vp9: add fullpel (avg) MC SIMD for 10/12bpp.

1 files changed, 42 insertions, 16 deletions

diff --git a/libavcodec/x86/vp9dsp_init_16bpp.c b/libavcodec/x86/vp9dsp_init_16bpp.c
index 6f2c50d04a..daf9cc5d6d 100644
--- a/libavcodec/x86/vp9dsp_init_16bpp.c
+++ b/libavcodec/x86/vp9dsp_init_16bpp.c
@@ -29,14 +29,22 @@
 
 #if HAVE_YASM
 
-decl_fpel_func(put,   8, mmx);
-decl_fpel_func(put,  16, sse);
-decl_fpel_func(put,  32, sse);
-decl_fpel_func(put,  64, sse);
-decl_fpel_func(put, 128, sse);
-decl_fpel_func(put,  32, avx);
-decl_fpel_func(put,  64, avx);
-decl_fpel_func(put, 128, avx);
+decl_fpel_func(put,   8,    , mmx);
+decl_fpel_func(avg,   8, _16, mmxext);
+decl_fpel_func(put,  16,    , sse);
+decl_fpel_func(put,  32,    , sse);
+decl_fpel_func(put,  64,    , sse);
+decl_fpel_func(put, 128,    , sse);
+decl_fpel_func(avg,  16, _16, sse2);
+decl_fpel_func(avg,  32, _16, sse2);
+decl_fpel_func(avg,  64, _16, sse2);
+decl_fpel_func(avg, 128, _16, sse2);
+decl_fpel_func(put,  32,    , avx);
+decl_fpel_func(put,  64,    , avx);
+decl_fpel_func(put, 128,    , avx);
+decl_fpel_func(avg,  32, _16, avx2);
+decl_fpel_func(avg,  64, _16, avx2);
+decl_fpel_func(avg, 128, _16, avx2);
 
 #endif /* HAVE_YASM */
 
@@ -46,19 +54,37 @@ av_cold void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp, int bpp)
     int cpu_flags = av_get_cpu_flags();
 
     if (EXTERNAL_MMX(cpu_flags)) {
-        init_fpel_func(4, 0,   8, put, mmx);
+        init_fpel_func(4, 0,   8, put, , mmx);
+    }
+
+    if (EXTERNAL_MMXEXT(cpu_flags)) {
+        init_fpel_func(4, 1,   8, avg, _16, mmxext);
     }
 
     if (EXTERNAL_SSE(cpu_flags)) {
-        init_fpel_func(3, 0,  16, put, sse);
-        init_fpel_func(2, 0,  32, put, sse);
-        init_fpel_func(1, 0,  64, put, sse);
-        init_fpel_func(0, 0, 128, put, sse);
+        init_fpel_func(3, 0,  16, put, , sse);
+        init_fpel_func(2, 0,  32, put, , sse);
+        init_fpel_func(1, 0,  64, put, , sse);
+        init_fpel_func(0, 0, 128, put, , sse);
+    }
+
+    if (EXTERNAL_SSE2(cpu_flags)) {
+        init_fpel_func(3, 1,  16, avg, _16, sse2);
+        init_fpel_func(2, 1,  32, avg, _16, sse2);
+        init_fpel_func(1, 1,  64, avg, _16, sse2);
+        init_fpel_func(0, 1, 128, avg, _16, sse2);
     }
+
     if (EXTERNAL_AVX_FAST(cpu_flags)) {
-        init_fpel_func(2, 0,  32, put, avx);
-        init_fpel_func(1, 0,  64, put, avx);
-        init_fpel_func(0, 0, 128, put, avx);
+        init_fpel_func(2, 0,  32, put, , avx);
+        init_fpel_func(1, 0,  64, put, , avx);
+        init_fpel_func(0, 0, 128, put, , avx);
+    }
+
+    if (EXTERNAL_AVX2(cpu_flags)) {
+        init_fpel_func(2, 1,  32, avg, _16, avx2);
+        init_fpel_func(1, 1,  64, avg, _16, avx2);
+        init_fpel_func(0, 1, 128, avg, _16, avx2);
     }
 
 #endif /* HAVE_YASM */