h264: Add x86 assembly for 10-bit weight/biweight H.264 functions.
authorDaniel Kang <daniel.d.kang@gmail.com>
Tue, 21 Jun 2011 04:30:23 +0000 (00:30 -0400)
committerDiego Biurrun <diego@biurrun.de>
Tue, 21 Jun 2011 13:24:13 +0000 (15:24 +0200)
Mainly ported from 8-bit H.264 weight/biweight.

Signed-off-by: Diego Biurrun <diego@biurrun.de>
libavcodec/x86/Makefile
libavcodec/x86/h264_weight_10bit.asm [new file with mode: 0644]
libavcodec/x86/h264dsp_mmx.c

index ea57bd1..022ab27 100644 (file)
@@ -15,6 +15,7 @@ YASM-OBJS-$(CONFIG_H264DSP)            += x86/h264_deblock.o            \
                                           x86/h264_idct.o               \
                                           x86/h264_idct_10bit.o         \
                                           x86/h264_weight.o             \
+                                          x86/h264_weight_10bit.o       \
 
 YASM-OBJS-$(CONFIG_H264PRED)           += x86/h264_intrapred.o          \
                                           x86/h264_intrapred_10bit.o
diff --git a/libavcodec/x86/h264_weight_10bit.asm b/libavcodec/x86/h264_weight_10bit.asm
new file mode 100644 (file)
index 0000000..1c58d72
--- /dev/null
@@ -0,0 +1,321 @@
+;*****************************************************************************
+;* MMX/SSE2/AVX-optimized 10-bit H.264 weighted prediction code
+;*****************************************************************************
+;* Copyright (C) 2005-2011 x264 project
+;*
+;* Authors: Daniel Kang <daniel.d.kang@gmail.com>
+;*
+;* 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 "x86inc.asm"
+%include "x86util.asm"
+
+SECTION_RODATA 32
+
+pw_pixel_max: times 8 dw ((1 << 10)-1)
+sq_1: dq 1
+      dq 0
+
+cextern pw_1
+
+SECTION .text
+
+;-----------------------------------------------------------------------------
+; void h264_weight(uint8_t *dst, int stride, int log2_denom,
+;                  int weight, int offset);
+;-----------------------------------------------------------------------------
+%ifdef ARCH_X86_32
+DECLARE_REG_TMP 2
+%else
+DECLARE_REG_TMP 10
+%endif
+
+%macro WEIGHT_PROLOGUE 1
+    mov t0, %1
+.prologue
+    PROLOGUE 0,5,8
+    movifnidn  r0, r0mp
+    movifnidn r1d, r1m
+    movifnidn r3d, r3m
+    movifnidn r4d, r4m
+%endmacro
+
+%macro WEIGHT_SETUP 1
+    mova       m0, [pw_1]
+    movd       m2, r2m
+    pslld      m0, m2       ; 1<<log2_denom
+    SPLATW     m0, m0
+    shl        r4, 19       ; *8, move to upper half of dword
+    lea        r4, [r4+r3*2+0x10000]
+    movd       m3, r4d      ; weight<<1 | 1+(offset<<(3))
+    pshufd     m3, m3, 0
+    mova       m4, [pw_pixel_max]
+    paddw      m2, [sq_1]   ; log2_denom+1
+%ifnidn %1, sse4
+    pxor       m7, m7
+%endif
+%endmacro
+
+%macro WEIGHT_OP 2-3
+%if %0==2
+    mova        m5, [r0+%2]
+    punpckhwd   m6, m5, m0
+    punpcklwd   m5, m0
+%else
+    movq        m5, [r0+%2]
+    movq        m6, [r0+%3]
+    punpcklwd   m5, m0
+    punpcklwd   m6, m0
+%endif
+    pmaddwd     m5, m3
+    pmaddwd     m6, m3
+    psrad       m5, m2
+    psrad       m6, m2
+%ifidn %1, sse4
+    packusdw    m5, m6
+    pminsw      m5, m4
+%else
+    packssdw    m5, m6
+    CLIPW       m5, m7, m4
+%endif
+%endmacro
+
+%macro WEIGHT_FUNC_DBL 1
+cglobal h264_weight_16x16_10_%1
+    WEIGHT_PROLOGUE 16
+    WEIGHT_SETUP %1
+.nextrow
+    WEIGHT_OP %1,  0
+    mova [r0   ], m5
+    WEIGHT_OP %1, 16
+    mova [r0+16], m5
+    add       r0, r1
+    dec       t0
+    jnz .nextrow
+    REP_RET
+
+cglobal h264_weight_16x8_10_%1
+    mov t0, 8
+    jmp mangle(ff_h264_weight_16x16_10_%1.prologue)
+%endmacro
+
+INIT_XMM
+WEIGHT_FUNC_DBL sse2
+WEIGHT_FUNC_DBL sse4
+
+
+%macro WEIGHT_FUNC_MM 1
+cglobal h264_weight_8x16_10_%1
+    WEIGHT_PROLOGUE 16
+    WEIGHT_SETUP %1
+.nextrow
+    WEIGHT_OP  %1, 0
+    mova     [r0], m5
+    add        r0, r1
+    dec        t0
+    jnz .nextrow
+    REP_RET
+
+cglobal h264_weight_8x8_10_%1
+    mov t0, 8
+    jmp mangle(ff_h264_weight_8x16_10_%1.prologue)
+
+cglobal h264_weight_8x4_10_%1
+    mov t0, 4
+    jmp mangle(ff_h264_weight_8x16_10_%1.prologue)
+%endmacro
+
+INIT_XMM
+WEIGHT_FUNC_MM sse2
+WEIGHT_FUNC_MM sse4
+
+
+%macro WEIGHT_FUNC_HALF_MM 1
+cglobal h264_weight_4x8_10_%1
+    WEIGHT_PROLOGUE 4
+    WEIGHT_SETUP %1
+    lea         r3, [r1*2]
+.nextrow
+    WEIGHT_OP   %1, 0, r1
+    movh      [r0], m5
+    movhps [r0+r1], m5
+    add         r0, r3
+    dec         t0
+    jnz .nextrow
+    REP_RET
+
+cglobal h264_weight_4x4_10_%1
+    mov t0, 2
+    jmp mangle(ff_h264_weight_4x8_10_%1.prologue)
+
+cglobal h264_weight_4x2_10_%1
+    mov t0, 1
+    jmp mangle(ff_h264_weight_4x8_10_%1.prologue)
+%endmacro
+
+INIT_XMM
+WEIGHT_FUNC_HALF_MM sse2
+WEIGHT_FUNC_HALF_MM sse4
+
+
+;-----------------------------------------------------------------------------
+; void h264_biweight(uint8_t *dst, uint8_t *src, int stride, int log2_denom,
+;                    int weightd, int weights, int offset);
+;-----------------------------------------------------------------------------
+%ifdef ARCH_X86_32
+DECLARE_REG_TMP 2,3
+%else
+DECLARE_REG_TMP 10,2
+%endif
+
+%macro BIWEIGHT_PROLOGUE 1
+    mov t0, %1
+.prologue
+    PROLOGUE 0,7,8
+    movifnidn  r0, r0mp
+    movifnidn  r1, r1mp
+    movifnidn t1d, r2m
+    movifnidn r4d, r4m
+    movifnidn r5d, r5m
+    movifnidn r6d, r6m
+%endmacro
+
+%macro BIWEIGHT_SETUP 1
+    lea        r6, [r6*4+1] ; (offset<<2)+1
+    or         r6, 1
+    shl        r5, 16
+    or         r4, r5
+    movd       m4, r4d      ; weightd | weights
+    movd       m5, r6d      ; (offset+1)|1
+    movd       m6, r3m      ; log2_denom
+    pslld      m5, m6       ; (((offset<<2)+1)|1)<<log2_denom
+    paddd      m6, [sq_1]
+    pshufd     m4, m4, 0
+    pshufd     m5, m5, 0
+    mova       m3, [pw_pixel_max]
+%ifnidn %1, sse4
+    pxor       m7, m7
+%endif
+%endmacro
+
+%macro BIWEIGHT 2-3
+%if %0==2
+    mova       m0, [r0+%2]
+    mova       m1, [r1+%2]
+    punpckhwd  m2, m0, m1
+    punpcklwd  m0, m1
+%else
+    movq       m0, [r0+%2]
+    movq       m1, [r1+%2]
+    punpcklwd  m0, m1
+    movq       m2, [r0+%3]
+    movq       m1, [r1+%3]
+    punpcklwd  m2, m1
+%endif
+    pmaddwd    m0, m4
+    pmaddwd    m2, m4
+    paddd      m0, m5
+    paddd      m2, m5
+    psrad      m0, m6
+    psrad      m2, m6
+%ifidn %1, sse4
+    packusdw   m0, m2
+    pminsw     m0, m3
+%else
+    packssdw   m0, m2
+    CLIPW      m0, m7, m3
+%endif
+%endmacro
+
+%macro BIWEIGHT_FUNC_DBL 1
+cglobal h264_biweight_16x16_10_%1
+    BIWEIGHT_PROLOGUE 16
+    BIWEIGHT_SETUP %1
+.nextrow
+    BIWEIGHT  %1,  0
+    mova [r0   ], m0
+    BIWEIGHT  %1, 16
+    mova [r0+16], m0
+    add       r0, t1
+    add       r1, t1
+    dec       t0
+    jnz .nextrow
+    REP_RET
+
+cglobal h264_biweight_16x8_10_%1
+    mov t0, 8
+    jmp mangle(ff_h264_biweight_16x16_10_%1.prologue)
+%endmacro
+
+INIT_XMM
+BIWEIGHT_FUNC_DBL sse2
+BIWEIGHT_FUNC_DBL sse4
+
+%macro BIWEIGHT_FUNC 1
+cglobal h264_biweight_8x16_10_%1
+    BIWEIGHT_PROLOGUE 16
+    BIWEIGHT_SETUP %1
+.nextrow
+    BIWEIGHT %1, 0
+    mova   [r0], m0
+    add      r0, t1
+    add      r1, t1
+    dec      t0
+    jnz .nextrow
+    REP_RET
+
+cglobal h264_biweight_8x8_10_%1
+    mov t0, 8
+    jmp mangle(ff_h264_biweight_8x16_10_%1.prologue)
+
+cglobal h264_biweight_8x4_10_%1
+    mov t0, 4
+    jmp mangle(ff_h264_biweight_8x16_10_%1.prologue)
+%endmacro
+
+INIT_XMM
+BIWEIGHT_FUNC sse2
+BIWEIGHT_FUNC sse4
+
+%macro BIWEIGHT_FUNC_HALF 1
+cglobal h264_biweight_4x8_10_%1
+    BIWEIGHT_PROLOGUE 4
+    BIWEIGHT_SETUP %1
+    lea        r4, [t1*2]
+.nextrow
+    BIWEIGHT    %1, 0, t1
+    movh   [r0   ], m0
+    movhps [r0+t1], m0
+    add         r0, r4
+    add         r1, r4
+    dec         t0
+    jnz .nextrow
+    REP_RET
+
+cglobal h264_biweight_4x4_10_%1
+    mov t0, 2
+    jmp mangle(ff_h264_biweight_4x8_10_%1.prologue)
+
+cglobal h264_biweight_4x2_10_%1
+    mov t0, 1
+    jmp mangle(ff_h264_biweight_4x8_10_%1.prologue)
+%endmacro
+
+INIT_XMM
+BIWEIGHT_FUNC_HALF sse2
+BIWEIGHT_FUNC_HALF sse4
index 3fccd08..1042552 100644 (file)
@@ -326,6 +326,30 @@ H264_BIWEIGHT_MMX    ( 4,  8)
 H264_BIWEIGHT_MMX    ( 4,  4)
 H264_BIWEIGHT_MMX    ( 4,  2)
 
+#define H264_WEIGHT_10(W, H, DEPTH, OPT) \
+void ff_h264_weight_ ## W ## x ## H ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
+    int stride, int log2_denom, int weight, int offset);
+
+#define H264_BIWEIGHT_10(W, H, DEPTH, OPT) \
+void ff_h264_biweight_ ## W ## x ## H ## _ ## DEPTH ## _ ## OPT \
+    (uint8_t *dst, uint8_t *src, int stride, int log2_denom, \
+     int weightd, int weights, int offset);
+
+#define H264_BIWEIGHT_10_SSE(W, H, DEPTH) \
+H264_WEIGHT_10  (W, H, DEPTH, sse2) \
+H264_WEIGHT_10  (W, H, DEPTH, sse4) \
+H264_BIWEIGHT_10(W, H, DEPTH, sse2) \
+H264_BIWEIGHT_10(W, H, DEPTH, sse4)
+
+H264_BIWEIGHT_10_SSE(16, 16, 10)
+H264_BIWEIGHT_10_SSE(16,  8, 10)
+H264_BIWEIGHT_10_SSE( 8, 16, 10)
+H264_BIWEIGHT_10_SSE( 8,  8, 10)
+H264_BIWEIGHT_10_SSE( 8,  4, 10)
+H264_BIWEIGHT_10_SSE( 4,  8, 10)
+H264_BIWEIGHT_10_SSE( 4,  4, 10)
+H264_BIWEIGHT_10_SSE( 4,  2, 10)
+
 void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
 {
     int mm_flags = av_get_cpu_flags();
@@ -454,6 +478,24 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
                 c->h264_idct8_add4     = ff_h264_idct8_add4_10_sse2;
 #endif
 
+                c->weight_h264_pixels_tab[0] = ff_h264_weight_16x16_10_sse2;
+                c->weight_h264_pixels_tab[1] = ff_h264_weight_16x8_10_sse2;
+                c->weight_h264_pixels_tab[2] = ff_h264_weight_8x16_10_sse2;
+                c->weight_h264_pixels_tab[3] = ff_h264_weight_8x8_10_sse2;
+                c->weight_h264_pixels_tab[4] = ff_h264_weight_8x4_10_sse2;
+                c->weight_h264_pixels_tab[5] = ff_h264_weight_4x8_10_sse2;
+                c->weight_h264_pixels_tab[6] = ff_h264_weight_4x4_10_sse2;
+                c->weight_h264_pixels_tab[7] = ff_h264_weight_4x2_10_sse2;
+
+                c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16x16_10_sse2;
+                c->biweight_h264_pixels_tab[1] = ff_h264_biweight_16x8_10_sse2;
+                c->biweight_h264_pixels_tab[2] = ff_h264_biweight_8x16_10_sse2;
+                c->biweight_h264_pixels_tab[3] = ff_h264_biweight_8x8_10_sse2;
+                c->biweight_h264_pixels_tab[4] = ff_h264_biweight_8x4_10_sse2;
+                c->biweight_h264_pixels_tab[5] = ff_h264_biweight_4x8_10_sse2;
+                c->biweight_h264_pixels_tab[6] = ff_h264_biweight_4x4_10_sse2;
+                c->biweight_h264_pixels_tab[7] = ff_h264_biweight_4x2_10_sse2;
+
                 c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_sse2;
                 c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_sse2;
 #if HAVE_ALIGNED_STACK
@@ -463,6 +505,25 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
 #endif
             }
+            if (mm_flags&AV_CPU_FLAG_SSE4) {
+                c->weight_h264_pixels_tab[0] = ff_h264_weight_16x16_10_sse4;
+                c->weight_h264_pixels_tab[1] = ff_h264_weight_16x8_10_sse4;
+                c->weight_h264_pixels_tab[2] = ff_h264_weight_8x16_10_sse4;
+                c->weight_h264_pixels_tab[3] = ff_h264_weight_8x8_10_sse4;
+                c->weight_h264_pixels_tab[4] = ff_h264_weight_8x4_10_sse4;
+                c->weight_h264_pixels_tab[5] = ff_h264_weight_4x8_10_sse4;
+                c->weight_h264_pixels_tab[6] = ff_h264_weight_4x4_10_sse4;
+                c->weight_h264_pixels_tab[7] = ff_h264_weight_4x2_10_sse4;
+
+                c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16x16_10_sse4;
+                c->biweight_h264_pixels_tab[1] = ff_h264_biweight_16x8_10_sse4;
+                c->biweight_h264_pixels_tab[2] = ff_h264_biweight_8x16_10_sse4;
+                c->biweight_h264_pixels_tab[3] = ff_h264_biweight_8x8_10_sse4;
+                c->biweight_h264_pixels_tab[4] = ff_h264_biweight_8x4_10_sse4;
+                c->biweight_h264_pixels_tab[5] = ff_h264_biweight_4x8_10_sse4;
+                c->biweight_h264_pixels_tab[6] = ff_h264_biweight_4x4_10_sse4;
+                c->biweight_h264_pixels_tab[7] = ff_h264_biweight_4x2_10_sse4;
+            }
 #if HAVE_AVX
             if (mm_flags&AV_CPU_FLAG_AVX) {
                 c->h264_idct_dc_add    =