[FFmpeg-cvslog] x86: hevc: adding transform_add

Pierre Edouard Lepere git at videolan.org
Wed Aug 20 01:37:54 CEST 2014


ffmpeg | branch: master | Pierre Edouard Lepere <Pierre-Edouard.Lepere at insa-rennes.fr> | Mon Aug 18 10:01:09 2014 +0200| [a6af4bf64dae46356a5f91537a1c8c5f86456b37] | committer: Michael Niedermayer

x86: hevc: adding transform_add

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

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=a6af4bf64dae46356a5f91537a1c8c5f86456b37
---

 libavcodec/x86/Makefile         |    3 +-
 libavcodec/x86/hevc_res_add.asm |  384 +++++++++++++++++++++++++++++++++++++++
 libavcodec/x86/hevcdsp.h        |   16 ++
 libavcodec/x86/hevcdsp_init.c   |   16 +-
 4 files changed, 417 insertions(+), 2 deletions(-)

diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 7ceb7e2..fbb9abf 100644
--- a/libavcodec/x86/Makefile
+++ b/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
diff --git a/libavcodec/x86/hevc_res_add.asm b/libavcodec/x86/hevc_res_add.asm
new file mode 100644
index 0000000..20dae1c
--- /dev/null
+++ b/libavcodec/x86/hevc_res_add.asm
@@ -0,0 +1,384 @@
+; /*
+; * Provide SIMD optimizations for transform_add functions for HEVC decoding
+; * Copyright (c) 2014 Pierre-Edouard LEPERE
+; *
+; * This file is part of FFmpeg.
+; *
+; * FFmpeg 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.
+; *
+; * FFmpeg 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 FFmpeg; 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_RODATA 32
+max_pixels_10:          times 16  dw ((1 << 10)-1)
+
+
+SECTION .text
+
+;the tr_add macros and functions were largely inspired by x264 project's code in the h264_idct.asm file
+%macro TR_ADD_MMX_4_8 0
+    mova              m2, [r1]
+    mova              m4, [r1+8]
+    pxor              m3, m3
+    psubw             m3, m2
+    packuswb          m2, m2
+    packuswb          m3, m3
+    pxor              m5, m5
+    psubw             m5, m4
+    packuswb          m4, m4
+    packuswb          m5, m5
+
+    movh              m0, [r0     ]
+    movh              m1, [r0+r2  ]
+    paddusb           m0, m2
+    paddusb           m1, m4
+    psubusb           m0, m3
+    psubusb           m1, m5
+    movh       [r0     ], m0
+    movh       [r0+r2  ], m1
+%endmacro
+
+
+INIT_MMX mmxext
+; void ff_hevc_tranform_add_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
+cglobal hevc_transform_add4_8, 3, 4, 6
+    TR_ADD_MMX_4_8
+    add               r1, 16
+    lea               r0, [r0+r2*2]
+    TR_ADD_MMX_4_8
+    RET
+
+%macro TR_ADD_SSE_8_8 0
+    pxor              m3, m3
+    mova              m4, [r1]
+    mova              m6, [r1+16]
+    mova              m0, [r1+32]
+    mova              m2, [r1+48]
+    psubw             m5, m3, m4
+    psubw             m7, m3, m6
+    psubw             m1, m3, m0
+    packuswb          m4, m0
+    packuswb          m5, m1
+    psubw             m3, m2
+    packuswb          m6, m2
+    packuswb          m7, m3
+
+    movq                m0, [r0     ]
+    movq                m1, [r0+r2  ]
+    movhps              m0, [r0+r2*2]
+    movhps              m1, [r0+r3  ]
+    paddusb             m0, m4
+    paddusb             m1, m6
+    psubusb             m0, m5
+    psubusb             m1, m7
+    movq         [r0     ], m0
+    movq         [r0+r2  ], m1
+    movhps       [r0+2*r2], m0
+    movhps       [r0+r3  ], m1
+%endmacro
+
+%macro TR_ADD_INIT_SSE_8 0
+    pxor              m0, m0
+
+    mova              m4, [r1]
+    mova              m1, [r1+16]
+    psubw             m2, m0, m1
+    psubw             m5, m0, m4
+    packuswb          m4, m1
+    packuswb          m5, m2
+
+    mova              m6, [r1+32]
+    mova              m1, [r1+48]
+    psubw             m2, m0, m1
+    psubw             m7, m0, m6
+    packuswb          m6, m1
+    packuswb          m7, m2
+
+    mova              m8, [r1+64]
+    mova              m1, [r1+80]
+    psubw             m2, m0, m1
+    psubw             m9, m0, m8
+    packuswb          m8, m1
+    packuswb          m9, m2
+
+    mova             m10, [r1+96]
+    mova              m1, [r1+112]
+    psubw             m2, m0, m1
+    psubw            m11, m0, m10
+    packuswb         m10, m1
+    packuswb         m11, m2
+%endmacro
+
+
+%macro TR_ADD_SSE_16_8 0
+    TR_ADD_INIT_SSE_8
+
+    paddusb           m0, m4, [r0     ]
+    paddusb           m1, m6, [r0+r2  ]
+    paddusb           m2, m8, [r0+r2*2]
+    paddusb           m3, m10,[r0+r3  ]
+    psubusb           m0, m5
+    psubusb           m1, m7
+    psubusb           m2, m9
+    psubusb           m3, m11
+    mova       [r0     ], m0
+    mova       [r0+r2  ], m1
+    mova       [r0+2*r2], m2
+    mova       [r0+r3  ], m3
+%endmacro
+
+%macro TR_ADD_SSE_32_8 0
+    TR_ADD_INIT_SSE_8
+
+    paddusb           m0, m4, [r0      ]
+    paddusb           m1, m6, [r0+16   ]
+    paddusb           m2, m8, [r0+r2   ]
+    paddusb           m3, m10,[r0+r2+16]
+    psubusb           m0, m5
+    psubusb           m1, m7
+    psubusb           m2, m9
+    psubusb           m3, m11
+    mova      [r0      ], m0
+    mova      [r0+16   ], m1
+    mova      [r0+r2   ], m2
+    mova      [r0+r2+16], m3
+%endmacro
+
+
+INIT_XMM sse2
+; void ff_hevc_transform_add8_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
+cglobal hevc_transform_add8_8, 3, 4, 8
+    lea               r3, [r2*3]
+    TR_ADD_SSE_8_8
+    add               r1, 64
+    lea               r0, [r0+r2*4]
+    TR_ADD_SSE_8_8
+    RET
+
+%if ARCH_X86_64
+; void ff_hevc_transform_add16_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
+cglobal hevc_transform_add16_8, 3, 4, 12
+    lea               r3, [r2*3]
+    TR_ADD_SSE_16_8
+%rep 3
+    add                r1, 128
+    lea                r0, [r0+r2*4]
+    TR_ADD_SSE_16_8
+%endrep
+    RET
+
+; void ff_hevc_transform_add16_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
+cglobal hevc_transform_add32_8, 3, 4, 12
+
+    TR_ADD_SSE_32_8
+%rep 15
+    add                r1, 128
+    lea                r0, [r0+r2*2]
+    TR_ADD_SSE_32_8
+%endrep
+    RET
+
+%endif ;ARCH_X86_64
+;-----------------------------------------------------------------------------
+; void ff_hevc_transform_add_10(pixel *dst, int16_t *block, int stride)
+;-----------------------------------------------------------------------------
+%macro TR_ADD_SSE_8_10 4
+    mova              m0, [%4]
+    mova              m1, [%4+16]
+    mova              m2, [%4+32]
+    mova              m3, [%4+48]
+    paddw             m0, [%1+0   ]
+    paddw             m1, [%1+%2  ]
+    paddw             m2, [%1+%2*2]
+    paddw             m3, [%1+%3  ]
+    CLIPW             m0, m4, m5
+    CLIPW             m1, m4, m5
+    CLIPW             m2, m4, m5
+    CLIPW             m3, m4, m5
+    mova       [%1+0   ], m0
+    mova       [%1+%2  ], m1
+    mova       [%1+%2*2], m2
+    mova       [%1+%3  ], m3
+%endmacro
+
+%macro TR_ADD_MMX4_10 3
+    mova              m0, [%1+0   ]
+    mova              m1, [%1+%2  ]
+    paddw             m0, [%3]
+    paddw             m1, [%3+8]
+    CLIPW             m0, m2, m3
+    CLIPW             m1, m2, m3
+    mova       [%1+0   ], m0
+    mova       [%1+%2  ], m1
+%endmacro
+
+%macro TRANS_ADD_SSE_16_10 3
+    mova              m0, [%3]
+    mova              m1, [%3+16]
+    mova              m2, [%3+32]
+    mova              m3, [%3+48]
+    paddw             m0, [%1      ]
+    paddw             m1, [%1+16   ]
+    paddw             m2, [%1+%2   ]
+    paddw             m3, [%1+%2+16]
+    CLIPW             m0, m4, m5
+    CLIPW             m1, m4, m5
+    CLIPW             m2, m4, m5
+    CLIPW             m3, m4, m5
+    mova      [%1      ], m0
+    mova      [%1+16   ], m1
+    mova      [%1+%2   ], m2
+    mova      [%1+%2+16], m3
+%endmacro
+
+%macro TRANS_ADD_SSE_32_10 2
+    mova              m0, [%2]
+    mova              m1, [%2+16]
+    mova              m2, [%2+32]
+    mova              m3, [%2+48]
+
+    paddw             m0, [%1   ]
+    paddw             m1, [%1+16]
+    paddw             m2, [%1+32]
+    paddw             m3, [%1+48]
+    CLIPW             m0, m4, m5
+    CLIPW             m1, m4, m5
+    CLIPW             m2, m4, m5
+    CLIPW             m3, m4, m5
+    mova         [%1   ], m0
+    mova         [%1+16], m1
+    mova         [%1+32], m2
+    mova         [%1+48], m3
+%endmacro
+
+%macro TRANS_ADD16_AVX2 4
+    mova              m0, [%4]
+    mova              m1, [%4+32]
+    mova              m2, [%4+64]
+    mova              m3, [%4+96]
+
+    paddw             m0, [%1+0   ]
+    paddw             m1, [%1+%2  ]
+    paddw             m2, [%1+%2*2]
+    paddw             m3, [%1+%3  ]
+
+    CLIPW             m0, m4, m5
+    CLIPW             m1, m4, m5
+    CLIPW             m2, m4, m5
+    CLIPW             m3, m4, m5
+    mova       [%1+0   ], m0
+    mova       [%1+%2  ], m1
+    mova       [%1+%2*2], m2
+    mova       [%1+%3  ], m3
+%endmacro
+
+%macro TRANS_ADD32_AVX2 3
+    mova              m0, [%3]
+    mova              m1, [%3+32]
+    mova              m2, [%3+64]
+    mova              m3, [%3+96]
+
+    paddw             m0, [%1      ]
+    paddw             m1, [%1+32   ]
+    paddw             m2, [%1+%2   ]
+    paddw             m3, [%1+%2+32]
+
+    CLIPW             m0, m4, m5
+    CLIPW             m1, m4, m5
+    CLIPW             m2, m4, m5
+    CLIPW             m3, m4, m5
+    mova      [%1      ], m0
+    mova      [%1+32   ], m1
+    mova      [%1+%2   ], m2
+    mova      [%1+%2+32], m3
+%endmacro
+
+
+INIT_MMX mmxext
+cglobal hevc_transform_add4_10,3,4, 6
+    pxor              m2, m2
+    mova              m3, [max_pixels_10]
+    TR_ADD_MMX4_10     r0, r2, r1
+    add               r1, 16
+    lea               r0, [r0+2*r2]
+    TR_ADD_MMX4_10     r0, r2, r1
+    RET
+
+;-----------------------------------------------------------------------------
+; void ff_hevc_transform_add_10(pixel *dst, int16_t *block, int stride)
+;-----------------------------------------------------------------------------
+INIT_XMM sse2
+cglobal hevc_transform_add8_10,3,4,6
+    pxor              m4, m4
+    mova              m5, [max_pixels_10]
+    lea               r3, [r2*3]
+
+    TR_ADD_SSE_8_10      r0, r2, r3, r1
+    lea               r0, [r0+r2*4]
+    add               r1, 64
+    TR_ADD_SSE_8_10      r0, r2, r3, r1
+    RET
+
+cglobal hevc_transform_add16_10,3,4,6
+    pxor              m4, m4
+    mova              m5, [max_pixels_10]
+
+    TRANS_ADD_SSE_16_10 r0, r2, r1
+%rep 7
+    lea                 r0, [r0+r2*2]
+    add                 r1, 64
+    TRANS_ADD_SSE_16_10 r0, r2, r1
+%endrep
+    RET
+
+cglobal hevc_transform_add32_10,3,4,6
+    pxor              m4, m4
+    mova              m5, [max_pixels_10]
+
+    TRANS_ADD_SSE_32_10 r0, r1
+%rep 31
+    lea                 r0, [r0+r2]
+    add                 r1, 64
+    TRANS_ADD_SSE_32_10 r0, r1
+%endrep
+    RET
+
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+
+cglobal hevc_transform_add16_10,3,4,10
+    pxor              m4, m4
+    mova              m5, [max_pixels_10]
+    lea               r3, [r2*3]
+
+    TRANS_ADD16_AVX2  r0, r2, r3, r1
+%rep 3
+    lea               r0, [r0+r2*4]
+    add               r1, 128
+    TRANS_ADD16_AVX2  r0, r2, r3, r1
+%endrep
+    RET
+
+cglobal hevc_transform_add32_10,3,4,10
+    pxor              m4, m4
+    mova              m5, [max_pixels_10]
+
+    TRANS_ADD32_AVX2  r0, r2, r1
+%rep 15
+    lea               r0, [r0+r2*2]
+    add               r1, 128
+    TRANS_ADD32_AVX2  r0, r2, r1
+%endrep
+    RET
+%endif ;HAVE_AVX_EXTERNAL
diff --git a/libavcodec/x86/hevcdsp.h b/libavcodec/x86/hevcdsp.h
index 4bcc8dc..7ced22c 100644
--- a/libavcodec/x86/hevcdsp.h
+++ b/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
diff --git a/libavcodec/x86/hevcdsp_init.c b/libavcodec/x86/hevcdsp_init.c
index 828c081..0f9fe7d 100644
--- a/libavcodec/x86/hevcdsp_init.c
+++ b/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)) {



More information about the ffmpeg-cvslog mailing list