Add apply_window_int16() to DSPContext with x86-optimized versions and use it
[ffmpeg.git] / libavcodec / x86 / dsputil_mmx.c
index a611a71..4d1a305 100644 (file)
@@ -2388,6 +2388,20 @@ int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2, int or
 int32_t ff_scalarproduct_and_madd_int16_mmx2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
 int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
 int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul);
+
+void ff_apply_window_int16_mmxext    (int16_t *output, const int16_t *input,
+                                      const int16_t *window, unsigned int len);
+void ff_apply_window_int16_mmxext_ba (int16_t *output, const int16_t *input,
+                                      const int16_t *window, unsigned int len);
+void ff_apply_window_int16_sse2      (int16_t *output, const int16_t *input,
+                                      const int16_t *window, unsigned int len);
+void ff_apply_window_int16_sse2_ba   (int16_t *output, const int16_t *input,
+                                      const int16_t *window, unsigned int len);
+void ff_apply_window_int16_ssse3     (int16_t *output, const int16_t *input,
+                                      const int16_t *window, unsigned int len);
+void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
+                                      const int16_t *window, unsigned int len);
+
 void ff_add_hfyu_median_prediction_mmx2(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);
@@ -2749,6 +2763,11 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 #if HAVE_YASM
             c->scalarproduct_int16 = ff_scalarproduct_int16_mmx2;
             c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmx2;
+            if (avctx->flags & CODEC_FLAG_BITEXACT) {
+                c->apply_window_int16 = ff_apply_window_int16_mmxext_ba;
+            } else {
+                c->apply_window_int16 = ff_apply_window_int16_mmxext;
+            }
 #endif
         }
         if(mm_flags & AV_CPU_FLAG_SSE){
@@ -2771,13 +2790,30 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 #if HAVE_YASM
             c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
             c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
+            if (avctx->flags & CODEC_FLAG_BITEXACT) {
+                c->apply_window_int16 = ff_apply_window_int16_sse2_ba;
+            } else {
+                if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
+                    c->apply_window_int16 = ff_apply_window_int16_sse2;
+                }
+            }
 
             c->emulated_edge_mc = emulated_edge_mc_sse;
             c->gmc= gmc_sse;
 #endif
         }
-        if((mm_flags & AV_CPU_FLAG_SSSE3) && !(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW)) && HAVE_YASM) // cachesplit
-            c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
+        if (mm_flags & AV_CPU_FLAG_SSSE3) {
+#if HAVE_YASM
+            if (mm_flags & AV_CPU_FLAG_ATOM) {
+                c->apply_window_int16 = ff_apply_window_int16_ssse3_atom;
+            } else {
+                c->apply_window_int16 = ff_apply_window_int16_ssse3;
+            }
+            if (!(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW))) { // cachesplit
+                c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
+            }
+#endif
+        }
     }
 
     if (CONFIG_ENCODERS)