x86: hevc: adding transform_add

Reviewed-by: James Almer <jamrial@gmail.com>
Approved-by: Ronald S. Bultje
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/x86/Makefile

@@ -131,7 +131,8 @@ YASM-OBJS-$(CONFIG_AAC_DECODER)        += x86/sbrdsp.o
 YASM-OBJS-$(CONFIG_DCA_DECODER)        += x86/dcadsp.o
 YASM-OBJS-$(CONFIG_HEVC_DECODER)       += x86/hevc_mc.o                 \
                                           x86/hevc_deblock.o            \
-                                          x86/hevc_idct.o
+                                          x86/hevc_idct.o               \
+                                          x86/hevc_res_add.o
 YASM-OBJS-$(CONFIG_PNG_DECODER)        += x86/pngdsp.o
 YASM-OBJS-$(CONFIG_PRORES_DECODER)     += x86/proresdsp.o
 YASM-OBJS-$(CONFIG_PRORES_LGPL_DECODER) += x86/proresdsp.o

libavcodec/x86/hevcdsp.h

@@ -131,4 +131,20 @@ WEIGHTING_PROTOTYPES(8, sse4);
 WEIGHTING_PROTOTYPES(10, sse4);
 WEIGHTING_PROTOTYPES(12, sse4);
 
+///////////////////////////////////////////////////////////////////////////////
+// TRANSFORM_ADD
+///////////////////////////////////////////////////////////////////////////////
+void ff_hevc_transform_add4_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+void ff_hevc_transform_add8_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+void ff_hevc_transform_add16_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+void ff_hevc_transform_add32_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+
+void ff_hevc_transform_add4_10_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+void ff_hevc_transform_add8_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+void ff_hevc_transform_add16_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+void ff_hevc_transform_add32_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+
+void ff_hevc_transform_add16_10_avx2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+void ff_hevc_transform_add32_10_avx2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
+
 #endif // AVCODEC_X86_HEVCDSP_H

libavcodec/x86/hevcdsp_init.c

@@ -469,6 +469,7 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
         if (EXTERNAL_MMXEXT(cpu_flags)) {
             c->idct_dc[0] = ff_hevc_idct4x4_dc_8_mmxext;
             c->idct_dc[1] = ff_hevc_idct8x8_dc_8_mmxext;
+            c->transform_add[0]    =  ff_hevc_transform_add4_8_mmxext;
         }
         if (EXTERNAL_SSE2(cpu_flags)) {
             c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
@@ -476,11 +477,15 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
             if (ARCH_X86_64) {
                 c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_sse2;
                 c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_sse2;
-            }
 
+                c->transform_add[2]    = ff_hevc_transform_add16_8_sse2;
+                c->transform_add[3]    = ff_hevc_transform_add32_8_sse2;
+            }
             c->idct_dc[1] = ff_hevc_idct8x8_dc_8_sse2;
             c->idct_dc[2] = ff_hevc_idct16x16_dc_8_sse2;
             c->idct_dc[3] = ff_hevc_idct32x32_dc_8_sse2;
+
+            c->transform_add[1]    = ff_hevc_transform_add8_8_sse2;
         }
         if (EXTERNAL_SSSE3(cpu_flags) && ARCH_X86_64) {
             c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_ssse3;
@@ -512,6 +517,7 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
         }
     } else if (bit_depth == 10) {
         if (EXTERNAL_MMXEXT(cpu_flags)) {
+            c->transform_add[0] = ff_hevc_transform_add4_10_mmxext;
             c->idct_dc[0] = ff_hevc_idct4x4_dc_10_mmxext;
             c->idct_dc[1] = ff_hevc_idct8x8_dc_10_mmxext;
         }
@@ -526,6 +532,10 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
             c->idct_dc[1] = ff_hevc_idct8x8_dc_10_sse2;
             c->idct_dc[2] = ff_hevc_idct16x16_dc_10_sse2;
             c->idct_dc[3] = ff_hevc_idct32x32_dc_10_sse2;
+
+            c->transform_add[1]    = ff_hevc_transform_add8_10_sse2;
+            c->transform_add[2]    = ff_hevc_transform_add16_10_sse2;
+            c->transform_add[3]    = ff_hevc_transform_add32_10_sse2;
         }
         if (EXTERNAL_SSSE3(cpu_flags) && ARCH_X86_64) {
             c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3;
@@ -551,9 +561,13 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
             }
         }
         if (EXTERNAL_AVX2(cpu_flags)) {
+
             c->idct_dc[2] = ff_hevc_idct16x16_dc_10_avx2;
             c->idct_dc[3] = ff_hevc_idct32x32_dc_10_avx2;
 
+            c->transform_add[2] = ff_hevc_transform_add16_10_avx2;
+            c->transform_add[3] = ff_hevc_transform_add32_10_avx2;
+
         }
     } else if (bit_depth == 12) {
         if (EXTERNAL_MMXEXT(cpu_flags)) {