From: James Almer <[email protected]>
---
libavcodec/x86/hevc_idct.asm | 16 ++++++++++++++++
libavcodec/x86/hevcdsp_init.c | 18 +++++++++++++++++-
2 files changed, 33 insertions(+), 1 deletion(-)
diff --git a/libavcodec/x86/hevc_idct.asm b/libavcodec/x86/hevc_idct.asm
index 46457b7..937ef15 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 47cd247..33bed7a 100644
--- a/libavcodec/x86/hevcdsp_init.c
+++ b/libavcodec/x86/hevcdsp_init.c
@@ -68,7 +68,8 @@ idct_dc_proto(32,10, avx2);
#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);
@@ -309,6 +310,12 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int
bit_depth)
SET_CHROMA_FUNCS(put_unweighted_pred_chroma,
ff_hevc_put_unweighted_pred, 10, sse2);
SET_CHROMA_FUNCS(put_unweighted_pred_avg_chroma,
ff_hevc_put_unweighted_pred_avg, 10, sse2);
}
+ } else if (bit_depth == 12) {
+ if (EXTERNAL_SSE2(cpu_flags)) {
+ 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 ARCH_X86_64
@@ -360,6 +367,15 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int
bit_depth)
c->idct_dc[2] = ff_hevc_idct16x16_dc_10_avx2;
c->idct_dc[3] = ff_hevc_idct32x32_dc_10_avx2;
}
+ } else if (bit_depth == 12) {
+ if (EXTERNAL_MMXEXT(cpu_flags)) {
+ c->idct_dc[0] = ff_hevc_idct4x4_dc_12_mmxext;
+ c->idct_dc[1] = ff_hevc_idct8x8_dc_12_mmxext;
+ }
+ if (EXTERNAL_AVX2(cpu_flags)) {
+ c->idct_dc[2] = ff_hevc_idct16x16_dc_12_avx2;
+ c->idct_dc[3] = ff_hevc_idct32x32_dc_12_avx2;
+ }
}
#endif /* ARCH_X86_64 */
}
--
2.6.4 (Apple Git-63)
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
