[FFmpeg-devel] [PATCH] x86/hevc_idct: add 12bit idct_dc

[James Almer]

Signed-off-by: James Almer <jamrial at gmail.com>
---
Didn't realize that the newest HEVC fate tests were 12 bit, so i didn't 
bother adding what i thought i couldn't feasibly test.
And as i mentioned in the previous patch, adding this is trivial.

 libavcodec/x86/hevc_idct.asm  | 16 ++++++++++++++++
 libavcodec/x86/hevcdsp_init.c | 15 ++++++++++++++-
 2 files changed, 30 insertions(+), 1 deletion(-)

diff --git a/libavcodec/x86/hevc_idct.asm b/libavcodec/x86/hevc_idct.asm
index 7a44f08..2ad6e97 100644
--- a/libavcodec/x86/hevc_idct.asm
+++ b/libavcodec/x86/hevc_idct.asm
@@ -104,3 +104,19 @@ INIT_YMM avx2
 IDCT_DC    16,  2, 10
 IDCT_DC    32,  8, 10
 %endif ;HAVE_AVX2_EXTERNAL
+
+; 12-bit
+INIT_MMX mmxext
+IDCT_DC_NL  4,     12
+IDCT_DC     8,  2, 12
+
+INIT_XMM sse2
+IDCT_DC_NL  8,     12
+IDCT_DC    16,  4, 12
+IDCT_DC    32, 16, 12
+
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+IDCT_DC    16,  2, 12
+IDCT_DC    32,  8, 12
+%endif ;HAVE_AVX2_EXTERNAL
diff --git a/libavcodec/x86/hevcdsp_init.c b/libavcodec/x86/hevcdsp_init.c
index fb3357b..e993962 100644
--- a/libavcodec/x86/hevcdsp_init.c
+++ b/libavcodec/x86/hevcdsp_init.c
@@ -56,7 +56,8 @@ LFL_FUNCS(uint8_t,  12, ssse3)
 
 #define IDCT_FUNCS(W, opt) \
 void ff_hevc_idct##W##_dc_8_##opt(int16_t *coeffs); \
-void ff_hevc_idct##W##_dc_10_##opt(int16_t *coeffs)
+void ff_hevc_idct##W##_dc_10_##opt(int16_t *coeffs); \
+void ff_hevc_idct##W##_dc_12_##opt(int16_t *coeffs)
 
 IDCT_FUNCS(4x4,   mmxext);
 IDCT_FUNCS(8x8,   mmxext);
@@ -535,6 +536,10 @@ void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth)
 
         }
     } else if (bit_depth == 12) {
+        if (EXTERNAL_MMXEXT(mm_flags)) {
+            c->idct_dc[0] = ff_hevc_idct4x4_dc_12_mmxext;
+            c->idct_dc[1] = ff_hevc_idct8x8_dc_12_mmxext;
+        }
         if (EXTERNAL_SSE2(mm_flags)) {
             c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_12_sse2;
             c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_12_sse2;
@@ -542,6 +547,10 @@ void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth)
                 c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_12_sse2;
                 c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_12_sse2;
             }
+
+            c->idct_dc[1] = ff_hevc_idct8x8_dc_12_sse2;
+            c->idct_dc[2] = ff_hevc_idct16x16_dc_12_sse2;
+            c->idct_dc[3] = ff_hevc_idct32x32_dc_12_sse2;
         }
         if (EXTERNAL_SSSE3(mm_flags) && ARCH_X86_64) {
             c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_12_ssse3;
@@ -558,5 +567,9 @@ void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth)
             QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v,     12, sse4);
             QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv,    12, sse4);
         }
+        if (EXTERNAL_AVX2(mm_flags)) {
+            c->idct_dc[2] = ff_hevc_idct16x16_dc_12_avx2;
+            c->idct_dc[3] = ff_hevc_idct32x32_dc_12_avx2;
+        }
     }
 }
-- 
1.8.5.5