hevc: Add AVX2 DC IDCT

Originally written by Pierre Edouard Lepere <pierre-edouard.lepere@insa-rennes.fr>.
Integrated to Libav by Josh de Kock <josh@itanimul.li>.

Signed-off-by: Alexandra Hájková <alexandra@khirnov.net>

4 files changed, 167 insertions, 2 deletions

diff --git a/libavcodec/hevc.c b/libavcodec/hevc.c
index 5d58b5212a..961991f403 100644
--- a/libavcodec/hevc.c
+++ b/libavcodec/hevc.c
@@ -900,7 +900,7 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
     int vshift       = s->ps.sps->vshift[c_idx];
     uint8_t *dst     = &s->frame->data[c_idx][(y0 >> vshift) * stride +
                                               ((x0 >> hshift) << s->ps.sps->pixel_shift)];
-    DECLARE_ALIGNED(16, int16_t, coeffs[MAX_TB_SIZE * MAX_TB_SIZE]) = { 0 };
+    DECLARE_ALIGNED(32, int16_t, coeffs[MAX_TB_SIZE * MAX_TB_SIZE]) = { 0 };
     DECLARE_ALIGNED(8, uint8_t, significant_coeff_group_flag[8][8]) = { { 0 } };
 
     int trafo_size = 1 << log2_trafo_size;
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index cdf7758858..1460197248 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -115,7 +115,8 @@ YASM-OBJS-$(CONFIG_APE_DECODER)        += x86/apedsp.o
 YASM-OBJS-$(CONFIG_DCA_DECODER)        += x86/dcadsp.o
 YASM-OBJS-$(CONFIG_DNXHD_ENCODER)      += x86/dnxhdenc.o
 YASM-OBJS-$(CONFIG_HEVC_DECODER)       += x86/hevc_deblock.o            \
-                                          x86/hevc_mc.o
+                                          x86/hevc_mc.o                 \
+                                          x86/hevc_idct.o
 YASM-OBJS-$(CONFIG_PNG_DECODER)        += x86/pngdsp.o
 YASM-OBJS-$(CONFIG_PRORES_DECODER)     += x86/proresdsp.o
 YASM-OBJS-$(CONFIG_RV40_DECODER)       += x86/rv40dsp.o
diff --git a/libavcodec/x86/hevc_idct.asm b/libavcodec/x86/hevc_idct.asm
new file mode 100644
index 0000000000..d662aa90ba
--- /dev/null
+++ b/libavcodec/x86/hevc_idct.asm
@@ -0,0 +1,107 @@
+;*******************************************************************************
+;* SIMD-optimized IDCT functions for HEVC decoding
+;* Copyright (c) 2014 Pierre-Edouard LEPERE
+;* Copyright (c) 2014 James Almer
+;*
+;* This file is part of Libav.
+;*
+;* Libav is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* Libav is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with Libav; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+section .text
+
+; void ff_hevc_idctHxW_dc_{8,10}_<opt>(int16_t *coeffs)
+; %1 = HxW
+; %2 = number of loops
+; %3 = bitdepth
+%macro IDCT_DC 3
+cglobal hevc_idct_%1x%1_dc_%3, 1, 2, 1, coeff, tmp
+    movsx             tmpd, word [coeffq]
+    add               tmpd, (1 << (14 - %3)) + 1
+    sar               tmpd, (15 - %3)
+    movd               xm0, tmpd
+    SPLATW              m0, xm0
+    DEFINE_ARGS coeff, cnt
+    mov               cntd, %2
+.loop:
+    mova [coeffq+mmsize*0], m0
+    mova [coeffq+mmsize*1], m0
+    mova [coeffq+mmsize*2], m0
+    mova [coeffq+mmsize*3], m0
+    add  coeffq, mmsize*8
+    mova [coeffq+mmsize*-4], m0
+    mova [coeffq+mmsize*-3], m0
+    mova [coeffq+mmsize*-2], m0
+    mova [coeffq+mmsize*-1], m0
+    dec  cntd
+    jg  .loop
+    RET
+%endmacro
+
+; %1 = HxW
+; %2 = bitdepth
+%macro IDCT_DC_NL 2 ; No loop
+cglobal hevc_idct_%1x%1_dc_%2, 1, 2, 1, coeff, tmp
+    movsx             tmpd, word [coeffq]
+    add               tmpd, (1 << (14 - %2)) + 1
+    sar               tmpd, (15 - %2)
+    movd                m0, tmpd
+    SPLATW              m0, xm0
+    mova [coeffq+mmsize*0], m0
+    mova [coeffq+mmsize*1], m0
+    mova [coeffq+mmsize*2], m0
+    mova [coeffq+mmsize*3], m0
+%if mmsize == 16
+    mova [coeffq+mmsize*4], m0
+    mova [coeffq+mmsize*5], m0
+    mova [coeffq+mmsize*6], m0
+    mova [coeffq+mmsize*7], m0
+%endif
+    RET
+%endmacro
+
+; 8-bit
+INIT_MMX mmxext
+IDCT_DC_NL  4,      8
+IDCT_DC     8,  2,  8
+
+INIT_XMM sse2
+IDCT_DC_NL  8,      8
+IDCT_DC    16,  4,  8
+IDCT_DC    32, 16,  8
+
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+IDCT_DC    16,  2,  8
+IDCT_DC    32,  8,  8
+%endif ;HAVE_AVX2_EXTERNAL
+
+; 10-bit
+INIT_MMX mmxext
+IDCT_DC_NL  4,     10
+IDCT_DC     8,  2, 10
+
+INIT_XMM sse2
+IDCT_DC_NL  8,     10
+IDCT_DC    16,  4, 10
+IDCT_DC    32, 16, 10
+
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+IDCT_DC    16,  2, 10
+IDCT_DC    32,  8, 10
+%endif ;HAVE_AVX2_EXTERNAL
diff --git a/libavcodec/x86/hevcdsp_init.c b/libavcodec/x86/hevcdsp_init.c
index fd22fc3c1a..1a675ab64d 100644
--- a/libavcodec/x86/hevcdsp_init.c
+++ b/libavcodec/x86/hevcdsp_init.c
@@ -45,6 +45,39 @@ LFC_FUNCS(uint8_t, 10)
 LFL_FUNCS(uint8_t, 8)
 LFL_FUNCS(uint8_t, 10)
 
+#define idct_dc_proto(size, bitd, opt) \
+                void ff_hevc_idct_ ## size ## _dc_add_ ## bitd ## _ ## opt(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
+
+idct_dc_proto(4, 8,mmxext);
+idct_dc_proto(8, 8,mmxext);
+idct_dc_proto(16,8,  sse2);
+idct_dc_proto(32,8,  sse2);
+
+idct_dc_proto(32,8,  avx2);
+
+idct_dc_proto(4, 10,mmxext);
+idct_dc_proto(8, 10,  sse2);
+idct_dc_proto(16,10,  sse2);
+idct_dc_proto(32,10,  sse2);
+idct_dc_proto(8, 10,   avx);
+idct_dc_proto(16,10,   avx);
+idct_dc_proto(32,10,   avx);
+
+idct_dc_proto(16,10,  avx2);
+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)
+
+IDCT_FUNCS(4x4,   mmxext);
+IDCT_FUNCS(8x8,   mmxext);
+IDCT_FUNCS(8x8,   sse2);
+IDCT_FUNCS(16x16, sse2);
+IDCT_FUNCS(32x32, sse2);
+IDCT_FUNCS(16x16, avx2);
+IDCT_FUNCS(32x32, avx2);
+
 #define GET_PIXELS(width, depth, cf)                                                                      \
 void ff_hevc_get_pixels_ ## width ## _ ## depth ## _ ## cf(int16_t *dst, ptrdiff_t dststride,             \
                                                            uint8_t *src, ptrdiff_t srcstride,             \
@@ -229,10 +262,17 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
 #define SET_EPEL_FUNCS(v, h, depth, cf, name) SET_CHROMA_FUNCS(put_hevc_epel[v][h], name, depth, cf)
 
     if (bit_depth == 8) {
+        if (EXTERNAL_MMXEXT(cpu_flags)) {
+            c->idct_dc[0] = ff_hevc_idct_4x4_dc_8_mmxext;
+            c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_mmxext;
+        }
         if (EXTERNAL_SSE2(cpu_flags)) {
             c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
             c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_sse2;
 
+            c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_sse2;
+            c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_sse2;
+            c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_sse2;
             SET_QPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
             SET_EPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
 
@@ -246,12 +286,21 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
             SET_QPEL_FUNCS(1, 0, 8, ssse3, ff_hevc_qpel_v);
             SET_EPEL_FUNCS(0, 1, 8, ssse3, ff_hevc_epel_h);
             SET_EPEL_FUNCS(1, 0, 8, ssse3, ff_hevc_epel_v);
+
         }
     } else if (bit_depth == 10) {
+        if (EXTERNAL_MMXEXT(cpu_flags)) {
+            c->idct_dc[0] = ff_hevc_idct_4x4_dc_10_mmxext;
+            c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_mmxext;
+        }
         if (EXTERNAL_SSE2(cpu_flags)) {
             c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_sse2;
             c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_sse2;
 
+            c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_sse2;
+            c->idct_dc[2] = ff_hevc_idct_16x16_dc_10_sse2;
+            c->idct_dc[3] = ff_hevc_idct_32x32_dc_10_sse2;
+
             SET_QPEL_FUNCS(0, 0, 10, sse2, ff_hevc_get_pixels);
             SET_EPEL_FUNCS(0, 0, 10, sse2, ff_hevc_get_pixels);
 
@@ -282,6 +331,10 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
             SET_EPEL_FUNCS(1, 1, 8, avx, hevc_epel_hv);
 #endif /* HAVE_AVX_EXTERNAL */
         }
+        if (EXTERNAL_AVX2(cpu_flags)) {
+            c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_avx2;
+            c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_avx2;
+        }
     } else if (bit_depth == 10) {
         if (EXTERNAL_SSSE3(cpu_flags)) {
             c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3;
@@ -303,6 +356,10 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
             SET_EPEL_FUNCS(1, 1, 10, avx, hevc_epel_hv);
 #endif /* HAVE_AVX_EXTERNAL */
         }
+        if (EXTERNAL_AVX2(cpu_flags)) {
+            c->idct_dc[2] = ff_hevc_idct_16x16_dc_10_avx2;
+            c->idct_dc[3] = ff_hevc_idct_32x32_dc_10_avx2;
+        }
     }
 #endif /* ARCH_X86_64 */
 }