H.264: Add x86 assembly for 10-bit MC Chroma H.264 functions.

Mainly ported from 8-bit H.264 MC Chroma.

Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>

3 files changed, 306 insertions, 0 deletions

diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 1c451c8352..ea57bd1db6 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -44,6 +44,7 @@ MMX-OBJS-$(HAVE_YASM)                  += x86/dsputil_yasm.o            \
                                           x86/deinterlace.o             \
                                           x86/fmtconvert.o              \
                                           x86/h264_chromamc.o           \
+                                          x86/h264_chromamc_10bit.o     \
                                           $(YASM-OBJS-yes)
 
 MMX-OBJS-$(CONFIG_FFT)                 += x86/fft.o
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index 214c6a3945..b174b8393f 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -1938,6 +1938,19 @@ void ff_avg_h264_chroma_mc8_ssse3_rnd (uint8_t *dst, uint8_t *src,
 void ff_avg_h264_chroma_mc4_ssse3     (uint8_t *dst, uint8_t *src,
                                        int stride, int h, int x, int y);
 
+#define CHROMA_MC(OP, NUM, DEPTH, OPT) \
+void ff_ ## OP ## _h264_chroma_mc ## NUM ## _ ## DEPTH ## _ ## OPT \
+                                      (uint8_t *dst, uint8_t *src,\
+                                       int stride, int h, int x, int y);
+
+CHROMA_MC(put, 2, 10, mmxext)
+CHROMA_MC(avg, 2, 10, mmxext)
+CHROMA_MC(put, 4, 10, mmxext)
+CHROMA_MC(avg, 4, 10, mmxext)
+CHROMA_MC(put, 8, 10, sse2)
+CHROMA_MC(avg, 8, 10, sse2)
+CHROMA_MC(put, 8, 10, avx)
+CHROMA_MC(avg, 8, 10, avx)
 
 /* CAVS specific */
 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
@@ -2420,6 +2433,7 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 {
     int mm_flags = av_get_cpu_flags();
     const int high_bit_depth = avctx->codec_id == CODEC_ID_H264 && avctx->bits_per_raw_sample > 8;
+    const int bit_depth = avctx->bits_per_raw_sample;
 
     if (avctx->dsp_mask) {
         if (avctx->dsp_mask & AV_CPU_FLAG_FORCE)
@@ -2651,6 +2665,12 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_mmx2;
             c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_mmx2;
             }
+            if (bit_depth == 10) {
+                c->put_h264_chroma_pixels_tab[2]= ff_put_h264_chroma_mc2_10_mmxext;
+                c->avg_h264_chroma_pixels_tab[2]= ff_avg_h264_chroma_mc2_10_mmxext;
+                c->put_h264_chroma_pixels_tab[1]= ff_put_h264_chroma_mc4_10_mmxext;
+                c->avg_h264_chroma_pixels_tab[1]= ff_avg_h264_chroma_mc4_10_mmxext;
+            }
 
             c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
 #endif
@@ -2756,6 +2776,10 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             H264_QPEL_FUNCS(3, 2, sse2);
             H264_QPEL_FUNCS(3, 3, sse2);
             }
+            if (bit_depth == 10) {
+                c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_sse2;
+                c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_sse2;
+            }
         }
 #if HAVE_SSSE3
         if(mm_flags & AV_CPU_FLAG_SSSE3){
@@ -2854,6 +2878,14 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             }
 #endif
         }
+#if HAVE_AVX
+        if (mm_flags & AV_CPU_FLAG_AVX) {
+            if (bit_depth == 10) {
+                c->put_h264_chroma_pixels_tab[0]= ff_put_h264_chroma_mc8_10_avx;
+                c->avg_h264_chroma_pixels_tab[0]= ff_avg_h264_chroma_mc8_10_avx;
+            }
+        }
+#endif
     }
 
     if (CONFIG_ENCODERS)
diff --git a/libavcodec/x86/h264_chromamc_10bit.asm b/libavcodec/x86/h264_chromamc_10bit.asm
new file mode 100644
index 0000000000..9d075434fe
--- /dev/null
+++ b/libavcodec/x86/h264_chromamc_10bit.asm
@@ -0,0 +1,273 @@
+;*****************************************************************************
+;* MMX/SSE2/AVX-optimized 10-bit H.264 chroma MC 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
+
+cextern pw_4
+cextern pw_8
+cextern pw_32
+cextern pw_64
+
+SECTION .text
+
+
+%macro MV0_PIXELS_MC8 0
+    lea           r4, [r2*3   ]
+    lea           r5, [r2*4   ]
+.next4rows
+    movu          m0, [r1     ]
+    movu          m1, [r1+r2  ]
+    CHROMAMC_AVG  m0, [r0     ]
+    CHROMAMC_AVG  m1, [r0+r2  ]
+    mova   [r0     ], m0
+    mova   [r0+r2  ], m1
+    movu          m0, [r1+r2*2]
+    movu          m1, [r1+r4  ]
+    CHROMAMC_AVG  m0, [r0+r2*2]
+    CHROMAMC_AVG  m1, [r0+r4  ]
+    mova   [r0+r2*2], m0
+    mova   [r0+r4  ], m1
+    add           r1, r5
+    add           r0, r5
+    sub          r3d, 4
+    jne .next4rows
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void put/avg_h264_chroma_mc8(pixel *dst, pixel *src, int stride, int h, int mx, int my)
+;-----------------------------------------------------------------------------
+%macro CHROMA_MC8 2
+; put/avg_h264_chroma_mc8_*(uint8_t *dst /*align 8*/, uint8_t *src /*align 1*/,
+;                              int stride, int h, int mx, int my)
+cglobal %1_h264_chroma_mc8_10_%2, 6,7,8
+    movsxdifnidn  r2, r2d
+    mov          r6d, r5d
+    or           r6d, r4d
+    jne .at_least_one_non_zero
+    ; mx == 0 AND my == 0 - no filter needed
+    MV0_PIXELS_MC8
+    REP_RET
+
+.at_least_one_non_zero
+    mov          r6d, 2
+    test         r5d, r5d
+    je .x_interpolation
+    mov           r6, r2        ; dxy = x ? 1 : stride
+    test         r4d, r4d
+    jne .xy_interpolation
+.x_interpolation
+    ; mx == 0 XOR my == 0 - 1 dimensional filter only
+    or           r4d, r5d       ; x + y
+    movd          m5, r4d
+    mova          m4, [pw_8]
+    mova          m6, [pw_4]    ; mm6 = rnd >> 3
+    SPLATW        m5, m5        ; mm5 = B = x
+    psubw         m4, m5        ; mm4 = A = 8-x
+
+.next1drow
+    movu          m0, [r1   ]   ; mm0 = src[0..7]
+    movu          m2, [r1+r6]   ; mm2 = src[1..8]
+
+    pmullw        m0, m4        ; mm0 = A * src[0..7]
+    pmullw        m2, m5        ; mm2 = B * src[1..8]
+
+    paddw         m0, m6
+    paddw         m0, m2
+    psrlw         m0, 3
+    CHROMAMC_AVG  m0, [r0]
+    mova        [r0], m0        ; dst[0..7] = (A * src[0..7] + B * src[1..8] + (rnd >> 3)) >> 3
+
+    add           r0, r2
+    add           r1, r2
+    dec           r3d
+    jne .next1drow
+    REP_RET
+
+.xy_interpolation ; general case, bilinear
+    movd          m4, r4m         ; x
+    movd          m6, r5m         ; y
+
+    SPLATW        m4, m4          ; mm4 = x words
+    SPLATW        m6, m6          ; mm6 = y words
+    psllw         m5, m4, 3       ; mm5 = 8x
+    pmullw        m4, m6          ; mm4 = x * y
+    psllw         m6, 3           ; mm6 = 8y
+    paddw         m1, m5, m6      ; mm7 = 8x+8y
+    mova          m7, m4          ; DD = x * y
+    psubw         m5, m4          ; mm5 = B = 8x - xy
+    psubw         m6, m4          ; mm6 = C = 8y - xy
+    paddw         m4, [pw_64]
+    psubw         m4, m1          ; mm4 = A = xy - (8x+8y) + 64
+
+    movu          m0, [r1  ]      ; mm0 = src[0..7]
+    movu          m1, [r1+2]      ; mm1 = src[1..8]
+.next2drow
+    add           r1, r2
+
+    pmullw        m2, m0, m4
+    pmullw        m1, m5
+    paddw         m2, m1          ; mm2 = A * src[0..7] + B * src[1..8]
+
+    movu          m0, [r1]
+    movu          m1, [r1+2]
+    pmullw        m3, m0, m6
+    paddw         m2, m3          ; mm2 += C * src[0..7+strde]
+    pmullw        m3, m1, m7
+    paddw         m2, m3          ; mm2 += D * src[1..8+strde]
+
+    paddw         m2, [pw_32]
+    psrlw         m2, 6
+    CHROMAMC_AVG  m2, [r0]
+    mova        [r0], m2          ; dst[0..7] = (mm2 + 32) >> 6
+
+    add           r0, r2
+    dec          r3d
+    jne .next2drow
+    REP_RET
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void put/avg_h264_chroma_mc4(pixel *dst, pixel *src, int stride, int h, int mx, int my)
+;-----------------------------------------------------------------------------
+;TODO: xmm mc4
+%macro MC4_OP 2
+    movq          %1, [r1  ]
+    movq          m1, [r1+2]
+    add           r1, r2
+    pmullw        %1, m4
+    pmullw        m1, m2
+    paddw         m1, %1
+    mova          %1, m1
+
+    pmullw        %2, m5
+    pmullw        m1, m3
+    paddw         %2, [pw_32]
+    paddw         m1, %2
+    psrlw         m1, 6
+    CHROMAMC_AVG  m1, %2, [r0]
+    movq        [r0], m1
+    add           r0, r2
+%endmacro
+
+%macro CHROMA_MC4 2
+cglobal %1_h264_chroma_mc4_10_%2, 6,6,7
+    movsxdifnidn  r2, r2d
+    movd          m2, r4m         ; x
+    movd          m3, r5m         ; y
+    mova          m4, [pw_8]
+    mova          m5, m4
+    SPLATW        m2, m2
+    SPLATW        m3, m3
+    psubw         m4, m2
+    psubw         m5, m3
+
+    movq          m0, [r1  ]
+    movq          m6, [r1+2]
+    add           r1, r2
+    pmullw        m0, m4
+    pmullw        m6, m2
+    paddw         m6, m0
+
+.next2rows
+    MC4_OP m0, m6
+    MC4_OP m6, m0
+    sub   r3d, 2
+    jnz .next2rows
+    REP_RET
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void put/avg_h264_chroma_mc2(pixel *dst, pixel *src, int stride, int h, int mx, int my)
+;-----------------------------------------------------------------------------
+%macro CHROMA_MC2 2
+cglobal %1_h264_chroma_mc2_10_%2, 6,7
+    movsxdifnidn  r2, r2d
+    mov          r6d, r4d
+    shl          r4d, 16
+    sub          r4d, r6d
+    add          r4d, 8
+    imul         r5d, r4d         ; x*y<<16 | y*(8-x)
+    shl          r4d, 3
+    sub          r4d, r5d         ; x*(8-y)<<16 | (8-x)*(8-y)
+
+    movd          m5, r4d
+    movd          m6, r5d
+    punpckldq     m5, m5          ; mm5 = {A,B,A,B}
+    punpckldq     m6, m6          ; mm6 = {C,D,C,D}
+    pxor          m7, m7
+    pshufw        m2, [r1], 0x94    ; mm0 = src[0,1,1,2]
+
+.nextrow
+    add           r1, r2
+    movq          m1, m2
+    pmaddwd       m1, m5          ; mm1 = A * src[0,1] + B * src[1,2]
+    pshufw        m0, [r1], 0x94    ; mm0 = src[0,1,1,2]
+    movq          m2, m0
+    pmaddwd       m0, m6
+    paddw         m1, [pw_32]
+    paddw         m1, m0          ; mm1 += C * src[0,1] + D * src[1,2]
+    psrlw         m1, 6
+    packssdw      m1, m7
+    CHROMAMC_AVG  m1, m3, [r0]
+    movd        [r0], m1
+    add           r0, r2
+    dec          r3d
+    jnz .nextrow
+    REP_RET
+%endmacro
+
+%macro NOTHING 2-3
+%endmacro
+%macro AVG 2-3
+%if %0==3
+    movq          %2, %3
+%endif
+    PAVG          %1, %2
+%endmacro
+
+%define CHROMAMC_AVG  NOTHING
+INIT_XMM
+CHROMA_MC8 put, sse2
+%ifdef HAVE_AVX
+INIT_AVX
+CHROMA_MC8 put, avx
+%endif
+INIT_MMX
+CHROMA_MC4 put, mmxext
+CHROMA_MC2 put, mmxext
+
+%define CHROMAMC_AVG  AVG
+%define PAVG          pavgw
+INIT_XMM
+CHROMA_MC8 avg, sse2
+%ifdef HAVE_AVX
+INIT_AVX
+CHROMA_MC8 avg, avx
+%endif
+INIT_MMX
+CHROMA_MC4 avg, mmxext
+CHROMA_MC2 avg, mmxext