Port x86 10-bit H.264 deblock asm from x264

libavcodec/x86/Makefile

@@ -9,6 +9,7 @@ YASM-OBJS-$(CONFIG_FFT)                += x86/fft_mmx.o                 \
 
 MMX-OBJS-$(CONFIG_H264DSP)             += x86/h264dsp_mmx.o
 YASM-OBJS-$(CONFIG_H264DSP)            += x86/h264_deblock.o            \
+                                          x86/h264_deblock_10bit.o      \
                                           x86/h264_weight.o             \
                                           x86/h264_idct.o               \
 

libavcodec/x86/dsputil_mmx.c

@@ -43,6 +43,7 @@ DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
 {0x8000000080000000ULL, 0x8000000080000000ULL};
 
 DECLARE_ALIGNED(8,  const uint64_t, ff_pw_1  ) = 0x0001000100010001ULL;
+DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_2  ) = {0x0002000200020002ULL, 0x0002000200020002ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_3  ) = {0x0003000300030003ULL, 0x0003000300030003ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_4  ) = {0x0004000400040004ULL, 0x0004000400040004ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_5  ) = {0x0005000500050005ULL, 0x0005000500050005ULL};

libavcodec/x86/h264_deblock.asm

@@ -324,7 +324,7 @@ cextern pb_A1
 ; void deblock_v_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
 %macro DEBLOCK_LUMA 1
-cglobal deblock_v_luma_%1, 5,5,10
+cglobal deblock_v_luma_8_%1, 5,5,10
     movd    m8, [r4] ; tc0
     lea     r4, [r1*3]
     dec     r2d        ; alpha-1
@@ -369,7 +369,7 @@ cglobal deblock_v_luma_%1, 5,5,10
 ; void deblock_h_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
 INIT_MMX
-cglobal deblock_h_luma_%1, 5,7
+cglobal deblock_h_luma_8_%1, 5,7
     movsxd r10, r1d
     lea    r11, [r10+r10*2]
     lea    r6,  [r0-4]
@@ -396,7 +396,7 @@ cglobal deblock_h_luma_%1, 5,7
 %ifdef WIN64
     mov    [rsp+0x20], r4
 %endif
-    call   deblock_v_luma_%1
+    call   deblock_v_luma_8_%1
 
     ; transpose 16x4 -> original space  (only the middle 4 rows were changed by the filter)
     add    r6, 2
@@ -436,7 +436,7 @@ DEBLOCK_LUMA avx
 ;-----------------------------------------------------------------------------
 ; void deblock_v8_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
-cglobal deblock_%2_luma_%1, 5,5
+cglobal deblock_%2_luma_8_%1, 5,5
     lea     r4, [r1*3]
     dec     r2     ; alpha-1
     neg     r4
@@ -489,7 +489,7 @@ cglobal deblock_%2_luma_%1, 5,5
 ; void deblock_h_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
 INIT_MMX
-cglobal deblock_h_luma_%1, 0,5
+cglobal deblock_h_luma_8_%1, 0,5
     mov    r0, r0mp
     mov    r3, r1m
     lea    r4, [r3*3]
@@ -512,11 +512,11 @@ cglobal deblock_h_luma_%1, 0,5
     PUSH   dword r2m
     PUSH   dword 16
     PUSH   dword r0
-    call   deblock_%2_luma_%1
+    call   deblock_%2_luma_8_%1
 %ifidn %2, v8
     add    dword [esp   ], 8 ; pix_tmp+0x38
     add    dword [esp+16], 2 ; tc0+2
-    call   deblock_%2_luma_%1
+    call   deblock_%2_luma_8_%1
 %endif
     ADD    esp, 20
 
@@ -685,7 +685,7 @@ DEBLOCK_LUMA avx, v, 16
 ;-----------------------------------------------------------------------------
 ; void deblock_v_luma_intra( uint8_t *pix, int stride, int alpha, int beta )
 ;-----------------------------------------------------------------------------
-cglobal deblock_%2_luma_intra_%1, 4,6,16
+cglobal deblock_%2_luma_intra_8_%1, 4,6,16
 %ifndef ARCH_X86_64
     sub     esp, 0x60
 %endif
@@ -747,7 +747,7 @@ INIT_MMX
 ;-----------------------------------------------------------------------------
 ; void deblock_h_luma_intra( uint8_t *pix, int stride, int alpha, int beta )
 ;-----------------------------------------------------------------------------
-cglobal deblock_h_luma_intra_%1, 4,7
+cglobal deblock_h_luma_intra_8_%1, 4,7
     movsxd r10, r1d
     lea    r11, [r10*3]
     lea    r6,  [r0-4]
@@ -763,7 +763,7 @@ cglobal deblock_h_luma_intra_%1, 4,7
 
     lea    r0,  [pix_tmp+0x40]
     mov    r1,  0x10
-    call   deblock_v_luma_intra_%1
+    call   deblock_v_luma_intra_8_%1
 
     ; transpose 16x6 -> original space (but we can't write only 6 pixels, so really 16x8)
     lea    r5, [r6+r11]
@@ -776,7 +776,7 @@ cglobal deblock_h_luma_intra_%1, 4,7
     add    rsp, 0x88
     RET
 %else
-cglobal deblock_h_luma_intra_%1, 2,4
+cglobal deblock_h_luma_intra_8_%1, 2,4
     lea    r3,  [r1*3]
     sub    r0,  4
     lea    r2,  [r0+r3]
@@ -795,10 +795,10 @@ cglobal deblock_h_luma_intra_%1, 2,4
     PUSH   dword r2m
     PUSH   dword 16
     PUSH   r0
-    call   deblock_%2_luma_intra_%1
+    call   deblock_%2_luma_intra_8_%1
 %ifidn %2, v8
     add    dword [rsp], 8 ; pix_tmp+8
-    call   deblock_%2_luma_intra_%1
+    call   deblock_%2_luma_intra_8_%1
 %endif
     ADD    esp, 16
 
@@ -851,7 +851,7 @@ INIT_MMX
 ;-----------------------------------------------------------------------------
 ; void ff_deblock_v_chroma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
-cglobal deblock_v_chroma_mmxext, 5,6
+cglobal deblock_v_chroma_8_mmxext, 5,6
     CHROMA_V_START
     movq  m0, [t5]
     movq  m1, [t5+r1]
@@ -865,7 +865,7 @@ cglobal deblock_v_chroma_mmxext, 5,6
 ;-----------------------------------------------------------------------------
 ; void ff_deblock_h_chroma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
 ;-----------------------------------------------------------------------------
-cglobal deblock_h_chroma_mmxext, 5,7
+cglobal deblock_h_chroma_8_mmxext, 5,7
 %ifdef ARCH_X86_64
     %define buf0 [rsp-24]
     %define buf1 [rsp-16]
@@ -911,7 +911,7 @@ ff_chroma_inter_body_mmxext:
 ;-----------------------------------------------------------------------------
 ; void ff_deblock_v_chroma_intra( uint8_t *pix, int stride, int alpha, int beta )
 ;-----------------------------------------------------------------------------
-cglobal deblock_v_chroma_intra_mmxext, 4,5
+cglobal deblock_v_chroma_intra_8_mmxext, 4,5
     CHROMA_V_START
     movq  m0, [t5]
     movq  m1, [t5+r1]
@@ -925,7 +925,7 @@ cglobal deblock_v_chroma_intra_mmxext, 4,5
 ;-----------------------------------------------------------------------------
 ; void ff_deblock_h_chroma_intra( uint8_t *pix, int stride, int alpha, int beta )
 ;-----------------------------------------------------------------------------
-cglobal deblock_h_chroma_intra_mmxext, 4,6
+cglobal deblock_h_chroma_intra_8_mmxext, 4,6
     CHROMA_H_START
     TRANSPOSE4x8_LOAD  bw, wd, dq, PASS8ROWS(t5, r0, r1, t6)
     call ff_chroma_intra_body_mmxext

libavcodec/x86/h264dsp_mmx.c

@@ -218,45 +218,49 @@ static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40]
     );
 }
 
-#define LF_FUNC(DIR, TYPE, OPT) \
-void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## OPT (uint8_t *pix, int stride, \
-                                                  int alpha, int beta, int8_t *tc0);
-#define LF_IFUNC(DIR, TYPE, OPT) \
-void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## OPT (uint8_t *pix, int stride, \
-                                                  int alpha, int beta);
-
-LF_FUNC (h,  chroma,       mmxext)
-LF_IFUNC(h,  chroma_intra, mmxext)
-LF_FUNC (v,  chroma,       mmxext)
-LF_IFUNC(v,  chroma_intra, mmxext)
-
-LF_FUNC (h,  luma,         mmxext)
-LF_IFUNC(h,  luma_intra,   mmxext)
-#if HAVE_YASM && ARCH_X86_32
-LF_FUNC (v8, luma,         mmxext)
-static void ff_deblock_v_luma_mmxext(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
+#define LF_FUNC(DIR, TYPE, DEPTH, OPT) \
+void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *pix, int stride, \
+                                                                int alpha, int beta, int8_t *tc0);
+#define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
+void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *pix, int stride, \
+                                                                int alpha, int beta);
+
+#define LF_FUNCS(type, depth)\
+LF_FUNC (h,  chroma,       depth, mmxext)\
+LF_IFUNC(h,  chroma_intra, depth, mmxext)\
+LF_FUNC (v,  chroma,       depth, mmxext)\
+LF_IFUNC(v,  chroma_intra, depth, mmxext)\
+LF_FUNC (h,  luma,         depth, mmxext)\
+LF_IFUNC(h,  luma_intra,   depth, mmxext)\
+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,  luma,         depth,  avx)\
+LF_IFUNC(h,  luma_intra,   depth,  avx)\
+LF_FUNC (v,  luma,         depth,  avx)\
+LF_IFUNC(v,  luma_intra,   depth,  avx)
+
+LF_FUNCS( uint8_t,  8)
+LF_FUNCS(uint16_t, 10)
+
+LF_FUNC (v8, luma,             8, mmxext)
+static void ff_deblock_v_luma_8_mmxext(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
     if((tc0[0] & tc0[1]) >= 0)
-        ff_deblock_v8_luma_mmxext(pix+0, stride, alpha, beta, tc0);
+        ff_deblock_v8_luma_8_mmxext(pix+0, stride, alpha, beta, tc0);
     if((tc0[2] & tc0[3]) >= 0)
-        ff_deblock_v8_luma_mmxext(pix+8, stride, alpha, beta, tc0+2);
+        ff_deblock_v8_luma_8_mmxext(pix+8, stride, alpha, beta, tc0+2);
 }
-LF_IFUNC(v8, luma_intra,   mmxext)
-static void ff_deblock_v_luma_intra_mmxext(uint8_t *pix, int stride, int alpha, int beta)
+LF_IFUNC(v8, luma_intra,        8, mmxext)
+static void ff_deblock_v_luma_intra_8_mmxext(uint8_t *pix, int stride, int alpha, int beta)
 {
-    ff_deblock_v8_luma_intra_mmxext(pix+0, stride, alpha, beta);
-    ff_deblock_v8_luma_intra_mmxext(pix+8, stride, alpha, beta);
+    ff_deblock_v8_luma_intra_8_mmxext(pix+0, stride, alpha, beta);
+    ff_deblock_v8_luma_intra_8_mmxext(pix+8, stride, alpha, beta);
 }
-#endif
 
-LF_FUNC (h,  luma,         sse2)
-LF_IFUNC(h,  luma_intra,   sse2)
-LF_FUNC (v,  luma,         sse2)
-LF_IFUNC(v,  luma_intra,   sse2)
-LF_FUNC (h,  luma,         avx)
-LF_IFUNC(h,  luma_intra,   avx)
-LF_FUNC (v,  luma,         avx)
-LF_IFUNC(v,  luma_intra,   avx)
+LF_FUNC (v,  luma,            10, mmxext)
+LF_IFUNC(v,  luma_intra,      10, mmxext)
 
 /***********************************/
 /* weighted prediction */
@@ -318,15 +322,15 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
             c->h264_idct_add8      = ff_h264_idct_add8_mmx2;
             c->h264_idct_add16intra= ff_h264_idct_add16intra_mmx2;
 
-            c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_mmxext;
-            c->h264_h_loop_filter_chroma= ff_deblock_h_chroma_mmxext;
-            c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_mmxext;
-            c->h264_h_loop_filter_chroma_intra= ff_deblock_h_chroma_intra_mmxext;
+            c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_8_mmxext;
+            c->h264_h_loop_filter_chroma= ff_deblock_h_chroma_8_mmxext;
+            c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_8_mmxext;
+            c->h264_h_loop_filter_chroma_intra= ff_deblock_h_chroma_intra_8_mmxext;
 #if ARCH_X86_32
-            c->h264_v_loop_filter_luma= ff_deblock_v_luma_mmxext;
-            c->h264_h_loop_filter_luma= ff_deblock_h_luma_mmxext;
-            c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_mmxext;
-            c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_mmxext;
+            c->h264_v_loop_filter_luma= ff_deblock_v_luma_8_mmxext;
+            c->h264_h_loop_filter_luma= ff_deblock_h_luma_8_mmxext;
+            c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmxext;
+            c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext;
 #endif
             c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;
             c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;
@@ -364,10 +368,10 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
                 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_sse2;
 
 #if HAVE_ALIGNED_STACK
-                c->h264_v_loop_filter_luma = ff_deblock_v_luma_sse2;
-                c->h264_h_loop_filter_luma = ff_deblock_h_luma_sse2;
-                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_sse2;
-                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_sse2;
+                c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2;
+                c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2;
+                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
+                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
 #endif
 
                 c->h264_idct_add16 = ff_h264_idct_add16_sse2;
@@ -383,10 +387,39 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
             }
             if (mm_flags&AV_CPU_FLAG_AVX) {
 #if HAVE_ALIGNED_STACK
-                c->h264_v_loop_filter_luma = ff_deblock_v_luma_avx;
-                c->h264_h_loop_filter_luma = ff_deblock_h_luma_avx;
-                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_avx;
-                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_avx;
+                c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx;
+                c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx;
+                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
+                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
+#endif
+            }
+        }
+    }
+#endif
+    } else if (bit_depth == 10) {
+#if HAVE_YASM
+    if (mm_flags & AV_CPU_FLAG_MMX) {
+        if (mm_flags & AV_CPU_FLAG_MMX2) {
+#if ARCH_X86_32
+            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) {
+#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;
+                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
+                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
+#endif
+            }
+            if (mm_flags&AV_CPU_FLAG_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;
+                c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx;
+                c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx;
 #endif
             }
         }

libavcodec/x86/x86util.asm

@@ -457,3 +457,8 @@
     pshufw     %1, %2, (%3)*0x55
 %endif
 %endmacro
+
+%macro CLIPW 3 ;(dst, min, max)
+    pmaxsw %1, %2
+    pminsw %1, %3
+%endmacro