x86: hpeldsp: propagate changes across codecs

Some codecs still use mmx versions, so have them use the versions
with newer instruction sets.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/x86/cavsdsp.c

@@ -474,6 +474,12 @@ static void avg_cavs_qpel8_mc00_mmx(uint8_t *dst, uint8_t *src,
     ff_avg_pixels8_mmx(dst, src, stride, 8);
 }
 
+static void avg_cavs_qpel8_mc00_mmxext(uint8_t *dst, uint8_t *src,
+                                       ptrdiff_t stride)
+{
+    ff_avg_pixels8_mmxext(dst, src, stride, 8);
+}
+
 static void put_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src,
                                      ptrdiff_t stride)
 {
@@ -485,6 +491,24 @@ static void avg_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src,
 {
     ff_avg_pixels16_mmx(dst, src, stride, 16);
 }
+
+static void avg_cavs_qpel16_mc00_mmxext(uint8_t *dst, uint8_t *src,
+                                        ptrdiff_t stride)
+{
+    ff_avg_pixels16_mmxext(dst, src, stride, 16);
+}
+
+static void put_cavs_qpel16_mc00_sse2(uint8_t *dst, uint8_t *src,
+                                      ptrdiff_t stride)
+{
+    ff_put_pixels16_sse2(dst, src, stride, 16);
+}
+
+static void avg_cavs_qpel16_mc00_sse2(uint8_t *dst, uint8_t *src,
+                                      ptrdiff_t stride)
+{
+    ff_avg_pixels16_sse2(dst, src, stride, 16);
+}
 #endif
 
 static av_cold void cavsdsp_init_mmx(CAVSDSPContext *c,
@@ -517,15 +541,6 @@ CAVS_MC(put_,  8, mmxext)
 CAVS_MC(put_, 16, mmxext)
 CAVS_MC(avg_,  8, mmxext)
 CAVS_MC(avg_, 16, mmxext)
-
-static av_cold void cavsdsp_init_mmxext(CAVSDSPContext *c,
-                                        AVCodecContext *avctx)
-{
-    DSPFUNC(put, 0, 16, mmxext);
-    DSPFUNC(put, 1,  8, mmxext);
-    DSPFUNC(avg, 0, 16, mmxext);
-    DSPFUNC(avg, 1,  8, mmxext);
-}
 #endif /* HAVE_MMXEXT_INLINE */
 
 #if HAVE_AMD3DNOW_INLINE
@@ -557,7 +572,23 @@ av_cold void ff_cavsdsp_init_x86(CAVSDSPContext *c, AVCodecContext *avctx)
         cavsdsp_init_3dnow(c, avctx);
 #endif /* HAVE_AMD3DNOW_INLINE */
 #if HAVE_MMXEXT_INLINE
-    if (INLINE_MMXEXT(cpu_flags))
-        cavsdsp_init_mmxext(c, avctx);
-#endif /* HAVE_MMXEXT_INLINE */
+    if (INLINE_MMXEXT(cpu_flags)) {
+        DSPFUNC(put, 0, 16, mmxext);
+        DSPFUNC(put, 1,  8, mmxext);
+        DSPFUNC(avg, 0, 16, mmxext);
+        DSPFUNC(avg, 1,  8, mmxext);
+    }
+#endif
+#if HAVE_MMX_EXTERNAL
+    if (EXTERNAL_MMXEXT(cpu_flags)) {
+        c->avg_cavs_qpel_pixels_tab[0][0] = avg_cavs_qpel16_mc00_mmxext;
+        c->avg_cavs_qpel_pixels_tab[1][0] = avg_cavs_qpel8_mc00_mmxext;
+    }
+#endif
+#if HAVE_SSE2_EXTERNAL
+    if (EXTERNAL_SSE2(cpu_flags)) {
+        c->put_cavs_qpel_pixels_tab[0][0] = put_cavs_qpel16_mc00_sse2;
+        c->avg_cavs_qpel_pixels_tab[0][0] = avg_cavs_qpel16_mc00_sse2;
+    }
+#endif
 }

libavcodec/x86/diracdsp_mmx.c

@@ -83,9 +83,6 @@ void ff_ ## OPNAME2 ## _dirac_pixels32_ ## EXT(uint8_t *dst, const uint8_t *src[
     }\
 }
 
-void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
-                            ptrdiff_t line_size, int h);
-
 DIRAC_PIXOP(put, ff_put, mmx)
 DIRAC_PIXOP(avg, ff_avg, mmx)
 DIRAC_PIXOP(avg, ff_avg, mmxext)

libavcodec/x86/dsputil_init.c

@@ -115,9 +115,6 @@ void ff_vector_clip_int32_sse4(int32_t *dst, const int32_t *src,
 #define ff_put_pixels16_mmxext ff_put_pixels16_mmx
 #define ff_put_pixels8_mmxext  ff_put_pixels8_mmx
 
-void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
-                            ptrdiff_t line_size, int h);
-
 #define QPEL_OP(OPNAME, RND, MMX)                                       \
 static void OPNAME ## qpel8_mc00_ ## MMX(uint8_t *dst, uint8_t *src,    \
                                          ptrdiff_t stride)              \

libavcodec/x86/fpel.h

@@ -28,6 +28,8 @@ void ff_avg_pixels8_mmxext(uint8_t *block, const uint8_t *pixels,
                            ptrdiff_t line_size, int h);
 void ff_avg_pixels16_mmx(uint8_t *block, const uint8_t *pixels,
                          ptrdiff_t line_size, int h);
+void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
+                            ptrdiff_t line_size, int h);
 void ff_avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
                           ptrdiff_t line_size, int h);
 void ff_put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,

libavcodec/x86/h264_qpel.c

@@ -53,9 +53,6 @@ void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
 #define ff_put_pixels8_mmxext   ff_put_pixels8_mmx
 #define ff_put_pixels4_mmxext   ff_put_pixels4_mmx
 
-void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
-                            ptrdiff_t line_size, int h);
-
 #define DEF_QPEL(OPNAME)\
 void ff_ ## OPNAME ## _h264_qpel4_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\
 void ff_ ## OPNAME ## _h264_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, int dstStride, int srcStride);\

libavcodec/x86/hpeldsp.h

@@ -27,12 +27,27 @@ void ff_avg_pixels8_x2_mmx(uint8_t *block, const uint8_t *pixels,
 
 void ff_avg_pixels8_xy2_mmx(uint8_t *block, const uint8_t *pixels,
                             ptrdiff_t line_size, int h);
+void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
+                               ptrdiff_t line_size, int h);
+void ff_avg_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
+                               ptrdiff_t line_size, int h);
+
 void ff_avg_pixels16_xy2_mmx(uint8_t *block, const uint8_t *pixels,
                              ptrdiff_t line_size, int h);
+void ff_avg_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void ff_avg_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
+                               ptrdiff_t line_size, int h);
 
 void ff_put_pixels8_xy2_mmx(uint8_t *block, const uint8_t *pixels,
                             ptrdiff_t line_size, int h);
+void ff_put_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
 void ff_put_pixels16_xy2_mmx(uint8_t *block, const uint8_t *pixels,
                              ptrdiff_t line_size, int h);
+void ff_put_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void ff_put_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
+                               ptrdiff_t line_size, int h);
 
 #endif /* AVCODEC_X86_HPELDSP_H */

libavcodec/x86/hpeldsp_init.c

@@ -48,10 +48,6 @@ void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
                              ptrdiff_t line_size, int h);
 void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
                              ptrdiff_t line_size, int h);
-void ff_put_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels,
-                              ptrdiff_t line_size, int h);
-void ff_avg_pixels16_xy2_sse2(uint8_t *block, const uint8_t *pixels,
-                              ptrdiff_t line_size, int h);
 void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
                                      ptrdiff_t line_size, int h);
 void ff_put_no_rnd_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
@@ -86,8 +82,6 @@ void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
                               ptrdiff_t line_size, int h);
 void ff_avg_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
                              ptrdiff_t line_size, int h);
-void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
-                               ptrdiff_t line_size, int h);
 void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
                               ptrdiff_t line_size, int h);
 void ff_avg_approx_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
@@ -95,15 +89,6 @@ void ff_avg_approx_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
 void ff_avg_approx_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
                                      ptrdiff_t line_size, int h);
 
-void ff_put_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
-                              ptrdiff_t line_size, int h);
-void ff_avg_pixels8_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
-                              ptrdiff_t line_size, int h);
-void ff_put_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
-                               ptrdiff_t line_size, int h);
-void ff_avg_pixels16_xy2_ssse3(uint8_t *block, const uint8_t *pixels,
-                               ptrdiff_t line_size, int h);
-
 #define avg_pixels8_mmx         ff_avg_pixels8_mmx
 #define avg_pixels8_x2_mmx      ff_avg_pixels8_x2_mmx
 #define avg_pixels16_mmx        ff_avg_pixels16_mmx

libavcodec/x86/rv40dsp_init.c

@@ -32,6 +32,13 @@
 #include "libavutil/x86/cpu.h"
 #include "hpeldsp.h"
 
+#define DEFINE_FN(op, size, insn) \
+static void op##_rv40_qpel##size##_mc33_##insn(uint8_t *dst, uint8_t *src, \
+                                               ptrdiff_t stride) \
+{ \
+    ff_##op##_pixels##size##_xy2_##insn(dst, src, stride, size); \
+}
+
 #if HAVE_YASM
 void ff_put_rv40_chroma_mc8_mmx  (uint8_t *dst, uint8_t *src,
                                   int stride, int h, int x, int y);
@@ -186,30 +193,24 @@ QPEL_FUNCS_SET (OP, 3, 1, OPT) \
 QPEL_FUNCS_SET (OP, 3, 2, OPT)
 /** @} */
 
+DEFINE_FN(put, 8, ssse3)
+
+DEFINE_FN(put, 16, sse2)
+DEFINE_FN(put, 16, ssse3)
+
+DEFINE_FN(avg, 8, mmxext)
+DEFINE_FN(avg, 8, ssse3)
+
+DEFINE_FN(avg, 16, sse2)
+DEFINE_FN(avg, 16, ssse3)
 #endif /* HAVE_YASM */
 
 #if HAVE_MMX_INLINE
-static void put_rv40_qpel8_mc33_mmx(uint8_t *dst, uint8_t *src,
-                                    ptrdiff_t stride)
-{
-    ff_put_pixels8_xy2_mmx(dst, src, stride, 8);
-}
-static void put_rv40_qpel16_mc33_mmx(uint8_t *dst, uint8_t *src,
-                                     ptrdiff_t stride)
-{
-    ff_put_pixels16_xy2_mmx(dst, src, stride, 16);
-}
-static void avg_rv40_qpel8_mc33_mmx(uint8_t *dst, uint8_t *src,
-                                    ptrdiff_t stride)
-{
-    ff_avg_pixels8_xy2_mmx(dst, src, stride, 8);
-}
-static void avg_rv40_qpel16_mc33_mmx(uint8_t *dst, uint8_t *src,
-                                     ptrdiff_t stride)
-{
-    ff_avg_pixels16_xy2_mmx(dst, src, stride, 16);
-}
-#endif /* HAVE_MMX_INLINE */
+DEFINE_FN(put, 8, mmx)
+DEFINE_FN(avg, 8, mmx)
+DEFINE_FN(put, 16, mmx)
+DEFINE_FN(avg, 16, mmx)
+#endif
 
 av_cold void ff_rv40dsp_init_x86(RV34DSPContext *c)
 {
@@ -240,6 +241,7 @@ av_cold void ff_rv40dsp_init_x86(RV34DSPContext *c)
 #endif
     }
     if (EXTERNAL_MMXEXT(cpu_flags)) {
+        c->avg_pixels_tab[1][15]        = avg_rv40_qpel8_mc33_mmxext;
         c->avg_chroma_pixels_tab[0]     = ff_avg_rv40_chroma_mc8_mmxext;
         c->avg_chroma_pixels_tab[1]     = ff_avg_rv40_chroma_mc4_mmxext;
         c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_mmxext;
@@ -251,6 +253,8 @@ av_cold void ff_rv40dsp_init_x86(RV34DSPContext *c)
 #endif
     }
     if (EXTERNAL_SSE2(cpu_flags)) {
+        c->put_pixels_tab[0][15]        = put_rv40_qpel16_mc33_sse2;
+        c->avg_pixels_tab[0][15]        = avg_rv40_qpel16_mc33_sse2;
         c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_sse2;
         c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_sse2;
         c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_sse2;
@@ -259,6 +263,10 @@ av_cold void ff_rv40dsp_init_x86(RV34DSPContext *c)
         QPEL_MC_SET(avg_, _sse2)
     }
     if (EXTERNAL_SSSE3(cpu_flags)) {
+        c->put_pixels_tab[0][15]        = put_rv40_qpel16_mc33_ssse3;
+        c->put_pixels_tab[1][15]        = put_rv40_qpel8_mc33_ssse3;
+        c->avg_pixels_tab[0][15]        = avg_rv40_qpel16_mc33_ssse3;
+        c->avg_pixels_tab[1][15]        = avg_rv40_qpel8_mc33_ssse3;
         c->rv40_weight_pixels_tab[0][0] = ff_rv40_weight_func_rnd_16_ssse3;
         c->rv40_weight_pixels_tab[0][1] = ff_rv40_weight_func_rnd_8_ssse3;
         c->rv40_weight_pixels_tab[1][0] = ff_rv40_weight_func_nornd_16_ssse3;