Merge commit 'b5aa48551300eed678aaea86ced7086758598a35'

[ffmpeg] / libavcodec / x86 / dsputil_init.c

diff --git a/libavcodec/x86/dsputil_init.c b/libavcodec/x86/dsputil_init.c
index 6c7b5218c46e9b439bd0d58ed238c78d65803b48..9fc92fe7992c656fa9d665b6a83fac0f96e48381 100644 (file)
--- a/libavcodec/x86/dsputil_init.c
+++ b/libavcodec/x86/dsputil_init.c
@@ -23,7 +23,6 @@
 #include "libavutil/attributes.h"
 #include "libavutil/cpu.h"
 #include "libavutil/internal.h"
-#include "libavutil/x86/asm.h"
 #include "libavutil/x86/cpu.h"
 #include "libavcodec/avcodec.h"
 #include "libavcodec/dsputil.h"
@@ -73,8 +72,8 @@ 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
+#define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmx
+#define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmx
 
 int32_t ff_scalarproduct_int16_mmxext(const int16_t *v1, const int16_t *v2,
                                       int order);
@@ -93,14 +92,6 @@ int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2,
 void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w);
 void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w);
 
-void ff_add_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *top,
-                                          const uint8_t *diff, int w,
-                                          int *left, int *left_top);
-int ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src,
-                                      int w, int left);
-int ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src,
-                                     int w, int left);
-
 void ff_vector_clip_int32_mmx(int32_t *dst, const int32_t *src,
                               int32_t min, int32_t max, unsigned int len);
 void ff_vector_clip_int32_sse2(int32_t *dst, const int32_t *src,
@@ -112,8 +103,8 @@ void ff_vector_clip_int32_sse4(int32_t *dst, const int32_t *src,
 
 #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 ff_put_pixels16_mmxext ff_put_pixels16_mmx
+#define ff_put_pixels8_mmxext  ff_put_pixels8_mmx
 
 #define QPEL_OP(OPNAME, RND, MMX)                                       \
 static void OPNAME ## qpel8_mc00_ ## MMX(uint8_t *dst, uint8_t *src,    \
@@ -527,24 +518,24 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
 {
 #if HAVE_MMX_INLINE
     c->put_pixels_clamped        = ff_put_pixels_clamped_mmx;
-    c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
     c->add_pixels_clamped        = ff_add_pixels_clamped_mmx;
 
     if (!high_bit_depth) {
-        c->clear_block  = ff_clear_block_mmx;
-        c->clear_blocks = ff_clear_blocks_mmx;
         c->draw_edges   = ff_draw_edges_mmx;
     }
 
 #if CONFIG_VIDEODSP && (ARCH_X86_32 || !HAVE_YASM)
     c->gmc = ff_gmc_mmx;
 #endif
-
-    c->add_bytes = ff_add_bytes_mmx;
 #endif /* HAVE_MMX_INLINE */
 
 #if HAVE_MMX_EXTERNAL
+    if (!high_bit_depth) {
+        c->clear_block  = ff_clear_block_mmx;
+        c->clear_blocks = ff_clear_blocks_mmx;
+    }
     c->vector_clip_int32 = ff_vector_clip_int32_mmx;
+    c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
 #endif /* HAVE_MMX_EXTERNAL */
 }
 
@@ -568,10 +559,6 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
     SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, );
     SET_QPEL_FUNCS(put_no_rnd_qpel, 1,  8, mmxext, );
 
-    /* slower than cmov version on AMD */
-    if (!(cpu_flags & AV_CPU_FLAG_3DNOW))
-        c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmxext;
-
     c->scalarproduct_int16          = ff_scalarproduct_int16_mmxext;
     c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmxext;
 #endif /* HAVE_MMXEXT_EXTERNAL */
@@ -580,7 +567,8 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
 static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx,
                                      int cpu_flags, unsigned high_bit_depth)
 {
-#if HAVE_SSE_INLINE
+#if HAVE_YASM
+#if HAVE_SSE_EXTERNAL
     c->vector_clipf = ff_vector_clipf_sse;
 
     /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
@@ -591,9 +579,7 @@ static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx,
         c->clear_block  = ff_clear_block_sse;
         c->clear_blocks = ff_clear_blocks_sse;
     }
-#endif /* HAVE_SSE_INLINE */
-
-#if HAVE_YASM
+#endif
 #if HAVE_INLINE_ASM && CONFIG_VIDEODSP
     c->gmc = ff_gmc_sse;
 #endif
@@ -621,6 +607,7 @@ static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
         c->vector_clip_int32 = ff_vector_clip_int32_sse2;
     }
     c->bswap_buf = ff_bswap32_buf_sse2;
+    c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
 #endif /* HAVE_SSE2_EXTERNAL */
 }
 
@@ -628,10 +615,6 @@ static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx,
                                        int cpu_flags, unsigned high_bit_depth)
 {
 #if HAVE_SSSE3_EXTERNAL
-    c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
-    if (cpu_flags & AV_CPU_FLAG_SSE4) // not really SSE4, just slow on Conroe
-        c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;
-
     if (!(cpu_flags & (AV_CPU_FLAG_SSE42 | AV_CPU_FLAG_3DNOW))) // cachesplit
         c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
     c->bswap_buf = ff_bswap32_buf_ssse3;
@@ -651,11 +634,6 @@ av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx,
 {
     int cpu_flags = av_get_cpu_flags();
 
-#if HAVE_7REGS && HAVE_INLINE_ASM
-    if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_CMOV)
-        c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_cmov;
-#endif
-
     if (X86_MMX(cpu_flags)) {
 #if HAVE_INLINE_ASM
         const int idct_algo = avctx->idct_algo;