10-bit H.264 x86 chroma v loopfilter asm

Also delete some unused deblock asm macros.

3 files changed, 121 insertions, 42 deletions

diff --git a/libavcodec/x86/h264_deblock.asm b/libavcodec/x86/h264_deblock.asm
index 37866812e7..0cf013f58f 100644
--- a/libavcodec/x86/h264_deblock.asm
+++ b/libavcodec/x86/h264_deblock.asm
@@ -106,47 +106,6 @@ cextern pb_A1
     TRANSPOSE4x8_LOAD bw, wd, dq, %1, %2, %3, %4, %5, %6, %7, %8
 %endmacro
 
-%macro TRANSPOSE4x8W_LOAD 8
-%if mmsize==16
-    TRANSPOSE4x8_LOAD wd, dq, qdq, %1, %2, %3, %4, %5, %6, %7, %8
-%else
-    SWAP  1, 4, 2, 3
-    mova  m0, [t5]
-    mova  m1, [t5+r1]
-    mova  m2, [t5+r1*2]
-    mova  m3, [t5+t6]
-    TRANSPOSE4x4W 0, 1, 2, 3, 4
-%endif
-%endmacro
-
-%macro TRANSPOSE8x2W_STORE 8
-    punpckhwd  m0, m1, m2
-    punpcklwd  m1, m2
-%if mmsize==8
-    movd       %3, m0
-    movd       %1, m1
-    psrlq      m1, 32
-    psrlq      m0, 32
-    movd       %2, m1
-    movd       %4, m0
-%else
-    movd       %5, m0
-    movd       %1, m1
-    psrldq     m1, 4
-    psrldq     m0, 4
-    movd       %2, m1
-    movd       %6, m0
-    psrldq     m1, 4
-    psrldq     m0, 4
-    movd       %3, m1
-    movd       %7, m0
-    psrldq     m1, 4
-    psrldq     m0, 4
-    movd       %4, m1
-    movd       %8, m0
-%endif
-%endmacro
-
 %macro SBUTTERFLY3 4
     punpckh%1  %4, %2, %3
     punpckl%1  %2, %3
diff --git a/libavcodec/x86/h264_deblock_10bit.asm b/libavcodec/x86/h264_deblock_10bit.asm
index 402ed9bfac..c253d02954 100644
--- a/libavcodec/x86/h264_deblock_10bit.asm
+++ b/libavcodec/x86/h264_deblock_10bit.asm
@@ -34,6 +34,7 @@ pw_pixel_max: times 8 dw ((1 << 10)-1)
 SECTION .text
 
 cextern pw_2
+cextern pw_3
 cextern pw_4
 
 ; out: %4 = |%1-%2|-%3
@@ -802,3 +803,108 @@ INIT_AVX
 DEBLOCK_LUMA avx
 DEBLOCK_LUMA_INTRA avx
 %endif
+
+; in: %1=p0, %2=q0, %3=p1, %4=q1, %5=mask, %6=tmp, %7=tmp
+; out: %1=p0', %2=q0'
+%macro CHROMA_DEBLOCK_P0_Q0_INTRA 7
+    mova    %6, [pw_2]
+    paddw   %6, %3
+    paddw   %6, %4
+    paddw   %7, %6, %2
+    paddw   %6, %1
+    paddw   %6, %3
+    paddw   %7, %4
+    psraw   %6, 2
+    psraw   %7, 2
+    psubw   %6, %1
+    psubw   %7, %2
+    pand    %6, %5
+    pand    %7, %5
+    paddw   %1, %6
+    paddw   %2, %7
+%endmacro
+
+%macro CHROMA_V_LOAD 1
+    mova        m0, [r0]    ; p1
+    mova        m1, [r0+r1] ; p0
+    mova        m2, [%1]    ; q0
+    mova        m3, [%1+r1] ; q1
+%endmacro
+
+%macro CHROMA_V_STORE 0
+    mova [r0+1*r1], m1
+    mova [r0+2*r1], m2
+%endmacro
+
+%macro DEBLOCK_CHROMA 1
+;-----------------------------------------------------------------------------
+; void deblock_v_chroma( uint16_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
+;-----------------------------------------------------------------------------
+cglobal deblock_v_chroma_10_%1, 5,7-(mmsize/16),8*(mmsize/16)
+    mov         r5, r0
+    sub         r0, r1
+    sub         r0, r1
+    shl        r2d, 2
+    shl        r3d, 2
+%if mmsize < 16
+    mov         r6, 16/mmsize
+.loop:
+%endif
+    CHROMA_V_LOAD r5
+    LOAD_AB     m4, m5, r2, r3
+    LOAD_MASK   m0, m1, m2, m3, m4, m5, m7, m6, m4
+    pxor        m4, m4
+    LOAD_TC     m6, r4
+    psubw       m6, [pw_3]
+    pmaxsw      m6, m4
+    pand        m7, m6
+    DEBLOCK_P0_Q0 m1, m2, m0, m3, m7, m5, m6
+    CHROMA_V_STORE
+%if mmsize < 16
+    add         r0, mmsize
+    add         r5, mmsize
+    add         r4, mmsize/8
+    dec         r6
+    jg .loop
+    REP_RET
+%else
+    RET
+%endif
+
+;-----------------------------------------------------------------------------
+; void deblock_v_chroma_intra( uint16_t *pix, int stride, int alpha, int beta )
+;-----------------------------------------------------------------------------
+cglobal deblock_v_chroma_intra_10_%1, 4,6-(mmsize/16),8*(mmsize/16)
+    mov         r4, r0
+    sub         r0, r1
+    sub         r0, r1
+    shl        r2d, 2
+    shl        r3d, 2
+%if mmsize < 16
+    mov         r5, 16/mmsize
+.loop:
+%endif
+    CHROMA_V_LOAD r4
+    LOAD_AB     m4, m5, r2, r3
+    LOAD_MASK   m0, m1, m2, m3, m4, m5, m7, m6, m4
+    CHROMA_DEBLOCK_P0_Q0_INTRA m1, m2, m0, m3, m7, m5, m6
+    CHROMA_V_STORE
+%if mmsize < 16
+    add         r0, mmsize
+    add         r4, mmsize
+    dec         r5
+    jg .loop
+    REP_RET
+%else
+    RET
+%endif
+%endmacro
+
+%ifndef ARCH_X86_64
+INIT_MMX
+DEBLOCK_CHROMA mmxext
+%endif
+INIT_XMM
+DEBLOCK_CHROMA sse2
+INIT_AVX
+DEBLOCK_CHROMA avx
diff --git a/libavcodec/x86/h264dsp_mmx.c b/libavcodec/x86/h264dsp_mmx.c
index 42dae93f2d..01b11163c8 100644
--- a/libavcodec/x86/h264dsp_mmx.c
+++ b/libavcodec/x86/h264dsp_mmx.c
@@ -236,10 +236,18 @@ LF_FUNC (h,  luma,         depth, sse2)\
 LF_IFUNC(h,  luma_intra,   depth, sse2)\
 LF_FUNC (v,  luma,         depth, sse2)\
 LF_IFUNC(v,  luma_intra,   depth, sse2)\
+LF_FUNC (h,  chroma,       depth, sse2)\
+LF_IFUNC(h,  chroma_intra, depth, sse2)\
+LF_FUNC (v,  chroma,       depth, sse2)\
+LF_IFUNC(v,  chroma_intra, depth, sse2)\
 LF_FUNC (h,  luma,         depth,  avx)\
 LF_IFUNC(h,  luma_intra,   depth,  avx)\
 LF_FUNC (v,  luma,         depth,  avx)\
-LF_IFUNC(v,  luma_intra,   depth,  avx)
+LF_IFUNC(v,  luma_intra,   depth,  avx)\
+LF_FUNC (h,  chroma,       depth,  avx)\
+LF_IFUNC(h,  chroma_intra, depth,  avx)\
+LF_FUNC (v,  chroma,       depth,  avx)\
+LF_IFUNC(v,  chroma_intra, depth,  avx)
 
 LF_FUNCS( uint8_t,  8)
 LF_FUNCS(uint16_t, 10)
@@ -401,12 +409,16 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
     if (mm_flags & AV_CPU_FLAG_MMX) {
         if (mm_flags & AV_CPU_FLAG_MMX2) {
 #if ARCH_X86_32
+            c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_mmxext;
+            c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_mmxext;
             c->h264_v_loop_filter_luma= ff_deblock_v_luma_10_mmxext;
             c->h264_h_loop_filter_luma= ff_deblock_h_luma_10_mmxext;
             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmxext;
             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmxext;
 #endif
             if (mm_flags&AV_CPU_FLAG_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
                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2;
                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2;
@@ -415,6 +427,8 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
 #endif
             }
             if (mm_flags&AV_CPU_FLAG_AVX) {
+                c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_avx;
+                c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_avx;
 #if HAVE_ALIGNED_STACK
                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx;
                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx;