[MPlayer-dev-eng] [PATCH 5/9] libavcodec: add AVR32 specific optimizations

Hans-Christian Egtvedt hans-christian.egtvedt at atmel.com
Mon Feb 16 17:16:54 CET 2009


Implemented by Ronny Pedersen.

Signed-off-by: Hans-Christian Egtvedt <hans-christian.egtvedt at atmel.com>
---
 cfg-common-opts.h                |    5 +-
 cfg-common.h                     |    4 +
 libavcodec/Makefile              |    6 +-
 libavcodec/avr32/dsputil_avr32.c | 2492 ++++++++++++++++++++++++++++++++++++++
 libavcodec/avr32/fdct.S          |  538 ++++++++
 libavcodec/avr32/h264idct.S      |  451 +++++++
 libavcodec/avr32/idct.S          |  793 ++++++++++++
 libavcodec/avr32/mc.S            |  434 +++++++
 libavcodec/avr32/pico.h          |  255 ++++
 libavcodec/bitstream.h           |   76 ++-
 libavcodec/dsputil.c             |    1 +
 libavcodec/h264.c                |   15 +
 12 files changed, 5065 insertions(+), 5 deletions(-)
 create mode 100644 libavcodec/avr32/dsputil_avr32.c
 create mode 100644 libavcodec/avr32/fdct.S
 create mode 100644 libavcodec/avr32/h264idct.S
 create mode 100644 libavcodec/avr32/idct.S
 create mode 100644 libavcodec/avr32/mc.S
 create mode 100644 libavcodec/avr32/pico.h

diff --git a/cfg-common-opts.h b/cfg-common-opts.h
index 13d3f38..a7ec2f3 100644
--- a/cfg-common-opts.h
+++ b/cfg-common-opts.h
@@ -250,7 +250,10 @@
 	{"tsprobe", &ts_probe, CONF_TYPE_POSITION, 0, 0, TS_MAX_PROBE_SIZE, NULL},
 	{"psprobe", &ps_probe, CONF_TYPE_POSITION, 0, 0, TS_MAX_PROBE_SIZE, NULL},
 	{"tskeepbroken", &ts_keep_broken, CONF_TYPE_FLAG, 0, 0, 1, NULL},
-
+#ifdef ARCH_AVR32
+        {"use-pico", &avr32_use_pico, CONF_TYPE_FLAG, 0, 0, 1, NULL},
+        {"nouse-pico", &avr32_use_pico, CONF_TYPE_FLAG, 0, 1, 0, NULL},
+#endif
 	// draw by slices or whole frame (useful with libmpeg2/libavcodec)
 	{"slices", &vd_use_slices, CONF_TYPE_FLAG, 0, 0, 1, NULL},
 	{"noslices", &vd_use_slices, CONF_TYPE_FLAG, 0, 1, 0, NULL},
diff --git a/cfg-common.h b/cfg-common.h
index ea5fce6..6330979 100644
--- a/cfg-common.h
+++ b/cfg-common.h
@@ -6,6 +6,10 @@
 #include "m_config.h"
 #include "m_option.h"
 
+#ifdef ARCH_AVR32
+extern int avr32_use_pico;
+#endif
+
 extern char *mp_msg_charset;
 extern int mp_msg_color;
 extern int mp_msg_module;
diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 7026ada..40cbc66 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -462,6 +462,10 @@ OBJS-$(HAVE_ARMV6)                     += arm/simple_idct_armv6.o       \
 OBJS-$(HAVE_ARMVFP)                    += arm/dsputil_vfp.o             \
                                           arm/float_arm_vfp.o           \
 
+OBJS-$(ARCH_AVR32)                     += avr32/dsputil_avr32.o         \
+                                          avr32/idct.o avr32/fdct.o     \
+                                          avr32/mc.o avr32/h264idct.o   \
+
 OBJS-$(HAVE_IWMMXT)                    += arm/dsputil_iwmmxt.o          \
                                           arm/mpegvideo_iwmmxt.o        \
 
@@ -517,7 +521,7 @@ TESTS-$(CONFIG_OLDSCALER) += imgresample-test$(EXESUF)
 TESTS-$(ARCH_X86) += x86/cpuid-test$(EXESUF) motion-test$(EXESUF)
 
 CLEANFILES = apiexample$(EXESUF)
-DIRS = alpha arm bfin mlib ppc ps2 sh4 sparc x86
+DIRS = alpha arm avr32 bfin mlib ppc ps2 sh4 sparc x86
 
 include $(SUBDIR)../subdir.mak
 
diff --git a/libavcodec/avr32/dsputil_avr32.c b/libavcodec/avr32/dsputil_avr32.c
new file mode 100644
index 0000000..2bc0d27
--- /dev/null
+++ b/libavcodec/avr32/dsputil_avr32.c
@@ -0,0 +1,2492 @@
+/*
+ * Copyright (c) 2007 Atmel Corporation. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials provided
+ * with the distribution.
+ *
+ * 3. The name of ATMEL may not be used to endorse or promote products
+ * derived from this software without specific prior written
+ * permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ATMEL ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE EXPRESSLY AND SPECIFICALLY DISCLAIMED. IN NO EVENT SHALL ATMEL
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
+ * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
+ * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+#include "../dsputil.h"
+#include "pico.h"
+
+int avr32_use_pico = 1;
+
+#ifdef CHECK_DSP_FUNCS_AGAINST_C
+#define DSP_FUNC_NAME(name) test_ ## name
+#else
+#define DSP_FUNC_NAME(name) name
+#endif
+
+union doubleword {
+  int64_t doubleword;
+  struct {
+    int32_t top;
+    int32_t bottom;
+  } words;
+};
+
+#undef  LD16
+#undef  LD32
+#undef  LD64
+
+#define LD16(a) (*((uint16_t*)(a)))
+#define LD32(a) (*((uint32_t*)(a)))
+#define LD64(a) (*((uint64_t*)(a)))
+#define LD64_UNALIGNED(a) \
+  ({ union doubleword __tmp__; \
+   __tmp__.words.top = LD32(a); \
+   __tmp__.words.bottom = LD32(a + 4); \
+   __tmp__.doubleword; })
+
+#undef  ST32
+#undef  ST16
+
+#define ST16(a, b) *((uint16_t*)(a)) = (b)
+#define ST32(a, b) *((uint32_t*)(a)) = (b)
+
+#undef rnd_avg32
+#define rnd_avg32(a, b) \
+  ({ uint32_t __tmp__;\
+     __asm__("pavg.ub\t%0, %1, %2" : "=r"(__tmp__) : "r"(a), "r"(b));\
+     __tmp__;})
+
+void idct_avr32(DCTELEM *data);
+void fdct_avr32(DCTELEM *data);
+
+void idct_put_avr32(uint8_t *dest, int line_size, DCTELEM *data);
+void idct_add_avr32(uint8_t *dest, int line_size, DCTELEM *data);
+
+void h264_idct_add_avr32(uint8_t *dest, DCTELEM *data, int stride);
+void h264_idct8_add_avr32(uint8_t *dest, DCTELEM *data, int stride);
+
+#define extern_dspfunc(PFX, NUM) \
+    void PFX ## _pixels ## NUM ## _avr32(uint8_t *dst, const uint8_t *pixels, int line_size, int h );     \
+    void PFX ## _pixels ## NUM ## _h_avr32(uint8_t *dst, const uint8_t *pixels, int line_size, int h );  \
+    void PFX ## _pixels ## NUM ## _v_avr32(uint8_t *dst, const uint8_t *pixels, int line_size, int h );  \
+    void PFX ## _pixels ## NUM ## _hv_avr32(uint8_t *dst, const uint8_t *pixels, int line_size, int h )
+
+extern_dspfunc(put, 8);
+extern_dspfunc(put_no_rnd, 8);
+extern_dspfunc(avg, 8);
+extern_dspfunc(avg_no_rnd, 8);
+#undef extern_dspfunc
+
+#ifdef CHECK_DSP_FUNCS_AGAINST_C
+#define extern_dspfunc(PFX, NUM)                                        \
+  void PFX ## _pixels ## NUM ## _c(uint8_t *dst, const uint8_t *pixels, int line_size, int h ); \
+  void PFX ## _pixels ## NUM ## _x2_c(uint8_t *dst, const uint8_t *pixels, int line_size, int h ); \
+  void PFX ## _pixels ## NUM ## _y2_c(uint8_t *dst, const uint8_t *pixels, int line_size, int h ); \
+  void PFX ## _pixels ## NUM ## _xy2_c(uint8_t *dst, const uint8_t *pixels, int line_size, int h )
+
+extern_dspfunc(put, 4);
+extern_dspfunc(put_no_rnd, 4);
+extern_dspfunc(put, 8);
+extern_dspfunc(put_no_rnd, 8);
+extern_dspfunc(put, 16);
+extern_dspfunc(put_no_rnd, 16);
+extern_dspfunc(avg, 8);
+extern_dspfunc(avg_no_rnd, 8);
+extern_dspfunc(avg, 16);
+extern_dspfunc(avg_no_rnd, 16);
+
+
+#undef extern_dspfunc
+#define extern_dspfunc(PFX, NUM) \
+void PFX ## NUM ## _mc00_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc10_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc20_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc30_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc01_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc11_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc21_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc31_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc02_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc12_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc22_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc32_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc03_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc13_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc23_c(uint8_t *dst, uint8_t *src, int stride);  \
+void PFX ## NUM ## _mc33_c(uint8_t *dst, uint8_t *src, int stride);  \
+
+extern_dspfunc(put_h264_qpel,  16);
+extern_dspfunc(put_h264_qpel,  8);
+extern_dspfunc(put_h264_qpel,  4);
+extern_dspfunc(avg_h264_qpel,  16);
+extern_dspfunc(avg_h264_qpel,  8);
+extern_dspfunc(avg_h264_qpel,  4);
+
+#undef extern_dspfunc
+
+void put_h264_chroma_mc2_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y);
+void put_h264_chroma_mc4_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y);
+void put_h264_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y);
+
+void avg_h264_chroma_mc2_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y);
+void avg_h264_chroma_mc4_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y);
+void avg_h264_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y);
+
+
+void dump_block8(uint8_t *block, int line_size, int h);
+void dump_block4(uint8_t *block, int line_size, int h);
+void dump_block(uint8_t *block, int line_size, int h, int w);
+
+void check_block8(uint8_t *test, uint8_t *correct, int line_size_test, int line_size_correct,
+                  int h, char *name, int max_dev);
+void check_block4(uint8_t *test, uint8_t *correct, int line_size_test, int line_size_correct,
+                  int h, char *name, int max_dev);
+void check_block(uint8_t *test, uint8_t *correct, int line_size_test, int line_size_correct,
+                 int h, int width, char *name, int max_dev);
+
+#define PIXOP2( OPNAME, OP ) \
+void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        OP(*((uint32_t*)(block  )), LD32(pixels  ));\
+        pixels+=line_size;\
+        block +=line_size;\
+    }\
+}\
+void OPNAME ## _pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \
+                                                int src_stride1, int src_stride2, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        uint32_t a,b;\
+        a= LD32(&src1[i*src_stride1  ]);\
+        b= LD32(&src2[i*src_stride2  ]);\
+        OP(*((uint32_t*)&dst[i*dst_stride  ]), rnd_avg32(a, b));\
+        a= LD32(&src1[i*src_stride1+4]);\
+        b= LD32(&src2[i*src_stride2+4]);\
+        OP(*((uint32_t*)&dst[i*dst_stride+4]), rnd_avg32(a, b));\
+    }\
+}\
+\
+void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \
+                                                int src_stride1, int src_stride2, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        uint32_t a,b;\
+        a= LD32(&src1[i*src_stride1  ]);\
+        b= LD32(&src2[i*src_stride2  ]);\
+        OP(*((uint32_t*)&dst[i*dst_stride  ]), rnd_avg32(a, b));\
+    }\
+}\
+\
+void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \
+                                                int src_stride1, int src_stride2, int h){\
+    OPNAME ## _pixels8_l2(dst  , src1  , src2  , dst_stride, src_stride1, src_stride2, h);\
+    OPNAME ## _pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\
+}\
+
+#else
+#define PIXOP2( OPNAME, OP ) \
+static void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        OP(*((uint32_t*)(block  )), LD32(pixels  ));\
+        pixels+=line_size;\
+        block +=line_size;\
+    }\
+}\
+static void OPNAME ## _pixels8_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        OP(*((uint32_t*)(block  )), LD32(pixels  ));\
+        OP(*((uint32_t*)(block+4)), LD32(pixels+4));\
+        pixels+=line_size;\
+        block +=line_size;\
+    }\
+}\
+static void OPNAME ## _pixels16_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        OP(*((uint32_t*)(block  )), LD32(pixels  ));\
+        OP(*((uint32_t*)(block+4)), LD32(pixels+4));\
+        OP(*((uint32_t*)(block+8)), LD32(pixels+8));\
+        OP(*((uint32_t*)(block+12)), LD32(pixels+12));\
+        pixels+=line_size;\
+        block +=line_size;\
+    }\
+}\
+static void OPNAME ## _pixels8_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \
+                                                int src_stride1, int src_stride2, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        uint32_t a,b;\
+        a= LD32(&src1[i*src_stride1  ]);\
+        b= LD32(&src2[i*src_stride2  ]);\
+        OP(*((uint32_t*)&dst[i*dst_stride  ]), rnd_avg32(a, b));\
+        a= LD32(&src1[i*src_stride1+4]);\
+        b= LD32(&src2[i*src_stride2+4]);\
+        OP(*((uint32_t*)&dst[i*dst_stride+4]), rnd_avg32(a, b));\
+    }\
+}\
+\
+static void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \
+                                                int src_stride1, int src_stride2, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        uint32_t a,b;\
+        a= LD32(&src1[i*src_stride1  ]);\
+        b= LD32(&src2[i*src_stride2  ]);\
+        OP(*((uint32_t*)&dst[i*dst_stride  ]), rnd_avg32(a, b));\
+    }\
+}\
+\
+static void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \
+                                                int src_stride1, int src_stride2, int h){\
+    OPNAME ## _pixels8_l2(dst  , src1  , src2  , dst_stride, src_stride1, src_stride2, h);\
+    OPNAME ## _pixels8_l2(dst+8, src1+8, src2+8, dst_stride, src_stride1, src_stride2, h);\
+}\
+
+#endif
+
+#define op_avg(a, b) a = rnd_avg32(a, b)
+#define op_put(a, b) a = b
+
+PIXOP2(avg, op_avg)
+PIXOP2(put, op_put)
+#undef op_avg
+#undef op_put
+
+
+
+static void clear_blocks_avr32(DCTELEM *blocks)
+{
+  int n = 12;
+  uint64_t tmp1, tmp2;
+  blocks += 6*64;
+  __asm__ volatile ( "mov\t%1, 0\n"
+                 "mov\t%m1, 0\n"
+                 "mov\t%2, 0\n"
+                 "mov\t%m2, 0\n"
+                 "0:\n"
+                 "stm\t--%3, %1, %m1, %2, %m2\n"
+                 "stm\t--%3, %1, %m1, %2, %m2\n"
+                 "stm\t--%3, %1, %m1, %2, %m2\n"
+                 "stm\t--%3, %1, %m1, %2, %m2\n"
+                 "sub\t%0, 1\n"
+                 "brne\t0b\n"
+                 : "+r"(n), "=&r"(tmp1), "=&r"(tmp2),
+                 "+r"(blocks));
+}
+
+
+static void put_h264_chroma_mc2_pico(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
+  const int A=(8-x)*(8-y);
+  const int B=(  x)*(8-y);
+  const int C=(8-x)*(  y);
+  const int D=(  x)*(  y);
+  int i;
+
+  PICO_PUT_W(PICO_COEFF0_A, (A << 16) | (B & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF0_B, 32);
+  PICO_PUT_W(PICO_COEFF1_A, (C << 16) | (D & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF1_B, 0);
+  PICO_PUT_W(PICO_COEFF2_A, 0);
+  PICO_PUT_W(PICO_COEFF2_B, 0);
+  PICO_PUT_W(PICO_CONFIG,
+             PICO_OUTPUT_MODE(PICO_PLANAR_MODE)
+             | PICO_INPUT_MODE(PICO_HOR_FILTER_MODE)
+             | PICO_COEFF_FRAC_BITS(6)
+             | PICO_OFFSET_FRAC_BITS(6));
+
+  for(i=0; i<h; i++)
+    {
+
+      int src0 = LD32(src);
+      int src1 = LD32(src + stride);
+
+      PICO_MVRC_W(PICO_INPIX0, src0);
+      PICO_MVRC_W(PICO_INPIX1, src1);
+      PICO_OP(PICO_SINGLE_VECTOR, 2, 0, 4, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 3, 1, 5, 0);
+      src += stride;
+      ST16(dst,(short)PICO_GET_W(PICO_OUTPIX0));
+      dst += stride;
+    }
+}
+
+
+static void put_h264_chroma_mc4_pico(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
+  const int A=(8-x)*(8-y);\
+  const int B=(  x)*(8-y);
+  const int C=(8-x)*(  y);
+  const int D=(  x)*(  y);
+  int i;
+
+  PICO_PUT_W(PICO_COEFF0_A, (A << 16) | (B & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF0_B, 32);
+  PICO_PUT_W(PICO_COEFF1_A, (C << 16) | (D & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF1_B, 0);
+  PICO_PUT_W(PICO_COEFF2_A, 0);
+  PICO_PUT_W(PICO_COEFF2_B, 0);
+  PICO_PUT_W(PICO_CONFIG,
+             PICO_OUTPUT_MODE(PICO_PLANAR_MODE)
+             | PICO_INPUT_MODE(PICO_HOR_FILTER_MODE)
+             | PICO_COEFF_FRAC_BITS(6)
+             | PICO_OFFSET_FRAC_BITS(6));
+
+  for(i=0; i<h; i++)
+    {
+      /*
+        OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));
+        OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));
+        OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));
+        OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));
+        dst+= stride;
+        src+= stride;
+      */
+
+      int src0 = LD32(src);
+      int src1 = (((int)src[4] << 24) | (int)src[stride]);
+      int src2 = LD32(src + stride + 1);
+
+      PICO_MVRC_W(PICO_INPIX0, src0);
+      PICO_MVRC_W(PICO_INPIX1, src1);
+      PICO_MVRC_W(PICO_INPIX2, src2);
+      PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 7, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 8, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 9, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 10, 0);
+      src += stride;
+      ST32(dst, PICO_GET_W(PICO_OUTPIX0));
+
+      dst += stride;
+    }
+}
+
+static void put_h264_chroma_mc8_pico(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
+  const int A=(8-x)*(8-y);
+  const int B=(  x)*(8-y);
+  const int C=(8-x)*(  y);
+  const int D=(  x)*(  y);
+  int i;
+
+  PICO_PUT_W(PICO_COEFF0_A, (A << 16) | (B & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF0_B, 32);
+  PICO_PUT_W(PICO_COEFF1_A, (C << 16) | (D & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF1_B, 0);
+  PICO_PUT_W(PICO_COEFF2_A, 0);
+  PICO_PUT_W(PICO_COEFF2_B, 0);
+  PICO_PUT_W(PICO_CONFIG,
+             PICO_OUTPUT_MODE(PICO_PLANAR_MODE)
+             | PICO_INPUT_MODE(PICO_HOR_FILTER_MODE)
+             | PICO_COEFF_FRAC_BITS(6)
+             | PICO_OFFSET_FRAC_BITS(6));
+
+  for(i=0; i<h; i++)
+    {
+      /*
+        OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));
+        OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));
+        OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));
+        OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));
+        OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5]));
+        OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6]));
+        OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7]));
+        OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8]));
+        dst+= stride;
+        src+= stride;
+      */
+      int src0 = LD32(src);
+      int src1 = (((int)src[4] << 24) | (int)src[stride]);
+      int src2 = LD32(src + stride + 1);
+
+      PICO_MVRC_W(PICO_INPIX0, src0);
+      PICO_MVRC_W(PICO_INPIX1, src1);
+      PICO_MVRC_W(PICO_INPIX2, src2);
+      PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 7, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 8, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 9, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 10, 0);
+      ST32(dst, PICO_GET_W(PICO_OUTPIX0));
+
+      src0 = LD32(src + 4);
+      src1 = (src[8] << 24) | src[stride + 4];
+      src2 = LD32(src + stride + 5);
+
+      PICO_MVRC_W(PICO_INPIX0, src0);
+      PICO_MVRC_W(PICO_INPIX1, src1);
+      PICO_MVRC_W(PICO_INPIX2, src2);
+      PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 7, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 8, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 9, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 10, 0);
+      src += stride;
+      ST32(dst + 4, PICO_GET_W(PICO_OUTPIX0));
+
+      dst += stride;
+    }
+}
+
+
+static void avg_h264_chroma_mc2_pico(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
+  const int A=(8-x)*(8-y);
+  const int B=(  x)*(8-y);
+  const int C=(8-x)*(  y);
+  const int D=(  x)*(  y);
+  int i;
+
+  PICO_PUT_W(PICO_COEFF0_A, (A << 16) | (B & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF0_B, 32);
+  PICO_PUT_W(PICO_COEFF1_A, (C << 16) | (D & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF1_B, 0);
+  PICO_PUT_W(PICO_COEFF2_A, 0);
+  PICO_PUT_W(PICO_COEFF2_B, 0);
+  PICO_PUT_W(PICO_CONFIG,
+             PICO_OUTPUT_MODE(PICO_PLANAR_MODE)
+             | PICO_INPUT_MODE(PICO_HOR_FILTER_MODE)
+             | PICO_COEFF_FRAC_BITS(6)
+             | PICO_OFFSET_FRAC_BITS(6));
+
+  for(i=0; i<h; i++)
+    {
+      int src0 = LD32(src);
+      int src1 = LD32(src + stride);
+
+      PICO_MVRC_W(PICO_INPIX0, src0);
+      PICO_MVRC_W(PICO_INPIX1, src1);
+      PICO_OP(PICO_SINGLE_VECTOR, 2, 0, 4, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 3, 1, 5, 0);
+      src += stride;
+      ST16(dst, rnd_avg32(LD16(dst), PICO_GET_W(PICO_OUTPIX0)));
+      dst += stride;
+    }
+}
+
+
+static void avg_h264_chroma_mc4_pico(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
+  const int A=(8-x)*(8-y);\
+  const int B=(  x)*(8-y);
+  const int C=(8-x)*(  y);
+  const int D=(  x)*(  y);
+  int i;
+
+  PICO_PUT_W(PICO_COEFF0_A, (A << 16) | (B & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF0_B, 32);
+  PICO_PUT_W(PICO_COEFF1_A, (C << 16) | (D & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF1_B, 0);
+  PICO_PUT_W(PICO_COEFF2_A, 0);
+  PICO_PUT_W(PICO_COEFF2_B, 0);
+  PICO_PUT_W(PICO_CONFIG,
+             PICO_OUTPUT_MODE(PICO_PLANAR_MODE)
+             | PICO_INPUT_MODE(PICO_HOR_FILTER_MODE)
+             | PICO_COEFF_FRAC_BITS(6)
+             | PICO_OFFSET_FRAC_BITS(6));
+
+  for(i=0; i<h; i++)
+    {
+      /*
+        OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));
+        OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));
+        OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));
+        OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));
+        dst+= stride;
+        src+= stride;
+      */
+
+      int src0 = *((int *)src);
+      int src1 = (int)((src[4] << 24) | src[stride]);
+      int src2 = *((int *)(src + stride + 1));
+
+      PICO_MVRC_W(PICO_INPIX0, src0);
+      PICO_MVRC_W(PICO_INPIX1, src1);
+      PICO_MVRC_W(PICO_INPIX2, src2);
+      PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 7, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 8, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 9, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 10, 0);
+      src += stride;
+      ST32(dst, rnd_avg32(LD32(dst), PICO_GET_W(PICO_OUTPIX0)));
+      dst += stride;
+    }
+}
+
+static void avg_h264_chroma_mc8_pico(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
+  const int A=(8-x)*(8-y);
+  const int B=(  x)*(8-y);
+  const int C=(8-x)*(  y);
+  const int D=(  x)*(  y);
+  int i;
+
+  PICO_PUT_W(PICO_COEFF0_A, (A << 16) | (B & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF0_B, 32);
+  PICO_PUT_W(PICO_COEFF1_A, (C << 16) | (D & 0xFFFF));
+  PICO_PUT_W(PICO_COEFF1_B, 0);
+  PICO_PUT_W(PICO_COEFF2_A, 0);
+  PICO_PUT_W(PICO_COEFF2_B, 0);
+  PICO_PUT_W(PICO_CONFIG,
+             PICO_OUTPUT_MODE(PICO_PLANAR_MODE)
+             | PICO_INPUT_MODE(PICO_HOR_FILTER_MODE)
+             | PICO_COEFF_FRAC_BITS(6)
+             | PICO_OFFSET_FRAC_BITS(6));
+
+  for(i=0; i<h; i++)
+    {
+      /*
+        OP(dst[0], (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1]));
+        OP(dst[1], (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2]));
+        OP(dst[2], (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3]));
+        OP(dst[3], (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4]));
+        OP(dst[4], (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5]));
+        OP(dst[5], (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6]));
+        OP(dst[6], (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7]));
+        OP(dst[7], (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8]));
+        dst+= stride;
+        src+= stride;
+      */
+      int src0 = *((int *)src);
+      int src1 = (volatile int)((src[4] << 24) | src[stride]);
+      int src2 = *((int *)(src + stride + 1));
+
+      PICO_MVRC_W(PICO_INPIX0, src0);
+      PICO_MVRC_W(PICO_INPIX1, src1);
+      PICO_MVRC_W(PICO_INPIX2, src2);
+      PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 7, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 8, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 9, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 10, 0);
+      ST32(dst, rnd_avg32(LD32(dst), PICO_GET_W(PICO_OUTPIX0)));
+
+      src0 = *((int *)(src + 4));
+      src1 = (int)((src[8] << 24) | src[stride + 4]);
+      src2 = *((int *)(src + stride + 5));
+
+      PICO_MVRC_W(PICO_INPIX0, src0);
+      PICO_MVRC_W(PICO_INPIX1, src1);
+      PICO_MVRC_W(PICO_INPIX2, src2);
+      PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 7, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 8, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 9, 0);
+      PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 10, 0);
+      src += stride;
+      ST32(dst + 4, rnd_avg32(LD32(dst + 4), PICO_GET_W(PICO_OUTPIX0)));
+      dst += stride;
+    }
+}
+
+static struct pico_config_t h264_qpel4_h_lowpass_config = {
+  .input_mode = PICO_HOR_FILTER_MODE,
+  .output_mode = PICO_PLANAR_MODE,
+  .coeff_frac_bits = 5,
+  .offset_frac_bits = 5,
+  .coeff0_0 = 1,
+  .coeff0_1 = -5,
+  .coeff0_2 = 20,
+  .coeff0_3 = 16,
+  .coeff1_0 = 20,
+  .coeff1_1 = -5,
+  .coeff1_2 = 1,
+  .coeff1_3 = 0,
+  .coeff2_0 = 0,
+  .coeff2_1 = 0,
+  .coeff2_2 = 0,
+  .coeff2_3 = 0
+};
+
+
+
+static void put_h264_qpel4_h_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  const int h=4;
+  int i;
+
+  set_pico_config(&h264_qpel4_h_lowpass_config);
+
+  for(i=0; i<h; i++){
+
+    /*
+      OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\
+      OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\
+      OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]));\
+      OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]));\
+      dst+=dstStride;\
+      src+=srcStride;\ */
+    PICO_MVRC_W(PICO_INPIX0, LD32(src - 2));
+    PICO_MVRC_D(PICO_INPIX2, LD64_UNALIGNED(src + 2));
+    PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 3, 6);
+    PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 4, 7);
+    PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 5, 8);
+    PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 6, 9);
+    src += srcStride;
+    ST32(dst, PICO_GET_W(PICO_OUTPIX0));
+    dst += dstStride;
+  }
+}
+
+static void avg_h264_qpel4_h_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  const int h=4;
+  int i;
+
+  set_pico_config(&h264_qpel4_h_lowpass_config);
+
+  for(i=0; i<h; i++){
+
+    /*
+      OP(dst[0], (src[0]+src[1])*20 - (src[-1]+src[2])*5 + (src[-2]+src[3]));\
+      OP(dst[1], (src[1]+src[2])*20 - (src[0 ]+src[3])*5 + (src[-1]+src[4]));\
+      OP(dst[2], (src[2]+src[3])*20 - (src[1 ]+src[4])*5 + (src[0 ]+src[5]));\
+      OP(dst[3], (src[3]+src[4])*20 - (src[2 ]+src[5])*5 + (src[1 ]+src[6]));\
+      dst+=dstStride;\
+      src+=srcStride;\
+    */
+
+    PICO_MVRC_W(PICO_INPIX0, LD32(src - 2));
+    PICO_MVRC_D(PICO_INPIX2, LD64_UNALIGNED(src + 2));
+    PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 3, 6);
+    PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 4, 7);
+    PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 5, 8);
+    PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 6, 9);
+    src += srcStride;
+    ST32(dst, rnd_avg32(LD32(dst), PICO_GET_W(PICO_OUTPIX0)));
+    dst += dstStride;
+  }
+}
+
+static struct pico_config_t h264_qpel4_v_lowpass_config1 = {
+  .input_mode = PICO_VERT_FILTER_MODE,
+  .output_mode = PICO_PACKED_MODE,
+  .coeff_frac_bits = 5,
+  .offset_frac_bits = 5,
+  .coeff0_0 = 1,
+  .coeff0_1 = -5,
+  .coeff0_2 = 20,
+  .coeff0_3 = 16,
+  .coeff1_0 = 1,
+  .coeff1_1 = -5,
+  .coeff1_2 = 20,
+  .coeff1_3 = 16,
+  .coeff2_0 = 1,
+  .coeff2_1 = -5,
+  .coeff2_2 = 20,
+  .coeff2_3 = 16
+};
+
+
+
+static struct pico_config_t h264_qpel4_v_lowpass_config2 = {
+  .input_mode = PICO_VERT_FILTER_MODE,
+  .output_mode = PICO_PLANAR_MODE,
+  .coeff_frac_bits = 5,
+  .offset_frac_bits = 5,
+  .coeff0_0 = 1,
+  .coeff0_1 = -5,
+  .coeff0_2 = 20,
+  .coeff0_3 = 16,
+  .coeff1_0 = 20,
+  .coeff1_1 = -5,
+  .coeff1_2 = 1,
+  .coeff1_3 = 0,
+  .coeff2_0 = 0,
+  .coeff2_1 = 0,
+  .coeff2_2 = 0,
+  .coeff2_3 = 0
+};
+
+static void put_h264_qpel4_v_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+
+  /*
+    const int w=4;
+    uint8_t *cm = cropTbl + MAX_NEG_CROP;
+    int i;
+    for(i=0; i<w; i++)
+    {
+    const int srcB= src[-2*srcStride];\
+    const int srcA= src[-1*srcStride];\
+    const int src0= src[0 *srcStride];\
+    const int src1= src[1 *srcStride];\
+    const int src2= src[2 *srcStride];\
+    const int src3= src[3 *srcStride];\
+    const int src4= src[4 *srcStride];\
+    const int src5= src[5 *srcStride];\
+    const int src6= src[6 *srcStride];\
+    OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\
+    OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\
+    OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\
+    OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\
+    dst++;\
+    src++;\
+  */
+
+  set_pico_config(&h264_qpel4_v_lowpass_config1);
+
+  {
+    int srcB= LD32(src - 2*srcStride);
+    int srcA= LD32(src - 1*srcStride);
+    int src0= LD32(src + 0 *srcStride);
+    int src1= LD32(src + 1 *srcStride);
+    int src2= LD32(src + 2 *srcStride);
+    int src3= LD32(src + 3 *srcStride);
+    int src4= LD32(src + 4 *srcStride);
+    int src5= LD32(src + 5 *srcStride);
+    int src6= LD32(src + 6 *srcStride);
+
+    union wordbytes {
+      int word;
+      struct  {
+        unsigned int t:8;
+        unsigned int u:8;
+        unsigned int l:8;
+        unsigned int b:8;
+      } bytes;
+    } tmp1, tmp2, tmp3;
+
+    /* First compute the leftmost three colums */
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(0, 0, 0, 3, 6);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX0, src3);
+    PICO_OP(PICO_USE_ACC, 0, 0, 3, 6);
+    ST32(dst, PICO_GET_W(PICO_OUTPIX0));
+    dst += dstStride;
+    PICO_MVRC_W(PICO_INPIX0, srcA);
+    PICO_MVRC_W(PICO_INPIX1, src0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_OP(0, 0, 0, 3, 6);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_MVRC_W(PICO_INPIX1, src3);
+    PICO_MVRC_W(PICO_INPIX0, src4);
+    PICO_OP(PICO_USE_ACC, 0, 0, 3, 6);
+    ST32(dst, PICO_GET_W(PICO_OUTPIX0));
+    dst += dstStride;
+    PICO_MVRC_W(PICO_INPIX0, src0);
+    PICO_MVRC_W(PICO_INPIX1, src1);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_OP(0, 0, 0, 3, 6);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_MVRC_W(PICO_INPIX1, src4);
+    PICO_MVRC_W(PICO_INPIX0, src5);
+    PICO_OP(PICO_USE_ACC, 0, 0, 3, 6);
+    ST32(dst, PICO_GET_W(PICO_OUTPIX0));
+    dst += dstStride;
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_OP(0, 0, 0, 3, 6);
+    PICO_MVRC_W(PICO_INPIX2, src4);
+    PICO_MVRC_W(PICO_INPIX1, src5);
+    PICO_MVRC_W(PICO_INPIX0, src6);
+    PICO_OP(PICO_USE_ACC, 0, 0, 3, 6);
+    ST32(dst, PICO_GET_W(PICO_OUTPIX0));
+    /* Now compute the last column */
+
+    tmp1.bytes.t = srcB;
+    tmp1.bytes.u = src1;
+    tmp1.bytes.l = src4;
+
+    tmp2.bytes.t = srcA;
+    tmp2.bytes.u = src2;
+    tmp2.bytes.l = src5;
+
+    tmp3.bytes.t = src0;
+    tmp3.bytes.u = src3;
+    tmp3.bytes.l = src6;
+
+    PICO_MVRC_W(PICO_INPIX0, tmp1.word);
+    PICO_MVRC_W(PICO_INPIX1, tmp2.word);
+    PICO_MVRC_W(PICO_INPIX2, tmp3.word);
+    set_pico_config(&h264_qpel4_v_lowpass_config2);
+
+    PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 3, 6);
+    PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 4, 7);
+    PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 5, 8);
+    PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 6, 9);
+
+    PICO_MVCR_W(tmp1.word, PICO_OUTPIX0);
+    dst[3] = (char)(tmp1.bytes.b);
+    dst[3 - dstStride] = (char)(tmp1.bytes.l);
+    dst[3 - 2*dstStride] = (char)(tmp1.bytes.u);
+    dst[3 - 3*dstStride] = (char)(tmp1.bytes.t);
+
+  }
+    /*}
+
+
+    }*/
+}
+
+static void avg_h264_qpel4_v_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+
+  /*
+    const int w=4;
+    uint8_t *cm = cropTbl + MAX_NEG_CROP;
+    int i;
+    for(i=0; i<w; i++)
+    {
+    const int srcB= src[-2*srcStride];\
+    const int srcA= src[-1*srcStride];\
+    const int src0= src[0 *srcStride];\
+    const int src1= src[1 *srcStride];\
+    const int src2= src[2 *srcStride];\
+    const int src3= src[3 *srcStride];\
+    const int src4= src[4 *srcStride];\
+    const int src5= src[5 *srcStride];\
+    const int src6= src[6 *srcStride];\
+    OP(dst[0*dstStride], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\
+    OP(dst[1*dstStride], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\
+    OP(dst[2*dstStride], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\
+    OP(dst[3*dstStride], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\
+    dst++;\
+    src++;\
+  */
+  uint8_t tmp_block[4*4];
+
+  set_pico_config(&h264_qpel4_v_lowpass_config1);
+
+  {
+    int srcB= LD32(src - 2*srcStride);
+    int srcA= LD32(src - 1*srcStride);
+    int src0= LD32(src + 0 *srcStride);
+    int src1= LD32(src + 1 *srcStride);
+    int src2= LD32(src + 2 *srcStride);
+    int src3= LD32(src + 3 *srcStride);
+    int src4= LD32(src + 4 *srcStride);
+    int src5= LD32(src + 5 *srcStride);
+    int src6= LD32(src + 6 *srcStride);
+
+    union wordbytes {
+      int word;
+      struct {
+        unsigned int t:8;
+        unsigned int u:8;
+        unsigned int l:8;
+        unsigned int b:8;
+      } bytes;
+    } tmp1, tmp2, tmp3;
+
+    /* First compute the leftmost three colums */
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(0, 0, 0, 3, 6);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX0, src3);
+    PICO_OP(PICO_USE_ACC, 0, 0, 3, 6);
+    ST32(tmp_block, PICO_GET_W(PICO_OUTPIX0));
+    PICO_MVRC_W(PICO_INPIX0, srcA);
+    PICO_MVRC_W(PICO_INPIX1, src0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_OP(0, 0, 0, 3, 6);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_MVRC_W(PICO_INPIX1, src3);
+    PICO_MVRC_W(PICO_INPIX0, src4);
+    PICO_OP(PICO_USE_ACC, 0, 0, 3, 6);
+    ST32(tmp_block + 4, PICO_GET_W(PICO_OUTPIX0));
+    PICO_MVRC_W(PICO_INPIX0, src0);
+    PICO_MVRC_W(PICO_INPIX1, src1);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_OP(0, 0, 0, 3, 6);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_MVRC_W(PICO_INPIX1, src4);
+    PICO_MVRC_W(PICO_INPIX0, src5);
+    PICO_OP(PICO_USE_ACC, 0, 0, 3, 6);
+    ST32(tmp_block + 8, PICO_GET_W(PICO_OUTPIX0));
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_OP(0, 0, 0, 3, 6);
+    PICO_MVRC_W(PICO_INPIX2, src4);
+    PICO_MVRC_W(PICO_INPIX1, src5);
+    PICO_MVRC_W(PICO_INPIX0, src6);
+    PICO_OP(PICO_USE_ACC, 0, 0, 3, 6);
+    ST32(tmp_block + 12, PICO_GET_W(PICO_OUTPIX0));
+    /* Now compute the last column */
+
+    tmp1.bytes.t = srcB;
+    tmp1.bytes.u = src1;
+    tmp1.bytes.l = src4;
+
+    tmp2.bytes.t = srcA;
+    tmp2.bytes.u = src2;
+    tmp2.bytes.l = src5;
+
+    tmp3.bytes.t = src0;
+    tmp3.bytes.u = src3;
+    tmp3.bytes.l = src6;
+
+    PICO_MVRC_W(PICO_INPIX0, tmp1.word);
+    PICO_MVRC_W(PICO_INPIX1, tmp2.word);
+    PICO_MVRC_W(PICO_INPIX2, tmp3.word);
+    set_pico_config(&h264_qpel4_v_lowpass_config2);
+
+    PICO_OP(PICO_SINGLE_VECTOR, 0, 0, 3, 6);
+    PICO_OP(PICO_SINGLE_VECTOR, 1, 1, 4, 7);
+    PICO_OP(PICO_SINGLE_VECTOR, 2, 2, 5, 8);
+    PICO_OP(PICO_SINGLE_VECTOR, 3, 3, 6, 9);
+
+    PICO_MVCR_W(tmp1.word, PICO_OUTPIX0);
+    tmp_block[3 + 3*4] = (char)(tmp1.bytes.b);
+    tmp_block[3 + 2*4] = (char)(tmp1.bytes.l);
+    tmp_block[3 + 1*4] = (char)(tmp1.bytes.u);
+    tmp_block[3] = (char)(tmp1.bytes.t);
+
+    /* Compute the average */
+    srcB= LD32(dst);
+    srcA= LD32(dst + dstStride);
+    src0= LD32(dst + dstStride*2);
+    src1= LD32(dst + dstStride*3);
+
+    src2= LD32(tmp_block);
+    src3= LD32(tmp_block + 4);
+    src4= LD32(tmp_block + 8);
+    src5= LD32(tmp_block + 12);
+
+    ST32(dst, rnd_avg32(srcB, src2));
+    ST32(dst + dstStride, rnd_avg32(srcA, src3));
+    ST32(dst + 2*dstStride, rnd_avg32(src0, src4));
+    ST32(dst + 3*dstStride, rnd_avg32(src1, src5));
+  }
+}
+
+static struct pico_config_t h264_qpel4_hv_lowpass_config = {
+  .input_mode = PICO_HOR_FILTER_MODE,
+  .output_mode = PICO_PACKED_MODE,
+  .coeff_frac_bits = 10,
+  .offset_frac_bits = 10,
+  .coeff0_0 = 1,
+  .coeff0_1 = -5,
+  .coeff0_2 = 20,
+  .coeff0_3 = 512,
+  .coeff1_0 = -5,
+  .coeff1_1 = 25,
+  .coeff1_2 = -100,
+  .coeff1_3 = 0,
+  .coeff2_0 = 20,
+  .coeff2_1 = -100,
+  .coeff2_2 = 400,
+  .coeff2_3 = 0
+};
+
+static void put_h264_qpel4_hv_lowpass_pico(uint8_t *dst,  uint8_t *src, int dstStride, int srcStride){
+
+  int32_t tmp_block[48];
+  int32_t *tmp = tmp_block;
+  int i;
+
+  set_pico_config(&h264_qpel4_hv_lowpass_config);
+
+  src -= 2;
+  for ( i = 0; i < 2; i++ ){
+    int srcB= LD32(src - 2*srcStride);
+    int srcA= LD32(src - 1*srcStride);
+    int src0= LD32(src + 0 *srcStride);
+    int src1= LD32(src + 1 *srcStride);
+    int src2= LD32(src + 2 *srcStride);
+    int src3= LD32(src + 3 *srcStride);
+    int src4= LD32(src + 4 *srcStride);
+    int src5= LD32(src + 5 *srcStride);
+    int src6= LD32(src + 6 *srcStride);
+
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(0, 0, 0, 4, 8);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX0, src3);
+    PICO_OP(PICO_USE_ACC, 0, 0, 4, 8);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_OP(0, 0, 1, 5, 9);
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(PICO_USE_ACC, 0, 1, 5, 9);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_OP(0, 0, 4, 8, 0);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_MVRC_W(PICO_INPIX1, src3);
+    PICO_MVRC_W(PICO_INPIX0, src4);
+    PICO_OP(PICO_USE_ACC, 0, 0, 4, 8);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_OP(0, 0, 1, 5, 9);
+    PICO_MVRC_W(PICO_INPIX0, srcA);
+    PICO_MVRC_W(PICO_INPIX1, src0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_OP(PICO_USE_ACC, 0, 1, 5, 9);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_MVRC_W(PICO_INPIX0, src2);
+    PICO_OP(0, 0, 4, 8, 0);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_MVRC_W(PICO_INPIX1, src4);
+    PICO_MVRC_W(PICO_INPIX0, src5);
+    PICO_OP(PICO_USE_ACC, 0, 0, 4, 8);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_OP(0, 0, 1, 5, 9);
+    PICO_MVRC_W(PICO_INPIX0, src0);
+    PICO_MVRC_W(PICO_INPIX1, src1);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_OP(PICO_USE_ACC, 0, 1, 5, 9);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_MVRC_W(PICO_INPIX0, src3);
+    PICO_OP(0, 0, 4, 8, 0);
+    PICO_MVRC_W(PICO_INPIX2, src4);
+    PICO_MVRC_W(PICO_INPIX1, src5);
+    PICO_MVRC_W(PICO_INPIX0, src6);
+    PICO_OP(PICO_USE_ACC, 0, 0, 4, 8);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_OP(0, 0, 1, 5, 9);
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_OP(PICO_USE_ACC, 0, 1, 5, 9);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+    src += 2;
+  }
+
+  src -= 1;
+  tmp -= 48;
+
+  PICO_PUT_W(PICO_CONFIG,
+             PICO_OUTPUT_MODE(PICO_PLANAR_MODE)
+             | PICO_INPUT_MODE(PICO_VERT_FILTER_MODE)
+             | PICO_COEFF_FRAC_BITS(10)
+             | PICO_OFFSET_FRAC_BITS(10));
+
+  for ( i = 0; i < 2; i++ ){
+    int srcB= LD32(src - 2*srcStride);
+    int srcA= LD32(src - 1*srcStride);
+    int src0= LD32(src + 0 *srcStride);
+    int src1= LD32(src + 1 *srcStride);
+    int src2= LD32(src + 2 *srcStride);
+    int src3= LD32(src + 3 *srcStride);
+    int src4= LD32(src + 4 *srcStride);
+    int src5= LD32(src + 5 *srcStride);
+    int src6= LD32(src + 6 *srcStride);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(PICO_USE_ACC, 0, 6, 3, 0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX0, src3);
+    PICO_OP(PICO_USE_ACC | PICO_SINGLE_VECTOR, 0, 6, 3, 0);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_OP(PICO_USE_ACC, 1, 9, 6, 3);
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(PICO_USE_ACC | PICO_SINGLE_VECTOR, 1, 9, 6, 3);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_MVRC_W(PICO_INPIX0, srcA);
+    PICO_MVRC_W(PICO_INPIX1, src0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_OP(PICO_USE_ACC, 2, 6, 3, 0);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_MVRC_W(PICO_INPIX1, src3);
+    PICO_MVRC_W(PICO_INPIX0, src4);
+    PICO_OP(PICO_USE_ACC | PICO_SINGLE_VECTOR, 2, 6, 3, 0);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_OP(PICO_USE_ACC, 3, 9, 6, 3);
+    PICO_MVRC_W(PICO_INPIX0, srcA);
+    PICO_MVRC_W(PICO_INPIX1, src0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_OP(PICO_USE_ACC | PICO_SINGLE_VECTOR, 3, 9, 6, 3);
+
+    ST16(dst + 0*dstStride, (short)(PICO_GET_W(PICO_OUTPIX0) >> 16));
+    ST16(dst + 1*dstStride, (short)PICO_GET_W(PICO_OUTPIX0));
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_MVRC_W(PICO_INPIX0, src0);
+    PICO_MVRC_W(PICO_INPIX1, src1);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_OP(PICO_USE_ACC, 0, 6, 3, 0);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_MVRC_W(PICO_INPIX1, src4);
+    PICO_MVRC_W(PICO_INPIX0, src5);
+    PICO_OP(PICO_USE_ACC| PICO_SINGLE_VECTOR, 0, 6, 3, 0);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_OP(PICO_USE_ACC, 1, 9, 6, 3);
+    PICO_MVRC_W(PICO_INPIX0, src0);
+    PICO_MVRC_W(PICO_INPIX1, src1);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_OP(PICO_USE_ACC| PICO_SINGLE_VECTOR, 1, 9, 6, 3);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_OP(PICO_USE_ACC, 2, 6, 3, 0);
+    PICO_MVRC_W(PICO_INPIX2, src4);
+    PICO_MVRC_W(PICO_INPIX1, src5);
+    PICO_MVRC_W(PICO_INPIX0, src6);
+    PICO_OP(PICO_USE_ACC| PICO_SINGLE_VECTOR, 2, 6, 3, 0);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_OP(PICO_USE_ACC, 3, 9, 6, 3);
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_OP(PICO_USE_ACC| PICO_SINGLE_VECTOR, 3, 9, 6, 3);
+
+    ST16(dst + 2*dstStride, (short)(PICO_GET_W(PICO_OUTPIX0) >> 16));
+    ST16(dst + 3*dstStride, (short)PICO_GET_W(PICO_OUTPIX0));
+
+    dst += 2;
+    src += 2;
+  }
+}
+
+
+
+
+static void avg_h264_qpel4_hv_lowpass_pico(uint8_t *dst,  uint8_t *src, int dstStride, int srcStride){
+
+  int32_t tmp_block[48];
+  int32_t *tmp = tmp_block;
+  int i;
+
+  set_pico_config(&h264_qpel4_hv_lowpass_config);
+
+  src -= 2;
+  for ( i = 0; i < 2; i++ ){
+    int srcB= LD32(src - 2*srcStride);
+    int srcA= LD32(src - 1*srcStride);
+    int src0= LD32(src + 0 *srcStride);
+    int src1= LD32(src + 1 *srcStride);
+    int src2= LD32(src + 2 *srcStride);
+    int src3= LD32(src + 3 *srcStride);
+    int src4= LD32(src + 4 *srcStride);
+    int src5= LD32(src + 5 *srcStride);
+    int src6= LD32(src + 6 *srcStride);
+
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(0, 0, 0, 4, 8);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX0, src3);
+    PICO_OP(PICO_USE_ACC, 0, 0, 4, 8);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_OP(0, 0, 1, 5, 9);
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(PICO_USE_ACC, 0, 1, 5, 9);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_OP(0, 0, 4, 8, 0);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_MVRC_W(PICO_INPIX1, src3);
+    PICO_MVRC_W(PICO_INPIX0, src4);
+    PICO_OP(PICO_USE_ACC, 0, 0, 4, 8);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_OP(0, 0, 1, 5, 9);
+    PICO_MVRC_W(PICO_INPIX0, srcA);
+    PICO_MVRC_W(PICO_INPIX1, src0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_OP(PICO_USE_ACC, 0, 1, 5, 9);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_MVRC_W(PICO_INPIX0, src2);
+    PICO_OP(0, 0, 4, 8, 0);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_MVRC_W(PICO_INPIX1, src4);
+    PICO_MVRC_W(PICO_INPIX0, src5);
+    PICO_OP(PICO_USE_ACC, 0, 0, 4, 8);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_OP(0, 0, 1, 5, 9);
+    PICO_MVRC_W(PICO_INPIX0, src0);
+    PICO_MVRC_W(PICO_INPIX1, src1);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_OP(PICO_USE_ACC, 0, 1, 5, 9);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_MVRC_W(PICO_INPIX0, src3);
+    PICO_OP(0, 0, 4, 8, 0);
+    PICO_MVRC_W(PICO_INPIX2, src4);
+    PICO_MVRC_W(PICO_INPIX1, src5);
+    PICO_MVRC_W(PICO_INPIX0, src6);
+    PICO_OP(PICO_USE_ACC, 0, 0, 4, 8);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+
+    PICO_OP(0, 0, 1, 5, 9);
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_OP(PICO_USE_ACC, 0, 1, 5, 9);
+    PICO_STCM_W(tmp,
+                PICO_REGVECT_VMU0_OUT,
+                PICO_REGVECT_VMU1_OUT,
+                PICO_REGVECT_VMU2_OUT);
+    tmp += 3;
+    src += 2;
+  }
+
+  src -= 1;
+  tmp -= 48;
+
+  PICO_PUT_W(PICO_CONFIG,
+             PICO_OUTPUT_MODE(PICO_PLANAR_MODE)
+             | PICO_INPUT_MODE(PICO_VERT_FILTER_MODE)
+             | PICO_COEFF_FRAC_BITS(10)
+             | PICO_OFFSET_FRAC_BITS(10));
+
+  for ( i = 0; i < 2; i++ ){
+    int srcB= LD32(src - 2*srcStride);
+    int srcA= LD32(src - 1*srcStride);
+    int src0= LD32(src + 0 *srcStride);
+    int src1= LD32(src + 1 *srcStride);
+    int src2= LD32(src + 2 *srcStride);
+    int src3= LD32(src + 3 *srcStride);
+    int src4= LD32(src + 4 *srcStride);
+    int src5= LD32(src + 5 *srcStride);
+    int src6= LD32(src + 6 *srcStride);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(PICO_USE_ACC, 0, 6, 3, 0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX0, src3);
+    PICO_OP(PICO_USE_ACC | PICO_SINGLE_VECTOR, 0, 6, 3, 0);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_OP(PICO_USE_ACC, 1, 9, 6, 3);
+    PICO_MVRC_W(PICO_INPIX0, srcB);
+    PICO_MVRC_W(PICO_INPIX1, srcA);
+    PICO_MVRC_W(PICO_INPIX2, src0);
+    PICO_OP(PICO_USE_ACC | PICO_SINGLE_VECTOR, 1, 9, 6, 3);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_MVRC_W(PICO_INPIX0, srcA);
+    PICO_MVRC_W(PICO_INPIX1, src0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_OP(PICO_USE_ACC, 2, 6, 3, 0);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_MVRC_W(PICO_INPIX1, src3);
+    PICO_MVRC_W(PICO_INPIX0, src4);
+    PICO_OP(PICO_USE_ACC | PICO_SINGLE_VECTOR, 2, 6, 3, 0);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_OP(PICO_USE_ACC, 3, 9, 6, 3);
+    PICO_MVRC_W(PICO_INPIX0, srcA);
+    PICO_MVRC_W(PICO_INPIX1, src0);
+    PICO_MVRC_W(PICO_INPIX2, src1);
+    PICO_OP(PICO_USE_ACC | PICO_SINGLE_VECTOR, 3, 9, 6, 3);
+
+    ST16(dst + 0*dstStride, rnd_avg32(LD16(dst + 0*dstStride), PICO_GET_W(PICO_OUTPIX0) >> 16));
+    ST16(dst + 1*dstStride, rnd_avg32(LD16(dst + 1*dstStride), PICO_GET_W(PICO_OUTPIX0)));
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_MVRC_W(PICO_INPIX0, src0);
+    PICO_MVRC_W(PICO_INPIX1, src1);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_OP(PICO_USE_ACC, 0, 6, 3, 0);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_MVRC_W(PICO_INPIX1, src4);
+    PICO_MVRC_W(PICO_INPIX0, src5);
+    PICO_OP(PICO_USE_ACC| PICO_SINGLE_VECTOR, 0, 6, 3, 0);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_OP(PICO_USE_ACC, 1, 9, 6, 3);
+    PICO_MVRC_W(PICO_INPIX0, src0);
+    PICO_MVRC_W(PICO_INPIX1, src1);
+    PICO_MVRC_W(PICO_INPIX2, src2);
+    PICO_OP(PICO_USE_ACC| PICO_SINGLE_VECTOR, 1, 9, 6, 3);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_OP(PICO_USE_ACC, 2, 6, 3, 0);
+    PICO_MVRC_W(PICO_INPIX2, src4);
+    PICO_MVRC_W(PICO_INPIX1, src5);
+    PICO_MVRC_W(PICO_INPIX0, src6);
+    PICO_OP(PICO_USE_ACC| PICO_SINGLE_VECTOR, 2, 6, 3, 0);
+
+    PICO_LDCM_W_INC(tmp,
+                    PICO_REGVECT_VMU0_OUT,
+                    PICO_REGVECT_VMU1_OUT,
+                    PICO_REGVECT_VMU2_OUT);
+    PICO_OP(PICO_USE_ACC, 3, 9, 6, 3);
+    PICO_MVRC_W(PICO_INPIX0, src1);
+    PICO_MVRC_W(PICO_INPIX1, src2);
+    PICO_MVRC_W(PICO_INPIX2, src3);
+    PICO_OP(PICO_USE_ACC| PICO_SINGLE_VECTOR, 3, 9, 6, 3);
+
+    ST16(dst + 2*dstStride, rnd_avg32(LD16(dst + 2*dstStride), PICO_GET_W(PICO_OUTPIX0) >> 16));
+    ST16(dst + 3*dstStride, rnd_avg32(LD16(dst + 3*dstStride), PICO_GET_W(PICO_OUTPIX0)));
+
+    dst += 2;
+    src += 2;
+  }
+}
+
+
+static void put_h264_qpel8_v_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  put_h264_qpel4_v_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel4_v_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+  src += 4*srcStride;
+  dst += 4*dstStride;
+  put_h264_qpel4_v_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel4_v_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+}
+
+static void avg_h264_qpel8_v_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  avg_h264_qpel4_v_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel4_v_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+  src += 4*srcStride;
+  dst += 4*dstStride;
+  avg_h264_qpel4_v_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel4_v_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+}
+
+static void put_h264_qpel8_h_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  put_h264_qpel4_h_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel4_h_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+  src += 4*srcStride;
+  dst += 4*dstStride;
+  put_h264_qpel4_h_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel4_h_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+}
+
+static void avg_h264_qpel8_h_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  avg_h264_qpel4_h_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel4_h_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+  src += 4*srcStride;
+  dst += 4*dstStride;
+  avg_h264_qpel4_h_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel4_h_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+}
+
+static void put_h264_qpel8_hv_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  put_h264_qpel4_hv_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel4_hv_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+  src += 4*srcStride;
+  dst += 4*dstStride;
+  put_h264_qpel4_hv_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel4_hv_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+}
+
+static void avg_h264_qpel8_hv_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  avg_h264_qpel4_hv_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel4_hv_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+  src += 4*srcStride;
+  dst += 4*dstStride;
+  avg_h264_qpel4_hv_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel4_hv_lowpass_pico(dst+4, src+4, dstStride, srcStride);
+}
+
+static void put_h264_qpel16_v_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  put_h264_qpel8_v_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel8_v_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+  src += 8*srcStride;
+  dst += 8*dstStride;
+  put_h264_qpel8_v_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel8_v_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+}
+
+static void avg_h264_qpel16_v_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  avg_h264_qpel8_v_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel8_v_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+  src += 8*srcStride;
+  dst += 8*dstStride;
+  avg_h264_qpel8_v_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel8_v_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+}
+
+static void put_h264_qpel16_h_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  put_h264_qpel8_h_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel8_h_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+  src += 8*srcStride;
+  dst += 8*dstStride;
+  put_h264_qpel8_h_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel8_h_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+}
+
+static void avg_h264_qpel16_h_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  avg_h264_qpel8_h_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel8_h_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+  src += 8*srcStride;
+  dst += 8*dstStride;
+  avg_h264_qpel8_h_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel8_h_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+}
+
+static void put_h264_qpel16_hv_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  put_h264_qpel8_hv_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel8_hv_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+  src += 8*srcStride;
+  dst += 8*dstStride;
+  put_h264_qpel8_hv_lowpass_pico(dst  , src  , dstStride, srcStride);
+  put_h264_qpel8_hv_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+}
+
+static void avg_h264_qpel16_hv_lowpass_pico(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){
+  avg_h264_qpel8_hv_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel8_hv_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+  src += 8*srcStride;
+  dst += 8*dstStride;
+  avg_h264_qpel8_hv_lowpass_pico(dst  , src  , dstStride, srcStride);
+  avg_h264_qpel8_hv_lowpass_pico(dst+8, src+8, dstStride, srcStride);
+}
+
+
+#define H264_MC(OPNAME, SIZE) \
+static void OPNAME ## h264_qpel ## SIZE ## _mc00_pico (uint8_t *dst, uint8_t *src, int stride){\
+    OPNAME ## pixels ## SIZE ## _c(dst, src, stride, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc10_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t half[SIZE*SIZE];\
+    put_h264_qpel ## SIZE ## _h_lowpass_pico(half, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, src, half, stride, stride, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc20_pico(uint8_t *dst, uint8_t *src, int stride){\
+    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_pico(dst, src, stride, stride);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc30_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t half[SIZE*SIZE];\
+    put_h264_qpel ## SIZE ## _h_lowpass_pico(half, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, src+1, half, stride, stride, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc01_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t full[SIZE*(SIZE+5)];\
+    uint8_t * const full_mid= full + SIZE*2;\
+    uint8_t half[SIZE*SIZE];\
+    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
+    put_h264_qpel ## SIZE ## _v_lowpass_pico(half, full_mid, SIZE, SIZE);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, full_mid, half, stride, SIZE, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc02_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t full[SIZE*(SIZE+5)];\
+    uint8_t * const full_mid= full + SIZE*2;\
+    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
+    OPNAME ## h264_qpel ## SIZE ## _v_lowpass_pico(dst, full_mid, stride, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc03_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t full[SIZE*(SIZE+5)];\
+    uint8_t * const full_mid= full + SIZE*2;\
+    uint8_t half[SIZE*SIZE];\
+    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
+    put_h264_qpel ## SIZE ## _v_lowpass_pico(half, full_mid, SIZE, SIZE);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, full_mid+SIZE, half, stride, SIZE, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc11_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t full[SIZE*(SIZE+5)];\
+    uint8_t * const full_mid= full + SIZE*2;\
+    uint8_t halfH[SIZE*SIZE];\
+    uint8_t halfV[SIZE*SIZE];\
+    put_h264_qpel ## SIZE ## _h_lowpass_pico(halfH, src, SIZE, stride);\
+    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
+    put_h264_qpel ## SIZE ## _v_lowpass_pico(halfV, full_mid, SIZE, SIZE);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc31_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t full[SIZE*(SIZE+5)];\
+    uint8_t * const full_mid= full + SIZE*2;\
+    uint8_t halfH[SIZE*SIZE];\
+    uint8_t halfV[SIZE*SIZE];\
+    put_h264_qpel ## SIZE ## _h_lowpass_pico(halfH, src, SIZE, stride);\
+    copy_block ## SIZE (full, src - stride*2 + 1, SIZE,  stride, SIZE + 5);\
+    put_h264_qpel ## SIZE ## _v_lowpass_pico(halfV, full_mid, SIZE, SIZE);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc13_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t full[SIZE*(SIZE+5)];\
+    uint8_t * const full_mid= full + SIZE*2;\
+    uint8_t halfH[SIZE*SIZE];\
+    uint8_t halfV[SIZE*SIZE];\
+    put_h264_qpel ## SIZE ## _h_lowpass_pico(halfH, src + stride, SIZE, stride);\
+    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
+    put_h264_qpel ## SIZE ## _v_lowpass_pico(halfV, full_mid, SIZE, SIZE);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc33_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t full[SIZE*(SIZE+5)];\
+    uint8_t * const full_mid= full + SIZE*2;\
+    uint8_t halfH[SIZE*SIZE];\
+    uint8_t halfV[SIZE*SIZE];\
+    put_h264_qpel ## SIZE ## _h_lowpass_pico(halfH, src + stride, SIZE, stride);\
+    copy_block ## SIZE (full, src - stride*2 + 1, SIZE,  stride, SIZE + 5);\
+    put_h264_qpel ## SIZE ## _v_lowpass_pico(halfV, full_mid, SIZE, SIZE);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc22_pico(uint8_t *dst, uint8_t *src, int stride){\
+    OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_pico(dst, src, stride, stride);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc21_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t halfH[SIZE*SIZE];\
+    uint8_t halfHV[SIZE*SIZE];\
+    put_h264_qpel ## SIZE ## _h_lowpass_pico(halfH, src, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _hv_lowpass_pico(halfHV, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc23_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t halfH[SIZE*SIZE];\
+    uint8_t halfHV[SIZE*SIZE];\
+    put_h264_qpel ## SIZE ## _h_lowpass_pico(halfH, src + stride, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _hv_lowpass_pico(halfHV, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc12_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t full[SIZE*(SIZE+5)];\
+    uint8_t * const full_mid= full + SIZE*2;\
+    uint8_t halfV[SIZE*SIZE];\
+    uint8_t halfHV[SIZE*SIZE];\
+    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
+    put_h264_qpel ## SIZE ## _v_lowpass_pico(halfV, full_mid, SIZE, SIZE);\
+    put_h264_qpel ## SIZE ## _hv_lowpass_pico(halfHV, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc32_pico(uint8_t *dst, uint8_t *src, int stride){\
+    uint8_t full[SIZE*(SIZE+5)];\
+    uint8_t * const full_mid= full + SIZE*2;\
+    uint8_t halfV[SIZE*SIZE];\
+    uint8_t halfHV[SIZE*SIZE];\
+    copy_block ## SIZE (full, src - stride*2 + 1, SIZE,  stride, SIZE + 5);\
+    put_h264_qpel ## SIZE ## _v_lowpass_pico(halfV, full_mid, SIZE, SIZE);\
+    put_h264_qpel ## SIZE ## _hv_lowpass_pico(halfHV, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\
+}\
+
+H264_MC(put_, 4)
+H264_MC(put_, 8)
+H264_MC(put_, 16)
+H264_MC(avg_, 4)
+H264_MC(avg_, 8)
+H264_MC(avg_, 16)
+
+
+#define dspfunc16(PFX) \
+    void PFX ## _pixels16_avr32(uint8_t *dst, const uint8_t *pixels, int line_size, int h ){ \
+          PFX ## _pixels8_avr32(dst, pixels, line_size, h);\
+          PFX ## _pixels8_avr32(dst + 8, pixels + 8, line_size, h);\
+    }\
+    void PFX ## _pixels16_h_avr32(uint8_t *dst, const uint8_t *pixels, int line_size, int h ){ \
+          PFX ## _pixels8_h_avr32(dst, pixels, line_size, h);\
+          PFX ## _pixels8_h_avr32(dst + 8, pixels + 8, line_size, h);\
+    }\
+    void PFX ## _pixels16_v_avr32(uint8_t *dst, const uint8_t *pixels, int line_size, int h ){ \
+          PFX ## _pixels8_v_avr32(dst, pixels, line_size, h);\
+          PFX ## _pixels8_v_avr32(dst + 8, pixels + 8, line_size, h);\
+    }\
+    void PFX ## _pixels16_hv_avr32(uint8_t *dst, const uint8_t *pixels, int line_size, int h ){ \
+          PFX ## _pixels8_hv_avr32(dst, pixels, line_size, h);\
+          PFX ## _pixels8_hv_avr32(dst + 8, pixels + 8, line_size, h);\
+    }\
+
+
+dspfunc16(put)
+dspfunc16(put_no_rnd)
+dspfunc16(avg)
+dspfunc16(avg_no_rnd)
+#undef dspfunc16
+
+
+/* Returns zero in each byte where the absolute difference between <a> and <b>
+   is not less than <compare> */
+#define PABS_DIFF_LESS_THAN( a, b, compare) \
+  ({  uint32_t __tmp__, __tmp2__, __mask__; \
+      __asm__ ( \
+      /* Check ABS( a - b ) < compare */  \
+        "psubs.ub\t%[tmp], %[opa], %[opb]\n"  \
+        "psubs.ub\t%[tmp2], %[opb], %[opa]\n" \
+        "or\t%[tmp], %[tmp2]\n" /* ABS ( a - b ) */ \
+        /* This produces 0 for all bytes where the comparison is not true */ \
+        "psubs.ub\t%[mask], %[cmp], %[tmp]\n" \
+        : [tmp] "=&r"(__tmp__), [tmp2] "=&r"(__tmp2__), [mask] "=&r"(__mask__)  \
+        : [opa] "r"(a), [opb] "r"(b), [cmp] "r"(compare)  ); \
+      __mask__; })
+
+/*
+   Set all bytes containing zero in <value> to 255 and the rest to zero.
+
+   Add with saturation 254 to all bytes making all bytes different from
+   zero become 255. Then add one without saturation to make all bytes
+   originally containing zero 255 and the rest 0. */
+#define SET_ALL_BITS_IN_ZERO_BYTES(value) \
+  ({  uint32_t __tmp__; \
+      __asm__ ( \
+        "padds.ub\t%[tmp], %[val], %[max_minus_one]\n"  \
+        "padd.b\t%[tmp], %[tmp], %[all_ones]\n" \
+        : [tmp] "=r"(__tmp__) \
+        : [val] "r"(value), [max_minus_one] "r"(0xFEFEFEFE), [all_ones] "r"(0x01010101) ); \
+      __tmp__; })
+
+#define  PACKW_SH(upper, lower) \
+  ({  uint32_t __tmp__; \
+      __asm__ ( \
+        "packw.sh\t%[tmp], %[u], %[l]\n"  \
+        : [tmp] "=r"(__tmp__) \
+        : [u] "r"(upper), [l] "r"(lower) ); \
+      __tmp__; })
+
+#define  PACKSH_UB(upper, lower) \
+  ({  uint32_t __tmp__; \
+      __asm__ ( \
+        "packsh.sb\t%[tmp], %[u], %[l]\n"  \
+        : [tmp] "=r"(__tmp__) \
+        : [u] "r"(upper), [l] "r"(lower) ); \
+      __tmp__; })
+
+static  void h264_v_loop_filter_luma_avr32(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
+{
+  int i;
+
+  if ( alpha == 0 )
+    return;
+
+  alpha = PACKW_SH(alpha, alpha);
+  alpha = PACKSH_UB(alpha, alpha);
+  beta = PACKW_SH(beta, beta);
+  beta = PACKSH_UB(beta, beta);
+
+  for( i = 0; i < 4; i++ ) {
+    int tc, tc0_p, tc0_m;
+    uint32_t p0, p1, p2, q0, q1, q2;
+    uint32_t old_p0, old_q0;
+    uint32_t mask, mask2;
+    uint32_t tmp, tmp2, tmp3, tmp4;
+
+    if( tc0[i] < 0 ) {
+      pix += 4;
+      continue;
+    }
+
+    p0 = LD32(pix - stride);
+    p1 = LD32(pix - 2*stride);
+    q0 = LD32(pix);
+    q1 = LD32(pix + stride);
+
+    /* Check which of the columns should be filtered, if any. */
+    mask = PABS_DIFF_LESS_THAN(p0, q0, alpha);
+    mask |= PABS_DIFF_LESS_THAN(p1, p0, beta);
+    mask |= PABS_DIFF_LESS_THAN(q1, q0, beta);
+
+    if ( !mask )
+      continue;
+
+    mask = SET_ALL_BITS_IN_ZERO_BYTES(mask);
+
+    tc = PACKW_SH(tc0[i], tc0[i]);
+    tc0_p = tc;
+    tc0_m = PACKW_SH(-tc0[i], -tc0[i]);
+
+    /*
+      int i_delta;
+      if( ABS( p2 - p0 ) < beta ) {
+      pix[-2*stride] = p1 + clip( (( p2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - p1, -tc0[i], tc0[i] );
+      tc++;
+      }
+    */
+
+    p2 = LD32(pix - 3*stride);
+    mask2 = PABS_DIFF_LESS_THAN(p2, p0, beta) & ~mask;
+
+    if ( mask2 ){
+      mask2 = SET_ALL_BITS_IN_ZERO_BYTES(mask2);
+      __asm__ ("pavg.ub\t%[tmp], %[p0], %[q0]\n"
+           "paddh.ub\t%[tmp], %[tmp], %[p2]\n"
+           "punpckub.h\t%[tmp2], %[tmp]:t\n"
+           "punpckub.h\t%[tmp], %[tmp]:b\n"
+           "punpckub.h\t%[tmp3], %[p1]:t\n"
+           "punpckub.h\t%[tmp4], %[p1]:b\n"
+           "psub.h\t%[tmp2], %[tmp2], %[tmp3]\n"
+           "psub.h\t%[tmp], %[tmp], %[tmp4]\n"
+           "pmin.sh\t%[tmp2], %[tmp2], %[tc0_p]\n"
+           "pmin.sh\t%[tmp], %[tmp], %[tc0_p]\n"
+           "pmax.sh\t%[tmp2], %[tmp2], %[tc0_m]\n"
+           "pmax.sh\t%[tmp], %[tmp], %[tc0_m]\n"
+           "padd.h\t%[tmp2], %[tmp2], %[tmp3]\n"
+           "padd.h\t%[tmp], %[tmp], %[tmp4]\n"
+           "packsh.ub\t%[tmp], %[tmp2], %[tmp]\n"
+           "andn\t%[tmp], %[mask2]\n"
+           "and\t%[tmp2], %[q1], %[mask2]\n"
+           "or\t%[tmp], %[tmp2]\n"
+           : [tmp]"=&r"(tmp), [tmp2]"=&r"(tmp2), [tmp3]"=&r"(tmp3),
+           [tmp4]"=&r"(tmp4)
+           : [q0]"r"(q0), [p2]"r"(p2), [p1]"r"(p1), [p0]"r"(p0), [q1]"r"(q1), [tc0_p]"r"(tc0_p),
+           [tc0_m]"r"(tc0_m), [mask2]"r"(mask2));
+      ST32(pix - 2*stride, tmp);
+      tc += 0x00010001;
+    }
+
+    q2 = LD32(pix + 2*stride);
+
+    /*
+      if( ABS( q2 - q0 ) < beta ) {
+      pix[   stride] = q1 + clip( (( q2 + ( ( p0 + q0 + 1 ) >> 1 ) ) >> 1) - q1, -tc0[i], tc0[i] );
+      tc++;
+      }
+    */
+    mask2 = PABS_DIFF_LESS_THAN(q2, q0, beta) & ~mask;
+
+    if ( mask2 ){
+      mask2 = SET_ALL_BITS_IN_ZERO_BYTES(mask2);
+      __asm__ ("pavg.ub\t%[tmp], %[p0], %[q0]\n"
+           "paddh.ub\t%[tmp], %[tmp], %[q2]\n"
+           "punpckub.h\t%[tmp2], %[tmp]:t\n"
+           "punpckub.h\t%[tmp], %[tmp]:b\n"
+           "punpckub.h\t%[tmp3], %[q1]:t\n"
+           "punpckub.h\t%[tmp4], %[q1]:b\n"
+           "psub.h\t%[tmp2], %[tmp2], %[tmp3]\n"
+           "psub.h\t%[tmp], %[tmp], %[tmp4]\n"
+           "pmin.sh\t%[tmp2], %[tmp2], %[tc0_p]\n"
+           "pmin.sh\t%[tmp], %[tmp], %[tc0_p]\n"
+           "pmax.sh\t%[tmp2], %[tmp2], %[tc0_m]\n"
+           "pmax.sh\t%[tmp], %[tmp], %[tc0_m]\n"
+           "padd.h\t%[tmp2], %[tmp2], %[tmp3]\n"
+           "padd.h\t%[tmp], %[tmp], %[tmp4]\n"
+           "packsh.ub\t%[tmp], %[tmp2], %[tmp]\n"
+           "andn\t%[tmp], %[mask2]\n"
+           "and\t%[tmp2], %[q1], %[mask2]\n"
+           "or\t%[tmp], %[tmp2]\n"
+           : [tmp]"=&r"(tmp), [tmp2]"=&r"(tmp2), [tmp3]"=&r"(tmp3),
+           [tmp4]"=&r"(tmp4)
+           : [q0]"r"(q0), [q2]"r"(q2), [q1]"r"(q1), [p0]"r"(p0), [tc0_p]"r"(tc0_p),
+           [tc0_m]"r"(tc0_m), [mask2]"r"(mask2));
+      ST32(pix + stride, tmp);
+      tc += 0x00010001;
+    }
+
+    old_p0 = p0;
+    old_q0 = q0;
+
+    /* i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
+       pix[-stride] = clip_uint8( p0 + i_delta );
+       pix[0]        = clip_uint8( q0 - i_delta );
+    */
+
+    __asm__ (
+         /* Check if the two upper pixels should be filtered */
+         "lsr\t%[tmp], %[inv_mask], 16\n"
+         "breq\t0f\n"
+
+         "punpckub.h\t%[tmp], %[p1]:t\n"
+         "punpckub.h\t%[tmp2], %[q1]:t\n"
+
+         /* p1 - q1 */
+         "psub.h\t%[tmp], %[tmp], %[tmp2]\n"
+
+         "punpckub.h\t%[tmp3], %[q0]:t\n"
+         "punpckub.h\t%[tmp4], %[p0]:t\n"
+
+         /* q0 - p0 */
+         "psub.h\t%[tmp2], %[tmp3], %[tmp4]\n"
+
+         /* (q0 - p0) << 2 */
+         "plsl.h\t%[tmp2], %[tmp2], 2\n"
+
+         /* ((q0 - p0) << 2) + (p1 - q1) */
+         "padd.h\t%[tmp2], %[tmp2], %[tmp]\n"
+
+         "mov\t%[tmp], 0x00040004\n"
+         /* ((q0 - p0) << 2) + (p1 - q1) + 4*/
+         "padd.h\t%[tmp2], %[tmp2], %[tmp]\n"
+
+         /* (((q0 - p0) << 2) + (p1 - q1) + 4) >> 3*/
+         "pasr.h\t%[tmp2], %[tmp2], 3\n"
+
+         "mov\t%[tmp], 0\n"
+         "psub.h\t%[tmp], %[tmp], %[tc]\n"
+
+         /* i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); */
+         "pmin.sh\t%[tmp2], %[tmp2], %[tc]\n"
+         "pmax.sh\t%[tmp2], %[tmp2], %[tmp]\n"
+
+         /* pix[-stride] = clip_uint8( p0 + i_delta ); */
+         "padd.h\t%[tmp4], %[tmp4], %[tmp2]\n"
+
+         /* pix[0]        = clip_uint8( q0 - i_delta ); */
+         "psub.h\t%[tmp3], %[tmp3], %[tmp2]\n"
+
+         /* Check if the two lower pixels should be filtered */
+         "lsl\t%[tmp2], %[inv_mask], 16\n"
+         "breq\t1f\n"
+
+         "0:\n"
+         "punpckub.h\t%[p1], %[p1]:b\n"
+         "punpckub.h\t%[q1], %[q1]:b\n"
+
+         /* p1 - q1 */
+         "psub.h\t%[p1], %[p1], %[q1]\n"
+
+         "punpckub.h\t%[q0], %[q0]:b\n"
+         "punpckub.h\t%[p0], %[p0]:b\n"
+
+         /* q0 - p0 */
+         "psub.h\t%[tmp2], %[q0], %[p0]\n"
+
+         /* (q0 - p0) << 2 */
+         "plsl.h\t%[tmp2], %[tmp2], 2\n"
+
+         /* ((q0 - p0) << 2) + (p1 - q1) */
+         "padd.h\t%[tmp2], %[tmp2], %[p1]\n"
+
+         "mov\t%[q1], 0x00040004\n"
+         /* ((q0 - p0) << 2) + (p1 - q1) + 4*/
+         "padd.h\t%[tmp2], %[tmp2], %[q1]\n"
+
+         /* (((q0 - p0) << 2) + (p1 - q1) + 4) >> 3*/
+         "pasr.h\t%[tmp2], %[tmp2], 3\n"
+
+         /* i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc ); */
+         "pmin.sh\t%[tmp2], %[tmp2], %[tc]\n"
+         "pmax.sh\t%[tmp2], %[tmp2], %[tmp]\n"
+
+         /* pix[-stride] = clip_uint8( p0 + i_delta ); */
+         "padd.h\t%[p0], %[p0], %[tmp2]\n"
+
+         /* pix[0]        = clip_uint8( q0 - i_delta ); */
+         "psub.h\t%[q0], %[q0], %[tmp2]\n"
+
+         "1:\n"
+         "packsh.ub\t%[p0], %[tmp4], %[p0]\n"
+         "packsh.ub\t%[q0], %[tmp3], %[tmp4]\n"
+
+         : [tmp]"=&r"(tmp), [tmp2]"=&r"(tmp2), [tmp3]"=&r"(tmp3),
+         [tmp4]"=&r"(tmp4), [q0]"=&r"(q0), [q1]"=&r"(q1), [p0]"=&r"(p0), [p1]"=&r"(p1)
+         : [tc]"r"(tc), [inv_mask]"r"(~mask));
+
+    ST32(pix - stride, (mask & old_p0) | (p0 & ~mask));
+    ST32(pix, (mask & old_q0) | (q0 & ~mask));
+
+  }
+  pix += 1;
+}
+
+
+
+
+#ifdef CHECK_DSP_FUNCS_AGAINST_C
+
+void dump_block8(uint8_t *block, int line_size, int h){
+  int i, j;
+
+  for ( i = 0; i < h ; i++ ){
+    av_log(NULL, AV_LOG_ERROR, "\t"); 
+    for ( j = 0; j < 8 ; j++ ){
+      av_log(NULL, AV_LOG_ERROR, "%d ", block[j + i*line_size]); 
+    }
+    av_log(NULL, AV_LOG_ERROR, "\n"); 
+  }
+}
+
+void dump_block4(uint8_t *block, int line_size, int h){
+  int i, j;
+
+  for ( i = 0; i < h ; i++ ){
+    av_log(NULL, AV_LOG_ERROR, "\t"); 
+    for ( j = 0; j < 4 ; j++ ){
+      av_log(NULL, AV_LOG_ERROR, "%d ", block[j + i*line_size]); 
+    }
+    av_log(NULL, AV_LOG_ERROR, "\n"); 
+  }
+}
+
+void dump_block(uint8_t *block, int line_size, int h, int w){
+  int i, j;
+
+  for ( i = 0; i < h ; i++ ){
+    av_log(NULL, AV_LOG_ERROR, "\t"); 
+    for ( j = 0; j < w ; j++ ){
+      av_log(NULL, AV_LOG_ERROR, "%d ", block[j + i*line_size]); 
+    }
+    av_log(NULL, AV_LOG_ERROR, "\n"); 
+  }
+}
+
+void check_block8(uint8_t *test, uint8_t *correct, int line_size_test, int line_size_correct, 
+                  int h, char *name, int max_dev){
+  int i,j;
+  for ( i = 0; i < 8 ; i++ ){
+    for ( j = 0; j < h ; j++ ){
+      int diff = test[i + line_size_test*j] - correct[i + line_size_correct*j];
+      diff = diff < 0 ? -diff : diff;
+      if ( diff > max_dev ){
+        av_log(NULL, AV_LOG_ERROR, "Error pixel x=%i, y=%i differs. Is 0x%x should be 0x%x\n", 
+               i, j, test[i + line_size_test*j], correct[i + j*line_size_correct]);        
+        av_log(NULL, AV_LOG_ERROR, "Error resulting block from %s is:\n", name);
+        dump_block8(test, line_size_test, h); 
+        av_log(NULL, AV_LOG_ERROR, "But should be equal to:\n");
+        dump_block8(correct, line_size_correct, h);         
+        exit(1);
+      }
+    }
+  }  
+}
+
+void check_block4(uint8_t *test, uint8_t *correct, int line_size_test, int line_size_correct, 
+                  int h, char *name, int max_dev){
+  int i,j;
+  for ( i = 0; i < 4 ; i++ ){
+    for ( j = 0; j < h ; j++ ){
+      int diff = test[i + line_size_test*j] - correct[i + line_size_correct*j];
+      diff = diff < 0 ? -diff : diff;
+      if ( diff > max_dev ){
+        av_log(NULL, AV_LOG_ERROR, "Error pixel x=%i, y=%i differs. Is 0x%x should be 0x%x\n", 
+               i, j, test[i + line_size_test*j], correct[i + j*line_size_correct]);        
+        av_log(NULL, AV_LOG_ERROR, "Error resulting block from %s is:\n", name);
+        dump_block8(test, line_size_test, h); 
+        av_log(NULL, AV_LOG_ERROR, "But should be equal to:\n");
+        dump_block4(correct, line_size_correct, h);         
+        exit(1);
+      }
+    }
+  }  
+}
+
+void check_block(uint8_t *test, uint8_t *correct, int line_size_test, int line_size_correct, 
+                 int h, int width, char *name, int max_dev){
+  int i,j;
+  for ( i = 0; i < width ; i++ ){
+    for ( j = 0; j < h ; j++ ){
+      int diff = test[i + line_size_test*j] - correct[i + line_size_correct*j];
+      diff = diff < 0 ? -diff : diff;
+      if ( diff > max_dev ){
+        av_log(NULL, AV_LOG_ERROR, "Error pixel x=%i, y=%i differs. Is 0x%x should be 0x%x\n", 
+               i, j, test[i + line_size_test*j], correct[i + j*line_size_correct]);        
+        av_log(NULL, AV_LOG_ERROR, "Error resulting block from %s is:\n", name);
+        dump_block(test, line_size_test, h, width); 
+        av_log(NULL, AV_LOG_ERROR, "But should be equal to:\n");
+        dump_block(correct, line_size_correct, h, width);         
+        exit(1);
+      }
+    }
+  }  
+}
+
+void dump_dct_block(DCTELEM *block){
+  int i, j;
+
+  for ( i = 0; i < 8 ; i++ ){
+    av_log(NULL, AV_LOG_ERROR, "\t"); 
+    for ( j = 0; j < 8 ; j++ ){
+      av_log(NULL, AV_LOG_ERROR, "0x%x ", block[j + i*8]); 
+    }
+    av_log(NULL, AV_LOG_ERROR, "\n"); 
+  }
+}
+
+void test_idct_avr32(DCTELEM *block){
+  DCTELEM testBlock[64];
+  int i, j;
+
+  /* Copy transposed block to testBlock */ 
+  for ( i = 0; i < 8 ; i++ ){
+    for ( j = 0; j < 8 ; j++ ){
+      testBlock[i + 8*j] = block[j + i*8]; 
+    }
+  }
+  
+  idct_avr32(block);
+  simple_idct(&testBlock);
+  
+  for ( i = 0; i < 64 ; i++ ){
+    if ( block[i] != testBlock[i] ){
+      av_log(NULL, AV_LOG_ERROR, "Error resulting block from idct is:\n");
+      dump_dct_block(block); 
+      av_log(NULL, AV_LOG_ERROR, "But should be equal to the transposed of:\n");
+      dump_dct_block(testBlock);         
+      exit(1);
+    }
+  }
+}
+
+void test_idct_put_avr32(uint8_t *dest, int line_size, DCTELEM *block){
+  uint8_t testBlock[64];
+  DCTELEM blockCopy[64];
+  int i, j;
+
+  /* Copy transposed block to blockCopy */ 
+  for ( i = 0; i < 8 ; i++ ){
+    for ( j = 0; j < 8 ; j++ ){
+      blockCopy[i + 8*j] = block[j + i*8]; 
+    }
+  }
+
+  idct_put_avr32(dest, line_size, block);
+  simple_idct_put(&testBlock, 8, blockCopy);
+  
+  check_block8(dest, testBlock, line_size, 8, 8, "idct_put", 1);
+}
+
+
+void test_idct_add_avr32(uint8_t *dest, int line_size, DCTELEM *block){
+  uint8_t testBlock[64];
+  DCTELEM blockCopy[64];
+  int i, j;
+  
+  /* Copy dest to testBlock */ 
+  for ( i = 0; i < 8 ; i++ ){
+    for ( j = 0; j < 8 ; j++ ){
+      testBlock[i + 8*j] = dest[i + j*line_size]; 
+    }
+  }
+
+  /* Copy transposed block to blockCopy */ 
+  for ( i = 0; i < 8 ; i++ ){
+    for ( j = 0; j < 8 ; j++ ){
+      blockCopy[i + 8*j] = block[j + i*8]; 
+    }
+  }
+
+  idct_add_avr32(dest, line_size, block);
+  simple_idct_add(&testBlock, 8, blockCopy);
+  
+  check_block8(dest, testBlock, line_size, 8, 8, "idct_add", 1);
+}
+
+void test_h264_idct_add_avr32(uint8_t *dest, DCTELEM *block, int stride){
+  uint8_t testBlock[16];
+  DCTELEM blockCopy[16];
+  int i, j;
+  
+  /* Copy dest to testBlock */ 
+  for ( i = 0; i < 4 ; i++ ){
+    for ( j = 0; j < 4 ; j++ ){
+      testBlock[i + 4*j] = dest[i + j*stride]; 
+    }
+  }
+
+  /* Copy transposed block to blockCopy */ 
+  for ( i = 0; i < 16 ; i++ ){
+    blockCopy[i] = block[i]; 
+  }
+
+  ff_h264_idct_add_c(dest, block, stride);
+  
+  h264_idct_add_avr32(testBlock, blockCopy, 4);
+  
+  check_block(dest, testBlock, stride, 4, 4, 4, "h264_idct_add", 0);
+}
+
+void test_h264_idct8_add_avr32(uint8_t *dest, DCTELEM *block, int stride){
+  uint8_t testBlock[8*8];
+  DCTELEM blockCopy[8*8];
+  int i, j;
+  
+  /* Copy dest to testBlock */ 
+  for ( i = 0; i < 8 ; i++ ){
+    for ( j = 0; j < 8 ; j++ ){
+      testBlock[i + 8*j] = dest[i + j*stride]; 
+    }
+  }
+
+  /* Copy source block to blockCopy */ 
+  for ( i = 0; i < 8*8 ; i++ ){
+    blockCopy[i] = block[i]; 
+  }
+
+  ff_h264_idct8_add_c(dest, block, stride);
+  h264_idct8_add_avr32(testBlock, blockCopy, 8);
+  
+  check_block(dest, testBlock, stride, 8, 8, 8, "h264_idct8_add", 0);
+}
+
+void test_put_pixels_funcs8(op_pixels_func test, op_pixels_func correct, uint8_t *block, 
+                       const uint8_t *pixels, int line_size, int h, char *name, int in_h_size, int in_v_size){
+  uint8_t *testBlock, *testBlock2;
+  int i, j;
+  int input_v_size = h + in_v_size;
+  int input_h_size = 8 + in_h_size;
+
+  testBlock = alloca(input_h_size*input_v_size);
+  testBlock2 = alloca(input_h_size*input_v_size);
+
+  for ( i = 0; i < input_h_size ; i++ ){
+    for ( j = 0; j < input_v_size ; j++ ){
+      testBlock[i + input_h_size*j] = pixels[i + j*line_size]; 
+    }
+  }
+
+  test(block, pixels, line_size, h);
+  correct(testBlock2, testBlock, input_h_size, h);
+  
+  check_block8(block, testBlock2, line_size, input_h_size, h, name, 0);
+
+} 
+
+void test_h264_chroma_mc_funcs(h264_chroma_mc_func test, h264_chroma_mc_func correct, uint8_t *dst, 
+                               uint8_t *src, int stride, int h, int w, int x, int y, char *name){
+  uint8_t *testBlock, *testBlock2;
+  int i, j;
+  int input_v_size = h + 1;
+  int input_h_size = ((w + 1) + 3) & ~3;
+
+  testBlock = alloca(input_h_size*input_v_size);
+  testBlock2 = alloca(input_h_size*input_v_size);
+
+  for ( i = 0; i < w + 1 ; i++ ){
+    for ( j = 0; j < h + 1 ; j++ ){
+      testBlock[i + input_h_size*j] = src[i + j*stride]; 
+    }
+  }
+
+  for ( i = 0; i < w ; i++ ){
+    for ( j = 0; j < h ; j++ ){
+      testBlock2[i + input_h_size*j] = dst[i + j*stride]; 
+    }
+  }
+
+  test(dst, src, stride, h, x, y);
+  correct(testBlock2, testBlock, input_h_size, h, x, y);
+  
+  check_block(dst, testBlock2, stride, input_h_size, h, w, name, 0);
+
+} 
+
+void test_qpel_mc_funcs(qpel_mc_func test, qpel_mc_func correct, uint8_t *dst, 
+                        uint8_t *src, int stride, int size, char *name){
+  uint8_t *testBlock, *testBlock2;
+  int i, j;
+  int test_stride = size + 8;
+
+  testBlock = alloca(test_stride*(size+8)) + 4 + test_stride*4;
+  testBlock2 = alloca(test_stride*size);
+  
+  for ( i = -4; i < size+4 ; i++ ){
+    for ( j = -4; j < size+4 ; j++ ){
+      testBlock[i + test_stride*j] = src[i + j*stride]; 
+    }
+  }
+  
+  for ( i = 0; i < size ; i++ ){
+    for ( j = 0; j < size ; j++ ){
+      testBlock2[i + test_stride*j] = dst[i + j*stride]; 
+    }
+  }
+
+  correct(dst, src, stride);
+  test(testBlock2, testBlock, test_stride);
+  
+  check_block(testBlock2, dst, test_stride, stride, size, size, name, 0);
+
+} 
+
+
+#define test_pixels_funcs(PFX, NUM ) \
+void test_ ## PFX ## _pixels ## NUM ## _avr32( uint8_t *block, const uint8_t *pixels, int line_size, int h){ \
+  test_put_pixels_funcs8(PFX ## _pixels ## NUM ## _avr32, PFX ## _pixels ## NUM ## _c, \
+                         block, pixels, line_size, h, "test_" #PFX "_pixels", 0, 0); } \
+void test_ ## PFX ## _pixels ## NUM ## _h_avr32( uint8_t *block, const uint8_t *pixels, int line_size, int h){ \
+  test_put_pixels_funcs8(PFX ## _pixels ## NUM ## _h_avr32, PFX ## _pixels ## NUM ## _x2_c, \
+                         block, pixels, line_size, h, "test_" #PFX "_pixels_h", 1, 0); } \
+void test_ ## PFX ## _pixels ## NUM ## _v_avr32( uint8_t *block, const uint8_t *pixels, int line_size, int h){ \
+  test_put_pixels_funcs8(PFX ## _pixels ## NUM ## _v_avr32, PFX ## _pixels ## NUM ## _y2_c, \
+                         block, pixels, line_size, h, "test_" #PFX "_pixels_v", 0, 1); } \
+void test_ ## PFX ## _pixels ## NUM ## _hv_avr32( uint8_t *block, const uint8_t *pixels, int line_size, int h){ \
+  test_put_pixels_funcs8(PFX ## _pixels ## NUM ## _hv_avr32, PFX ## _pixels ## NUM ## _xy2_c, \
+                         block, pixels, line_size, h, "test_" #PFX "_pixels_hv", 1, 1); }
+
+test_pixels_funcs(put, 8);
+test_pixels_funcs(put_no_rnd, 8);
+test_pixels_funcs(put, 16);
+test_pixels_funcs(put_no_rnd, 16);
+
+test_pixels_funcs(avg, 8);
+test_pixels_funcs(avg_no_rnd, 8);
+test_pixels_funcs(avg, 16);
+test_pixels_funcs(avg_no_rnd, 16);
+
+#define test_h264_chroma_mc_funcs(PFX, NUM ) \
+void test_ ## PFX ## _h264_chroma_mc ## NUM ## _pico( uint8_t *dst, uint8_t *src, int stride, int h, int x, int y){ \
+  test_h264_chroma_mc_funcs(PFX ## _h264_chroma_mc ## NUM ## _pico, PFX ## _h264_chroma_mc ## NUM ## _c, \
+                         dst, src, stride, h, NUM, x, y, "test_" #PFX "_h264_chroma_mc" #NUM "_pico"); } \
+
+test_h264_chroma_mc_funcs(put, 2);
+test_h264_chroma_mc_funcs(put, 4);
+test_h264_chroma_mc_funcs(put, 8);
+test_h264_chroma_mc_funcs(avg, 2);
+test_h264_chroma_mc_funcs(avg, 4);
+test_h264_chroma_mc_funcs(avg, 8);
+
+#define test_qpel_mc_funcs_type(PFX, NUM, TYPE ) \
+void test_ ## PFX ## NUM ## _ ## TYPE ## _pico( uint8_t *dst, uint8_t *src, int stride){ \
+  test_qpel_mc_funcs(PFX ## NUM ## _ ## TYPE ## _pico, PFX ## NUM ## _ ## TYPE ## _c, \
+                         dst, src, stride, NUM, "test_" #PFX #NUM "_" #TYPE "_pico"); } 
+
+#define test_qpel_mc_funcs(PFX, NUM) \
+  test_qpel_mc_funcs_type(PFX, NUM, mc00);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc10);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc20);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc30);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc01);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc11);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc21);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc31);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc02);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc12);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc22);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc32);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc03);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc13);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc23);\
+  test_qpel_mc_funcs_type(PFX, NUM, mc33)
+
+test_qpel_mc_funcs(put_h264_qpel, 4);
+test_qpel_mc_funcs(put_h264_qpel, 8);
+test_qpel_mc_funcs(put_h264_qpel, 16);
+test_qpel_mc_funcs(avg_h264_qpel, 4);
+test_qpel_mc_funcs(avg_h264_qpel, 8);
+test_qpel_mc_funcs(avg_h264_qpel, 16);
+
+
+#define dspfunc(PFX, IDX, NUM) \
+    c->PFX ## _pixels_tab[IDX][ 0] = DSP_FUNC_NAME( PFX ## NUM ## _mc00_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 1] = DSP_FUNC_NAME( PFX ## NUM ## _mc10_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 2] = DSP_FUNC_NAME( PFX ## NUM ## _mc20_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 3] = DSP_FUNC_NAME( PFX ## NUM ## _mc30_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 4] = DSP_FUNC_NAME( PFX ## NUM ## _mc01_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 5] = DSP_FUNC_NAME( PFX ## NUM ## _mc11_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 6] = DSP_FUNC_NAME( PFX ## NUM ## _mc21_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 7] = DSP_FUNC_NAME( PFX ## NUM ## _mc31_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 8] = DSP_FUNC_NAME( PFX ## NUM ## _mc02_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 9] = DSP_FUNC_NAME( PFX ## NUM ## _mc12_pico ); \
+    c->PFX ## _pixels_tab[IDX][10] = DSP_FUNC_NAME( PFX ## NUM ## _mc22_pico ); \
+    c->PFX ## _pixels_tab[IDX][11] = DSP_FUNC_NAME( PFX ## NUM ## _mc32_pico ); \
+    c->PFX ## _pixels_tab[IDX][12] = DSP_FUNC_NAME( PFX ## NUM ## _mc03_pico ); \
+    c->PFX ## _pixels_tab[IDX][13] = DSP_FUNC_NAME( PFX ## NUM ## _mc13_pico ); \
+    c->PFX ## _pixels_tab[IDX][14] = DSP_FUNC_NAME( PFX ## NUM ## _mc23_pico ); \
+    c->PFX ## _pixels_tab[IDX][15] = DSP_FUNC_NAME( PFX ## NUM ## _mc33_pico )
+
+#endif
+
+void dsputil_init_avr32(DSPContext* c, AVCodecContext *avctx)
+{
+
+  /* H264 */
+
+  if ( 0 /*avr32_use_pico*/ ){
+    c->put_h264_chroma_pixels_tab[0]= DSP_FUNC_NAME(put_h264_chroma_mc8_pico);
+    c->put_h264_chroma_pixels_tab[1]= DSP_FUNC_NAME(put_h264_chroma_mc4_pico);
+    c->put_h264_chroma_pixels_tab[2]= DSP_FUNC_NAME(put_h264_chroma_mc2_pico);
+
+    c->avg_h264_chroma_pixels_tab[0]= DSP_FUNC_NAME(avg_h264_chroma_mc8_pico);
+    c->avg_h264_chroma_pixels_tab[1]= DSP_FUNC_NAME(avg_h264_chroma_mc4_pico);
+    c->avg_h264_chroma_pixels_tab[2]= DSP_FUNC_NAME(avg_h264_chroma_mc2_pico);
+  }
+
+#define dspfunc(PFX, IDX, NUM) \
+    c->PFX ## _pixels_tab[IDX][ 0] = DSP_FUNC_NAME( PFX ## NUM ## _mc00_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 1] = DSP_FUNC_NAME( PFX ## NUM ## _mc10_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 2] = DSP_FUNC_NAME( PFX ## NUM ## _mc20_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 3] = DSP_FUNC_NAME( PFX ## NUM ## _mc30_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 4] = DSP_FUNC_NAME( PFX ## NUM ## _mc01_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 5] = DSP_FUNC_NAME( PFX ## NUM ## _mc11_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 6] = DSP_FUNC_NAME( PFX ## NUM ## _mc21_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 7] = DSP_FUNC_NAME( PFX ## NUM ## _mc31_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 8] = DSP_FUNC_NAME( PFX ## NUM ## _mc02_pico ); \
+    c->PFX ## _pixels_tab[IDX][ 9] = DSP_FUNC_NAME( PFX ## NUM ## _mc12_pico ); \
+    c->PFX ## _pixels_tab[IDX][10] = DSP_FUNC_NAME( PFX ## NUM ## _mc22_pico ); \
+    c->PFX ## _pixels_tab[IDX][11] = DSP_FUNC_NAME( PFX ## NUM ## _mc32_pico ); \
+    c->PFX ## _pixels_tab[IDX][12] = DSP_FUNC_NAME( PFX ## NUM ## _mc03_pico ); \
+    c->PFX ## _pixels_tab[IDX][13] = DSP_FUNC_NAME( PFX ## NUM ## _mc13_pico ); \
+    c->PFX ## _pixels_tab[IDX][14] = DSP_FUNC_NAME( PFX ## NUM ## _mc23_pico ); \
+    c->PFX ## _pixels_tab[IDX][15] = DSP_FUNC_NAME( PFX ## NUM ## _mc33_pico )
+
+  if ( avr32_use_pico ){
+    dspfunc(put_h264_qpel, 0, 16);
+    dspfunc(put_h264_qpel, 1, 8);
+    dspfunc(put_h264_qpel, 2, 4);
+    dspfunc(avg_h264_qpel, 0, 16);
+    dspfunc(avg_h264_qpel, 1, 8);
+    dspfunc(avg_h264_qpel, 2, 4);
+  }
+
+  c->idct_put= DSP_FUNC_NAME(idct_put_avr32);
+  c->idct_add= DSP_FUNC_NAME(idct_add_avr32);
+  c->idct    = DSP_FUNC_NAME(idct_avr32);
+  c->h264_idct_add = DSP_FUNC_NAME(h264_idct_add_avr32);
+  c->h264_idct8_add = DSP_FUNC_NAME(h264_idct8_add_avr32);
+
+  /*
+  c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_avr32;
+  */
+
+  c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
+
+  c->fdct = fdct_avr32;
+
+  c->clear_blocks = clear_blocks_avr32;
+
+#undef dspfunc
+#define dspfunc(PFX, IDX, NUM) \
+    c->PFX ## _pixels_tab[IDX][0] = DSP_FUNC_NAME( PFX ## _pixels ## NUM ## _avr32 );  \
+    c->PFX ## _pixels_tab[IDX][1] = DSP_FUNC_NAME( PFX ## _pixels ## NUM ## _h_avr32); \
+    c->PFX ## _pixels_tab[IDX][2] = DSP_FUNC_NAME( PFX ## _pixels ## NUM ## _v_avr32); \
+    c->PFX ## _pixels_tab[IDX][3] = DSP_FUNC_NAME( PFX ## _pixels ## NUM ## _hv_avr32)
+
+  dspfunc(put, 0, 16);
+  dspfunc(put_no_rnd, 0, 16);
+  dspfunc(put, 1, 8);
+  dspfunc(put_no_rnd, 1, 8);
+
+  dspfunc(avg, 1, 8);
+  dspfunc(avg_no_rnd, 1, 8);
+  dspfunc(avg, 0, 16);
+  dspfunc(avg_no_rnd, 0, 16);
+#undef dspfunc
+
+}
+
diff --git a/libavcodec/avr32/fdct.S b/libavcodec/avr32/fdct.S
new file mode 100644
index 0000000..ceede08
--- /dev/null
+++ b/libavcodec/avr32/fdct.S
@@ -0,0 +1,538 @@
+/*
+ * Copyright (c) 2007 Atmel Corporation. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials provided
+ * with the distribution.
+ *
+ * 3. The name of ATMEL may not be used to endorse or promote products
+ * derived from this software without specific prior written
+ * permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ATMEL ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE EXPRESSLY AND SPECIFICALLY DISCLAIMED. IN NO EVENT SHALL ATMEL
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
+ * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
+ * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+//**********************************************************
+//* 2-D fDCT, Based on:                                    *
+//* C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical *
+//* Fast 1-D DCT Algorithms with 11 Multiplications",      *
+//* Proc. Int'l. Conf. on Acoustics, Speech, and Signal    *
+//* Processing 1989 (ICASSP '89), pp. 988-991.             *
+//*                                                        *
+//* Fixed point implementation optimized for the AVR32     *
+//* instruction set. If a table is used for the            *
+//* coeffisients we can load two and two of them from      *
+//* This will give a reduction of                          *
+//*                                                        *
+//*                                                        *
+//**********************************************************
+
+
+/* This routine is a slow-but-accurate integer implementation of the
+ * forward DCT (Discrete Cosine Transform). Taken from the IJG software
+ *
+ * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
+ * on each column.  Direct algorithms are also available, but they are
+ * much more complex and seem not to be any faster when reduced to code.
+ *
+ * This implementation is based on an algorithm described in
+ *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
+ *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
+ *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
+ * The primary algorithm described there uses 11 multiplies and 29 adds.
+ * We use their alternate method with 12 multiplies and 32 adds.
+ * The advantage of this method is that no data path contains more than one
+ * multiplication; this allows a very simple and accurate implementation in
+ * scaled fixed-point arithmetic, with a minimal number of shifts.
+ *
+ * The poop on this scaling stuff is as follows:
+ *
+ * Each 1-D DCT step produces outputs which are a factor of sqrt(N)
+ * larger than the true DCT outputs.  The final outputs are therefore
+ * a factor of N larger than desired; since N=8 this can be cured by
+ * a simple right shift at the end of the algorithm.  The advantage of
+ * this arrangement is that we save two multiplications per 1-D DCT,
+ * because the y0 and y4 outputs need not be divided by sqrt(N).
+ * In the IJG code, this factor of 8 is removed by the quantization step
+ * (in jcdctmgr.c), here it is removed.
+ *
+ * We have to do addition and subtraction of the integer inputs, which
+ * is no problem, and multiplication by fractional constants, which is
+ * a problem to do in integer arithmetic.  We multiply all the constants
+ * by CONST_SCALE and convert them to integer constants (thus retaining
+ * CONST_BITS bits of precision in the constants).  After doing a
+ * multiplication we have to divide the product by CONST_SCALE, with proper
+ * rounding, to produce the correct output.  This division can be done
+ * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
+ * as long as possible so that partial sums can be added together with
+ * full fractional precision.
+ *
+ * The outputs of the first pass are scaled up by PASS1_BITS bits so that
+ * they are represented to better-than-integral precision.  These outputs
+ * require 8 + PASS1_BITS + 3 bits; this fits in a 16-bit word
+ * with the recommended scaling.  (For 12-bit sample data, the intermediate
+ * array is INT32 anyway.)
+ *
+ * To avoid overflow of the 32-bit intermediate results in pass 2, we must
+ * have 8 + CONST_BITS + PASS1_BITS <= 26.  Error analysis
+ * shows that the values given below are the most effective.
+ *
+ * We can gain a little more speed, with a further compromise in accuracy,
+ * by omitting the addition in a descaling shift.  This yields an incorrectly
+ * rounded result half the time...
+ */
+
+	.global		fdct_avr32
+
+
+
+#define CONST_BITS  13
+#define PASS1_BITS  2
+
+#define FIX_0_298631336  2446	/* FIX(0.298631336) */
+#define FIX_0_390180644  3196	/* FIX(0.390180644) */
+#define FIX_0_541196100  4433	/* FIX(0.541196100) */
+#define FIX_0_765366865  6270	/* FIX(0.765366865) */
+#define FIX_0_899976223  7373	/* FIX(0.899976223) */
+#define FIX_1_175875602  9633	/* FIX(1.175875602) */
+#define FIX_1_501321110  12299	/* FIX(1.501321110) */
+#define FIX_1_847759065  15137	/* FIX(1.847759065) */
+#define FIX_1_961570560  16069	/* FIX(1.961570560) */
+#define FIX_2_053119869  16819	/* FIX(2.053119869) */
+#define FIX_2_562915447  20995	/* FIX(2.562915447) */
+#define FIX_3_072711026  25172	/* FIX(3.072711026) */
+
+
+/*
+ * Perform an integer forward DCT on one block of samples.
+ */
+
+//void
+//fdct_int32(short *const block)
+//{
+//	int tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+//	int tmp10, tmp11, tmp12, tmp13;
+//	int z1, z2, z3, z4, z5;
+//	short *blkptr;
+//	int *dataptr;
+//	int data[64];
+//	int i;
+//
+//       /* Pass 1: process rows. */
+//       /* Note results are scaled up by sqrt(8) compared to a true DCT; */
+//       /* furthermore, we scale the results by 2**PASS1_BITS. */
+//
+//	dataptr = data;
+//	blkptr = block;
+
+	.text
+fdct_avr32:
+	pushm		r0-r3, r4-r7, lr
+#define		loop_ctr	r0
+#define		blkptr		r12
+#define		x0		r1
+#define		x1		r2
+#define		x2		r3
+#define		x3		r4
+#define		x4		r5
+#define		x5		r6
+#define		x6		r7
+#define		x7		r8
+#define		tmp0		r5
+#define		tmp7		r2
+#define		tmp1		r3
+#define		tmp6		r4
+#define		tmp2		r9
+#define		tmp5		r8
+#define		tmp3		r7
+#define		tmp4		r6
+
+
+	mov		loop_ctr, 8
+//	for (i = 0; i < 8; i++) {
+ROW_LOOP:
+
+	ldm		blkptr, r1, r2, r3, r4
+
+//		tmp2 = blkptr[2] + blkptr[5];
+//		tmp3 = blkptr[3] + blkptr[4];
+	paddx.h		r5, r3, r2
+//		tmp5 = blkptr[2] - blkptr[5];
+//		tmp4 = blkptr[3] - blkptr[4];
+	psubx.h		r6, r3, r2
+//		tmp0 = blkptr[0] + blkptr[7];
+//		tmp1 = blkptr[1] + blkptr[6];
+	paddx.h		r2, r4, r1
+//		tmp7 = blkptr[0] - blkptr[7];
+//		tmp6 = blkptr[1] - blkptr[6];
+	psubx.h		r3, r4, r1
+
+//		/* Even part per LL&M figure 1 --- note that published figure is faulty;
+//		 * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
+//		 */
+
+#define		tmp10		r1
+#define		tmp13		r5
+#define		tmp11		r7
+#define		tmp12		r3
+#define		z1		r9
+
+//		tmp10 = tmp0 + tmp3;
+//		tmp13 = tmp0 - tmp3;
+	paddsub.h	r1, r2:t, r5:b
+//		tmp11 = tmp1 + tmp2;
+//		tmp12 = tmp1 - tmp2;
+	paddsub.h	r4, r2:b, r5:t
+
+
+//		dataptr[0] = (tmp10 + tmp11) << PASS1_BITS;
+//		dataptr[4] = (tmp10 - tmp11) << PASS1_BITS;
+	paddsub.h	r7, r1:t, r4:t
+	ld.w		r10, pc[const_table - .]
+	plsl.h		r7, r7, PASS1_BITS
+
+//		z1 = (tmp12 + tmp13) * FIX_0_541196100;
+	addhh.w		r8, r4:b, r1:b
+	mulhh.w		r8, r8:b, r10:t
+
+//		dataptr[2] =
+//			DESCALE(z1 + tmp13 * FIX_0_765366865, CONST_BITS - PASS1_BITS);
+//		dataptr[6] =
+//			DESCALE(z1 + tmp12 * (-FIX_1_847759065), CONST_BITS - PASS1_BITS);
+	mulhh.w		r9, r1:b, r10:b
+	ld.w		r10, pc[const_table - . + 4]
+	add		r1, r8, r9
+	satrnds		r1 >> (CONST_BITS - PASS1_BITS), 31
+
+	mulhh.w		r9, r4:b, r10:t
+	add		r4, r8, r9
+	satrnds		r4 >> (CONST_BITS - PASS1_BITS), 31
+
+
+//		/* Odd part per figure 8 --- note paper omits factor of sqrt(2).
+//		 * cK represents cos(K*pi/16).
+//		 * i0..i3 in the paper are tmp4..tmp7 here.
+//		 */
+
+#define		z2		r5
+#define		z3		r6
+#define		z4		r7
+#define		z5		r8
+
+//		z4 = tmp5 + tmp7;
+//		z3 = tmp4 + tmp6;
+	padd.h		r2, r6, r3
+//		z2 = tmp5 + tmp6;
+//		z1 = tmp4 + tmp7;
+	paddx.h		r5, r6, r3
+
+	lddpc		r9, pc[const_table - . + 8]
+//		z5 = (z3 + z4) * FIX_1_175875602; /* sqrt(2) * c3 */
+	addhh.w		r8, r2:t, r2:b
+	mulhh.w		r8, r8:b, r10:b
+	lddpc		r10, pc[const_table - . + 12]
+
+
+//		tmp4 *= FIX_0_298631336; /* sqrt(2) * (-c1+c3+c5-c7) */
+	mulhh.w		r11, r6:b, r9:t
+
+//		tmp5 *= FIX_2_053119869; /* sqrt(2) * ( c1+c3-c5+c7) */
+	mulhh.w		r6, r6:t, r9:b
+
+//		tmp6 *= FIX_3_072711026; /* sqrt(2) * ( c1+c3+c5-c7) */
+	lddpc		r9, pc[const_table - . + 20]
+	mulhh.w		lr, r3:b, r10:t
+
+//		tmp7 *= FIX_1_501321110; /* sqrt(2) * ( c1+c3-c5-c7) */
+	mulhh.w		r3, r3:t, r10:b
+
+//		z3 *= -FIX_1_961570560;	/* sqrt(2) * (-c3-c5) */
+	mulhh.w		r10, r2:b, r9:t
+
+//		z4 *= -FIX_0_390180644;	/* sqrt(2) * (c5-c3) */
+	mulhh.w		r2, r2:t, r9:b
+	lddpc		r9, pc[const_table - . + 16]
+//		z3 += z5;
+//		z4 += z5;
+	add		r10, r8
+	add		r2, r8
+
+//		z1 *= -FIX_0_899976223;	/* sqrt(2) * (c7-c3) */
+	mulhh.w		r8, r5:b, r9:t
+
+//		z2 *= -FIX_2_562915447;	/* sqrt(2) * (-c1-c3) */
+	mulhh.w		r5, r5:t, r9:b
+
+//		dataptr[7] = DESCALE(tmp4 + z1 + z3, CONST_BITS - PASS1_BITS);
+	add		r11, r8
+	add		r11, r10
+	satrnds		r11 >> (CONST_BITS - PASS1_BITS), 31
+
+//		dataptr[5] = DESCALE(tmp5 + z2 + z4, CONST_BITS - PASS1_BITS);
+	add		r6, r5
+
+	sthh.w		blkptr[6*2], r4:b, r11:b
+	add		r6, r2
+	satrnds		r6 >> (CONST_BITS - PASS1_BITS), 31
+
+//		dataptr[3] = DESCALE(tmp6 + z2 + z3, CONST_BITS - PASS1_BITS);
+	add		lr, r5
+	sthh.w		blkptr[4*2], r7:b, r6:b
+	add		lr, r10
+	satrnds		lr >> (CONST_BITS - PASS1_BITS), 31
+
+//		dataptr[1] = DESCALE(tmp7 + z1 + z4, CONST_BITS - PASS1_BITS);
+	add		r3, r8
+	sthh.w		blkptr[2*2], r1:b, lr:b
+	add		r3, r2
+	satrnds		r3 >> (CONST_BITS - PASS1_BITS), 31
+
+
+
+//		dataptr += 8; /* advance pointer to next row */
+//		blkptr += 8;
+	sthh.w		blkptr[0], r7:t, r3:b
+	sub		blkptr, -16
+	sub		loop_ctr, 1
+	brne		ROW_LOOP
+
+//	}
+
+	/* Pass 2: process columns.
+	 * We remove the PASS1_BITS scaling, but leave the results scaled up
+	 * by an overall factor of 8.
+	 */
+
+//	dataptr = data;
+	sub		blkptr, 128
+
+	mov		loop_ctr, 4
+//	for (i = 0; i < 8; i++) {
+COLOUMN_LOOP:
+	ld.w		r1, blkptr[0]
+	ld.w		r2, blkptr[1*8*2]
+	ld.w		r3, blkptr[2*8*2]
+	ld.w		r4, blkptr[3*8*2]
+	ld.w		r5, blkptr[4*8*2]
+	ld.w		r6, blkptr[5*8*2]
+	ld.w		r7, blkptr[6*8*2]
+	ld.w		r8, blkptr[7*8*2]
+
+//		tmp0 = blkptr[0] + blkptr[7*8];
+	padds.sh	r9, r1, r8
+//		tmp7 = blkptr[0] - blkptr[7*8];
+	psubs.sh	r1, r1, r8
+//		tmp1 = blkptr[1*8] + blkptr[6*8];
+	padds.sh	r8, r2, r7
+//		tmp6 = blkptr[1*8] - blkptr[6*8];
+	psubs.sh	r2, r2, r7
+//		tmp2 = blkptr[2*8] + blkptr[5*8];
+	padds.sh	r7, r3, r6
+//		tmp5 = blkptr[2*8] - blkptr[5*8];
+	psubs.sh	r3, r3, r6
+//		tmp3 = blkptr[3*8] + blkptr[4*8];
+	padds.sh	r6, r4, r5
+//		tmp4 = blkptr[3*8] - blkptr[4*8];
+	psubs.sh	r4, r4, r5
+
+//		/* even part per ll&m figure 1 --- note that published figure is faulty;
+//		 * rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
+//		 */
+//
+//		tmp10 = tmp0 + tmp3;
+	padds.sh	r5, r9, r6
+//		tmp13 = tmp0 - tmp3;
+	psubs.sh	r9, r9, r6
+//		tmp11 = tmp1 + tmp2;
+	padds.sh	r6, r8, r7
+//		tmp12 = tmp1 - tmp2;
+	psubs.sh	r8, r8, r7
+
+//		dataptr[0] = DESCALE(tmp10 + tmp11, PASS1_BITS);
+//		dataptr[32] = DESCALE(tmp10 - tmp11, PASS1_BITS);
+//Might get an overflow here
+	padds.sh	r7, r5, r6
+	psubs.sh	r5, r5, r6
+
+	//Rounding
+	mov		lr,  (1 << (PASS1_BITS + 2))
+	orh		lr, hi(1 << (16 + PASS1_BITS + 2))
+	padds.sh	r7, r7, lr
+	padds.sh	r5, r5, lr
+
+	pasr.h		r7, r7, PASS1_BITS + 3
+	pasr.h		r5, r5, PASS1_BITS + 3
+	st.w		r12[0], r7
+	st.w		r12[4*8*2], r5
+
+	lddpc		r10, const_table2
+
+
+//		z1 = (tmp12 + tmp13) * FIX_0_541196100;
+	padds.sh	r5, r8, r9
+	mulhh.w		r6, r5:t, r10:t
+	mulhh.w		r7, r5:b, r10:t
+
+//		dataptr[16] =
+//			DESCALE(z1 + tmp13 * FIX_0_765366865, CONST_BITS + PASS1_BITS);
+	lddpc		r11, const_table2 + 4
+	mulhh.w		lr, r9:t, r10:b
+	mulhh.w		r9, r9:b, r10:b
+	add		lr, r6
+	add		r9, r7
+	satrnds		lr >> (CONST_BITS + PASS1_BITS + 3), 31
+	satrnds		r9 >> (CONST_BITS + PASS1_BITS + 3), 31
+	sthh.w		r12[2*8*2], lr:b, r9:b
+
+//		dataptr[48] =
+//			DESCALE(z1 + tmp12 * (-FIX_1_847759065), CONST_BITS + PASS1_BITS);
+	mulhh.w		lr, r8:t, r11:t
+	mulhh.w		r8, r8:b, r11:t
+	add		lr, r6
+	add		r8, r7
+	satrnds		lr >> (CONST_BITS + PASS1_BITS + 3), 31
+	satrnds		r8 >> (CONST_BITS + PASS1_BITS + 3), 31
+	sthh.w		r12[6*8*2], lr:b, r8:b
+
+//		/* Odd part per figure 8 --- note paper omits factor of sqrt(2).
+//		 * cK represents cos(K*pi/16).
+//		 * i0..i3 in the paper are tmp4..tmp7 here.
+//		 */
+//
+//		z2 = tmp5 + tmp6;
+//		z3 = tmp4 + tmp6;
+//		z4 = tmp5 + tmp7;
+	padds.sh	r5, r3, r2
+	padds.sh	r6, r4, r2
+	padds.sh	r7, r3, r1
+
+//		z5 = (z3 + z4) * FIX_1_175875602; /* sqrt(2) * c3 */
+	padds.sh	r8, r6, r7
+	mulhh.w		r9, r8:t, r11:b
+	mulhh.w		r8, r8:b, r11:b
+
+//		z3 *= -FIX_1_961570560; /* sqrt(2) * (-c3-c5) */
+//		z3 += z5;
+	lddpc		r11, const_table2 + 8
+	mulhh.w		r10, r6:t, r11:t
+	mulhh.w		r6, r6:b, r11:t
+	add		r10, r9
+	add		r6, r8
+
+//		z4 *= -FIX_0_390180644; /* sqrt(2) * (c5-c3) */
+//		z4 += z5;
+	mulhh.w		lr, r7:t, r11:b
+	mulhh.w		r7, r7:b, r11:b
+	lddpc		r11, const_table2 + 12
+	st.w		--sp,r0
+	add		lr, r9
+	add		r7, r8
+
+//		tmp6 *= FIX_3_072711026; /* sqrt(2) * ( c1+c3+c5-c7) */
+	mulhh.w		r0, r2:t, r11:t
+	machh.w		r0, r5:t, r11:b
+	mulhh.w		r2, r2:b, r11:t
+	machh.w		r2, r5:b, r11:b
+
+//		z2 *= -FIX_2_562915447;	/* sqrt(2) * (-c1-c3) */
+//		dataptr[24] = DESCALE(tmp6 + z2 + z3, CONST_BITS + PASS1_BITS);
+	add		r0, r10
+	lddpc		r11, const_table2 + 16
+	add		r2, r6
+	satrnds		r0 >> (CONST_BITS + PASS1_BITS + 3), 31
+	satrnds		r2 >> (CONST_BITS + PASS1_BITS + 3), 31
+	sthh.w		r12[3*8*2], r0:b, r2:b
+//		tmp5 *= FIX_2_053119869; /* sqrt(2) * ( c1+c3-c5+c7) */
+	mulhh.w		r0, r3:t, r11:t
+	machh.w		r0, r5:t, r11:b
+	mulhh.w		r2, r3:b, r11:t
+	machh.w		r2, r5:b, r11:b
+	add		r0, lr
+	lddpc		r11, const_table2 + 20
+	add		r2, r7
+
+//		dataptr[40] = DESCALE(tmp5 + z2 + z4, CONST_BITS + PASS1_BITS);
+	satrnds		r0 >> (CONST_BITS + PASS1_BITS + 3), 31
+	satrnds		r2 >> (CONST_BITS + PASS1_BITS + 3), 31
+	sthh.w		r12[5*8*2], r0:b, r2:b
+
+
+//		z1 = tmp4 + tmp7;
+	padds.sh	r2, r4, r1
+
+//		tmp4 *= FIX_0_298631336; /* sqrt(2) * (-c1+c3+c5-c7) */
+	mulhh.w		r3, r4:t, r11:t
+	machh.w		r3, r2:t, r11:b
+	mulhh.w		r4, r4:b, r11:t
+	machh.w		r4, r2:b, r11:b
+	add		r3, r10
+	lddpc		r11, const_table2 + 24
+	add		r4, r6
+
+//		z1 *= -FIX_0_899976223; /* sqrt(2) * (c7-c3) */
+//		dataptr[56] = DESCALE(tmp4 + z1 + z3, CONST_BITS + PASS1_BITS);
+	satrnds		r3 >> (CONST_BITS + PASS1_BITS + 3), 31
+	satrnds		r4 >> (CONST_BITS + PASS1_BITS + 3), 31
+	sthh.w		r12[7*8*2], r3:b, r4:b
+
+
+//		tmp7 *= FIX_1_501321110; /* sqrt(2) * ( c1+c3-c5-c7) */
+	mulhh.w		r3, r1:t, r11:t
+	machh.w		r3, r2:t, r11:b
+	mulhh.w		r4, r1:b, r11:t
+	machh.w		r4, r2:b, r11:b
+	add		r3, lr
+	add		r4, r7
+
+//		dataptr[8] = DESCALE(tmp7 + z1 + z4, CONST_BITS + PASS1_BITS);
+	satrnds		r3 >> (CONST_BITS + PASS1_BITS + 3), 31
+	satrnds		r4 >> (CONST_BITS + PASS1_BITS + 3), 31
+	sthh.w		r12[1*8*2], r3:b, r4:b
+	ld.w		r0, sp++
+
+//		dataptr++; /* advance pointer to next column */
+	sub		blkptr, -4
+	sub		loop_ctr, 1
+	brne		COLOUMN_LOOP
+
+//	}
+
+	popm		r0-r3, r4-r7, pc
+
+//	/* descale */
+//	for (i = 0; i < 64; i++)
+//		block[i] = (short int) DESCALE(data[i], 3);
+
+
+//}
+
+
+	.align	2
+const_table:	.short	FIX_0_541196100, FIX_0_765366865, -FIX_1_847759065, FIX_1_175875602
+		.short	FIX_0_298631336, FIX_2_053119869, FIX_3_072711026, FIX_1_501321110
+		.short	-FIX_0_899976223,-FIX_2_562915447, -FIX_1_961570560, -FIX_0_390180644
+
+const_table2:	.short	FIX_0_541196100, FIX_0_765366865, -FIX_1_847759065, FIX_1_175875602
+		.short	-FIX_1_961570560, -FIX_0_390180644, FIX_3_072711026, -FIX_2_562915447
+		.short	FIX_2_053119869, -FIX_2_562915447, FIX_0_298631336,  -FIX_0_899976223
+		.short	FIX_1_501321110, -FIX_0_899976223
+
diff --git a/libavcodec/avr32/h264idct.S b/libavcodec/avr32/h264idct.S
new file mode 100644
index 0000000..97009ee
--- /dev/null
+++ b/libavcodec/avr32/h264idct.S
@@ -0,0 +1,451 @@
+/*
+ * Copyright (c) 2007 Atmel Corporation. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials provided
+ * with the distribution.
+ *
+ * 3. The name of ATMEL may not be used to endorse or promote products
+ * derived from this software without specific prior written
+ * permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ATMEL ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE EXPRESSLY AND SPECIFICALLY DISCLAIMED. IN NO EVENT SHALL ATMEL
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
+ * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
+ * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+	.global	h264_idct_add_avr32
+
+	/* Macro for performing the 1-D transform on one row line.
+
+	   The register 'w01' should contain the first two pixels,
+	   and the register 'w23' should contain the last two pixels
+	   in the line. The resulting line is placed in p01 and p23
+	   so that { w01, w23 } = { x0, x1, x3, x2 }.
+	   'tmp' and 'tmp2' should be scratchpad registers. */
+	.macro	transform_row	w01, w23, tmp, tmp2
+	add		\tmp, \w23, \w01 << 1   /* tmp = { xxxx, 2*w1 + w3 } */
+	sub		\tmp2, \w01, \w23 << 1  /* tmp2 = { xxxx, w1 - 2*w3 } */
+	bfins		\tmp2, \tmp, 16, 16     /* tmp2 = { 2*w1 + w3, w1 - 2*w3 } */
+	pasr.h		\tmp2, \tmp2, 1	        /* tmp2 = { w1 + w3/2, w1/2 - w3 } */
+	paddsub.h	\tmp, \w01:t, \w23:t    /* tmp =  { w0 + w2, w0 - w2 }  */
+	padd.h		\w01, \tmp, \tmp2	/* w01 =  { w0 + w2 + w1 + w3/2, w0 - w2 + w1/2 - w3 } */
+	psub.h		\w23, \tmp, \tmp2	/* w23 =  { w0 + w2 - w1 - w3/2, w0 - w2 - w1/2 + w3 } */
+	.endm
+
+	/* Macro for performing the 1-D transform on two columns.
+
+	   The registers w0, w1, w2, w3 should each contain two
+	   packed samples from the two colomns to transform.
+	   tmp and tmp2 are scratchpad registers.
+
+	   The resulting transformed columns are placed in the
+	   same positions as the input columns.
+	*/
+	.macro	transform_2columns	w0, w1, w2, w3, tmp, tmp2
+	padd.h		\tmp, \w0, \w2  /* tmp = z0 = w0 + w2 */
+	psub.h		\w0, \w0, \w2   /* w0 = z1 = w0 - w2 */
+	pasr.h		\w2, \w1, 1	/* w2 = w1/2 */
+	pasr.h		\tmp2, \w3, 1	/* tmp2 = w3/2 */
+	psub.h		\w3, \w2, \w3	/* w3 = z2 = w1/2 - w3 */
+	padd.h		\tmp2, \w1, \tmp2/* tmp2 = z3 = w1 + w3/2 */
+	padd.h		\w1, \w0, \w3	/* w1 = x1 = z1 + z2 */
+	psub.h		\w2, \w0, \w3	/* w2 = x2 = z1 - z2 */
+	padd.h		\w0, \tmp, \tmp2/* w0 = x0 = z0 + z3 */
+	psub.h		\w3, \tmp, \tmp2/* w3 = x3 = z0 - z3 */
+	/* Scale down result. */
+	pasr.h		\w0, \w0, 6
+	pasr.h		\w1, \w1, 6
+	pasr.h		\w2, \w2, 6
+	pasr.h		\w3, \w3, 6
+	.endm
+
+/*void h264_idct_add_avr32(uint8_t *dst, DCTELEM *block, int stride)*/
+
+h264_idct_add_avr32:
+
+	stm	--sp,r0-r3,r4-r7, lr
+
+	/* Setup rounding factor. */
+	mov	r0, (1 << 5)
+	lsl	r0, 16
+
+	/* Load block */
+        ldm	r11,r2-r9
+        /* r9 = { w00, w01 },
+           r8 = { w02, w03 },
+           r7 = { w10, w11 },
+           r6 = { w12, w13 },
+           r5 = { w20, w21 },
+           r4 = { w22, w23 },
+           r3 = { w30, w31 },
+           r2 = { w32, w33 } */
+
+	/* Add the rounding factor to w00. */
+	add			r9, r0
+
+	/* Transform rows */
+	transform_row	        r9, r8, r0, r1
+	transform_row	        r7, r6, r0, r1
+	transform_row	        r5, r4, r0, r1
+	transform_row	        r3, r2, r0, r1
+
+	/* Transform columns */
+	transform_2columns	r9, r7, r5, r3, r0, r1
+	transform_2columns	r8, r6, r4, r2, r0, r1
+
+	/* Load predicted pixels.*/
+	ld.w			lr, r12[0]
+        ld.w			r11, r12[r10]
+
+	/* Unpack to halwords. */
+        punpckub.h		r0, lr:t
+        punpckub.h		r1, lr:b
+
+	/* Add with transformed row. */
+        padd.h			r0, r0, r9
+        paddx.h			r1, r1, r8
+	/* Pack and saturate back to 8-bit pixels. */
+        packsh.ub		r0, r0, r1
+
+	/* Unpack to halwords. */
+        punpckub.h		lr, r11:t
+        punpckub.h		r11, r11:b
+
+	/* Add with transformed row. */
+        padd.h			lr, lr, r7
+        paddx.h			r11, r11, r6
+	/* Pack and saturate back to 8-bit pixels. */
+        packsh.ub		r1, lr, r11
+
+	/* Store back to frame. */
+	st.w			r12[0], r0
+	st.w			r12[r10], r1
+
+	add			r12, r12, r10 << 1
+
+	/* Load predicted pixels.*/
+	ld.w			lr, r12[0]
+        ld.w			r11, r12[r10]
+
+	/* Unpack to halwords. */
+        punpckub.h		r0, lr:t
+        punpckub.h		r1, lr:b
+
+	/* Add with transformed row. */
+        padd.h			r0, r0, r5
+        paddx.h			r1, r1, r4
+	/* Pack and saturate back to 8-bit pixels. */
+        packsh.ub		r0, r0, r1
+
+	/* Unpack to halwords. */
+        punpckub.h		lr, r11:t
+        punpckub.h		r11, r11:b
+
+	/* Add with transformed row. */
+        padd.h			lr, lr, r3
+        paddx.h			r11, r11, r2
+	/* Pack and saturate back to 8-bit pixels. */
+        packsh.ub		r1, lr, r11
+
+	/* Store back to frame. */
+	st.w			r12[0], r0
+	st.w			r12[r10], r1
+
+	ldm			sp++,r0-r3,r4-r7, pc
+
+
+	.global	h264_idct8_add_avr32
+//void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride){
+
+h264_idct8_add_avr32:
+	stm         --sp,r0-r3,r4-r7, lr
+
+	/* Push dst and stride on stack */
+	stm         --sp,r10,r12
+
+//    int i;
+//    DCTELEM (*src)[8] = (DCTELEM(*)[8])block;
+//    uint8_t *cm = cropTbl + MAX_NEG_CROP;
+
+//    block[0] += 32;
+
+
+//    for( i = 0; i < 8; i++ )
+//    {
+	mov		lr, 4
+0:
+	ld.w		r7, r11[0*(8*2)]
+	ld.w		r6, r11[1*(8*2)]
+	ld.w		r5, r11[2*(8*2)]
+	ld.w		r4, r11[3*(8*2)]
+	ld.w		r3, r11[4*(8*2)]
+	ld.w		r2, r11[5*(8*2)]
+	ld.w		r1, r11[6*(8*2)]
+	ld.w		r0, r11[7*(8*2)]
+
+/*
+        const int a0 =  src[0][i] + src[4][i];
+        const int a2 =  src[0][i] - src[4][i];
+        const int a4 = (src[2][i]>>1) - src[6][i];
+        const int a6 = (src[6][i]>>1) + src[2][i];
+*/
+	padd.h		r8, r7, r3	/* r8 = a0 */
+	psub.h		r7, r7, r3	/* r7 = a2 */
+	pasr.h		r3, r5, 1	/* r3 = src[2][i] >> 1 */
+	pasr.h		r9, r1, 1	/* r9 = src[6][i] >> 1 */
+	psub.h		r3, r3, r1	/* r3 = a4 */
+	padd.h		r9, r9, r5	/* r9 = a6 */
+
+/*
+        const int b0 = a0 + a6;
+        const int b2 = a2 + a4;
+        const int b4 = a2 - a4;
+        const int b6 = a0 - a6;
+*/
+	padd.h		r1, r8, r9	/* r1 = b0 */
+	psub.h		r8, r8, r9	/* r8 = b6 */
+	padd.h		r5, r7, r3	/* r5 = b2 */
+	psub.h		r7, r7, r3	/* r7 = b4 */
+
+/*
+        const int a1 = -src[3][i] + src[5][i] - src[7][i] - (src[7][i]>>1);
+        const int a3 =  src[1][i] + src[7][i] - src[3][i] - (src[3][i]>>1);
+        const int a5 = -src[1][i] + src[7][i] + src[5][i] + (src[5][i]>>1);
+        const int a7 =  src[3][i] + src[5][i] + src[1][i] + (src[1][i]>>1);
+*/
+	pasr.h		r3, r0, 1
+	padd.h		r3, r3, r0
+	psub.h		r3, r2, r3
+	psub.h		r3, r3, r4 /* r3 = a1 */
+
+	pasr.h		r9, r4, 1
+	padd.h		r9, r9, r4
+	psub.h		r9, r0, r9
+	padd.h		r9, r6, r9 /* r9 = a3 */
+
+	pasr.h		r10, r2, 1
+	padd.h		r10, r10, r2
+	padd.h		r10, r10, r0
+	psub.h		r10, r10, r6 /* r10 = a5 */
+
+	pasr.h		r0, r6, 1
+	padd.h		r0, r0, r6
+	padd.h		r0, r0, r2
+	padd.h		r0, r0, r4 /* r0 = a7 */
+
+/*
+        const int b1 = (a7>>2) + a1;
+        const int b3 =  a3 + (a5>>2);
+        const int b5 = (a3>>2) - a5;
+        const int b7 =  a7 - (a1>>2);
+*/
+	pasr.h		r2, r0, 2
+	padd.h		r2, r2, r3 /* r2 = b1 */
+	pasr.h		r3, r3, 2
+	psub.h		r3, r0, r3 /* r3 = b7 */
+
+	pasr.h		r0, r10, 2
+	padd.h		r0, r0, r9 /* r0 = b3 */
+	pasr.h		r9, r9, 2
+	psub.h		r9, r9, r10 /* r9 = b5 */
+
+/*
+        src[0][i] = b0 + b7;
+        src[7][i] = b0 - b7;
+        src[1][i] = b2 + b5;
+        src[6][i] = b2 - b5;
+        src[2][i] = b4 + b3;
+        src[5][i] = b4 - b3;
+        src[3][i] = b6 + b1;
+        src[4][i] = b6 - b1;
+*/
+
+	padd.h		r4, r1, r3
+	psub.h		r1, r1, r3
+	st.w		r11[0*(8*2)], r4
+	st.w		r11[7*(8*2)], r1
+
+	padd.h		r3, r5, r9
+	psub.h		r5, r5, r9
+	st.w		r11[1*(8*2)], r3
+	st.w		r11[6*(8*2)], r5
+
+	padd.h		r9, r7, r0
+	psub.h		r7, r7, r0
+	st.w		r11[2*(8*2)], r9
+	st.w		r11[5*(8*2)], r7
+
+	padd.h		r0, r8, r2
+	psub.h		r8, r8, r2
+	st.w		r11[3*(8*2)], r0
+	st.w		r11[4*(8*2)], r8
+
+	sub		r11, -4
+	sub		lr, 1
+	brne		0b
+
+//    }
+
+	lddsp		r12, sp[0]	/* r12 = dst */
+	sub		r11, 4*4
+	ldm		r11++, r4-r7
+	mov		lr, 8
+	/* Push dst and stride on stack */
+
+1:
+//    for( i = 0; i < 8; i++ )
+//    {
+
+	/*
+           r7 = {src[i][0], src[i][1]}
+           r6 = {src[i][2], src[i][3]}
+           r5 = {src[i][4], src[i][5]}
+           r4 = {src[i][6], src[i][7]}
+	*/
+
+/*
+        const int a0 =  src[i][0] + src[i][4];
+        const int a2 =  src[i][0] - src[i][4];
+        const int a4 = (src[i][2]>>1) - src[i][6];
+        const int a6 = (src[i][6]>>1) + src[i][2];
+*/
+	pasr.h		r8, r6, 1
+	pasr.h		r9, r4, 1
+	addhh.w		r0, r7:t, r5:t	/* r0 = a0 */
+	subhh.w		r1, r7:t, r5:t	/* r1 = a2 */
+	subhh.w		r2, r8:t, r4:t	/* r2 = a4 */
+	addhh.w		r3, r9:t, r6:t	/* r3 = a6 */
+
+/*
+        const int b0 = a0 + a6;
+        const int b2 = a2 + a4;
+        const int b4 = a2 - a4;
+        const int b6 = a0 - a6;
+*/
+	add		r10, r0, r3	/* r10 = b0 */
+	sub		r0, r3		/* r0 = b6 */
+	add		r3, r1, r2	/* r3 = b2 */
+	sub		r1, r2		/* r1 = b4 */
+
+/*
+          const int a7 =  src[i][5] + src[i][3] + src[i][1] + (src[i][1]>>1);
+          const int a1 =  src[i][5] - src[i][3] - src[i][7] - (src[i][7]>>1);
+          const int a3 =  src[i][7] + src[i][1] - src[i][3] - (src[i][3]>>1);
+          const int a5 =  src[i][7] - src[i][1] + src[i][5] + (src[i][5]>>1);
+*/
+	addhh.w		r8, r8:b, r6:b
+	addhh.w		r2, r4:b, r7:b
+	sub		r2, r8		/* r2 = a3 */
+
+	addhh.w		r9, r9:b, r4:b
+	subhh.w		r8, r5:b, r6:b
+	sub		r8, r9		/* r8 = a1 */
+
+	pasr.h		r9, r7, 1
+	addhh.w		r9, r9:b, r7:b
+	addhh.w		r6, r5:b, r6:b
+	add		r6, r9		/* r6 = a7 */
+
+	pasr.h		r9, r5, 1
+	addhh.w		r9, r9:b, r5:b
+	subhh.w		r5, r4:b, r7:b
+	add		r5, r9		/* r5 = a5 */
+
+/*        const int b1 = (a7>>2) + a1;
+          const int b3 = (a5>>2) + a3;
+          const int b5 = (a3>>2) - a5;
+          const int b7 = -(a1>>2) + a7;
+*/
+	asr		r4, r6, 2
+	add		r4, r8		/* r4 = b1 */
+	asr		r8, 2
+	rsub		r8, r6		/* r8 = b7 */
+
+	asr		r6, r5, 2
+	add		r6, r2		/* r6 = b3 */
+	asr		r2, 2
+	sub		r2, r5		/* r2 = b5 */
+
+/*
+        dst[i*stride + 0] = cm[ dst[i*stride + 0] + ((b0 + b7) >> 6) ];
+        dst[i*stride + 1] = cm[ dst[i*stride + 1] + ((b2 + b5) >> 6) ];
+        dst[i*stride + 2] = cm[ dst[i*stride + 2] + ((b4 + b3) >> 6) ];
+        dst[i*stride + 3] = cm[ dst[i*stride + 3] + ((b6 + b1) >> 6) ];
+        dst[i*stride + 4] = cm[ dst[i*stride + 4] + ((b6 - b1) >> 6) ];
+        dst[i*stride + 5] = cm[ dst[i*stride + 5] + ((b4 - b3) >> 6) ];
+        dst[i*stride + 6] = cm[ dst[i*stride + 6] + ((b2 - b5) >> 6) ];
+        dst[i*stride + 7] = cm[ dst[i*stride + 7] + ((b0 - b7) >> 6) ];
+*/
+	add		r5, r10, r8
+	satrnds		r5 >> 6, 0	/* r5 = (b0 + b7) >> 6 */
+	sub		r10, r8
+	satrnds		r10 >> 6, 0	/* r10 = (b0 - b7) >> 6 */
+	add		r8, r3, r2
+	satrnds		r8 >> 6, 0	/* r8 = (b2 + b5) >> 6 */
+	sub		r3, r2
+	satrnds		r3 >> 6, 0	/* r3 = (b2 - b5) >> 6 */
+
+	add		r2, r1, r6
+	satrnds		r2 >> 6, 0	/* r2 = (b4 + b3) >> 6 */
+	sub		r1, r6
+	satrnds		r1 >> 6, 0	/* r1 = (b4 - b3) >> 6 */
+
+	add		r6, r0, r4
+	satrnds		r6 >> 6, 0	/* r6 = (b6 + b1) >> 6 */
+	sub		r0, r4
+	satrnds		r0 >> 6, 0	/* r0 = (b6 - b1) >> 6 */
+
+	ld.w		r4, r12[0]
+
+	packw.sh	r8, r5, r8
+	packw.sh	r7, r2, r6
+	ld.w		r9, r12[4]
+	packw.sh	r6, r0, r1
+	packw.sh	r5, r3, r10
+
+	punpckub.h	r10, r4:t
+	punpckub.h	r4, r4:b
+	punpckub.h	r3, r9:t
+	punpckub.h	r9, r9:b
+
+	padd.h		r8, r8, r10
+	padd.h		r7, r7, r4
+	padd.h		r6, r6, r3
+	padd.h		r5, r5, r9
+
+	lddsp		r10, sp[4]	/* r10 = stride */
+	packsh.ub	r0, r8, r7
+	packsh.ub	r1, r6, r5
+
+	st.w		r12[0], r0
+	st.w		r12[4], r1
+
+	ldm		r11++, r4-r7
+	add		r12, r10	/* dst += stride */
+
+	sub		lr, 1
+	brne		1b
+
+	sub		sp, -8
+	ldm		sp++,r0-r3,r4-r7, pc
+//    }
+//}
+
diff --git a/libavcodec/avr32/idct.S b/libavcodec/avr32/idct.S
new file mode 100644
index 0000000..19251cb
--- /dev/null
+++ b/libavcodec/avr32/idct.S
@@ -0,0 +1,793 @@
+/*
+ * Copyright (c) 2007 Atmel Corporation. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials provided
+ * with the distribution.
+ *
+ * 3. The name of ATMEL may not be used to endorse or promote products
+ * derived from this software without specific prior written
+ * permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ATMEL ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE EXPRESSLY AND SPECIFICALLY DISCLAIMED. IN NO EVENT SHALL ATMEL
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
+ * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
+ * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+	.global idct_add_avr32
+	.global idct_put_avr32
+	.global idct_avr32
+
+#define CONST_BITS  13
+#define PASS1_BITS  2
+
+#define ONE	((INT32) 1)
+
+#define CONST_SCALE (ONE << CONST_BITS)
+
+#define	LINE_SIZE	32
+
+#define FIX_0_298631336  (2446)	/* FIX(0.298631336) */
+#define FIX_0_390180644  (3196)	/* FIX(0.390180644) */
+#define FIX_0_541196100  (4433)	/* FIX(0.541196100) */
+#define FIX_0_765366865  (6270)	/* FIX(0.765366865) */
+#define FIX_0_899976223  (7373)	/* FIX(0.899976223) */
+#define FIX_1_175875602  (9633)	/* FIX(1.175875602) */
+#define FIX_1_501321110  (12299)/* FIX(1.501321110) */
+#define FIX_1_847759065  (15137)/* FIX(1.847759065) */
+#define FIX_1_961570560  (16069)/* FIX(1.961570560) */
+#define FIX_2_053119869  (16819)/* FIX(2.053119869) */
+#define FIX_2_562915447  (20995)/* FIX(2.562915447) */
+#define FIX_3_072711026  (25172)/* FIX(3.072711026) */
+
+
+#define loop_cnt	r11
+
+	.text
+
+idct_add_avr32:
+	pushm		r0-r3, r4-r7, lr	//Free up registers to use for local variables
+
+	// Give room for some variables on the stack
+	sub		sp, 8
+	stdsp		SP[0], r12 // rfp
+	stdsp		SP[4], r11 // iinc
+
+	mov		loop_cnt, 8		//Initialize loop counter
+
+FOR_ROW:
+
+	ldm		r10, r0, r1, r2, r3	//Load 8 DCT-coeffisients from the current row in the DCT-block
+	mov		r6, 0
+#ifdef USE_PREFETCH
+	pref		r10[LINE_SIZE]		//Prefetch next line
+#endif
+	or		r4, r2, r3 << 16
+	or		r4, r1			//Check if all DCT-coeffisients except the DC is zero
+	or		r4, r0
+	brne		AC_ROW			//If there are non-zero AC coeffisients perform row-transform
+
+	paddsub.h	r5, r3:t, r6:b		//Extract the DC-coeff from r5
+	plsl.h		r5, r5, PASS1_BITS
+	mov		r4, r5
+	st.d		r10++, r4
+	st.d		r10++, r4
+
+	sub		loop_cnt, 1		//Decrement loop counter
+	brne		FOR_ROW			//Perform loop one more time if loop_cnt is not zero
+
+	bral		COLOUMN_TRANSFORM	//Perform coloumn transform after row transform is computed
+
+
+AC_ROW:
+
+
+	ld.w		r12, pc[coef_table - .]
+	ld.w		r9, pc[coef_table - . + 4]
+
+	padd.h		r4, r2, r0 // r4:t = dataptr[2] + dataptr[6],r4:b = dataptr[3] + dataptr[7]
+	mulhh.w		r5, r4:t, r12:t
+	mulhh.w		r6, r0:t, r12:b
+	ld.w		r12, pc[coef_table - . + 8]
+	mulhh.w		r7, r2:t, r9:t
+	add		r6, r5	// tmp2
+	satrnds		r6 >> (CONST_BITS - PASS1_BITS), 31
+	add		r7, r5	// tmp3
+	satrnds		r7 >> (CONST_BITS - PASS1_BITS), 31
+
+	paddsub.h	r5, r3:t, r1:t
+	plsl.h		r5, r5, PASS1_BITS // r5:t = tmp0, r5:b = tmp1
+
+	paddsub.h	r7, r5:t, r7:b // r7:t = tmp10, r7:b = tmp13
+	paddsub.h	r6, r5:b, r6:b // r6:t = tmp11, r6:b = tmp12
+
+
+	addhh.w		lr, r3:b, r1:b // lr = z4
+	addhh.w		r5, r4:b, lr:b
+	mulhh.w		r5, r5:b, r9:b // r5 = z5
+
+	ld.w		r9, pc[coef_table - . + 12]
+	mulhh.w		r4, r4:b, r12:t // r4 = z3
+	mulhh.w		lr, lr:b, r12:b // lr = z4
+
+	add		r4, r5
+	add		lr, r5
+
+	addhh.w		r5, r2:b, r1:b // r5 = z2
+	addhh.w		r8, r3:b, r0:b // r8 = z1
+
+	mulhh.w		r0, r0:b, r9:t // r0 = tmp0
+	ld.w		r12, pc[coef_table - . + 16]
+	mulhh.w		r1, r1:b, r9:b // r1 = tmp1
+	ld.w		r9, pc[coef_table - . + 20]
+	mulhh.w		r2, r2:b, r12:t // r2 = tmp2
+	mulhh.w		r3, r3:b, r12:b // r3 = tmp3
+	mulhh.w		r8, r8:b, r9:t // r8 = z1
+	mulhh.w		r5, r5:b, r9:b // r5 = z2
+
+	add		r0, r8
+	add		r0, r4
+	add		r1, r5
+	add		r1, lr
+	add		r2, r5
+	add		r2, r4
+	add		r3, r8
+	add		r3, lr
+
+	satrnds		r0 >> (CONST_BITS - PASS1_BITS), 31
+	satrnds		r1 >> (CONST_BITS - PASS1_BITS), 31
+	satrnds		r2 >> (CONST_BITS - PASS1_BITS), 31
+	satrnds		r3 >> (CONST_BITS - PASS1_BITS), 31
+
+	paddsub.h	r5, r6:t, r2:b // r5:t = dataptr[1], r5:b = dataptr[6]
+	paddsub.h	r4, r7:t, r3:b // r4:t = dataptr[0], r4:b = dataptr[7]
+	paddsub.h	r3, r6:b, r1:b // r3:t = dataptr[2], r3:b = dataptr[5]
+	paddsub.h	r2, r7:b, r0:b // r2:t = dataptr[3], r2:b = dataptr[4]
+
+	sthh.w		r10[0], r4:t, r5:t
+	sthh.w		r10[4], r3:t, r2:t
+	sthh.w		r10[8], r2:b, r3:b
+	sthh.w		r10[12], r5:b, r4:b
+
+	sub		r10, -16
+	sub		loop_cnt, 1
+	brne		FOR_ROW, e
+
+COLOUMN_TRANSFORM:
+	sub		r10, 128	//Set pointer to start of DCT block
+	mov		loop_cnt, 8
+
+FOR_COLOUMN:
+	ldins.h		r3:t,r10[0]    // r3:t = dataptr[0]
+	ldins.h		r1:t,r10[1*8*2]// r1:t = dataptr[1]
+	ldins.h		r2:t,r10[2*8*2]// r2:t = dataptr[2]
+	ldins.h		r0:t,r10[5*8*2]// r0:t = dataptr[5]
+	ldins.h		r3:b,r10[4*8*2]// r3:b = dataptr[4]
+	ldins.h		r1:b,r10[3*8*2]// r1:b = dataptr[3]
+	ldins.h		r2:b,r10[6*8*2]// r2:b = dataptr[6]
+	ldins.h		r0:b,r10[7*8*2]// r0:b = dataptr[7]
+
+	or		r4, r1, r3 << 16
+	or		r4, r2
+	or		r4, r0
+	brne		AC_COLOUMN			//If there are non-zero AC coeffisients perform row-transform
+
+	lddsp		r12, SP[0]       // rfp
+	lddsp		r9, SP[4]	// iinc
+	satrnds		r3 >> ( PASS1_BITS + 3 + 16 ), 9
+	ld.d		r0, r12[0]
+	sub		r10, -2	// Increment the dataptr
+	bfins		r3, r3, 16, 16
+	punpckub.h	r2, r1:t
+	padd.h		r2, r2, r3
+	punpckub.h	r1, r1:b
+	padd.h		r1, r1, r3
+	packsh.ub	r1, r2, r1
+	punpckub.h	r2, r0:t
+	padd.h		r2, r2, r3
+	punpckub.h	r0, r0:b
+	padd.h		r0, r0, r3
+	packsh.ub	r0, r2, r0
+	st.d		r12[0], r0
+	add		r12, r9	// increment rfp
+	stdsp		SP[0], r12
+
+	sub		loop_cnt, 1//Decrement loop counter
+	brne		FOR_COLOUMN//Perform loop one more time if loop_cnt is not zero
+
+	sub		sp, -8
+	popm		r0-r3, r4-r7, pc//Pop back registers and PC
+
+AC_COLOUMN:
+	ld.w		r12, pc[coef_table - .]
+	ld.w		r9, pc[coef_table - . + 4]
+
+	addhh.w		r4, r2:t, r2:b
+	mulhh.w		r4, r4:b, r12:t	// r4 = z1
+	mulhh.w		r5, r2:b, r12:b
+	ld.w		r12, pc[coef_table - . + 8]
+	mulhh.w		r6, r2:t, r9:t
+	add		r5, r4	// r5 = tmp2
+	add		r6, r4	// r6 = tmp3
+
+	addhh.w		r7, r3:t, r3:b
+	subhh.w		r8, r3:t, r3:b
+
+	lsl		r7, CONST_BITS
+	lsl		r8, CONST_BITS
+
+	add		r2, r7, r6 // r2 = tmp10
+	sub		r3, r7, r6 // r3 = tmp13
+	add		r4, r8, r5 // r4 = tmp11
+	sub		r5, r8, r5 // r5 = tmp12
+
+	padd.h		r6, r0, r1 // r6:t = z4, r6:b = z3
+	addhh.w		r7, r6:t, r6:b
+	mulhh.w		r7, r7:b, r9:b // r7 = z5
+
+	ld.w		r9, pc[coef_table - . + 12]
+	mulhh.w		r8, r6:b, r12:t // r8 = z3
+	mulhh.w		r6, r6:t, r12:b // r6 = z4
+
+	add		r8, r7
+	add		r6, r7
+
+	paddx.h		r7, r0, r1 // r7:t = z2, r7:b = z1
+
+	mulhh.w		r12, r0:b, r9:t // r12 = tmp0
+	mulhh.w		r0, r0:t, r9:b // r0 = tmp1
+	ld.w		r9, pc[coef_table - . + 16]
+	add		r12, r8
+	add		r0, r6
+
+	ld.w		lr, pc[coef_table - . + 20]
+	machh.w		r8, r1:b, r9:t // r8 = tmp2
+	machh.w		r6, r1:t, r9:b // r6 = tmp3
+	mulhh.w		r9, r7:b, lr:t // r9 = z1
+	mulhh.w		r7, r7:t, lr:b // r7 = z2
+
+	add		r12, r9
+	add		r0, r7
+	add		r8, r7
+	add		r6, r9
+
+	add		r1, r2, r6 // r1 = dataptr[DCTSIZE*0]
+	sub		r2, r2, r6 // r2 = dataptr[DCTSIZE*7]
+	add		r6, r4, r8 // r6 = dataptr[DCTSIZE*1]
+	sub		r4, r4, r8 // r4 = dataptr[DCTSIZE*6]
+	add		r8, r5, r0 // r8 = dataptr[DCTSIZE*2]
+	sub		r5, r5, r0 // r5 = dataptr[DCTSIZE*5]
+	add		r0, r3, r12 // r0 = dataptr[DCTSIZE*3]
+	sub		r3, r3, r12 // r3 = dataptr[DCTSIZE*4]
+
+	satrnds		r1 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r2 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r6 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r4 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r8 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r5 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r0 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r3 >> (CONST_BITS+PASS1_BITS+3), 9
+
+	packw.sh	r1, r1, r6
+	packw.sh	r8, r8, r0
+	packw.sh	r3, r3, r5
+	packw.sh	r4, r4, r2
+
+	lddsp		r12, SP[0]       // rfp
+	lddsp		r9, SP[4]	// iinc
+	ld.d		r6, r12[0]
+	sub		r10, -2	// Increment the dataptr
+	punpckub.h	r0, r7:t
+	padd.h		r1, r1, r0
+	punpckub.h	r0, r7:b
+	padd.h		r8, r8, r0
+	packsh.ub	r7, r1, r8
+	punpckub.h	r0, r6:t
+	padd.h		r3, r3, r0
+	punpckub.h	r0, r6:b
+	padd.h		r4, r4, r0
+	packsh.ub	r6, r3, r4
+	st.d		r12[0], r6
+	add		r12, r9	// increment rfp
+	stdsp		SP[0], r12
+
+	sub		loop_cnt, 1		//Decrement loop counter
+	brne		FOR_COLOUMN		//Perform loop one more time if loop_cnt is not zero
+
+	sub		sp, -8
+	popm		r0-r3, r4-r7, pc	//Pop back registers and PC
+
+
+//Coeffisient Table:
+	.align	2
+coef_table:
+	.short	FIX_0_541196100, -FIX_1_847759065, FIX_0_765366865, FIX_1_175875602
+	.short	- FIX_1_961570560, - FIX_0_390180644, FIX_0_298631336, FIX_2_053119869
+	.short	FIX_3_072711026, FIX_1_501321110, - FIX_0_899976223, - FIX_2_562915447
+
+
+idct_put_avr32:
+	pushm		r0-r3, r4-r7, lr	//Free up registers to use for local variables
+
+		//; Give room for some variables on the stack
+	sub		sp, 8
+	stdsp		SP[0], r12 // rfp
+	stdsp		SP[4], r11 // iinc
+
+	mov		loop_cnt, 8		//Initialize loop counter
+
+0:
+
+	ldm		r10, r0, r1, r2, r3	//Load 8 DCT-coeffisients from the current row in the DCT-block
+	mov		r6, 0
+#ifdef USE_PREFETCH
+	pref		r10[LINE_SIZE]		//Prefetch next line
+#endif
+	or		r4, r2, r3 << 16
+	or		r4, r1			//Check if all DCT-coeffisients except the DC is zero
+	or		r4, r0
+	brne		1f			//If there are non-zero AC coeffisients perform row-transform
+
+	paddsub.h	r5, r3:t, r6:b		//Extract the DC-coeff from r5
+	plsl.h		r5, r5, PASS1_BITS
+	mov		r4, r5
+	st.d		r10++, r4
+	st.d		r10++, r4
+
+	sub		loop_cnt, 1		//Decrement loop counter
+	brne		0b			//Perform loop one more time if loop_cnt is not zero
+
+	bral		2f	                //Perform coloumn transform after row transform is computed
+
+1:
+	ld.w		r12, pc[coef_table_copy - .]
+	ld.w		r9, pc[coef_table_copy - . + 4]
+
+	padd.h		r4, r2, r0 // r4:t = dataptr[2] + dataptr[6],r4:b = dataptr[3] + dataptr[7]
+	mulhh.w		r5, r4:t, r12:t
+	mulhh.w		r6, r0:t, r12:b
+	ld.w		r12, pc[coef_table_copy - . + 8]
+	mulhh.w		r7, r2:t, r9:t
+	add		r6, r5	// tmp2
+	satrnds		r6 >> (CONST_BITS - PASS1_BITS), 31
+	add		r7, r5	// tmp3
+	satrnds		r7 >> (CONST_BITS - PASS1_BITS), 31
+
+	paddsub.h	r5, r3:t, r1:t
+	plsl.h		r5, r5, PASS1_BITS // r5:t = tmp0, r5:b = tmp1
+
+	paddsub.h	r7, r5:t, r7:b // r7:t = tmp10, r7:b = tmp13
+	paddsub.h	r6, r5:b, r6:b // r6:t = tmp11, r6:b = tmp12
+
+	addhh.w		lr, r3:b, r1:b // lr = z4
+	addhh.w		r5, r4:b, lr:b
+	mulhh.w		r5, r5:b, r9:b // r5 = z5
+
+	ld.w		r9, pc[coef_table_copy - . + 12]
+	mulhh.w		r4, r4:b, r12:t // r4 = z3
+	mulhh.w		lr, lr:b, r12:b // lr = z4
+
+	add		r4, r5
+	add		lr, r5
+
+	addhh.w		r5, r2:b, r1:b // r5 = z2
+	addhh.w		r8, r3:b, r0:b // r8 = z1
+
+	mulhh.w		r0, r0:b, r9:t // r0 = tmp0
+	ld.w		r12, pc[coef_table_copy - . + 16]
+	mulhh.w		r1, r1:b, r9:b // r1 = tmp1
+	ld.w		r9, pc[coef_table_copy - . + 20]
+	mulhh.w		r2, r2:b, r12:t // r2 = tmp2
+	mulhh.w		r3, r3:b, r12:b // r3 = tmp3
+	mulhh.w		r8, r8:b, r9:t // r8 = z1
+	mulhh.w		r5, r5:b, r9:b // r5 = z2
+
+	add		r0, r8
+	add		r0, r4
+	add		r1, r5
+	add		r1, lr
+	add		r2, r5
+	add		r2, r4
+	add		r3, r8
+	add		r3, lr
+
+	satrnds		r0 >> (CONST_BITS - PASS1_BITS), 31
+	satrnds		r1 >> (CONST_BITS - PASS1_BITS), 31
+	satrnds		r2 >> (CONST_BITS - PASS1_BITS), 31
+	satrnds		r3 >> (CONST_BITS - PASS1_BITS), 31
+
+	paddsub.h	r5, r6:t, r2:b // r5:t = dataptr[1], r5:b = dataptr[6]
+	paddsub.h	r4, r7:t, r3:b // r4:t = dataptr[0], r4:b = dataptr[7]
+	paddsub.h	r3, r6:b, r1:b // r3:t = dataptr[2], r3:b = dataptr[5]
+	paddsub.h	r2, r7:b, r0:b // r2:t = dataptr[3], r2:b = dataptr[4]
+
+	sthh.w		r10[0], r4:t, r5:t
+	sthh.w		r10[4], r3:t, r2:t
+	sthh.w		r10[8], r2:b, r3:b
+	sthh.w		r10[12], r5:b, r4:b
+
+	sub		r10, -16
+	sub		loop_cnt, 1
+	brne		0b
+
+2:
+
+	sub		r10, 128	//Set pointer to start of DCT block
+
+	mov		loop_cnt, 8
+
+0:
+	ldins.h		r3:t,r10[0]    // r3:t = dataptr[0]
+	ldins.h		r1:t,r10[1*8*2]// r1:t = dataptr[1]
+	ldins.h		r2:t,r10[2*8*2]// r2:t = dataptr[2]
+	ldins.h		r0:t,r10[5*8*2]// r0:t = dataptr[5]
+	ldins.h		r3:b,r10[4*8*2]// r3:b = dataptr[4]
+	ldins.h		r1:b,r10[3*8*2]// r1:b = dataptr[3]
+	ldins.h		r2:b,r10[6*8*2]// r2:b = dataptr[6]
+	ldins.h		r0:b,r10[7*8*2]// r0:b = dataptr[7]
+
+	or		r4, r1, r3 << 16
+	or		r4, r2
+	or		r4, r0
+	brne		1f			//If there are non-zero AC coeffisients perform row-transform
+
+	lddsp		r12, SP[0]       // rfp
+	lddsp		r9, SP[4]	// iinc
+	satrnds		r3 >> ( PASS1_BITS + 3 + 16 ), 31
+	packw.sh	r3, r3, r3
+	packsh.ub	r3, r3, r3
+	mov		r2, r3
+	st.d		r12[0], r2
+	add		r12, r9	// increment rfp
+	sub		r10, -2	// Increment the dataptr
+	stdsp		SP[0], r12
+
+	sub		loop_cnt, 1//Decrement loop counter
+	brne		0b         //Perform loop one more time if loop_cnt is not zero
+
+	sub		sp, -8
+	popm		r0-r3, r4-r7, pc//Pop back registers and PC
+
+1:
+	ld.w		r12, pc[coef_table_copy - .]
+	ld.w		r9, pc[coef_table_copy - . + 4]
+
+	addhh.w		r4, r2:t, r2:b
+	mulhh.w		r4, r4:b, r12:t	// r4 = z1
+	mulhh.w		r5, r2:b, r12:b
+	ld.w		r12, pc[coef_table_copy - . + 8]
+	mulhh.w		r6, r2:t, r9:t
+	add		r5, r4	// r5 = tmp2
+	add		r6, r4	// r6 = tmp3
+
+	addhh.w		r7, r3:t, r3:b
+	subhh.w		r8, r3:t, r3:b
+
+	lsl		r7, CONST_BITS
+	lsl		r8, CONST_BITS
+
+	add		r2, r7, r6 // r2 = tmp10
+	sub		r3, r7, r6 // r3 = tmp13
+	add		r4, r8, r5 // r4 = tmp11
+	sub		r5, r8, r5 // r5 = tmp12
+
+
+	padd.h		r6, r0, r1 // r6:t = z4, r6:b = z3
+	addhh.w		r7, r6:t, r6:b
+	mulhh.w		r7, r7:b, r9:b // r7 = z5
+
+	ld.w		r9, pc[coef_table_copy - . + 12]
+	mulhh.w		r8, r6:b, r12:t // r8 = z3
+	mulhh.w		r6, r6:t, r12:b // r6 = z4
+
+	add		r8, r7
+	add		r6, r7
+
+	paddx.h		r7, r0, r1 // r7:t = z2, r7:b = z1
+
+	mulhh.w		r12, r0:b, r9:t // r12 = tmp0
+	mulhh.w		r0, r0:t, r9:b // r0 = tmp1
+	ld.w		r9, pc[coef_table_copy - . + 16]
+	add		r12, r8
+	add		r0, r6
+
+	ld.w		lr, pc[coef_table_copy - . + 20]
+	machh.w		r8, r1:b, r9:t // r8 = tmp2
+	machh.w		r6, r1:t, r9:b // r6 = tmp3
+	mulhh.w		r9, r7:b, lr:t // r9 = z1
+	mulhh.w		r7, r7:t, lr:b // r7 = z2
+
+	add		r12, r9
+	add		r0, r7
+	add		r8, r7
+	add		r6, r9
+
+	add		r1, r2, r6 // r1 = dataptr[DCTSIZE*0]
+	sub		r2, r2, r6 // r2 = dataptr[DCTSIZE*7]
+	add		r6, r4, r8 // r6 = dataptr[DCTSIZE*1]
+	sub		r4, r4, r8 // r4 = dataptr[DCTSIZE*6]
+	add		r8, r5, r0 // r8 = dataptr[DCTSIZE*2]
+	sub		r5, r5, r0 // r5 = dataptr[DCTSIZE*5]
+	add		r0, r3, r12 // r0 = dataptr[DCTSIZE*3]
+	sub		r3, r3, r12 // r3 = dataptr[DCTSIZE*4]
+
+	satrnds		r1 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r2 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r6 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r4 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r8 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r5 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r0 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r3 >> (CONST_BITS+PASS1_BITS+3), 9
+
+	packw.sh	r1, r1, r6
+	packw.sh	r8, r8, r0
+	packw.sh	r3, r3, r5
+	packw.sh	r4, r4, r2
+
+	packsh.ub	r1, r1, r8
+	packsh.ub	r0, r3, r4
+	lddsp		r12, SP[0]       // rfp
+	lddsp		r9, SP[4]	// iinc
+	st.d		r12[0], r0
+	sub		r10, -2	// Increment the dataptr
+	add		r12, r9	// increment rfp
+	stdsp		SP[0], r12
+
+	sub		loop_cnt, 1		//Decrement loop counter
+	brne		0b			//Perform loop one more time if loop_cnt is not zero
+
+	sub		sp, -8
+	popm		r0-r3, r4-r7, pc	//Pop back registers and PC
+
+	.align 2
+coef_table_copy:
+	.short	FIX_0_541196100, -FIX_1_847759065, FIX_0_765366865, FIX_1_175875602
+	.short	- FIX_1_961570560, - FIX_0_390180644, FIX_0_298631336, FIX_2_053119869
+	.short	FIX_3_072711026, FIX_1_501321110, - FIX_0_899976223, - FIX_2_562915447
+
+idct_avr32:
+	pushm		r0-r3, r4-r7, lr	//Free up registers to use for local variables
+
+		//; Give room for a temporary block on the stack
+	sub		sp, 8*8*2
+
+	mov		loop_cnt, 8		//Initialize loop counter
+
+0:
+	ldm		r12++, r0, r1, r2, r3	//Load 8 DCT-coeffisients from the current row in the DCT-block
+	mov		r6, 0
+#ifdef USE_PREFETCH
+	pref		r12[LINE_SIZE]		//Prefetch next line
+#endif
+	or		r4, r2, r3 << 16
+	or		r4, r1			//Check if all DCT-coeffisients except the DC is zero
+	or		r4, r0
+	brne		1f			//If there are non-zero AC coeffisients perform row-transform
+
+	paddsub.h	r5, r3:t, r6:b		//Extract the DC-coeff from r5
+	plsl.h		r5, r5, PASS1_BITS
+	mov		r4, r5
+	st.d		sp++, r4
+	st.d		sp++, r4
+
+	sub		loop_cnt, 1		//Decrement loop counter
+	brne		0b			//Perform loop one more time if loop_cnt is not zero
+
+	bral		2f	                //Perform coloumn transform after row transform is computed
+
+1:
+	ld.w		r10, pc[coef_table_idct - .]
+	ld.w		r9, pc[coef_table_idct - . + 4]
+
+	padd.h		r4, r2, r0 // r4:t = dataptr[2] + dataptr[6],r4:b = dataptr[3] + dataptr[7]
+	mulhh.w		r5, r4:t, r10:t
+	mulhh.w		r6, r0:t, r10:b
+	ld.w		r10, pc[coef_table_idct - . + 8]
+	mulhh.w		r7, r2:t, r9:t
+	add		r6, r5	// tmp2
+	satrnds		r6 >> (CONST_BITS - PASS1_BITS), 31
+	add		r7, r5	// tmp3
+	satrnds		r7 >> (CONST_BITS - PASS1_BITS), 31
+
+	paddsub.h	r5, r3:t, r1:t
+	plsl.h		r5, r5, PASS1_BITS // r5:t = tmp0, r5:b = tmp1
+
+	paddsub.h	r7, r5:t, r7:b // r7:t = tmp10, r7:b = tmp13
+	paddsub.h	r6, r5:b, r6:b // r6:t = tmp11, r6:b = tmp12
+
+	addhh.w		lr, r3:b, r1:b // lr = z4
+	addhh.w		r5, r4:b, lr:b
+	mulhh.w		r5, r5:b, r9:b // r5 = z5
+
+	ld.w		r9, pc[coef_table_idct - . + 12]
+	mulhh.w		r4, r4:b, r10:t // r4 = z3
+	mulhh.w		lr, lr:b, r10:b // lr = z4
+
+	add		r4, r5
+	add		lr, r5
+	addhh.w		r5, r2:b, r1:b // r5 = z2
+	addhh.w		r8, r3:b, r0:b // r8 = z1
+
+	mulhh.w		r0, r0:b, r9:t // r0 = tmp0
+	ld.w		r10, pc[coef_table_idct - . + 16]
+	mulhh.w		r1, r1:b, r9:b // r1 = tmp1
+	ld.w		r9, pc[coef_table_idct - . + 20]
+	mulhh.w		r2, r2:b, r10:t // r2 = tmp2
+	mulhh.w		r3, r3:b, r10:b // r3 = tmp3
+	mulhh.w		r8, r8:b, r9:t // r8 = z1
+	mulhh.w		r5, r5:b, r9:b // r5 = z2
+
+	add		r0, r8
+	add		r0, r4
+	add		r1, r5
+	add		r1, lr
+	add		r2, r5
+	add		r2, r4
+	add		r3, r8
+	add		r3, lr
+
+	satrnds		r0 >> (CONST_BITS - PASS1_BITS), 31
+	satrnds		r1 >> (CONST_BITS - PASS1_BITS), 31
+	satrnds		r2 >> (CONST_BITS - PASS1_BITS), 31
+	satrnds		r3 >> (CONST_BITS - PASS1_BITS), 31
+
+	paddsub.h	r5, r6:t, r2:b // r5:t = dataptr[1], r5:b = dataptr[6]
+	paddsub.h	r4, r7:t, r3:b // r4:t = dataptr[0], r4:b = dataptr[7]
+	paddsub.h	r3, r6:b, r1:b // r3:t = dataptr[2], r3:b = dataptr[5]
+	paddsub.h	r2, r7:b, r0:b // r2:t = dataptr[3], r2:b = dataptr[4]
+
+	sthh.w		sp[0], r4:t, r5:t
+	sthh.w		sp[4], r3:t, r2:t
+	sthh.w		sp[8], r2:b, r3:b
+	sthh.w		sp[12], r5:b, r4:b
+
+	sub		sp, -16
+	sub		loop_cnt, 1
+	brne		0b
+
+2:
+	sub		sp, 8*8*2	//Set pointer to start of DCT block
+	sub		r12, 8*8*2	//Set pointer to start of DCT block
+
+	mov		loop_cnt, 8
+
+0:
+	ldins.h		r3:t,sp[0]    // r3:t = dataptr[0]
+	ldins.h		r1:t,sp[1*8*2]// r1:t = dataptr[1]
+	ldins.h		r2:t,sp[2*8*2]// r2:t = dataptr[2]
+	ldins.h		r0:t,sp[5*8*2]// r0:t = dataptr[5]
+	ldins.h		r3:b,sp[4*8*2]// r3:b = dataptr[4]
+	ldins.h		r1:b,sp[3*8*2]// r1:b = dataptr[3]
+	ldins.h		r2:b,sp[6*8*2]// r2:b = dataptr[6]
+	ldins.h		r0:b,sp[7*8*2]// r0:b = dataptr[7]
+
+	or		r4, r1, r3 << 16
+	or		r4, r2
+	or		r4, r0
+	brne		1f		//If there are non-zero AC coeffisients perform row-transform
+
+	satrnds		r3 >> ( PASS1_BITS + 3 + 16 ), 31
+	packw.sh	r3, r3, r3
+	mov		r2, r3
+	st.d		r12++, r2
+	st.d		r12++, r2
+	sub		sp, -2	// Increment the dataptr
+
+	sub		loop_cnt, 1//Decrement loop counter
+	brne		0b         //Perform loop one more time if loop_cnt is not zero
+
+	sub		sp, -(8*8*2 - 8)
+	popm		r0-r3, r4-r7, pc//Pop back registers and PC
+
+1:
+	ld.w		r10, pc[coef_table_idct - .]
+	ld.w		r9, pc[coef_table_idct - . + 4]
+
+	addhh.w		r4, r2:t, r2:b
+	mulhh.w		r4, r4:b, r10:t	// r4 = z1
+	mulhh.w		r5, r2:b, r10:b
+	ld.w		r10, pc[coef_table_idct - . + 8]
+	mulhh.w		r6, r2:t, r9:t
+	add		r5, r4	// r5 = tmp2
+	add		r6, r4	// r6 = tmp3
+
+	addhh.w		r7, r3:t, r3:b
+	subhh.w		r8, r3:t, r3:b
+
+	lsl		r7, CONST_BITS
+	lsl		r8, CONST_BITS
+
+	add		r2, r7, r6 // r2 = tmp10
+	sub		r3, r7, r6 // r3 = tmp13
+	add		r4, r8, r5 // r4 = tmp11
+	sub		r5, r8, r5 // r5 = tmp12
+
+	padd.h		r6, r0, r1 // r6:t = z4, r6:b = z3
+	addhh.w		r7, r6:t, r6:b
+	mulhh.w		r7, r7:b, r9:b // r7 = z5
+
+	ld.w		r9, pc[coef_table_idct - . + 12]
+	mulhh.w		r8, r6:b, r10:t // r8 = z3
+	mulhh.w		r6, r6:t, r10:b // r6 = z4
+
+	add		r8, r7
+	add		r6, r7
+
+	paddx.h		r7, r0, r1 // r7:t = z2, r7:b = z1
+
+	mulhh.w		r10, r0:b, r9:t // r10 = tmp0
+	mulhh.w		r0, r0:t, r9:b // r0 = tmp1
+	ld.w		r9, pc[coef_table_idct - . + 16]
+	add		r10, r8
+	add		r0, r6
+
+	ld.w		lr, pc[coef_table_idct - . + 20]
+	machh.w		r8, r1:b, r9:t // r8 = tmp2
+	machh.w		r6, r1:t, r9:b // r6 = tmp3
+	mulhh.w		r9, r7:b, lr:t // r9 = z1
+	mulhh.w		r7, r7:t, lr:b // r7 = z2
+
+	add		r10, r9
+	add		r0, r7
+	add		r8, r7
+	add		r6, r9
+
+	add		r1, r2, r6 // r1 = dataptr[DCTSIZE*0]
+	sub		r2, r2, r6 // r2 = dataptr[DCTSIZE*7]
+	add		r6, r4, r8 // r6 = dataptr[DCTSIZE*1]
+	sub		r4, r4, r8 // r4 = dataptr[DCTSIZE*6]
+	add		r8, r5, r0 // r8 = dataptr[DCTSIZE*2]
+	sub		r5, r5, r0 // r5 = dataptr[DCTSIZE*5]
+	add		r0, r3, r10 // r0 = dataptr[DCTSIZE*3]
+	sub		r3, r3, r10 // r3 = dataptr[DCTSIZE*4]
+
+	satrnds		r1 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r2 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r6 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r4 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r8 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r5 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r0 >> (CONST_BITS+PASS1_BITS+3), 9
+	satrnds		r3 >> (CONST_BITS+PASS1_BITS+3), 9
+
+	packw.sh	r7, r1, r6
+	packw.sh	r6, r8, r0
+	packw.sh	r5, r3, r5
+	packw.sh	r4, r4, r2
+
+	stm		r12, r4-r7
+	sub		sp, -2	// Increment the dataptr
+	sub		r12, -16
+
+	sub		loop_cnt, 1		//Decrement loop counter
+	brne		0b			//Perform loop one more time if loop_cnt is not zero
+
+	sub		sp, -(8*8*2 - 8)
+	popm		r0-r3, r4-r7, pc	//Pop back registers and PC
+
+
+	.align 2
+coef_table_idct:
+	.short	FIX_0_541196100, -FIX_1_847759065, FIX_0_765366865, FIX_1_175875602
+	.short	- FIX_1_961570560, - FIX_0_390180644, FIX_0_298631336, FIX_2_053119869
+	.short	FIX_3_072711026, FIX_1_501321110, - FIX_0_899976223, - FIX_2_562915447
+
diff --git a/libavcodec/avr32/mc.S b/libavcodec/avr32/mc.S
new file mode 100644
index 0000000..2ffcf7f
--- /dev/null
+++ b/libavcodec/avr32/mc.S
@@ -0,0 +1,434 @@
+/*
+ * Copyright (c) 2007 Atmel Corporation. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials provided
+ * with the distribution.
+ *
+ * 3. The name of ATMEL may not be used to endorse or promote products
+ * derived from this software without specific prior written
+ * permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ATMEL ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE EXPRESSLY AND SPECIFICALLY DISCLAIMED. IN NO EVENT SHALL ATMEL
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
+ * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
+ * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+
+
+	/* Macro for masking the lowest bit of each byte in a
+	   packed word */
+	.macro	packedmask1	reg, round
+	.if	\round
+	and	\reg, \reg, r8 >> 1
+	.else
+	and	\reg, r8
+	.endif
+	.endm
+
+	/* Macro for 8 pixel wide horizontal and vertical interpolation functions */
+	.macro	pixels8_hv	round, put
+
+
+	pushm	r0-r7, lr
+
+	/* R12 = uint8_t *block, R11 = uint8_t pixels, R10 = int line_size, R9 = int h */
+
+	/* Rounding immediate */
+	.if	\round
+	mov	r8, lo(0x02020202)
+	orh	r8, hi(0x02020202)
+	.else
+	mov	r8, lo(0x01010101)
+	orh	r8, hi(0x01010101)
+	.endif
+	mov	r7, 2
+
+	/* Pixel naming convention :
+
+		|-----------------------------------------------------|
+		| s00 | s01 | s02 | s03 | s04 | s05 | s06 | s07 | s08 |
+		|----d00---d01---d02---d03---d04---d05---d06---d07----|
+		| s10 | s11 | s12 | s13 | s14 | s15 | s16 | s17 | s18 |
+		|-----------------------------------------------------|
+	*/
+1:
+	ld.w	r0, r11[0]		// r0 = { s00, s01, s02, s03 }
+	ld.w	r1, r11[1]		// r1 = { s01, s02, s03, s04 }
+	mov	lr, r9
+	eor	r2, r0, r1
+	packedmask1	r2, \round
+	add	r2, r8
+
+	paddh.ub	r0, r0, r1	// r0 = {(s00+s01)/2,(s01+s02)/2,(s02+s03)/2,(s03+s04)/2}
+
+	add		r11, r10	// pixels += line_size
+	ld.w	r1, r11[0]		// r1 = { s10, s11, s12, s13 }
+	ld.w	r3, r11[1]		// r3 = { s11, s12, s13, s14 }
+0:
+	eor	r5, r1, r3
+	packedmask1	r5, \round
+	add	r2, r5
+
+	paddh.ub	r1, r1, r3	// r1 = {(s10+s11)/2,(s11+s12)/2,(s12+s13)/2,(s13+s14)/2}
+	eor	r6, r0, r1
+	packedmask1	r6, \round
+	add	r2, r2, r6 << 1
+
+	ld.w	r3, r11[r10]		// r3 = { s00, s01, s02, s03 }
+	add	r11, r10		// pixels += line_size
+	ld.w	r4, r11[1]		// r4 = { s01, s02, s03, s04 }
+
+	paddh.ub	r0, r0, r1
+	plsr.b		r2, r2, 2
+	padd.b		r0, r0, r2	// r0 = { d00, d01, d02, d03 }
+
+	/* Next row */
+	.if	\put
+	eor	r2, r3, r4
+	packedmask1	r2, \round
+	add	r2, r8
+	.else
+	ld.w	r6, r12[0]
+	eor	r2, r3, r4
+	packedmask1	r2, \round
+	add	r2, r8
+	pavg.ub	r0, r0, r6
+	.endif
+	st.w	r12[0], r0		// Put data into the block
+
+	add	r5, r2
+	paddh.ub	r0, r3, r4	// r0 = {(s00+s01)/2,(s01+s02)/2,(s02+s03)/2,(s03+s04)/2}
+
+	eor	r6, r0, r1
+	packedmask1	r6, \round
+	add	r5, r5, r6 << 1
+
+	.if	\put
+	paddh.ub	r1, r0, r1
+	plsr.b		r5, r5, 2
+	padd.b		r1, r1, r5	// r1 = { d10, d11, d12, d13 }
+	.else
+	ld.w		r3, r12[r10]
+	paddh.ub	r1, r0, r1
+	plsr.b		r5, r5, 2
+	padd.b		r1, r1, r5	// r1 = { d10, d11, d12, d13 }
+	pavg.ub		r1, r1, r3
+	.endif
+
+	st.w	r12[r10], r1		// Put data into the block
+
+	ld.w	r1, r11[r10]		// r1 = { s10, s11, s12, s13 }
+	add	r11, r10		// pixels += line_size
+	ld.w	r3, r11[1]		// r3 = { s11, s12, s13, s14 }
+	add	r12, r12, r10 << 1	// block += 2*line_size
+	sub	lr, 2
+	brne	0b
+
+	mul	r0, r10, r9		// r0 = line_size * h
+	rsub	r0, r0, 4		// r0 = 4 - (line_size * h)
+	add	r11, r0
+	sub	r11, r10		// pixels += 4 - (line_size * (h+1))
+	add	r12, r0			// pixels += 4 - (line_size * (h))
+	sub	r7, 1
+	brne	1b
+
+	popm	r0-r7, pc
+	.endm
+
+
+	/* Macro for 8 pixel wide vertical interpolation functions */
+
+	.macro	pixels8_v	round, put
+	pushm	r4-r7,lr
+	/* R12 = uint8_t *block, R11 = uint8_t pixels, R10 = int line_size, R9 = int h */
+
+	/*
+		Pixel Naming Convention :
+		|-----------------------------------------------|
+		| s00 | s01 | s02 | s03 | s04 | s05 | s06 | s07 |
+		|-d00---d01---d02---d03---d04---d05---d06---d07-|
+		| s10 | s11 | s12 | s13 | s14 | s15 | s16 | s17 |
+		|-----------------------------------------------|
+	*/
+	ld.w	r8, r11[r10]		// r8 = { s10, s11, s12, s13 }
+	ld.w	lr, r11++		// lr = { s00, s01, s02, s03 }, src += 4
+	ld.w	r7, r11[0]		// r7 = { s04, s05, s06, s07 }
+	ld.w	r6, r11[r10]		// r6 = { s14, s15, s16, s17 }
+	sub	r10, 4			// stride -= 4
+	add	r11, r11, r10 << 1	// src += 2*stride
+	sub	r11, -4			// src += 4
+
+0:
+	.if	\round
+	pavg.ub r5, r8, lr		// r5 = {(s10+s00)/2,(s11+s01)/2,(s12+s02)/2,(s13+s03)/2}
+	pavg.ub r4, r6, r7		// r4 = {(s14+s04)/2,(s15+s05)/2,(s16+s06)/2,(s17+s07)/2}
+	.else
+	paddh.ub r5, r8, lr		// r5 = {(s10+s00)/2,(s11+s01)/2,(s12+s02)/2,(s13+s03)/2}
+	paddh.ub r4, r6, r7		// r4 = {(s14+s04)/2,(s15+s05)/2,(s16+s06)/2,(s17+s07)/2}
+	.endif
+
+	.if	\put
+	st.w	r12++, r5		// *dst++ = { d00, d01, d02, d03 }
+	ld.w	lr, r11++		// lr = { s10, s11, s12, s13 }, src += 4
+	st.w	r12[0], r4		// *dst = { d04, d05, d06, d07 }
+	ld.w	r7, r11[0]		// r7 = { s14, s15, s16, s17 }
+	.else
+	ld.w	lr, r12[0]
+	ld.w	r7, r12[4]
+	pavg.ub	r5, r5, lr
+	pavg.ub	r4, r4, r7
+	st.w	r12++, r5		// *dst++ = { d00, d01, d02, d03 }
+	ld.w	lr, r11++		// lr = { s10, s11, s12, s13 }, src += 4
+	st.w	r12[0], r4		// *dst = { d04, d05, d06, d07 }
+	ld.w	r7, r11[0]		// r7 = { s14, s15, s16, s17 }
+	.endif
+	add	r11, r10		// src += stride
+#ifdef USE_PREFETCH
+	pref		r11[0]
+#endif
+	add	r12, r10		// dst += stride
+
+	.if	\round
+	pavg.ub r5, r8, lr		// r5 = {(s10+s00)/2,(s11+s01)/2,(s12+s02)/2,(s13+s03)/2}
+	pavg.ub r4, r6, r7		// r4 = {(s14+s04)/2,(s15+s05)/2,(s16+s06)/2,(s17+s07)/2}
+	.else
+	paddh.ub r5, r8, lr		// r5 = {(s10+s00)/2,(s11+s01)/2,(s12+s02)/2,(s13+s03)/2}
+	paddh.ub r4, r6, r7		// r4 = {(s14+s04)/2,(s15+s05)/2,(s16+s06)/2,(s17+s07)/2}
+	.endif
+	.if	\put
+	st.w	r12++, r5		// *dst++ = { d00, d01, d02, d03 }
+	ld.w	r8, r11++		// r8 = { s10, s11, s12, s13 }, src += 4
+	st.w	r12[0], r4		// *dst = { d04, d05, d06, d07 }
+	ld.w	r6, r11[0]		// r6 = { s14, s15, s16, s17 }
+	.else
+	ld.w	r8, r12[0]
+	ld.w	r6, r12[4]
+	pavg.ub	r5, r5, r8
+	pavg.ub	r4, r4, r6
+	st.w	r12++, r5		// *dst++ = { d00, d01, d02, d03 }
+	ld.w	r8, r11++		// r8 = { s10, s11, s12, s13 }, src += 4
+	st.w	r12[0], r4		// *dst = { d04, d05, d06, d07 }
+	ld.w	r6, r11[0]		// r6 = { s14, s15, s16, s17 }
+	.endif
+
+	add	r11, r10		// src += stride
+#ifdef USE_PREFETCH
+	pref		r11[0]
+#endif
+	add	r12, r10		// dst += stride
+	sub	r9, 2
+	brne	0b
+
+	popm	r4-r7,pc
+	.endm
+
+	/* Macro for 8 pixel wide horizontal interpolation functions */	
+
+	.macro	pixels8_h	round, put
+	pushm	r4-r7, lr
+
+	/* R12 = uint8_t *block, R11 = uint8_t pixels, R10 = int line_size, R9 = int h */
+	/*
+		 Pixel Naming Convention:
+		|--------------------------------------------------------------------|
+		| s00 d00 s01 d01 s02 d02 s03 d03 s04 d04 s05 d05 s06 d06 s07 d07 s08|
+		|------|-------|-------|-------|-------|-------|-------|-------|-----|
+		| s10 d10 s11 d11 s12 d12 s13 d13 s14 d14 s15 d15 s16 d16 s17 d17 s18|
+		|--------------------------------------------------------------------|
+	*/
+
+	ld.w	lr, r11[0]	// lr = { s00, s01, s02, s03 }
+	ld.w	r8, r11[1]	// r8 = { s01, s02, s03, s04 }
+	ld.w	r7, r11[4]	// r7 = { s04, s05, s06, s07 }
+	ld.w	r6, r11[5]	// r6 = { s05, s06, s07, s08 }
+	add	r11, r10	// src += stride
+
+0:
+	.if	\round
+	pavg.ub lr, r8, lr	// lr = {(s00+s01)/2,(s01+s02)/2,(s02+s03)/2,(s03+s04)/2}
+	pavg.ub r7, r6, r7	// r7 = {(s04+s05)/2,(s05+s06)/2,(s06+s07)/2,(s07+s08)/2}
+	.else
+	paddh.ub lr, r8, lr	// lr = {(s00+s01)/2,(s01+s02)/2,(s02+s03)/2,(s03+s04)/2}
+	paddh.ub r7, r6, r7	// r7 = {(s04+s05)/2,(s05+s06)/2,(s06+s07)/2,(s07+s08)/2}
+	.endif
+	.if	\put
+	ld.w	r5, r11[0]	// r5 = { s00, s01, s02, s03 }
+	ld.w	r4, r11[1]	// r4 = { s01, s02, s03, s04 }
+	.else
+	ld.w	r8, r12[0]
+	ld.w	r6, r12[4]
+	ld.w	r5, r11[0]	// r5 = { s00, s01, s02, s03 }
+	ld.w	r4, r11[1]	// r4 = { s01, s02, s03, s04 }
+	pavg.ub	lr, lr, r8
+	pavg.ub	r7, r7, r6
+	.endif
+	st.w	r12[0], lr	// dst = { d00, d01, d02, d03 }
+	st.w	r12[4], r7	// dst = { d04, d05, d06, d07 }
+	ld.w	r8, r11[4]	// r8 = { s04, s05, s06, s07 }
+	ld.w	r6, r11[5]	// r6 = { s05, s06, s07, s08 }
+	add	r11, r10	// src += stride
+#ifdef USE_PREFETCH
+	pref		r11[0]
+#endif
+	add	r12, r10	// dst += stride
+
+	.if	\round
+	pavg.ub r5, r4, r5	// r5 = {(s00+s01)/2,(s01+s02)/2,(s02+s03)/2,(s03+s04)/2}
+	pavg.ub r4, r6, r8	// r4 = {(s04+s05)/2,(s05+s06)/2,(s06+s07)/2,(s07+s08)/2}
+	.else
+	paddh.ub r5, r4, r5	// r5 = {(s00+s01)/2,(s01+s02)/2,(s02+s03)/2,(s03+s04)/2}
+	paddh.ub r4, r6, r8	// r4 = {(s04+s05)/2,(s05+s06)/2,(s06+s07)/2,(s07+s08)/2}
+	.endif
+	.if	\put
+	ld.w	lr, r11[0]	// lr = { s00, s01, s02, s03 }
+	ld.w	r8, r11[1]	// r8 = { s01, s02, s03, s04 }
+	.else
+	ld.w	r7, r12[0]
+	ld.w	r6, r12[4]
+	ld.w	lr, r11[0]	// lr = { s00, s01, s02, s03 }
+	ld.w	r8, r11[1]	// r8 = { s01, s02, s03, s04 }
+	pavg.ub	r5, r5, r7
+	pavg.ub	r4, r4, r6
+	.endif
+	st.w	r12[0], r5	// dst = { d00, d01, d02, d03 }
+	st.w	r12[4], r4	// dst = { d04, d05, d06, d07 }
+	ld.w	r7, r11[4]	// r7 = { s04, s05, s06, s07 }
+	ld.w	r6, r11[5]	// r6 = { s05, s06, s07, s08 }
+	add	r11, r10	// src += stride
+#ifdef USE_PREFETCH
+	pref		r11[0]
+#endif
+	add	r12, r10	// dst += stride
+	sub	r9, 2
+	brne	0b
+
+	popm	r4-r7, pc
+	.endm
+
+	/* Macro for 8 pixel wide copy functions */
+	.macro	pixels8	put
+	stm		--sp, r3-r7,lr
+	/* R12 = uint8_t *block, R11 = uint8_t pixels, R10 = int line_size, R9 = int h */
+	mov		lr, r9
+	sub		r3, r10, 2	        // stride2 = stride - 2
+0:
+	.if	\put
+	ld.w		r9, r11[r10]	        // r9 = { s10, s11, s12, s13 }
+	ld.w		r7, r11++	        // r7 = { s00, s01, s02, s03 }, src += 4
+	ld.w		r6, r11[0]	        // r6 = { s04, s05, s06, s07 }
+	ld.w		r8, r11[r10]	        // r8 = { s14, s15, s16, s17 }
+	.else
+	ld.w		r9, r11[r10]	        // r9 = { s10, s11, s12, s13 }
+	ld.d		r4, r12[0]
+	ld.w		r7, r11++	        // r7 = { s00, s01, s02, s03 }, src += 4
+	ld.w		r6, r11[0]	        // r6 = { s04, s05, s06, s07 }
+	ld.w		r8, r11[r10]	        // r8 = { s14, s15, s16, s17 }
+	pavg.ub		r6, r6, r4
+	pavg.ub		r7, r7, r5
+	ld.d		r4, r12[r10]
+	.endif
+	st.d		r12, r6			// *dst = { s00, s01, s02, s03, s04, s05, s06, s07 }
+	add		r11, r11, r3 << 1	// src += stride2 * 2
+	.ifeq	\put
+	pavg.ub		r8, r8, r4
+	pavg.ub		r9, r9, r5
+	.endif
+	st.d		r12[r10 << 0], r8	// *(dst + stride) = { s10, s11, s12, s13, s14, s15, s16, s17 }
+	add		r12, r12, r10 << 1	// dst += 2*stride
+	sub		lr, 2
+	brne		0b
+	ldm		sp++, r3-r7,pc
+
+	.endm
+
+	.global	put_no_rnd_pixels8_hv_avr32
+	.text
+put_no_rnd_pixels8_hv_avr32:
+	pixels8_hv	0, 1
+
+	.global	put_pixels8_hv_avr32
+	.text
+put_pixels8_hv_avr32:
+	pixels8_hv	1, 1
+
+	.global	avg_no_rnd_pixels8_hv_avr32
+	.text
+avg_no_rnd_pixels8_hv_avr32:
+	pixels8_hv	0, 0
+
+	.global	avg_pixels8_hv_avr32
+	.text
+avg_pixels8_hv_avr32:
+	pixels8_hv	1, 0
+
+	.global	put_no_rnd_pixels8_v_avr32
+	.text
+put_no_rnd_pixels8_v_avr32:
+	pixels8_v	0, 1
+
+	.global	put_pixels8_v_avr32
+	.text
+put_pixels8_v_avr32:
+	pixels8_v	1, 1
+
+	.global	avg_no_rnd_pixels8_v_avr32
+	.text
+avg_no_rnd_pixels8_v_avr32:
+	pixels8_v	0, 0
+
+	.global	avg_pixels8_v_avr32
+	.text
+avg_pixels8_v_avr32:
+	pixels8_v	1, 0
+
+	.global	put_no_rnd_pixels8_h_avr32
+	.text
+put_no_rnd_pixels8_h_avr32:
+	pixels8_h	0, 1
+
+	.global	put_pixels8_h_avr32
+	.text
+put_pixels8_h_avr32:
+	pixels8_h	1, 1
+
+	.global	avg_no_rnd_pixels8_h_avr32
+	.text
+avg_no_rnd_pixels8_h_avr32:
+	pixels8_h	0, 0
+
+	.global	avg_pixels8_h_avr32
+	.text
+avg_pixels8_h_avr32:
+	pixels8_h	1, 0
+
+	.global	put_pixels8_avr32
+	.global	put_no_rnd_pixels8_avr32
+	.text
+put_pixels8_avr32:
+put_no_rnd_pixels8_avr32:
+	pixels8	1
+
+	.global	avg_no_rnd_pixels8_avr32
+	.global	avg_pixels8_avr32
+	.text
+avg_pixels8_avr32:
+avg_no_rnd_pixels8_avr32:
+	pixels8	0
+
diff --git a/libavcodec/avr32/pico.h b/libavcodec/avr32/pico.h
new file mode 100644
index 0000000..f2020fd
--- /dev/null
+++ b/libavcodec/avr32/pico.h
@@ -0,0 +1,255 @@
+/*
+ * Copyright (c) 2007 Atmel Corporation. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following
+ * disclaimer in the documentation and/or other materials provided
+ * with the distribution.
+ *
+ * 3. The name of ATMEL may not be used to endorse or promote products
+ * derived from this software without specific prior written
+ * permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ATMEL ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE EXPRESSLY AND SPECIFICALLY DISCLAIMED. IN NO EVENT SHALL ATMEL
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
+ * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
+ * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+ * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+ * DAMAGE.
+ */
+#ifndef __PICO_H__
+#define __PICO_H__
+
+/* Coprocessor Number */
+#define PICO_CPNO  1
+
+/* Pixel Coprocessor Register file */
+#define PICO_REGVECT_INPIX2  cr0
+#define PICO_REGVECT_INPIX1  cr1
+#define PICO_REGVECT_INPIX0  cr2
+#define PICO_REGVECT_OUTPIX2 cr3
+#define PICO_REGVECT_OUTPIX1 cr4
+#define PICO_REGVECT_OUTPIX0 cr5
+#define PICO_REGVECT_COEFF0_A cr6
+#define PICO_REGVECT_COEFF0_B cr7
+#define PICO_REGVECT_COEFF1_A cr8
+#define PICO_REGVECT_COEFF1_B cr9
+#define PICO_REGVECT_COEFF2_A cr10
+#define PICO_REGVECT_COEFF2_B cr11
+#define PICO_REGVECT_VMU0_OUT cr12
+#define PICO_REGVECT_VMU1_OUT cr13
+#define PICO_REGVECT_VMU2_OUT cr14
+#define PICO_REGVECT_CONFIG   cr15
+
+#define PICO_INPIX2  0
+#define PICO_INPIX1  1
+#define PICO_INPIX0  2
+#define PICO_OUTPIX2 3
+#define PICO_OUTPIX1 4
+#define PICO_OUTPIX0 5
+#define PICO_COEFF0_A 6
+#define PICO_COEFF0_B 7
+#define PICO_COEFF1_A 8
+#define PICO_COEFF1_B 9
+#define PICO_COEFF2_A 10
+#define PICO_COEFF2_B 11
+#define PICO_VMU0_OUT 12
+#define PICO_VMU1_OUT 13
+#define PICO_VMU2_OUT 14
+#define PICO_CONFIG   15
+
+/* Config Register */
+#define PICO_COEFF_FRAC_BITS_OFFSET  0
+#define PICO_COEFF_FRAC_BITS_SIZE  4
+#define PICO_OFFSET_FRAC_BITS_OFFSET  4
+#define PICO_OFFSET_FRAC_BITS_SIZE  4
+#define PICO_INPUT_MODE_OFFSET  8
+#define PICO_INPUT_MODE_SIZE  2
+#define PICO_OUTPUT_MODE_OFFSET 10
+#define PICO_OUTPUT_MODE_SIZE 1
+
+struct pico_config_t {
+  unsigned int          : 32 - PICO_OUTPUT_MODE_OFFSET - PICO_OUTPUT_MODE_SIZE;
+  unsigned int          output_mode : PICO_OUTPUT_MODE_SIZE;
+  unsigned int          input_mode : PICO_INPUT_MODE_SIZE;
+  unsigned int          offset_frac_bits : PICO_OFFSET_FRAC_BITS_SIZE;
+  unsigned int          coeff_frac_bits : PICO_COEFF_FRAC_BITS_SIZE;
+  int                   vmu2_out;
+  int                   vmu1_out;
+  int                   vmu0_out;
+  short                 coeff2_2;
+  short                 coeff2_3;
+  short                 coeff2_0;
+  short                 coeff2_1;
+  short                 coeff1_2;
+  short                 coeff1_3;
+  short                 coeff1_0;
+  short                 coeff1_1;
+  short                 coeff0_2;
+  short                 coeff0_3;
+  short                 coeff0_0;
+  short                 coeff0_1;
+};
+
+
+#define PICO_COEFF_FRAC_BITS(x) (x << PICO_COEFF_FRAC_BITS_OFFSET)
+#define PICO_OFFSET_FRAC_BITS(x) (x << PICO_OFFSET_FRAC_BITS_OFFSET)
+#define PICO_INPUT_MODE(x) (x << PICO_INPUT_MODE_OFFSET)
+#define PICO_OUTPUT_MODE(x) (x << PICO_OUTPUT_MODE_OFFSET)
+
+#define GET_PICO_COEFF_FRAC_BITS(x) ((x >> PICO_COEFF_FRAC_BITS_OFFSET)&((1 << PICO_COEFF_FRAC_BITS_SIZE)-1))
+#define GET_PICO_OFFSET_FRAC_BITS(x) ((x >> PICO_OFFSET_FRAC_BITS_OFFSET)&((1 << PICO_OFFSET_FRAC_BITS_SIZE)-1))
+#define GET_PICO_INPUT_MODE(x) ((x >> PICO_INPUT_MODE_OFFSET)&((1 << PICO_INPUT_MODE_SIZE)-1))
+#define GET_PICO_OUTPUT_MODE(x) ((x >> PICO_OUTPUT_MODE_OFFSET)&((1 << PICO_OUTPUT_MODE_SIZE)-1))
+
+enum pico_input_mode { PICO_TRANSFORMATION_MODE,
+                       PICO_HOR_FILTER_MODE,
+                       PICO_VERT_FILTER_MODE };
+
+enum pico_output_mode { PICO_PACKED_MODE,
+                        PICO_PLANAR_MODE };
+
+/* Bits in coefficients */
+#define PICO_COEFF_BITS 12
+
+/* Operation bits */
+#define PICO_MATRIX (0)
+#define PICO_USE_ACC (1 << 2)
+#define PICO_SINGLE_VECTOR (1 << 3)
+
+
+#define __str(x...) #x
+#define __xstr(x...) __str(x)
+
+#define PICO_PUT_W(pico_reg, x) \
+  __builtin_mvrc_w(PICO_CPNO, pico_reg, x);
+#define PICO_GET_W(pico_reg) \
+  __builtin_mvcr_w(PICO_CPNO, pico_reg)
+
+#define PICO_MVCR_W(x, pico_reg) \
+  asm ("mvcr.w\tcp" __xstr(PICO_CPNO) ", %0, cr" __xstr(pico_reg) : "=r"(x));
+
+#define PICO_MVRC_W(pico_reg, x) \
+  asm  ("mvrc.w\tcp" __xstr(PICO_CPNO) ", cr" __xstr(pico_reg) ", %0" :: "r"(x));
+
+#define PICO_PUT_D(pico_reg, x) \
+  __builtin_mvrc_d(PICO_CPNO, pico_reg, x);
+#define PICO_GET_D(pico_reg) \
+  __builtin_mvcr_d(PICO_CPNO, pico_reg)
+
+#define PICO_MVCR_D(x, pico_reg) \
+  asm volatile ("mvcr.d\tcp" __xstr(PICO_CPNO) ", %0, cr" __xstr(pico_reg) : "=r"(x));
+#define PICO_MVRC_D(pico_reg, x) \
+  asm volatile ("mvrc.d\tcp" __xstr(PICO_CPNO) ", cr" __xstr(pico_reg) ", %0" :: "r"(x));
+
+#define PICO_STCM_W(ptr, pico_regs...) \
+  asm volatile ("stcm.w\tcp" __xstr(PICO_CPNO) ", %0," __xstr(pico_regs)  :: "r"(ptr));
+#define PICO_STCM_D(ptr, pico_regs...) \
+  asm volatile ("stcm.d\tcp" __xstr(PICO_CPNO) ", %0," __xstr(pico_regs)  :: "r"(ptr));
+
+#define PICO_STCM_W_DEC(ptr, pico_regs...) \
+  asm volatile ("stcm.w\tcp" __xstr(PICO_CPNO) ", --%0," __xstr(pico_regs)  : "+r"(ptr));
+#define PICO_STCM_D_DEC(ptr, pico_regs...) \
+  asm volatile ("stcm.d\tcp" __xstr(PICO_CPNO) ", --%0," __xstr(pico_regs)  : "+r"(ptr));
+
+#define PICO_LDCM_W(ptr, pico_regs...) \
+  asm volatile ("ldcm.w\tcp" __xstr(PICO_CPNO) ", %0," __xstr(pico_regs)  :: "r"(ptr));
+#define PICO_LDCM_D(ptr, pico_regs...) \
+  asm volatile ("ldcm.d\tcp" __xstr(PICO_CPNO) ", %0," __xstr(pico_regs)  :: "r"(ptr));
+
+#define PICO_LDCM_W_INC(ptr, pico_regs...) \
+  asm volatile ("ldcm.w\tcp" __xstr(PICO_CPNO) ", %0++," __xstr(pico_regs)  : "+r"(ptr));
+#define PICO_LDCM_D_INC(ptr, pico_regs...) \
+  asm volatile ("ldcm.d\tcp" __xstr(PICO_CPNO) ", %0++," __xstr(pico_regs)  : "+r"(ptr));
+
+#define PICO_OP(op, dst_addr, addr0, addr1, addr2) \
+  __builtin_cop(PICO_CPNO, addr0, addr1, addr2, op | dst_addr);
+
+static inline void set_pico_config(struct pico_config_t *config){
+  PICO_LDCM_D(config,
+              PICO_REGVECT_COEFF0_A, PICO_REGVECT_COEFF0_B,
+              PICO_REGVECT_COEFF1_A, PICO_REGVECT_COEFF1_B,
+              PICO_REGVECT_COEFF2_A, PICO_REGVECT_COEFF2_B,
+              PICO_REGVECT_VMU0_OUT, PICO_REGVECT_VMU1_OUT,
+              PICO_REGVECT_VMU2_OUT, PICO_REGVECT_CONFIG);
+}
+
+static inline void get_pico_config(struct pico_config_t *config){
+  PICO_STCM_D(config,
+              PICO_REGVECT_COEFF0_A, PICO_REGVECT_COEFF0_B,
+              PICO_REGVECT_COEFF1_A, PICO_REGVECT_COEFF1_B,
+              PICO_REGVECT_COEFF2_A, PICO_REGVECT_COEFF2_B,
+              PICO_REGVECT_VMU0_OUT, PICO_REGVECT_VMU1_OUT,
+              PICO_REGVECT_VMU2_OUT, PICO_REGVECT_CONFIG);
+}
+
+static inline void dump_pico_config(){
+  struct pico_config_t pico_config;
+  char *input_mode, *output_mode;
+  get_pico_config(&pico_config);
+
+  av_log(NULL, AV_LOG_INFO, "Dumping pico configuration:\n\n");
+  av_log(NULL, AV_LOG_INFO, "\tcoeff_frac_bits = %d\n", pico_config.coeff_frac_bits);
+  av_log(NULL, AV_LOG_INFO, "\toffset_frac_bits = %d\n", pico_config.offset_frac_bits);
+
+  switch ( pico_config.input_mode ){
+  case PICO_TRANSFORMATION_MODE:
+    input_mode = "Transformation Mode";
+    break;
+  case PICO_HOR_FILTER_MODE:
+    input_mode = "Horisontal Filter Mode";
+    break;
+  case PICO_VERT_FILTER_MODE:
+    input_mode = "Vertical Filter Mode";
+    break;
+  default:
+    input_mode = "Unknown Mode!!";
+    break;
+  }
+  av_log(NULL, AV_LOG_INFO, "\tinput_mode = %s\n", input_mode);
+
+  switch ( pico_config.output_mode ){
+  case PICO_PLANAR_MODE:
+    output_mode = "Planar Mode";
+    break;
+  case PICO_PACKED_MODE:
+    output_mode = "Packed Mode";
+    break;
+  default:
+    output_mode = "Unknown Mode!!";
+    break;
+  }
+
+  av_log(NULL, AV_LOG_INFO, "\toutput_mode = %s\n", output_mode);
+
+  av_log(NULL, AV_LOG_INFO, "\tCoeff0_0 = %f\n", (float)pico_config.coeff0_0/(float)(1 << pico_config.coeff_frac_bits));
+  av_log(NULL, AV_LOG_INFO, "\tCoeff0_1 = %f\n", (float)pico_config.coeff0_1/(float)(1 << pico_config.coeff_frac_bits));
+  av_log(NULL, AV_LOG_INFO, "\tCoeff0_2 = %f\n", (float)pico_config.coeff0_2/(float)(1 << pico_config.coeff_frac_bits));
+  av_log(NULL, AV_LOG_INFO, "\tCoeff0_3 = %f\n", (float)pico_config.coeff0_3/(float)(1 << pico_config.offset_frac_bits));
+
+  av_log(NULL, AV_LOG_INFO, "\tCoeff1_0 = %f\n", (float)pico_config.coeff1_0/(float)(1 << pico_config.coeff_frac_bits));
+  av_log(NULL, AV_LOG_INFO, "\tCoeff1_1 = %f\n", (float)pico_config.coeff1_1/(float)(1 << pico_config.coeff_frac_bits));
+  av_log(NULL, AV_LOG_INFO, "\tCoeff1_2 = %f\n", (float)pico_config.coeff1_2/(float)(1 << pico_config.coeff_frac_bits));
+  av_log(NULL, AV_LOG_INFO, "\tCoeff1_3 = %f\n", (float)pico_config.coeff1_3/(float)(1 << pico_config.offset_frac_bits));
+
+  av_log(NULL, AV_LOG_INFO, "\tCoeff2_0 = %f\n", (float)pico_config.coeff2_0/(float)(1 << pico_config.coeff_frac_bits));
+  av_log(NULL, AV_LOG_INFO, "\tCoeff2_1 = %f\n", (float)pico_config.coeff2_1/(float)(1 << pico_config.coeff_frac_bits));
+  av_log(NULL, AV_LOG_INFO, "\tCoeff2_2 = %f\n", (float)pico_config.coeff2_2/(float)(1 << pico_config.coeff_frac_bits));
+  av_log(NULL, AV_LOG_INFO, "\tCoeff2_3 = %f\n", (float)pico_config.coeff2_3/(float)(1 << pico_config.offset_frac_bits));
+}
+
+#endif
+
diff --git a/libavcodec/bitstream.h b/libavcodec/bitstream.h
index e1ec934..f2095be 100644
--- a/libavcodec/bitstream.h
+++ b/libavcodec/bitstream.h
@@ -181,6 +181,7 @@ typedef struct RL_VLC_ELEM {
     uint8_t run;
 } RL_VLC_ELEM;
 
+
 #ifndef ALT_BITSTREAM_WRITER
 static inline void put_bits(PutBitContext *s, int n, unsigned int value)
 {
@@ -799,6 +800,44 @@ void free_vlc(VLC *vlc);
  * if the vlc code is invalid and max_depth>1 than the number of bits removed
  * is undefined
  */
+
+#if defined(ARCH_AVR32)
+#define GET_VLC(code, name, gb, table, bits, max_depth)\
+{\
+    int n, index, nb_bits;\
+    union { VLC_TYPE vlc[2];\
+            uint32_t u32; } table_elem;\
+\
+    index= SHOW_UBITS(name, gb, bits);\
+    table_elem.u32 = (*((uint32_t*)(&table[index]))); \
+    code = table_elem.vlc[0];\
+    n    = table_elem.vlc[1];\
+\
+    if(max_depth > 1 && n < 0 ){\
+        LAST_SKIP_BITS(name, gb, bits)\
+        UPDATE_CACHE(name, gb)\
+\
+        nb_bits = -n;\
+\
+        index= SHOW_UBITS(name, gb, nb_bits) + code;\
+        table_elem.u32 = (*((uint32_t*)(&table[index]))); \
+        code = table_elem.vlc[0];\
+        n    = table_elem.vlc[1];\
+        if(max_depth > 2 && n < 0){\
+            LAST_SKIP_BITS(name, gb, nb_bits)\
+            UPDATE_CACHE(name, gb)\
+\
+            nb_bits = -n;\
+\
+            index= SHOW_UBITS(name, gb, nb_bits) + code;\
+            code = table[index][0];\
+            n    = table[index][1];\
+        }\
+    }\
+    SKIP_BITS(name, gb, n)\
+}
+
+#else
 #define GET_VLC(code, name, gb, table, bits, max_depth)\
 {\
     int n, index, nb_bits;\
@@ -807,7 +846,7 @@ void free_vlc(VLC *vlc);
     code = table[index][0];\
     n    = table[index][1];\
 \
-    if(max_depth > 1 && n < 0){\
+    if(max_depth > 1 && n < 0 ){\
         LAST_SKIP_BITS(name, gb, bits)\
         UPDATE_CACHE(name, gb)\
 \
@@ -829,7 +868,38 @@ void free_vlc(VLC *vlc);
     }\
     SKIP_BITS(name, gb, n)\
 }
+#endif
 
+#if defined(ARCH_AVR32)
+#define GET_RL_VLC(level, run, name, gb, table, bits, max_depth, need_update)\
+{\
+    int n, index, nb_bits;\
+    union { RL_VLC_ELEM vlc;\
+            uint32_t u32; } table_elem;\
+\
+    index= SHOW_UBITS(name, gb, bits);\
+    table_elem.u32 = (*((uint32_t*)(&table[index]))); \
+    level = table_elem.vlc.level;\
+    n     = table_elem.vlc.len;\
+\
+    if(max_depth > 1 && n < 0 ){\
+        SKIP_BITS(name, gb, bits)\
+        if(need_update){\
+            UPDATE_CACHE(name, gb)\
+        }\
+\
+        nb_bits = -n;\
+\
+        index= SHOW_UBITS(name, gb, nb_bits) + level;\
+        table_elem.u32 = (*((uint32_t*)(&table[index]))); \
+        level = table_elem.vlc.level;\
+        n     = table_elem.vlc.len;\
+    }\
+    run= table_elem.vlc.run;\
+    SKIP_BITS(name, gb, n)\
+}
+
+#else
 #define GET_RL_VLC(level, run, name, gb, table, bits, max_depth, need_update)\
 {\
     int n, index, nb_bits;\
@@ -838,7 +908,7 @@ void free_vlc(VLC *vlc);
     level = table[index].level;\
     n     = table[index].len;\
 \
-    if(max_depth > 1 && n < 0){\
+    if(max_depth > 1 && n < 0 ){\
         SKIP_BITS(name, gb, bits)\
         if(need_update){\
             UPDATE_CACHE(name, gb)\
@@ -853,7 +923,7 @@ void free_vlc(VLC *vlc);
     run= table[index].run;\
     SKIP_BITS(name, gb, n)\
 }
-
+#endif
 
 /**
  * parses a vlc code, faster then get_vlc()
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index b33fb50..30418bf 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -4644,6 +4644,7 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx)
 
     if (HAVE_MMX)        dsputil_init_mmx   (c, avctx);
     if (ARCH_ARM)        dsputil_init_arm   (c, avctx);
+    if (ARCH_AVR32)      dsputil_init_avr32 (c, avctx);
     if (CONFIG_MLIB)     dsputil_init_mlib  (c, avctx);
     if (HAVE_VIS)        dsputil_init_vis   (c, avctx);
     if (ARCH_ALPHA)      dsputil_init_alpha (c, avctx);
diff --git a/libavcodec/h264.c b/libavcodec/h264.c
index 285f9c9..133b918 100644
--- a/libavcodec/h264.c
+++ b/libavcodec/h264.c
@@ -2010,7 +2010,12 @@ static void free_tables(H264Context *h){
 
 static void init_dequant8_coeff_table(H264Context *h){
     int i,q,x;
+#ifdef ARCH_AVR32
+    const int transpose = 0;
+#else
     const int transpose = (h->s.dsp.h264_idct8_add != ff_h264_idct8_add_c); //FIXME ugly
+#endif
+
     h->dequant8_coeff[0] = h->dequant8_buffer[0];
     h->dequant8_coeff[1] = h->dequant8_buffer[1];
 
@@ -2033,7 +2038,13 @@ static void init_dequant8_coeff_table(H264Context *h){
 
 static void init_dequant4_coeff_table(H264Context *h){
     int i,j,q,x;
+    // Yes this is ugly as hell....
+#ifdef ARCH_AVR32
+    const int transpose = 0;
+#else
     const int transpose = (h->s.dsp.h264_idct_add != ff_h264_idct_add_c); //FIXME ugly
+#endif
+
     for(i=0; i<6; i++ ){
         h->dequant4_coeff[i] = h->dequant4_buffer[i];
         for(j=0; j<i; j++){
@@ -3560,7 +3571,11 @@ static int init_poc(H264Context *h){
 static void init_scan_tables(H264Context *h){
     MpegEncContext * const s = &h->s;
     int i;
+#ifdef ARCH_AVR32
+    if(1){
+#else
     if(s->dsp.h264_idct_add == ff_h264_idct_add_c){ //FIXME little ugly
+#endif
         memcpy(h->zigzag_scan, zigzag_scan, 16*sizeof(uint8_t));
         memcpy(h-> field_scan,  field_scan, 16*sizeof(uint8_t));
     }else{
-- 
1.5.6.3




More information about the MPlayer-dev-eng mailing list