dsputil: Split off quarterpel bits into their own context

4 files changed, 509 insertions, 470 deletions

diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 10242269c2..5fddf3fb83 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -20,6 +20,7 @@ OBJS-$(CONFIG_LPC)                     += x86/lpc.o
 OBJS-$(CONFIG_MPEGAUDIODSP)            += x86/mpegaudiodsp.o
 OBJS-$(CONFIG_MPEGVIDEO)               += x86/mpegvideo.o
 OBJS-$(CONFIG_MPEGVIDEOENC)            += x86/mpegvideoenc.o
+OBJS-$(CONFIG_QPELDSP)                 += x86/qpeldsp_init.o
 OBJS-$(CONFIG_VIDEODSP)                += x86/videodsp_init.o
 OBJS-$(CONFIG_VP3DSP)                  += x86/vp3dsp_init.o
 OBJS-$(CONFIG_XMM_CLOBBER_TEST)        += x86/w64xmmtest.o
@@ -44,13 +45,13 @@ OBJS-$(CONFIG_VP8_DECODER)             += x86/vp8dsp_init.o
 OBJS-$(CONFIG_VP9_DECODER)             += x86/vp9dsp_init.o
 
 MMX-OBJS-$(CONFIG_DSPUTIL)             += x86/dsputil_mmx.o             \
-                                          x86/fpel_mmx.o                \
                                           x86/idct_mmx_xvid.o           \
                                           x86/idct_sse2_xvid.o          \
                                           x86/simple_idct.o
 MMX-OBJS-$(CONFIG_HPELDSP)             += x86/fpel_mmx.o                \
                                           x86/hpeldsp_mmx.o
 MMX-OBJS-$(CONFIG_HUFFYUVDSP)          += x86/huffyuvdsp_mmx.o
+MMX-OBJS-$(CONFIG_QPELDSP)             += x86/fpel_mmx.o
 
 MMX-OBJS-$(CONFIG_SVQ1_ENCODER)        += x86/svq1enc_mmx.o
 MMX-OBJS-$(CONFIG_VC1_DECODER)         += x86/vc1dsp_mmx.o
@@ -61,10 +62,7 @@ YASM-OBJS                              += x86/deinterlace.o             \
 YASM-OBJS-$(CONFIG_AC3DSP)             += x86/ac3dsp.o
 YASM-OBJS-$(CONFIG_DCT)                += x86/dct32.o
 YASM-OBJS-$(CONFIG_DNXHD_ENCODER)      += x86/dnxhdenc.o
-YASM-OBJS-$(CONFIG_DSPUTIL)            += x86/dsputil.o                 \
-                                          x86/fpel.o                    \
-                                          x86/mpeg4qpel.o               \
-                                          x86/qpel.o
+YASM-OBJS-$(CONFIG_DSPUTIL)            += x86/dsputil.o
 YASM-OBJS-$(CONFIG_ENCODERS)           += x86/dsputilenc.o
 YASM-OBJS-$(CONFIG_FFT)                += x86/fft.o
 YASM-OBJS-$(CONFIG_H263DSP)            += x86/h263_loopfilter.o
@@ -86,6 +84,9 @@ YASM-OBJS-$(CONFIG_HPELDSP)            += x86/fpel.o                    \
                                           x86/hpeldsp.o
 YASM-OBJS-$(CONFIG_HUFFYUVDSP)         += x86/huffyuvdsp.o
 YASM-OBJS-$(CONFIG_MPEGAUDIODSP)       += x86/imdct36.o
+YASM-OBJS-$(CONFIG_QPELDSP)            += x86/qpeldsp.o                 \
+                                          x86/fpel.o                    \
+                                          x86/qpel.o
 YASM-OBJS-$(CONFIG_VIDEODSP)           += x86/videodsp.o
 YASM-OBJS-$(CONFIG_VP3DSP)             += x86/vp3dsp.o
 
diff --git a/libavcodec/x86/dsputil_init.c b/libavcodec/x86/dsputil_init.c
index 9b0788ff73..389e7634dd 100644
--- a/libavcodec/x86/dsputil_init.c
+++ b/libavcodec/x86/dsputil_init.c
@@ -23,55 +23,11 @@
 #include "libavutil/x86/cpu.h"
 #include "libavcodec/avcodec.h"
 #include "libavcodec/dsputil.h"
-#include "libavcodec/pixels.h"
 #include "libavcodec/simple_idct.h"
 #include "libavcodec/version.h"
 #include "dsputil_x86.h"
-#include "fpel.h"
 #include "idct_xvid.h"
 
-void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                              int dstStride, int src1Stride, int h);
-void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1,
-                                     uint8_t *src2, int dstStride,
-                                     int src1Stride, int h);
-void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                              int dstStride, int src1Stride, int h);
-void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                               int dstStride, int src1Stride, int h);
-void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                               int dstStride, int src1Stride, int h);
-void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                                      int dstStride, int src1Stride, int h);
-void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                          int dstStride, int srcStride, int h);
-void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                          int dstStride, int srcStride, int h);
-void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                                 int dstStride, int srcStride,
-                                                 int h);
-void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                         int dstStride, int srcStride, int h);
-void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                         int dstStride, int srcStride, int h);
-void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                                int dstStride, int srcStride,
-                                                int h);
-void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                          int dstStride, int srcStride);
-void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                          int dstStride, int srcStride);
-void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                                 int dstStride, int srcStride);
-void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                         int dstStride, int srcStride);
-void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                         int dstStride, int srcStride);
-void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                                int dstStride, int srcStride);
-#define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmxext
-#define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmxext
-
 int32_t ff_scalarproduct_int16_mmxext(const int16_t *v1, const int16_t *v2,
                                       int order);
 int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2,
@@ -89,418 +45,6 @@ void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src,
 void ff_vector_clip_int32_sse4(int32_t *dst, const int32_t *src,
                                int32_t min, int32_t max, unsigned int len);
 
-#if HAVE_YASM
-
-CALL_2X_PIXELS(ff_avg_pixels16_mmxext, ff_avg_pixels8_mmxext, 8)
-CALL_2X_PIXELS(ff_put_pixels16_mmxext, ff_put_pixels8_mmxext, 8)
-
-#define QPEL_OP(OPNAME, RND, MMX)                                       \
-static void OPNAME ## qpel8_mc00_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);              \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t temp[8];                                                   \
-    uint8_t *const half = (uint8_t *) temp;                             \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
-                                                   stride, 8);          \
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
-                                        stride, stride, 8);             \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride,    \
-                                                   stride, 8);          \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t temp[8];                                                   \
-    uint8_t *const half = (uint8_t *) temp;                             \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
-                                                   stride, 8);          \
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride,     \
-                                        stride, 8);                     \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t temp[8];                                                   \
-    uint8_t *const half = (uint8_t *) temp;                             \
-    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
-                                                   8, stride);          \
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
-                                        stride, stride, 8);             \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src,            \
-                                                   stride, stride);     \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t temp[8];                                                   \
-    uint8_t *const half = (uint8_t *) temp;                             \
-    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
-                                                   8, stride);          \
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\
-                                        stride, 8);                     \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t half[8 + 9];                                               \
-    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
-                                                   stride, 9);          \
-    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
-                                        stride, 9);                     \
-    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
-                                        stride, 8, 8);                  \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t half[8 + 9];                                               \
-    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
-                                                   stride, 9);          \
-    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
-                                        stride, 9);                     \
-    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
-                                        stride, 8, 8);                  \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t half[8 + 9];                                               \
-    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
-                                                   stride, 9);          \
-    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
-                                        stride, 9);                     \
-    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
-                                        stride, 8, 8);                  \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t half[8 + 9];                                               \
-    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
-                                                   stride, 9);          \
-    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
-                                        stride, 9);                     \
-    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
-                                        stride, 8, 8);                  \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t half[8 + 9];                                               \
-    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
-                                                   stride, 9);          \
-    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
-                                        stride, 8, 8);                  \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t half[8 + 9];                                               \
-    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
-                                                   stride, 9);          \
-    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
-                                        stride, 8, 8);                  \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t half[8 + 9];                                               \
-    uint8_t *const halfH = (uint8_t *) half;                            \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
-                                                   stride, 9);          \
-    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH,              \
-                                        8, stride, 9);                  \
-    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
-                                                   stride, 8);          \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t half[8 + 9];                                               \
-    uint8_t *const halfH = (uint8_t *) half;                            \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
-                                                   stride, 9);          \
-    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
-                                        stride, 9);                     \
-    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
-                                                   stride, 8);          \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src,    \
-                                         ptrdiff_t stride)              \
-{                                                                       \
-    uint64_t half[9];                                                   \
-    uint8_t *const halfH = (uint8_t *) half;                            \
-    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
-                                                   stride, 9);          \
-    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
-                                                   stride, 8);          \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc00_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);            \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t temp[32];                                                  \
-    uint8_t *const half = (uint8_t *) temp;                             \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
-                                                    stride, 16);        \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
-                                         stride, 16);                   \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    ff_ ## OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src,           \
-                                                    stride, stride, 16);\
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t temp[32];                                                  \
-    uint8_t *const half = (uint8_t*) temp;                              \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
-                                                    stride, 16);        \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half,            \
-                                         stride, stride, 16);           \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t temp[32];                                                  \
-    uint8_t *const half = (uint8_t *) temp;                             \
-    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
-                                                    stride);            \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
-                                         stride, 16);                   \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src,           \
-                                                    stride, stride);    \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t temp[32];                                                  \
-    uint8_t *const half = (uint8_t *) temp;                             \
-    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
-                                                    stride);            \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half,         \
-                                         stride, stride, 16);           \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t half[16 * 2 + 17 * 2];                                     \
-    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
-                                                    stride, 17);        \
-    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
-                                         stride, 17);                   \
-    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
-                                                    16, 16);            \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
-                                         stride, 16, 16);               \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t half[16 * 2 + 17 * 2];                                     \
-    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
-                                                    stride, 17);        \
-    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
-                                         stride, 17);                   \
-    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
-                                                    16, 16);            \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
-                                         stride, 16, 16);               \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t half[16 * 2 + 17 * 2];                                     \
-    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
-                                                    stride, 17);        \
-    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
-                                         stride, 17);                   \
-    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
-                                                    16, 16);            \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
-                                         stride, 16, 16);               \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t half[16 * 2 + 17 * 2];                                     \
-    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
-                                                    stride, 17);        \
-    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
-                                         stride, 17);                   \
-    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
-                                                    16, 16);            \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
-                                         stride, 16, 16);               \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t half[16 * 2 + 17 * 2];                                     \
-    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
-                                                    stride, 17);        \
-    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
-                                                    16, 16);            \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
-                                         stride, 16, 16);               \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t half[16 * 2 + 17 * 2];                                     \
-    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
-    uint8_t *const halfHV = (uint8_t *) half;                           \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
-                                                    stride, 17);        \
-    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
-                                                    16, 16);            \
-    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
-                                         stride, 16, 16);               \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t half[17 * 2];                                              \
-    uint8_t *const halfH = (uint8_t *) half;                            \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
-                                                    stride, 17);        \
-    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
-                                         stride, 17);                   \
-    ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
-                                                    stride, 16);        \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t half[17 * 2];                                              \
-    uint8_t *const halfH = (uint8_t *) half;                            \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
-                                                    stride, 17);        \
-    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
-                                         stride, 17);                   \
-    ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
-                                                    stride, 16);        \
-}                                                                       \
-                                                                        \
-static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src,   \
-                                          ptrdiff_t stride)             \
-{                                                                       \
-    uint64_t half[17 * 2];                                              \
-    uint8_t *const halfH = (uint8_t *) half;                            \
-    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
-                                                    stride, 17);        \
-    ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
-                                                    stride, 16);        \
-}
-
-QPEL_OP(put_,        _,        mmxext)
-QPEL_OP(avg_,        _,        mmxext)
-QPEL_OP(put_no_rnd_, _no_rnd_, mmxext)
-
-#endif /* HAVE_YASM */
-
-#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
-do {                                                                         \
-    c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
-    c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
-} while (0)
-
 static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
                                      int cpu_flags, unsigned high_bit_depth)
 {
@@ -550,14 +94,6 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
 #endif /* HAVE_MMXEXT_INLINE */
 
 #if HAVE_MMXEXT_EXTERNAL
-    SET_QPEL_FUNCS(avg_qpel,        0, 16, mmxext, );
-    SET_QPEL_FUNCS(avg_qpel,        1,  8, mmxext, );
-
-    SET_QPEL_FUNCS(put_qpel,        0, 16, mmxext, );
-    SET_QPEL_FUNCS(put_qpel,        1,  8, mmxext, );
-    SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, );
-    SET_QPEL_FUNCS(put_no_rnd_qpel, 1,  8, mmxext, );
-
     c->scalarproduct_int16          = ff_scalarproduct_int16_mmxext;
 #endif /* HAVE_MMXEXT_EXTERNAL */
 }
diff --git a/libavcodec/x86/mpeg4qpel.asm b/libavcodec/x86/qpeldsp.asm
index 2cdd84b76d..8f65550e60 100644
--- a/libavcodec/x86/mpeg4qpel.asm
+++ b/libavcodec/x86/qpeldsp.asm
@@ -1,5 +1,6 @@
 ;******************************************************************************
-;* mpeg4 qpel
+;* quarterpel DSP functions
+;*
 ;* Copyright (c) 2008 Loren Merritt
 ;*
 ;* This file is part of Libav.
diff --git a/libavcodec/x86/qpeldsp_init.c b/libavcodec/x86/qpeldsp_init.c
new file mode 100644
index 0000000000..435b7651fd
--- /dev/null
+++ b/libavcodec/x86/qpeldsp_init.c
@@ -0,0 +1,501 @@
+/*
+ * quarterpel DSP functions
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include "config.h"
+#include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/x86/cpu.h"
+#include "libavcodec/pixels.h"
+#include "libavcodec/qpeldsp.h"
+#include "fpel.h"
+
+void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                              int dstStride, int src1Stride, int h);
+void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1,
+                                     uint8_t *src2, int dstStride,
+                                     int src1Stride, int h);
+void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                              int dstStride, int src1Stride, int h);
+void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                               int dstStride, int src1Stride, int h);
+void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                               int dstStride, int src1Stride, int h);
+void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                                      int dstStride, int src1Stride, int h);
+void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                          int dstStride, int srcStride, int h);
+void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                          int dstStride, int srcStride, int h);
+void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                                 int dstStride, int srcStride,
+                                                 int h);
+void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                         int dstStride, int srcStride, int h);
+void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                         int dstStride, int srcStride, int h);
+void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                                int dstStride, int srcStride,
+                                                int h);
+void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                          int dstStride, int srcStride);
+void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                          int dstStride, int srcStride);
+void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                                 int dstStride, int srcStride);
+void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                         int dstStride, int srcStride);
+void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                         int dstStride, int srcStride);
+void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                                int dstStride, int srcStride);
+#define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmxext
+#define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmxext
+
+#if HAVE_YASM
+
+CALL_2X_PIXELS(ff_avg_pixels16_mmxext, ff_avg_pixels8_mmxext, 8)
+CALL_2X_PIXELS(ff_put_pixels16_mmxext, ff_put_pixels8_mmxext, 8)
+
+#define QPEL_OP(OPNAME, RND, MMX)                                       \
+static void OPNAME ## qpel8_mc00_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);              \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t temp[8];                                                   \
+    uint8_t *const half = (uint8_t *) temp;                             \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
+                                                   stride, 8);          \
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
+                                        stride, stride, 8);             \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride,    \
+                                                   stride, 8);          \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t temp[8];                                                   \
+    uint8_t *const half = (uint8_t *) temp;                             \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
+                                                   stride, 8);          \
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride,     \
+                                        stride, 8);                     \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t temp[8];                                                   \
+    uint8_t *const half = (uint8_t *) temp;                             \
+    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
+                                                   8, stride);          \
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
+                                        stride, stride, 8);             \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src,            \
+                                                   stride, stride);     \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t temp[8];                                                   \
+    uint8_t *const half = (uint8_t *) temp;                             \
+    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
+                                                   8, stride);          \
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\
+                                        stride, 8);                     \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t half[8 + 9];                                               \
+    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                   stride, 9);          \
+    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
+                                        stride, 9);                     \
+    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
+                                        stride, 8, 8);                  \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t half[8 + 9];                                               \
+    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                   stride, 9);          \
+    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
+                                        stride, 9);                     \
+    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
+                                        stride, 8, 8);                  \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t half[8 + 9];                                               \
+    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                   stride, 9);          \
+    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
+                                        stride, 9);                     \
+    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
+                                        stride, 8, 8);                  \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t half[8 + 9];                                               \
+    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                   stride, 9);          \
+    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
+                                        stride, 9);                     \
+    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
+                                        stride, 8, 8);                  \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t half[8 + 9];                                               \
+    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                   stride, 9);          \
+    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
+                                        stride, 8, 8);                  \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t half[8 + 9];                                               \
+    uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                   stride, 9);          \
+    ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+    ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
+                                        stride, 8, 8);                  \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t half[8 + 9];                                               \
+    uint8_t *const halfH = (uint8_t *) half;                            \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                   stride, 9);          \
+    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH,              \
+                                        8, stride, 9);                  \
+    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
+                                                   stride, 8);          \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t half[8 + 9];                                               \
+    uint8_t *const halfH = (uint8_t *) half;                            \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                   stride, 9);          \
+    ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
+                                        stride, 9);                     \
+    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
+                                                   stride, 8);          \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                         ptrdiff_t stride)              \
+{                                                                       \
+    uint64_t half[9];                                                   \
+    uint8_t *const halfH = (uint8_t *) half;                            \
+    ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                   stride, 9);          \
+    ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
+                                                   stride, 8);          \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc00_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);            \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t temp[32];                                                  \
+    uint8_t *const half = (uint8_t *) temp;                             \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
+                                                    stride, 16);        \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
+                                         stride, 16);                   \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    ff_ ## OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src,           \
+                                                    stride, stride, 16);\
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t temp[32];                                                  \
+    uint8_t *const half = (uint8_t*) temp;                              \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
+                                                    stride, 16);        \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half,            \
+                                         stride, stride, 16);           \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t temp[32];                                                  \
+    uint8_t *const half = (uint8_t *) temp;                             \
+    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
+                                                    stride);            \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
+                                         stride, 16);                   \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src,           \
+                                                    stride, stride);    \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t temp[32];                                                  \
+    uint8_t *const half = (uint8_t *) temp;                             \
+    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
+                                                    stride);            \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half,         \
+                                         stride, stride, 16);           \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t half[16 * 2 + 17 * 2];                                     \
+    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                    stride, 17);        \
+    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
+                                         stride, 17);                   \
+    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                    16, 16);            \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
+                                         stride, 16, 16);               \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t half[16 * 2 + 17 * 2];                                     \
+    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                    stride, 17);        \
+    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
+                                         stride, 17);                   \
+    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                    16, 16);            \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
+                                         stride, 16, 16);               \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t half[16 * 2 + 17 * 2];                                     \
+    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                    stride, 17);        \
+    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
+                                         stride, 17);                   \
+    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                    16, 16);            \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
+                                         stride, 16, 16);               \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t half[16 * 2 + 17 * 2];                                     \
+    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                    stride, 17);        \
+    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
+                                         stride, 17);                   \
+    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                    16, 16);            \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
+                                         stride, 16, 16);               \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t half[16 * 2 + 17 * 2];                                     \
+    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                    stride, 17);        \
+    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                    16, 16);            \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
+                                         stride, 16, 16);               \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t half[16 * 2 + 17 * 2];                                     \
+    uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+    uint8_t *const halfHV = (uint8_t *) half;                           \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                    stride, 17);        \
+    ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                    16, 16);            \
+    ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
+                                         stride, 16, 16);               \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t half[17 * 2];                                              \
+    uint8_t *const halfH = (uint8_t *) half;                            \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                    stride, 17);        \
+    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
+                                         stride, 17);                   \
+    ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
+                                                    stride, 16);        \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t half[17 * 2];                                              \
+    uint8_t *const halfH = (uint8_t *) half;                            \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                    stride, 17);        \
+    ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
+                                         stride, 17);                   \
+    ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
+                                                    stride, 16);        \
+}                                                                       \
+                                                                        \
+static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                          ptrdiff_t stride)             \
+{                                                                       \
+    uint64_t half[17 * 2];                                              \
+    uint8_t *const halfH = (uint8_t *) half;                            \
+    ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                    stride, 17);        \
+    ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
+                                                    stride, 16);        \
+}
+
+QPEL_OP(put_,        _,        mmxext)
+QPEL_OP(avg_,        _,        mmxext)
+QPEL_OP(put_no_rnd_, _no_rnd_, mmxext)
+
+#endif /* HAVE_YASM */
+
+#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
+do {                                                                         \
+    c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
+    c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
+} while (0)
+
+av_cold void ff_qpeldsp_init_x86(QpelDSPContext *c)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (X86_MMXEXT(cpu_flags)) {
+#if HAVE_MMXEXT_EXTERNAL
+        SET_QPEL_FUNCS(avg_qpel,        0, 16, mmxext, );
+        SET_QPEL_FUNCS(avg_qpel,        1,  8, mmxext, );
+
+        SET_QPEL_FUNCS(put_qpel,        0, 16, mmxext, );
+        SET_QPEL_FUNCS(put_qpel,        1,  8, mmxext, );
+        SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, );
+        SET_QPEL_FUNCS(put_no_rnd_qpel, 1,  8, mmxext, );
+#endif /* HAVE_MMXEXT_EXTERNAL */
+    }
+}