h264 qpel mc, size 16 patch by (Romain Dolbeau <dolbeau at caps-entreprise dot com>)

Originally committed as revision 3472 to svn://svn.ffmpeg.org/ffmpeg/trunk

4 files changed, 937 insertions, 7 deletions

diff --git a/libavcodec/ppc/dsputil_h264_altivec.c b/libavcodec/ppc/dsputil_h264_altivec.c
new file mode 100644
index 0000000000..1891e194ae
--- /dev/null
+++ b/libavcodec/ppc/dsputil_h264_altivec.c
@@ -0,0 +1,266 @@
+/*
+ * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+ 
+#include "../dsputil.h"
+
+#include "gcc_fixes.h"
+
+#include "dsputil_altivec.h"
+
+#define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
+#define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
+
+#define OP_U8_ALTIVEC                          PUT_OP_U8_ALTIVEC
+#define PREFIX_h264_chroma_mc8_altivec         put_h264_chroma_mc8_altivec
+#define PREFIX_h264_chroma_mc8_num             altivec_put_h264_chroma_mc8_num
+#define PREFIX_h264_qpel16_h_lowpass_altivec   put_h264_qpel16_h_lowpass_altivec
+#define PREFIX_h264_qpel16_h_lowpass_num       altivec_put_h264_qpel16_h_lowpass_num
+#define PREFIX_h264_qpel16_v_lowpass_altivec   put_h264_qpel16_v_lowpass_altivec
+#define PREFIX_h264_qpel16_v_lowpass_num       altivec_put_h264_qpel16_v_lowpass_num
+#define PREFIX_h264_qpel16_hv_lowpass_altivec  put_h264_qpel16_hv_lowpass_altivec
+#define PREFIX_h264_qpel16_hv_lowpass_num      altivec_put_h264_qpel16_hv_lowpass_num
+#include "dsputil_h264_template_altivec.c"
+#undef OP_U8_ALTIVEC
+#undef PREFIX_h264_chroma_mc8_altivec
+#undef PREFIX_h264_chroma_mc8_num
+#undef PREFIX_h264_qpel16_h_lowpass_altivec
+#undef PREFIX_h264_qpel16_h_lowpass_num
+#undef PREFIX_h264_qpel16_v_lowpass_altivec
+#undef PREFIX_h264_qpel16_v_lowpass_num
+#undef PREFIX_h264_qpel16_hv_lowpass_altivec
+#undef PREFIX_h264_qpel16_hv_lowpass_num
+
+#define OP_U8_ALTIVEC                          AVG_OP_U8_ALTIVEC
+#define PREFIX_h264_chroma_mc8_altivec         avg_h264_chroma_mc8_altivec
+#define PREFIX_h264_chroma_mc8_num             altivec_avg_h264_chroma_mc8_num
+#define PREFIX_h264_qpel16_h_lowpass_altivec   avg_h264_qpel16_h_lowpass_altivec
+#define PREFIX_h264_qpel16_h_lowpass_num       altivec_avg_h264_qpel16_h_lowpass_num
+#define PREFIX_h264_qpel16_v_lowpass_altivec   avg_h264_qpel16_v_lowpass_altivec
+#define PREFIX_h264_qpel16_v_lowpass_num       altivec_avg_h264_qpel16_v_lowpass_num
+#define PREFIX_h264_qpel16_hv_lowpass_altivec  avg_h264_qpel16_hv_lowpass_altivec
+#define PREFIX_h264_qpel16_hv_lowpass_num      altivec_avg_h264_qpel16_hv_lowpass_num
+#include "dsputil_h264_template_altivec.c"
+#undef OP_U8_ALTIVEC
+#undef PREFIX_h264_chroma_mc8_altivec
+#undef PREFIX_h264_chroma_mc8_num
+#undef PREFIX_h264_qpel16_h_lowpass_altivec
+#undef PREFIX_h264_qpel16_h_lowpass_num
+#undef PREFIX_h264_qpel16_v_lowpass_altivec
+#undef PREFIX_h264_qpel16_v_lowpass_num
+#undef PREFIX_h264_qpel16_hv_lowpass_altivec
+#undef PREFIX_h264_qpel16_hv_lowpass_num
+
+#define H264_MC(OPNAME, SIZE, CODETYPE) \
+static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\
+    OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \
+    uint64_t temp[SIZE*SIZE/8] __align16;\
+    uint8_t * const half= (uint8_t*)temp;\
+    put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*SIZE/8] __align16;\
+    uint8_t * const half= (uint8_t*)temp;\
+    put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*SIZE/8] __align16;\
+    uint8_t * const half= (uint8_t*)temp;\
+    put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*SIZE/8] __align16;\
+    uint8_t * const half= (uint8_t*)temp;\
+    put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*SIZE/4] __align16;\
+    uint8_t * const halfH= (uint8_t*)temp;\
+    uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
+    put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*SIZE/4] __align16;\
+    uint8_t * const halfH= (uint8_t*)temp;\
+    uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
+    put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*SIZE/4] __align16;\
+    uint8_t * const halfH= (uint8_t*)temp;\
+    uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
+    put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*SIZE/4] __align16;\
+    uint8_t * const halfH= (uint8_t*)temp;\
+    uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\
+    put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*(SIZE+8)/4] __align16;\
+    int16_t * const tmp= (int16_t*)temp;\
+    OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4] __align16;\
+    uint8_t * const halfH= (uint8_t*)temp;\
+    uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
+    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
+    put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4] __align16;\
+    uint8_t * const halfH= (uint8_t*)temp;\
+    uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
+    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
+    put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4] __align16;\
+    uint8_t * const halfV= (uint8_t*)temp;\
+    uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
+    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
+    put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
+}\
+\
+static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\
+    uint64_t temp[SIZE*(SIZE+8)/4 + SIZE*SIZE/4] __align16;\
+    uint8_t * const halfV= (uint8_t*)temp;\
+    uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\
+    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\
+    put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
+    put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
+    OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
+}\
+
+
+/* from dsputil.c */
+static inline void put_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 = (((const struct unaligned_32 *) (&src1[i * src_stride1]))->l);
+		b = (((const struct unaligned_32 *) (&src2[i * src_stride2]))->l);
+		*((uint32_t *) & dst[i * dst_stride]) = rnd_avg32(a, b);
+		a = (((const struct unaligned_32 *) (&src1[i * src_stride1 + 4]))->l);
+		b = (((const struct unaligned_32 *) (&src2[i * src_stride2 + 4]))->l);
+		*((uint32_t *) & dst[i * dst_stride + 4]) = rnd_avg32(a, b);
+	}
+} static inline void avg_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 = (((const struct unaligned_32 *) (&src1[i * src_stride1]))->l);
+		b = (((const struct unaligned_32 *) (&src2[i * src_stride2]))->l);
+		*((uint32_t *) & dst[i * dst_stride]) = rnd_avg32(*((uint32_t *) & dst[i * dst_stride]), rnd_avg32(a, b));
+		a = (((const struct unaligned_32 *) (&src1[i * src_stride1 + 4]))->l);
+		b = (((const struct unaligned_32 *) (&src2[i * src_stride2 + 4]))->l);
+		*((uint32_t *) & dst[i * dst_stride + 4]) = rnd_avg32(*((uint32_t *) & dst[i * dst_stride + 4]), rnd_avg32(a, b));
+	}
+} static inline void put_pixels16_l2(uint8_t * dst, const uint8_t * src1, const uint8_t * src2, int dst_stride, int src_stride1, int src_stride2, int h) {
+	put_pixels8_l2(dst, src1, src2, dst_stride, src_stride1, src_stride2, h);
+	put_pixels8_l2(dst + 8, src1 + 8, src2 + 8, dst_stride, src_stride1, src_stride2, h);
+} static inline void avg_pixels16_l2(uint8_t * dst, const uint8_t * src1, const uint8_t * src2, int dst_stride, int src_stride1, int src_stride2, int h) {
+	avg_pixels8_l2(dst, src1, src2, dst_stride, src_stride1, src_stride2, h);
+	avg_pixels8_l2(dst + 8, src1 + 8, src2 + 8, dst_stride, src_stride1, src_stride2, h);
+}
+
+/* UNIMPLEMENTED YET !! */
+#define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
+#define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
+
+H264_MC(put_, 16, altivec)
+     H264_MC(avg_, 16, altivec)
+
+void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
+    
+#ifdef HAVE_ALTIVEC
+  if (has_altivec()) {
+    c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
+    c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
+
+#define dspfunc(PFX, IDX, NUM) \
+    c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
+    c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
+    c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
+    c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
+    c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
+    c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
+    c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
+    c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
+    c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
+    c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
+    c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
+    c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
+    c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
+    c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
+    c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
+    c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
+    
+    dspfunc(put_h264_qpel, 0, 16);
+    dspfunc(avg_h264_qpel, 0, 16);
+#undef dspfunc
+    
+  } else
+#endif /* HAVE_ALTIVEC */
+  {
+    // Non-AltiVec PPC optimisations
+    
+    // ... pending ...
+  }
+}
diff --git a/libavcodec/ppc/dsputil_h264_template_altivec.c b/libavcodec/ppc/dsputil_h264_template_altivec.c
new file mode 100644
index 0000000000..cb0fa954d7
--- /dev/null
+++ b/libavcodec/ppc/dsputil_h264_template_altivec.c
@@ -0,0 +1,639 @@
+/*
+ * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/* this code assume that stride % 16 == 0 */
+void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
+  POWERPC_PERF_DECLARE(PREFIX_h264_chroma_mc8_num, 1);
+  POWERPC_PERF_START_COUNT(PREFIX_h264_chroma_mc8_num, 1);
+    signed int ABCD[4] __attribute__((aligned(16)));
+    register int i;
+    ABCD[0] = ((8 - x) * (8 - y));
+    ABCD[1] = ((x) * (8 - y));
+    ABCD[2] = ((8 - x) * (y));
+    ABCD[3] = ((x) * (y));
+    const vector signed int vABCD = vec_ld(0, ABCD);
+    const vector signed short vA = vec_splat((vector signed short)vABCD, 1);
+    const vector signed short vB = vec_splat((vector signed short)vABCD, 3);
+    const vector signed short vC = vec_splat((vector signed short)vABCD, 5);
+    const vector signed short vD = vec_splat((vector signed short)vABCD, 7);
+    const vector signed int vzero = vec_splat_s32(0);
+    const vector signed short v32ss = (const vector signed short)AVV(32);
+    const vector unsigned short v6us = vec_splat_u16(6);
+
+    vector unsigned char fperm;
+
+    if (((unsigned long)dst) % 16 == 0) {
+      fperm = (vector unsigned char)AVV(0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
+                                        0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F);
+    } else {
+      fperm = (vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
+                                        0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
+    }
+
+    register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
+    register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
+    
+    vector unsigned char vsrcAuc;
+    vector unsigned char vsrcBuc;
+    vector unsigned char vsrcperm0;
+    vector unsigned char vsrcperm1;
+    vsrcAuc = vec_ld(0, src);
+    if (loadSecond)
+      vsrcBuc = vec_ld(16, src);
+    vsrcperm0 = vec_lvsl(0, src);
+    vsrcperm1 = vec_lvsl(1, src);
+    
+    vector unsigned char vsrc0uc;
+    vector unsigned char vsrc1uc;
+    vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
+    if (reallyBadAlign)
+      vsrc1uc = vsrcBuc;
+    else
+      vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);
+    
+    vector signed short vsrc0ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc0uc);
+    vector signed short vsrc1ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc1uc);
+
+    if (!loadSecond) {// -> !reallyBadAlign
+      for (i = 0 ; i < h ; i++) {
+        vector unsigned char vsrcCuc;
+        vsrcCuc = vec_ld(stride + 0, src);
+        
+        vector unsigned char vsrc2uc;
+        vector unsigned char vsrc3uc;
+        vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
+        vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
+        
+        vector signed short vsrc2ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc2uc);
+        vector signed short vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc3uc);
+        
+        vector signed short psum;
+        
+        psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
+        psum = vec_mladd(vB, vsrc1ssH, psum);
+        psum = vec_mladd(vC, vsrc2ssH, psum);
+        psum = vec_mladd(vD, vsrc3ssH, psum);
+        psum = vec_add(v32ss, psum);
+        psum = vec_sra(psum, v6us);
+        
+        vector unsigned char vdst = vec_ld(0, dst);
+        vector unsigned char ppsum = (vector unsigned char)vec_packsu(psum, psum);
+        
+        vector unsigned char vfdst = vec_perm(vdst, ppsum, fperm);
+        vector unsigned char fsum;
+        
+        OP_U8_ALTIVEC(fsum, vfdst, vdst);
+
+        vec_st(fsum, 0, dst);
+        
+        vsrc0ssH = vsrc2ssH;
+        vsrc1ssH = vsrc3ssH;
+        
+        dst += stride;
+        src += stride;
+      }
+    } else {
+      for (i = 0 ; i < h ; i++) {
+        vector unsigned char vsrcCuc;
+        vector unsigned char vsrcDuc;
+        vsrcCuc = vec_ld(stride + 0, src);
+        vsrcDuc = vec_ld(stride + 16, src);
+        
+        vector unsigned char vsrc2uc;
+        vector unsigned char vsrc3uc;
+        vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
+        if (reallyBadAlign)
+          vsrc3uc = vsrcDuc;
+        else
+          vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
+        
+        vector signed short vsrc2ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc2uc);
+        vector signed short vsrc3ssH = (vector signed short)vec_mergeh((vector unsigned char)vzero, (vector unsigned char)vsrc3uc);
+        
+        vector signed short psum;
+      
+        psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
+        psum = vec_mladd(vB, vsrc1ssH, psum);
+        psum = vec_mladd(vC, vsrc2ssH, psum);
+        psum = vec_mladd(vD, vsrc3ssH, psum);
+        psum = vec_add(v32ss, psum);
+        psum = vec_sr(psum, v6us);
+        
+        vector unsigned char vdst = vec_ld(0, dst);
+        vector unsigned char ppsum = (vector unsigned char)vec_pack(psum, psum); 
+        
+        vector unsigned char vfdst = vec_perm(vdst, ppsum, fperm);
+        vector unsigned char fsum;
+        
+        OP_U8_ALTIVEC(fsum, vfdst, vdst);
+
+        vec_st(fsum, 0, dst);
+        
+        vsrc0ssH = vsrc2ssH;
+        vsrc1ssH = vsrc3ssH;
+        
+        dst += stride;
+        src += stride;
+      }
+    }
+    POWERPC_PERF_STOP_COUNT(PREFIX_h264_chroma_mc8_num, 1);
+}
+
+/* this code assume stride % 16 == 0 */
+static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
+  POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_h_lowpass_num, 1);
+  POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
+  register int i;
+  
+  const vector signed int vzero = vec_splat_s32(0);
+  const vector unsigned char permM2 = vec_lvsl(-2, src);
+  const vector unsigned char permM1 = vec_lvsl(-1, src);
+  const vector unsigned char permP0 = vec_lvsl(+0, src);
+  const vector unsigned char permP1 = vec_lvsl(+1, src);
+  const vector unsigned char permP2 = vec_lvsl(+2, src);
+  const vector unsigned char permP3 = vec_lvsl(+3, src);
+  const vector signed short v20ss = (const vector signed short)AVV(20);
+  const vector unsigned short v5us = vec_splat_u16(5);
+  const vector signed short v5ss = vec_splat_s16(5);
+  const vector signed short v16ss = (const vector signed short)AVV(16);
+  const vector unsigned char dstperm = vec_lvsr(0, dst);
+  const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
+  const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
+
+  register int align = ((((unsigned long)src) - 2) % 16);
+
+  for (i = 0 ; i < 16 ; i ++) {
+    vector unsigned char srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
+    vector unsigned char srcR1 = vec_ld(-2, src);
+    vector unsigned char srcR2 = vec_ld(14, src);
+
+    switch (align) {
+    default: {
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = vec_perm(srcR1, srcR2, permP0);
+      srcP1 = vec_perm(srcR1, srcR2, permP1);
+      srcP2 = vec_perm(srcR1, srcR2, permP2);
+      srcP3 = vec_perm(srcR1, srcR2, permP3);
+    } break;
+    case 11: {
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = vec_perm(srcR1, srcR2, permP0);
+      srcP1 = vec_perm(srcR1, srcR2, permP1);
+      srcP2 = vec_perm(srcR1, srcR2, permP2);
+      srcP3 = srcR2;
+    } break;
+    case 12: {
+      vector unsigned char srcR3 = vec_ld(30, src);
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = vec_perm(srcR1, srcR2, permP0);
+      srcP1 = vec_perm(srcR1, srcR2, permP1);
+      srcP2 = srcR2;
+      srcP3 = vec_perm(srcR2, srcR3, permP3);
+    } break;
+    case 13: {
+      vector unsigned char srcR3 = vec_ld(30, src);
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = vec_perm(srcR1, srcR2, permP0);
+      srcP1 = srcR2;
+      srcP2 = vec_perm(srcR2, srcR3, permP2);
+      srcP3 = vec_perm(srcR2, srcR3, permP3);
+    } break;
+    case 14: {
+      vector unsigned char srcR3 = vec_ld(30, src);
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = srcR2;
+      srcP1 = vec_perm(srcR2, srcR3, permP1);
+      srcP2 = vec_perm(srcR2, srcR3, permP2);
+      srcP3 = vec_perm(srcR2, srcR3, permP3);
+    } break;
+    case 15: {
+      vector unsigned char srcR3 = vec_ld(30, src);
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = srcR2;
+      srcP0 = vec_perm(srcR2, srcR3, permP0);
+      srcP1 = vec_perm(srcR2, srcR3, permP1);
+      srcP2 = vec_perm(srcR2, srcR3, permP2);
+      srcP3 = vec_perm(srcR2, srcR3, permP3);
+    } break;
+    }
+
+    const vector signed short srcP0A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
+    const vector signed short srcP0B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
+    const vector signed short srcP1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
+    const vector signed short srcP1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
+
+    const vector signed short srcP2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
+    const vector signed short srcP2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
+    const vector signed short srcP3A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
+    const vector signed short srcP3B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
+
+    const vector signed short srcM1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
+    const vector signed short srcM1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
+    const vector signed short srcM2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
+    const vector signed short srcM2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
+
+    const vector signed short sum1A = vec_adds(srcP0A, srcP1A);
+    const vector signed short sum1B = vec_adds(srcP0B, srcP1B);
+    const vector signed short sum2A = vec_adds(srcM1A, srcP2A);
+    const vector signed short sum2B = vec_adds(srcM1B, srcP2B);
+    const vector signed short sum3A = vec_adds(srcM2A, srcP3A);
+    const vector signed short sum3B = vec_adds(srcM2B, srcP3B);
+    
+    const vector signed short pp1A = vec_mladd(sum1A, v20ss, v16ss);
+    const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
+
+    const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
+    const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
+    
+    const vector signed short pp3A = vec_add(sum3A, pp1A);
+    const vector signed short pp3B = vec_add(sum3B, pp1B);
+
+    const vector signed short psumA = vec_sub(pp3A, pp2A);
+    const vector signed short psumB = vec_sub(pp3B, pp2B);
+
+    const vector signed short sumA = vec_sra(psumA, v5us);
+    const vector signed short sumB = vec_sra(psumB, v5us);
+
+    const vector unsigned char sum = vec_packsu(sumA, sumB);
+
+    const vector unsigned char dst1 = vec_ld(0, dst);
+    const vector unsigned char dst2 = vec_ld(16, dst);
+    const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
+
+    vector unsigned char fsum;
+    OP_U8_ALTIVEC(fsum, sum, vdst);
+
+    const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
+    const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
+    const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
+
+    vec_st(fdst1, 0, dst);
+    vec_st(fdst2, 16, dst);
+
+    src += srcStride;
+    dst += dstStride;
+  }
+POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
+}
+
+/* this code assume stride % 16 == 0 */
+static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
+  POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_v_lowpass_num, 1);
+  POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);
+  
+  register int i;
+
+  const vector signed int vzero = vec_splat_s32(0);
+  const vector unsigned char perm = vec_lvsl(0, src);
+  const vector signed short v20ss = (const vector signed short)AVV(20);
+  const vector unsigned short v5us = vec_splat_u16(5);
+  const vector signed short v5ss = vec_splat_s16(5);
+  const vector signed short v16ss = (const vector signed short)AVV(16);
+  const vector unsigned char dstperm = vec_lvsr(0, dst);
+  const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
+  const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
+  
+  uint8_t *srcbis = src - (srcStride * 2);
+
+  const vector unsigned char srcM2a = vec_ld(0, srcbis);
+  const vector unsigned char srcM2b = vec_ld(16, srcbis);
+  const vector unsigned char srcM2 = vec_perm(srcM2a, srcM2b, perm);
+  srcbis += srcStride;
+  const vector unsigned char srcM1a = vec_ld(0, srcbis);
+  const vector unsigned char srcM1b = vec_ld(16, srcbis);
+  const vector unsigned char srcM1 = vec_perm(srcM1a, srcM1b, perm);
+  srcbis += srcStride;
+  const vector unsigned char srcP0a = vec_ld(0, srcbis);
+  const vector unsigned char srcP0b = vec_ld(16, srcbis);
+  const vector unsigned char srcP0 = vec_perm(srcP0a, srcP0b, perm);
+  srcbis += srcStride;
+  const vector unsigned char srcP1a = vec_ld(0, srcbis);
+  const vector unsigned char srcP1b = vec_ld(16, srcbis);
+  const vector unsigned char srcP1 = vec_perm(srcP1a, srcP1b, perm);
+  srcbis += srcStride;
+  const vector unsigned char srcP2a = vec_ld(0, srcbis);
+  const vector unsigned char srcP2b = vec_ld(16, srcbis);
+  const vector unsigned char srcP2 = vec_perm(srcP2a, srcP2b, perm);
+  srcbis += srcStride;
+
+  vector signed short srcM2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
+  vector signed short srcM2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
+  vector signed short srcM1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
+  vector signed short srcM1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
+  vector signed short srcP0ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
+  vector signed short srcP0ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
+  vector signed short srcP1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
+  vector signed short srcP1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
+  vector signed short srcP2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
+  vector signed short srcP2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
+
+  for (i = 0 ; i < 16 ; i++) {
+    const vector unsigned char srcP3a = vec_ld(0, srcbis);
+    const vector unsigned char srcP3b = vec_ld(16, srcbis);
+    const vector unsigned char srcP3 = vec_perm(srcP3a, srcP3b, perm);
+    const vector signed short srcP3ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
+    const vector signed short srcP3ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
+    srcbis += srcStride;
+
+    const vector signed short sum1A = vec_adds(srcP0ssA, srcP1ssA);
+    const vector signed short sum1B = vec_adds(srcP0ssB, srcP1ssB);
+    const vector signed short sum2A = vec_adds(srcM1ssA, srcP2ssA);
+    const vector signed short sum2B = vec_adds(srcM1ssB, srcP2ssB);
+    const vector signed short sum3A = vec_adds(srcM2ssA, srcP3ssA);
+    const vector signed short sum3B = vec_adds(srcM2ssB, srcP3ssB);
+
+    srcM2ssA = srcM1ssA;
+    srcM2ssB = srcM1ssB;
+    srcM1ssA = srcP0ssA;
+    srcM1ssB = srcP0ssB;
+    srcP0ssA = srcP1ssA;
+    srcP0ssB = srcP1ssB;
+    srcP1ssA = srcP2ssA;
+    srcP1ssB = srcP2ssB;
+    srcP2ssA = srcP3ssA;
+    srcP2ssB = srcP3ssB;
+    
+    const vector signed short pp1A = vec_mladd(sum1A, v20ss, v16ss);
+    const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
+
+    const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
+    const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
+    
+    const vector signed short pp3A = vec_add(sum3A, pp1A);
+    const vector signed short pp3B = vec_add(sum3B, pp1B);
+
+    const vector signed short psumA = vec_sub(pp3A, pp2A);
+    const vector signed short psumB = vec_sub(pp3B, pp2B);
+
+    const vector signed short sumA = vec_sra(psumA, v5us);
+    const vector signed short sumB = vec_sra(psumB, v5us);
+
+    const vector unsigned char sum = vec_packsu(sumA, sumB);
+
+    const vector unsigned char dst1 = vec_ld(0, dst);
+    const vector unsigned char dst2 = vec_ld(16, dst);
+    const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
+
+    vector unsigned char fsum;
+    OP_U8_ALTIVEC(fsum, sum, vdst);
+
+    const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
+    const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
+    const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
+
+    vec_st(fdst1, 0, dst);
+    vec_st(fdst2, 16, dst);
+
+    dst += dstStride;
+  }
+  POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);
+}
+
+/* this code assume stride % 16 == 0 *and* tmp is properly aligned */
+static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp, uint8_t * src, int dstStride, int tmpStride, int srcStride) {
+  POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_hv_lowpass_num, 1);
+  POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1);
+  register int i;
+  const vector signed int vzero = vec_splat_s32(0);
+  const vector unsigned char permM2 = vec_lvsl(-2, src);
+  const vector unsigned char permM1 = vec_lvsl(-1, src);
+  const vector unsigned char permP0 = vec_lvsl(+0, src);
+  const vector unsigned char permP1 = vec_lvsl(+1, src);
+  const vector unsigned char permP2 = vec_lvsl(+2, src);
+  const vector unsigned char permP3 = vec_lvsl(+3, src);
+  const vector signed short v20ss = (const vector signed short)AVV(20);
+  const vector unsigned int v10ui = vec_splat_u32(10);
+  const vector signed short v5ss = vec_splat_s16(5);
+  const vector signed short v1ss = vec_splat_s16(1);
+  const vector signed int v512si = (const vector signed int)AVV(512);
+  const vector unsigned int v16ui = (const vector unsigned int)AVV(16);
+
+  register int align = ((((unsigned long)src) - 2) % 16);
+
+  src -= (2 * srcStride);
+
+  for (i = 0 ; i < 21 ; i ++) {
+    vector unsigned char srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;
+    vector unsigned char srcR1 = vec_ld(-2, src);
+    vector unsigned char srcR2 = vec_ld(14, src);
+
+    switch (align) {
+    default: {
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = vec_perm(srcR1, srcR2, permP0);
+      srcP1 = vec_perm(srcR1, srcR2, permP1);
+      srcP2 = vec_perm(srcR1, srcR2, permP2);
+      srcP3 = vec_perm(srcR1, srcR2, permP3);
+    } break;
+    case 11: {
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = vec_perm(srcR1, srcR2, permP0);
+      srcP1 = vec_perm(srcR1, srcR2, permP1);
+      srcP2 = vec_perm(srcR1, srcR2, permP2);
+      srcP3 = srcR2;
+    } break;
+    case 12: {
+      vector unsigned char srcR3 = vec_ld(30, src);
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = vec_perm(srcR1, srcR2, permP0);
+      srcP1 = vec_perm(srcR1, srcR2, permP1);
+      srcP2 = srcR2;
+      srcP3 = vec_perm(srcR2, srcR3, permP3);
+    } break;
+    case 13: {
+      vector unsigned char srcR3 = vec_ld(30, src);
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = vec_perm(srcR1, srcR2, permP0);
+      srcP1 = srcR2;
+      srcP2 = vec_perm(srcR2, srcR3, permP2);
+      srcP3 = vec_perm(srcR2, srcR3, permP3);
+    } break;
+    case 14: {
+      vector unsigned char srcR3 = vec_ld(30, src);
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = vec_perm(srcR1, srcR2, permM1);
+      srcP0 = srcR2;
+      srcP1 = vec_perm(srcR2, srcR3, permP1);
+      srcP2 = vec_perm(srcR2, srcR3, permP2);
+      srcP3 = vec_perm(srcR2, srcR3, permP3);
+    } break;
+    case 15: {
+      vector unsigned char srcR3 = vec_ld(30, src);
+      srcM2 = vec_perm(srcR1, srcR2, permM2);
+      srcM1 = srcR2;
+      srcP0 = vec_perm(srcR2, srcR3, permP0);
+      srcP1 = vec_perm(srcR2, srcR3, permP1);
+      srcP2 = vec_perm(srcR2, srcR3, permP2);
+      srcP3 = vec_perm(srcR2, srcR3, permP3);
+    } break;
+    }
+
+    const vector signed short srcP0A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
+    const vector signed short srcP0B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
+    const vector signed short srcP1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
+    const vector signed short srcP1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
+
+    const vector signed short srcP2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
+    const vector signed short srcP2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
+    const vector signed short srcP3A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
+    const vector signed short srcP3B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
+
+    const vector signed short srcM1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
+    const vector signed short srcM1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
+    const vector signed short srcM2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
+    const vector signed short srcM2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
+
+    const vector signed short sum1A = vec_adds(srcP0A, srcP1A);
+    const vector signed short sum1B = vec_adds(srcP0B, srcP1B);
+    const vector signed short sum2A = vec_adds(srcM1A, srcP2A);
+    const vector signed short sum2B = vec_adds(srcM1B, srcP2B);
+    const vector signed short sum3A = vec_adds(srcM2A, srcP3A);
+    const vector signed short sum3B = vec_adds(srcM2B, srcP3B);
+    
+    const vector signed short pp1A = vec_mladd(sum1A, v20ss, sum3A);
+    const vector signed short pp1B = vec_mladd(sum1B, v20ss, sum3B);
+
+    const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero);
+    const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero);
+
+    const vector signed short psumA = vec_sub(pp1A, pp2A);
+    const vector signed short psumB = vec_sub(pp1B, pp2B);
+
+    vec_st(psumA, 0, tmp);
+    vec_st(psumB, 16, tmp);
+    
+    src += srcStride;
+    tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */
+  }
+  
+  const vector unsigned char dstperm = vec_lvsr(0, dst);
+  const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1);
+  const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm);
+  const vector unsigned char mperm = (const vector unsigned char)
+    AVV(0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B,
+        0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F);
+  
+  int16_t *tmpbis = tmp - (tmpStride * 21);
+
+  vector signed short tmpM2ssA = vec_ld(0, tmpbis);
+  vector signed short tmpM2ssB = vec_ld(16, tmpbis);
+  tmpbis += tmpStride;
+  vector signed short tmpM1ssA = vec_ld(0, tmpbis);
+  vector signed short tmpM1ssB = vec_ld(16, tmpbis);
+  tmpbis += tmpStride;
+  vector signed short tmpP0ssA = vec_ld(0, tmpbis);
+  vector signed short tmpP0ssB = vec_ld(16, tmpbis);
+  tmpbis += tmpStride;
+  vector signed short tmpP1ssA = vec_ld(0, tmpbis);
+  vector signed short tmpP1ssB = vec_ld(16, tmpbis);
+  tmpbis += tmpStride;
+  vector signed short tmpP2ssA = vec_ld(0, tmpbis);
+  vector signed short tmpP2ssB = vec_ld(16, tmpbis);
+  tmpbis += tmpStride;
+
+  for (i = 0 ; i < 16 ; i++) {
+    const vector signed short tmpP3ssA = vec_ld(0, tmpbis);
+    const vector signed short tmpP3ssB = vec_ld(16, tmpbis);
+    tmpbis += tmpStride;
+
+    const vector signed short sum1A = vec_adds(tmpP0ssA, tmpP1ssA);
+    const vector signed short sum1B = vec_adds(tmpP0ssB, tmpP1ssB);
+    const vector signed short sum2A = vec_adds(tmpM1ssA, tmpP2ssA);
+    const vector signed short sum2B = vec_adds(tmpM1ssB, tmpP2ssB);
+    const vector signed short sum3A = vec_adds(tmpM2ssA, tmpP3ssA);
+    const vector signed short sum3B = vec_adds(tmpM2ssB, tmpP3ssB);
+
+    tmpM2ssA = tmpM1ssA;
+    tmpM2ssB = tmpM1ssB;
+    tmpM1ssA = tmpP0ssA;
+    tmpM1ssB = tmpP0ssB;
+    tmpP0ssA = tmpP1ssA;
+    tmpP0ssB = tmpP1ssB;
+    tmpP1ssA = tmpP2ssA;
+    tmpP1ssB = tmpP2ssB;
+    tmpP2ssA = tmpP3ssA;
+    tmpP2ssB = tmpP3ssB;
+
+    const vector signed int pp1Ae = vec_mule(sum1A, v20ss);
+    const vector signed int pp1Ao = vec_mulo(sum1A, v20ss);
+    const vector signed int pp1Be = vec_mule(sum1B, v20ss);
+    const vector signed int pp1Bo = vec_mulo(sum1B, v20ss);
+
+    const vector signed int pp2Ae = vec_mule(sum2A, v5ss);
+    const vector signed int pp2Ao = vec_mulo(sum2A, v5ss);
+    const vector signed int pp2Be = vec_mule(sum2B, v5ss);
+    const vector signed int pp2Bo = vec_mulo(sum2B, v5ss);
+
+    const vector signed int pp3Ae = vec_sra((vector signed int)sum3A, v16ui);
+    const vector signed int pp3Ao = vec_mulo(sum3A, v1ss);
+    const vector signed int pp3Be = vec_sra((vector signed int)sum3B, v16ui);
+    const vector signed int pp3Bo = vec_mulo(sum3B, v1ss);
+
+    const vector signed int pp1cAe = vec_add(pp1Ae, v512si);
+    const vector signed int pp1cAo = vec_add(pp1Ao, v512si);
+    const vector signed int pp1cBe = vec_add(pp1Be, v512si);
+    const vector signed int pp1cBo = vec_add(pp1Bo, v512si);
+
+    const vector signed int pp32Ae = vec_sub(pp3Ae, pp2Ae);
+    const vector signed int pp32Ao = vec_sub(pp3Ao, pp2Ao);
+    const vector signed int pp32Be = vec_sub(pp3Be, pp2Be);
+    const vector signed int pp32Bo = vec_sub(pp3Bo, pp2Bo);
+
+    const vector signed int sumAe = vec_add(pp1cAe, pp32Ae);
+    const vector signed int sumAo = vec_add(pp1cAo, pp32Ao);
+    const vector signed int sumBe = vec_add(pp1cBe, pp32Be);
+    const vector signed int sumBo = vec_add(pp1cBo, pp32Bo);
+    
+    const vector signed int ssumAe = vec_sra(sumAe, v10ui);
+    const vector signed int ssumAo = vec_sra(sumAo, v10ui);
+    const vector signed int ssumBe = vec_sra(sumBe, v10ui);
+    const vector signed int ssumBo = vec_sra(sumBo, v10ui);
+
+    const vector signed short ssume = vec_packs(ssumAe, ssumBe);
+    const vector signed short ssumo = vec_packs(ssumAo, ssumBo);
+
+    const vector unsigned char sumv = vec_packsu(ssume, ssumo);
+    const vector unsigned char sum = vec_perm(sumv, sumv, mperm);
+
+    const vector unsigned char dst1 = vec_ld(0, dst);
+    const vector unsigned char dst2 = vec_ld(16, dst);
+    const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst));
+
+    vector unsigned char fsum;
+    OP_U8_ALTIVEC(fsum, sum, vdst);
+
+    const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm);
+    const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask);
+    const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask);
+
+    vec_st(fdst1, 0, dst);
+    vec_st(fdst2, 16, dst);
+
+    dst += dstStride;
+  }
+  POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1);
+}
diff --git a/libavcodec/ppc/dsputil_ppc.c b/libavcodec/ppc/dsputil_ppc.c
index ea1d499e46..546fa0bd49 100644
--- a/libavcodec/ppc/dsputil_ppc.c
+++ b/libavcodec/ppc/dsputil_ppc.c
@@ -64,7 +64,16 @@ static unsigned char* perfname[] = {
   "hadamard8_diff16_altivec",
   "avg_pixels8_xy2_altivec",
   "clear_blocks_dcbz32_ppc",
-  "clear_blocks_dcbz128_ppc"
+  "clear_blocks_dcbz128_ppc",
+  "put_h264_chroma_mc8_altivec",
+  "avg_h264_chroma_mc8_altivec",
+  "put_h264_qpel16_h_lowpass_altivec",
+  "avg_h264_qpel16_h_lowpass_altivec",
+  "put_h264_qpel16_v_lowpass_altivec",
+  "avg_h264_qpel16_v_lowpass_altivec",
+  "put_h264_qpel16_hv_lowpass_altivec",
+  "avg_h264_qpel16_hv_lowpass_altivec",
+  ""
 };
 #include <stdio.h>
 #endif
@@ -228,6 +237,9 @@ long check_dcbzl_effect(void)
 }
 #endif
 
+
+void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx);
+
 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx)
 {
     // Common optimizations whether Altivec is available or not
@@ -242,6 +254,8 @@ void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx)
   default:
     break;
   }
+
+  dsputil_h264_init_ppc(c, avctx);
   
 #ifdef HAVE_ALTIVEC
     if (has_altivec()) {
@@ -310,10 +324,10 @@ void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx)
           {
 	    for (j = 0; j < POWERPC_NUM_PMC_ENABLED ; j++)
 	      {
-		perfdata[j][i][powerpc_data_min] = (unsigned long long)0xFFFFFFFFFFFFFFFF;
-		perfdata[j][i][powerpc_data_max] = (unsigned long long)0x0000000000000000;
-		perfdata[j][i][powerpc_data_sum] = (unsigned long long)0x0000000000000000;
-		perfdata[j][i][powerpc_data_num] = (unsigned long long)0x0000000000000000;
+		perfdata[j][i][powerpc_data_min] = 0xFFFFFFFFFFFFFFFFULL;
+		perfdata[j][i][powerpc_data_max] = 0x0000000000000000ULL;
+		perfdata[j][i][powerpc_data_sum] = 0x0000000000000000ULL;
+		perfdata[j][i][powerpc_data_num] = 0x0000000000000000ULL;
 	      }
 	  }
         }
diff --git a/libavcodec/ppc/dsputil_ppc.h b/libavcodec/ppc/dsputil_ppc.h
index 247264835d..7e01677f19 100644
--- a/libavcodec/ppc/dsputil_ppc.h
+++ b/libavcodec/ppc/dsputil_ppc.h
@@ -32,8 +32,11 @@
 
 #ifdef POWERPC_PERFORMANCE_REPORT
 void powerpc_display_perf_report(void);
-/* the 604* have 2, the G3* have 4, the G4s have 6 */
-#define POWERPC_NUM_PMC_ENABLED 4
+/* the 604* have 2, the G3* have 4, the G4s have 6,
+   and the G5 are completely different (they MUST use
+   POWERPC_MODE_64BITS, and let's hope all future 64 bis PPC
+   will use the same PMCs... */
+#define POWERPC_NUM_PMC_ENABLED 6
 /* if you add to the enum below, also add to the perfname array
    in dsputil_ppc.c */
 enum powerpc_perf_index {
@@ -55,6 +58,14 @@ enum powerpc_perf_index {
   altivec_avg_pixels8_xy2_num,
   powerpc_clear_blocks_dcbz32,
   powerpc_clear_blocks_dcbz128,
+  altivec_put_h264_chroma_mc8_num,
+  altivec_avg_h264_chroma_mc8_num,
+  altivec_put_h264_qpel16_h_lowpass_num,
+  altivec_avg_h264_qpel16_h_lowpass_num,
+  altivec_put_h264_qpel16_v_lowpass_num,
+  altivec_avg_h264_qpel16_v_lowpass_num,
+  altivec_put_h264_qpel16_hv_lowpass_num,
+  altivec_avg_h264_qpel16_hv_lowpass_num,
   powerpc_perf_total
 };
 enum powerpc_data_index {