r12, [r5]
+        add r5, r5, r4, lsl #2
+        bic r1, r1, #3
+        ldrne pc, [r5]
+1:
+        ldmia r1, {r4-r5, r10}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD_D 1, r6, r7, r4, r5, r10
+        pld [r1]
+        NO_RND_AVG32 r8, r9, r4, r5, r6, r7, r12
+        subs r3, r3, #1
+        stmia r0, {r8-r9}
+        add r0, r0, r2
+        bne 1b
+        ldmfd sp!, {r4-r10,pc}
+        .align 8
+2:
+        ldmia r1, {r4-r5, r10}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD_D 1, r6, r7, r4, r5, r10
+        ADJ_ALIGN_DOUBLEWORD_D 2, r8, r9, r4, r5, r10
+        pld [r1]
+        NO_RND_AVG32 r4, r5, r6, r7, r8, r9, r12
+        subs r3, r3, #1
+        stmia r0, {r4-r5}
+        add r0, r0, r2
+        bne 2b
+        ldmfd sp!, {r4-r10,pc}
+        .align 8
+3:
+        ldmia r1, {r4-r5, r10}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD_D 2, r6, r7, r4, r5, r10
+        ADJ_ALIGN_DOUBLEWORD_D 3, r8, r9, r4, r5, r10
+        pld [r1]
+        NO_RND_AVG32 r4, r5, r6, r7, r8, r9, r12
+        subs r3, r3, #1
+        stmia r0, {r4-r5}
+        add r0, r0, r2
+        bne 3b
+        ldmfd sp!, {r4-r10,pc}
+        .align 8
+4:
+        ldmia r1, {r4-r5, r10}
+        add r1, r1, r2
+        ADJ_ALIGN_DOUBLEWORD_D 3, r6, r7, r4, r5, r10
+        pld [r1]
+        NO_RND_AVG32 r8, r9, r6, r7, r5, r10, r12
+        subs r3, r3, #1
+        stmia r0, {r8-r9}
+        add r0, r0, r2
+        bne 4b
+        ldmfd sp!, {r4-r10,pc} @@ update PC with LR content.
+        .align 8
+5:
+        .word 0xFEFEFEFE
+        .word 2b
+        .word 3b
+        .word 4b
+
+
+@ ----------------------------------------------------------------
+        .align 8
+        .global put_pixels8_y2_arm
+put_pixels8_y2_arm:
+        @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+        @ block = word aligned, pixles = unaligned
+        pld [r1]
+        stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+        adr r5, 5f
+        ands r4, r1, #3
+        mov r3, r3, lsr #1
+        ldr r12, [r5]
+        add r5, r5, r4, lsl #2
+        bic r1, r1, #3
+        ldrne pc, [r5]
+1:
+        ldmia r1, {r4-r5}
+        add r1, r1, r2
+6:      ldmia r1, {r6-r7}
+        add r1, r1, r2
+        pld [r1]
+        RND_AVG32 r8, r9, r4, r5, r6, r7, r12
+        ldmia r1, {r4-r5}
+        add r1, r1, r2
+        stmia r0, {r8-r9}
+        add r0, r0, r2
+        pld [r1]
+        RND_AVG32 r8, r9, r6, r7, r4, r5, r12
+        subs r3, r3, #1
+        stmia r0, {r8-r9}
+        add r0, r0, r2
+        bne 6b
+        ldmfd sp!, {r4-r11,pc}
+        .align 8
+2:
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r6
+6:      ldmia r1, {r7-r9}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 1, r7, r8, r9
+        RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r6
+        subs r3, r3, #1
+        RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        bne 6b
+        ldmfd sp!, {r4-r11,pc}
+        .align 8
+3:
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r6
+6:      ldmia r1, {r7-r9}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 2, r7, r8, r9
+        RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r6
+        subs r3, r3, #1
+        RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        bne 6b
+        ldmfd sp!, {r4-r11,pc}
+        .align 8
+4:
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r6
+6:      ldmia r1, {r7-r9}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 3, r7, r8, r9
+        RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r6
+        subs r3, r3, #1
+        RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        bne 6b
+        ldmfd sp!, {r4-r11,pc}
+
+        .align 8
+5:
+        .word 0xFEFEFEFE
+        .word 2b
+        .word 3b
+        .word 4b
+
+        .align 8
+        .global put_no_rnd_pixels8_y2_arm
+put_no_rnd_pixels8_y2_arm:
+        @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+        @ block = word aligned, pixles = unaligned
+        pld [r1]
+        stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+        adr r5, 5f
+        ands r4, r1, #3
+        mov r3, r3, lsr #1
+        ldr r12, [r5]
+        add r5, r5, r4, lsl #2
+        bic r1, r1, #3
+        ldrne pc, [r5]
+1:
+        ldmia r1, {r4-r5}
+        add r1, r1, r2
+6:      ldmia r1, {r6-r7}
+        add r1, r1, r2
+        pld [r1]
+        NO_RND_AVG32 r8, r9, r4, r5, r6, r7, r12
+        ldmia r1, {r4-r5}
+        add r1, r1, r2
+        stmia r0, {r8-r9}
+        add r0, r0, r2
+        pld [r1]
+        NO_RND_AVG32 r8, r9, r6, r7, r4, r5, r12
+        subs r3, r3, #1
+        stmia r0, {r8-r9}
+        add r0, r0, r2
+        bne 6b
+        ldmfd sp!, {r4-r11,pc}
+        .align 8
+2:
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r6
+6:      ldmia r1, {r7-r9}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 1, r7, r8, r9
+        NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 1, r4, r5, r6
+        subs r3, r3, #1
+        NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        bne 6b
+        ldmfd sp!, {r4-r11,pc}
+        .align 8
+3:
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r6
+6:      ldmia r1, {r7-r9}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 2, r7, r8, r9
+        NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 2, r4, r5, r6
+        subs r3, r3, #1
+        NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        bne 6b
+        ldmfd sp!, {r4-r11,pc}
+        .align 8
+4:
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r6
+6:      ldmia r1, {r7-r9}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 3, r7, r8, r9
+        NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        ldmia r1, {r4-r6}
+        add r1, r1, r2
+        pld [r1]
+        ADJ_ALIGN_DOUBLEWORD 3, r4, r5, r6
+        subs r3, r3, #1
+        NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
+        stmia r0, {r10-r11}
+        add r0, r0, r2
+        bne 6b
+        ldmfd sp!, {r4-r11,pc}
+        .align 8
+5:
+        .word 0xFEFEFEFE
+        .word 2b
+        .word 3b
+        .word 4b
+
+@ ----------------------------------------------------------------
+.macro  RND_XY2_IT align, rnd
+        @ l1=  (a & 0x03030303) + (b & 0x03030303) ?(+ 0x02020202)
+        @ h1= ((a & 0xFCFCFCFCUL) >> 2) + ((b & 0xFCFCFCFCUL) >> 2)
+.if \align == 0
+        ldmia r1, {r6-r8}
+.elseif \align == 3
+        ldmia r1, {r5-r7}
+.else
+        ldmia r1, {r8-r10}
+.endif
+        add r1, r1, r2
+        pld [r1]
+.if \align == 0
+        ADJ_ALIGN_DOUBLEWORD_D 1, r4, r5, r6, r7, r8
+.elseif \align == 1
+        ADJ_ALIGN_DOUBLEWORD_D 1, r4, r5, r8, r9, r10
+        ADJ_ALIGN_DOUBLEWORD_D 2, r6, r7, r8, r9, r10
+.elseif \align == 2
+        ADJ_ALIGN_DOUBLEWORD_D 2, r4, r5, r8, r9, r10
+        ADJ_ALIGN_DOUBLEWORD_D 3, r6, r7, r8, r9, r10
+.elseif \align == 3
+        ADJ_ALIGN_DOUBLEWORD_D 3, r4, r5, r5, r6, r7
+.endif
+        ldr r14, [r12, #0]      @ 0x03030303
+        tst r3, #1
+        and r8, r4, r14
+        and r9, r5, r14
+        and r10, r6, r14
+        and r11, r7, r14
+.if \rnd == 1
+        ldreq r14, [r12, #16]   @ 0x02020202
+.else
+        ldreq r14, [r12, #28]   @ 0x01010101
+.endif
+        add r8, r8, r10
+        add r9, r9, r11
+        addeq r8, r8, r14
+        addeq r9, r9, r14
+        ldr r14, [r12, #20]     @ 0xFCFCFCFC >> 2
+        and r4, r14, r4, lsr #2
+        and r5, r14, r5, lsr #2
+        and r6, r14, r6, lsr #2
+        and r7, r14, r7, lsr #2
+        add r10, r4, r6
+        add r11, r5, r7
+.endm
+
+.macro RND_XY2_EXPAND align, rnd
+        RND_XY2_IT \align, \rnd
+6:      stmfd sp!, {r8-r11}
+        RND_XY2_IT \align, \rnd
+        ldmfd sp!, {r4-r7}
+        add r4, r4, r8
+        add r5, r5, r9
+        add r6, r6, r10
+        add r7, r7, r11
+        ldr r14, [r12, #24]     @ 0x0F0F0F0F
+        and r4, r14, r4, lsr #2
+        and r5, r14, r5, lsr #2
+        add r4, r4, r6
+        add r5, r5, r7
+        subs r3, r3, #1
+        stmia r0, {r4-r5}
+        add r0, r0, r2
+        bne 6b
+        ldmfd sp!, {r4-r11,pc}
+.endm
+
+        .align 8
+        .global put_pixels8_xy2_arm
+put_pixels8_xy2_arm:
+        @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+        @ block = word aligned, pixles = unaligned
+        pld [r1]
+        stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+        adrl r12, 5f
+        ands r4, r1, #3
+        add r5, r12, r4, lsl #2
+        bic r1, r1, #3
+        ldrne pc, [r5]
+1:
+        RND_XY2_EXPAND 0, 1
+
+        .align 8
+2:
+        RND_XY2_EXPAND 1, 1
+        
+        .align 8
+3:
+        RND_XY2_EXPAND 2, 1
+        
+        .align 8
+4:
+        RND_XY2_EXPAND 3, 1
+        
+5:
+        .word 0x03030303
+        .word 2b
+        .word 3b
+        .word 4b
+        .word 0x02020202
+        .word 0xFCFCFCFC >> 2
+        .word 0x0F0F0F0F
+        .word 0x01010101
+
+        .align 8
+        .global put_no_rnd_pixels8_xy2_arm
+put_no_rnd_pixels8_xy2_arm:
+        @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+        @ block = word aligned, pixles = unaligned
+        pld [r1]
+        stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+        adrl r12, 5f
+        ands r4, r1, #3
+        add r5, r12, r4, lsl #2
+        bic r1, r1, #3
+        ldrne pc, [r5]
+1:
+        RND_XY2_EXPAND 0, 0
+
+        .align 8
+2:
+        RND_XY2_EXPAND 1, 0
+        
+        .align 8
+3:
+        RND_XY2_EXPAND 2, 0
+        
+        .align 8
+4:
+        RND_XY2_EXPAND 3, 0
+        
+5:
+        .word 0x03030303
+        .word 2b
+        .word 3b
+        .word 4b
+        .word 0x02020202
+        .word 0xFCFCFCFC >> 2
+        .word 0x0F0F0F0F
+        .word 0x01010101

libavcodec/armv4l/dsputil_iwmmxt.c

@@ -0,0 +1,168 @@
+/*
+ * iWMMXt optimized DSP utils
+ * Copyright (c) 2004 AGAWA Koji
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "../dsputil.h"
+
+#define DEF(x, y) x ## _no_rnd_ ## y ##_iwmmxt
+#define SET_RND(regd)  __asm__ __volatile__ ("mov r12, #1 \n\t tbcsth " #regd ", r12":::"r12");
+#define WAVG2B "wavg2b"
+#include "dsputil_iwmmxt_rnd.h"
+#undef DEF
+#undef SET_RND
+#undef WAVG2B
+
+#define DEF(x, y) x ## _ ## y ##_iwmmxt
+#define SET_RND(regd)  __asm__ __volatile__ ("mov r12, #2 \n\t tbcsth " #regd ", r12":::"r12");
+#define WAVG2B "wavg2br"
+#include "dsputil_iwmmxt_rnd.h"
+#undef DEF
+#undef SET_RND
+#undef WAVG2BR
+
+// need scheduling
+#define OP(AVG)                                         \
+    asm volatile (                                      \
+        /* alignment */                                 \
+        "and r12, %[pixels], #7 \n\t"                   \
+        "bic %[pixels], %[pixels], #7 \n\t"             \
+        "tmcr wcgr1, r12 \n\t"                          \
+                                                        \
+        "wldrd wr0, [%[pixels]] \n\t"                   \
+        "wldrd wr1, [%[pixels], #8] \n\t"               \
+        "add %[pixels], %[pixels], %[line_size] \n\t"   \
+        "walignr1 wr4, wr0, wr1 \n\t"                   \
+                                                        \
+        "1: \n\t"                                       \
+                                                        \
+        "wldrd wr2, [%[pixels]] \n\t"                   \
+        "wldrd wr3, [%[pixels], #8] \n\t"               \
+        "add %[pixels], %[pixels], %[line_size] \n\t"   \
+        "pld [%[pixels]] \n\t"                          \
+        "walignr1 wr5, wr2, wr3 \n\t"                   \
+        AVG " wr6, wr4, wr5 \n\t"                       \
+        "wstrd wr6, [%[block]] \n\t"                    \
+        "add %[block], %[block], %[line_size] \n\t"     \
+                                                        \
+        "wldrd wr0, [%[pixels]] \n\t"                   \
+        "wldrd wr1, [%[pixels], #8] \n\t"               \
+        "add %[pixels], %[pixels], %[line_size] \n\t"   \
+        "walignr1 wr4, wr0, wr1 \n\t"                   \
+        "pld [%[pixels]] \n\t"                          \
+        AVG " wr6, wr4, wr5 \n\t"                       \
+        "wstrd wr6, [%[block]] \n\t"                    \
+        "add %[block], %[block], %[line_size] \n\t"     \
+                                                        \
+        "subs %[h], %[h], #2 \n\t"                      \
+        "bne 1b \n\t"                                   \
+        : [block]"+r"(block), [pixels]"+r"(pixels), [h]"+r"(h)  \
+        : [line_size]"r"(line_size) \
+        : "memory", "r12");
+void put_pixels8_y2_iwmmxt(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+    OP("wavg2br");
+}
+void put_no_rnd_pixels8_y2_iwmmxt(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
+{
+    OP("wavg2b");
+}
+#undef OP
+
+void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size)
+{
+    uint8_t *pixels2 = pixels + line_size;
+
+    __asm__ __volatile__ (
+        "mov            r12, #4                 \n\t"
+        "1:                                     \n\t"
+        "pld            [%[pixels], %[line_size2]]              \n\t"
+        "pld            [%[pixels2], %[line_size2]]             \n\t"
+        "wldrd          wr4, [%[pixels]]        \n\t"
+        "wldrd          wr5, [%[pixels2]]       \n\t"
+        "pld            [%[block], #32]         \n\t"
+        "wunpckelub     wr6, wr4                \n\t"
+        "wldrd          wr0, [%[block]]         \n\t"
+        "wunpckehub     wr7, wr4                \n\t"
+        "wldrd          wr1, [%[block], #8]     \n\t"
+        "wunpckelub     wr8, wr5                \n\t"
+        "wldrd          wr2, [%[block], #16]    \n\t"
+        "wunpckehub     wr9, wr5                \n\t"
+        "wldrd          wr3, [%[block], #24]    \n\t"
+        "add            %[block], %[block], #32 \n\t"
+        "waddhss        wr10, wr0, wr6          \n\t"
+        "waddhss        wr11, wr1, wr7          \n\t"
+        "waddhss        wr12, wr2, wr8          \n\t"
+        "waddhss        wr13, wr3, wr9          \n\t"
+        "wpackhus       wr14, wr10, wr11        \n\t"
+        "wpackhus       wr15, wr12, wr13        \n\t"
+        "wstrd          wr14, [%[pixels]]       \n\t"
+        "add            %[pixels], %[pixels], %[line_size2]     \n\t"
+        "subs           r12, r12, #1            \n\t"
+        "wstrd          wr15, [%[pixels2]]      \n\t"
+        "add            %[pixels2], %[pixels2], %[line_size2]   \n\t"
+        "bne            1b                      \n\t"
+        : [block]"+r"(block), [pixels]"+r"(pixels), [pixels2]"+r"(pixels2)
+        : [line_size2]"r"(line_size << 1)
+        : "cc", "memory", "r12");
+}
+
+static void nop(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+{
+    return;
+}
+
+void dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx)
+{
+    c->add_pixels_clamped = add_pixels_clamped_iwmmxt;
+
+    c->put_pixels_tab[0][0] = put_pixels16_iwmmxt;
+    c->put_pixels_tab[0][1] = put_pixels16_x2_iwmmxt;
+    c->put_pixels_tab[0][2] = put_pixels16_y2_iwmmxt;
+    c->put_pixels_tab[0][3] = put_pixels16_xy2_iwmmxt;
+    c->put_no_rnd_pixels_tab[0][0] = put_pixels16_iwmmxt;
+    c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_iwmmxt;
+    c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_iwmmxt;
+    c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_iwmmxt;
+
+    c->put_pixels_tab[1][0] = put_pixels8_iwmmxt;
+    c->put_pixels_tab[1][1] = put_pixels8_x2_iwmmxt;
+    c->put_pixels_tab[1][2] = put_pixels8_y2_iwmmxt;
+    c->put_pixels_tab[1][3] = put_pixels8_xy2_iwmmxt;
+    c->put_no_rnd_pixels_tab[1][0] = put_pixels8_iwmmxt;
+    c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_iwmmxt;
+    c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_iwmmxt;
+    c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_iwmmxt;
+
+    c->avg_pixels_tab[0][0] = avg_pixels16_iwmmxt;
+    c->avg_pixels_tab[0][1] = avg_pixels16_x2_iwmmxt;
+    c->avg_pixels_tab[0][2] = avg_pixels16_y2_iwmmxt;
+    c->avg_pixels_tab[0][3] = avg_pixels16_xy2_iwmmxt;
+    c->avg_no_rnd_pixels_tab[0][0] = avg_pixels16_iwmmxt;
+    c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_iwmmxt;
+    c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_iwmmxt;
+    c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_iwmmxt;
+
+    c->avg_pixels_tab[1][0] = avg_pixels8_iwmmxt;
+    c->avg_pixels_tab[1][1] = avg_pixels8_x2_iwmmxt;
+    c->avg_pixels_tab[1][2] = avg_pixels8_y2_iwmmxt;
+    c->avg_pixels_tab[1][3] = avg_pixels8_xy2_iwmmxt;
+    c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_iwmmxt;
+    c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_iwmmxt;
+    c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_iwmmxt;
+    c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_iwmmxt;
+}

libavcodec/armv4l/mpegvideo_arm.c

@@ -21,6 +21,13 @@
 #include "../mpegvideo.h"
 #include "../avcodec.h"
 
+#ifdef HAVE_IWMMXT
+extern void MPV_common_init_iwmmxt(MpegEncContext *s);
+#endif
+
 void MPV_common_init_armv4l(MpegEncContext *s)
 {
+#ifdef HAVE_IWMMXT
+    MPV_common_init_iwmmxt(s);
+#endif
 }

libavcodec/armv4l/mpegvideo_iwmmxt.c

@@ -0,0 +1,97 @@
+#include "../dsputil.h"
+#include "../mpegvideo.h"
+#include "../avcodec.h"
+
+static void dct_unquantize_h263_intra_iwmmxt(MpegEncContext *s,
+                                             DCTELEM *block, int n, int qscale)
+{
+    int level, qmul, qadd;
+    int nCoeffs;
+    DCTELEM *block_orig = block;
+
+    assert(s->block_last_index[n]>=0);
+
+    qmul = qscale << 1;
+
+    if (!s->h263_aic) {
+        if (n < 4)
+            level = block[0] * s->y_dc_scale;
+        else
+            level = block[0] * s->c_dc_scale;
+        qadd = (qscale - 1) | 1;
+    }else{
+        qadd = 0;
+        level = block[0];
+    }
+    if(s->ac_pred)
+        nCoeffs=63;
+    else
+        nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+
+    __asm__ __volatile__ (
+/*      "movd %1, %%mm6                 \n\t" //qmul */
+/*      "packssdw %%mm6, %%mm6          \n\t" */
+/*      "packssdw %%mm6, %%mm6          \n\t" */
+        "tbcsth wr6, %[qmul]            \n\t"
+/*      "movd %2, %%mm5                 \n\t" //qadd */
+/*      "packssdw %%mm5, %%mm5          \n\t" */
+/*      "packssdw %%mm5, %%mm5          \n\t" */
+        "tbcsth wr5, %[qadd]            \n\t"
+        "wzero wr7                      \n\t" /* "pxor %%mm7, %%mm7             \n\t" */
+        "wzero wr4                      \n\t" /* "pxor %%mm4, %%mm4             \n\t" */
+        "wsubh wr7, wr5, wr7            \n\t" /* "psubw %%mm5, %%mm7            \n\t" */
+        "1:                             \n\t"
+        "wldrd wr2, [%[block]]          \n\t" /* "movq (%0, %3), %%mm0          \n\t" */
+        "wldrd wr3, [%[block], #8]      \n\t" /* "movq 8(%0, %3), %%mm1         \n\t" */
+        "wmulsl wr0, wr6, wr2           \n\t" /* "pmullw %%mm6, %%mm0           \n\t" */
+        "wmulsl wr1, wr6, wr3           \n\t" /* "pmullw %%mm6, %%mm1           \n\t" */
+/*      "movq (%0, %3), %%mm2           \n\t" */
+/*      "movq 8(%0, %3), %%mm3          \n\t" */
+        "wcmpgtsh wr2, wr4, wr2         \n\t" /* "pcmpgtw %%mm4, %%mm2          \n\t" // block[i] < 0 ? -1 : 0 */
+        "wcmpgtsh wr3, wr4, wr2         \n\t" /* "pcmpgtw %%mm4, %%mm3          \n\t" // block[i] < 0 ? -1 : 0 */
+        "wxor wr0, wr2, wr0             \n\t" /* "pxor %%mm2, %%mm0             \n\t" */
+        "wxor wr1, wr3, wr1             \n\t" /* "pxor %%mm3, %%mm1             \n\t" */
+        "waddh wr0, wr7, wr0            \n\t" /* "paddw %%mm7, %%mm0            \n\t" */
+        "waddh wr1, wr7, wr1            \n\t" /* "paddw %%mm7, %%mm1            \n\t" */
+        "wxor wr2, wr0, wr2             \n\t" /* "pxor %%mm0, %%mm2             \n\t" */
+        "wxor wr3, wr1, wr3             \n\t" /* "pxor %%mm1, %%mm3             \n\t" */
+        "wcmpeqh wr0, wr7, wr0          \n\t" /* "pcmpeqw %%mm7, %%mm0          \n\t" // block[i] == 0 ? -1 : 0 */
+        "wcmpeqh wr1, wr7, wr1          \n\t" /* "pcmpeqw %%mm7, %%mm1          \n\t" // block[i] == 0 ? -1 : 0 */
+        "wandn wr0, wr2, wr0            \n\t" /* "pandn %%mm2, %%mm0            \n\t" */
+        "wandn wr1, wr3, wr1            \n\t" /* "pandn %%mm3, %%mm1            \n\t" */
+        "wstrd wr0, [%[block]]          \n\t" /* "movq %%mm0, (%0, %3)          \n\t" */
+        "wstrd wr1, [%[block], #8]      \n\t" /* "movq %%mm1, 8(%0, %3)         \n\t" */
+        "add %[block], %[block], #16    \n\t" /* "addl $16, %3                  \n\t" */
+        "subs %[i], %[i], #1            \n\t"
+        "bne 1b                         \n\t" /* "jng 1b                                \n\t" */
+        :[block]"+r"(block)
+        :[i]"r"((nCoeffs + 8) / 8), [qmul]"r"(qmul), [qadd]"r"(qadd)
+        :"memory");
+
+    block_orig[0] = level;
+}
+
+#if 0
+static void dct_unquantize_h263_inter_iwmmxt(MpegEncContext *s,
+                                             DCTELEM *block, int n, int qscale)
+{
+    int nCoeffs;
+
+    assert(s->block_last_index[n]>=0);
+
+    if(s->ac_pred)
+        nCoeffs=63;
+    else
+        nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+
+    ippiQuantInvInter_Compact_H263_16s_I(block, nCoeffs+1, qscale);
+}
+#endif
+
+void MPV_common_init_iwmmxt(MpegEncContext *s)
+{
+    s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_iwmmxt;
+#if 0
+    s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_iwmmxt;
+#endif
+}

libavcodec/avcodec.h

@@ -1180,6 +1180,7 @@ typedef struct AVCodecContext {
 #define FF_IDCT_SIMPLEARM    10
 #define FF_IDCT_H264         11
 #define FF_IDCT_VP3          12
+#define FP_IDCT_IPP          13
 
     /**
      * slice count.

libavcodec/bswap.h

@@ -94,10 +94,23 @@ static always_inline uint16_t bswap_16(uint16_t x){
     return (x>>8) | (x<<8);
 }
 
+#ifdef ARCH_ARM
+static always_inline uint32_t bswap_32(uint32_t x){
+    uint32_t t;
+    __asm__ (
+      "eor %1, %0, %0, ror #16 \n\t"
+      "bic %1, %1, #0xFF0000   \n\t"
+      "mov %0, %0, ror #8      \n\t"
+      "eor %0, %0, %1, lsr #8  \n\t"
+      : "+r"(x), "+r"(t));
+    return x;
+}
+#else
 static always_inline uint32_t bswap_32(uint32_t x){
     x= ((x<<8)&0xFF00FF00) | ((x>>8)&0x00FF00FF);
     return (x>>16) | (x<<16);
 }
+#endif
 
 static inline uint64_t bswap_64(uint64_t x)
 {