vp9: add subpel MC SIMD for 10/12bpp.

libavcodec/x86/Makefile

@@ -63,6 +63,8 @@ OBJS-$(CONFIG_VC1_DECODER)             += x86/vc1dsp_init.o
 OBJS-$(CONFIG_VORBIS_DECODER)          += x86/vorbisdsp_init.o
 OBJS-$(CONFIG_VP6_DECODER)             += x86/vp6dsp_init.o
 OBJS-$(CONFIG_VP9_DECODER)             += x86/vp9dsp_init.o            \
+                                          x86/vp9dsp_init_10bpp.o      \
+                                          x86/vp9dsp_init_12bpp.o      \
                                           x86/vp9dsp_init_16bpp.o
 OBJS-$(CONFIG_WEBP_DECODER)            += x86/vp8dsp_init.o
 
@@ -157,5 +159,6 @@ YASM-OBJS-$(CONFIG_VP6_DECODER)        += x86/vp6dsp.o
 YASM-OBJS-$(CONFIG_VP9_DECODER)        += x86/vp9intrapred.o            \
                                           x86/vp9itxfm.o                \
                                           x86/vp9lpf.o                  \
-                                          x86/vp9mc.o
+                                          x86/vp9mc.o                   \
+                                          x86/vp9mc_16bpp.o
 YASM-OBJS-$(CONFIG_WEBP_DECODER)       += x86/vp8dsp.o

libavcodec/x86/vp9dsp_init.h

@@ -28,12 +28,120 @@ void ff_vp9_##avg##sz##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
                                    const uint8_t *src, ptrdiff_t src_stride, \
                                    int h, int mx, int my)
 
+#define decl_mc_func(avg, sz, dir, opt, type, f_sz, bpp) \
+void ff_vp9_##avg##_8tap_1d_##dir##_##sz##_##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
+                                                         const uint8_t *src, ptrdiff_t src_stride, \
+                                                         int h, const type (*filter)[f_sz])
+
+#define decl_mc_funcs(sz, opt, type, fsz, bpp) \
+decl_mc_func(put, sz, h, opt, type, fsz, bpp); \
+decl_mc_func(avg, sz, h, opt, type, fsz, bpp); \
+decl_mc_func(put, sz, v, opt, type, fsz, bpp); \
+decl_mc_func(avg, sz, v, opt, type, fsz, bpp)
+
+#define mc_rep_func(avg, sz, hsz, hszb, dir, opt, type, f_sz, bpp) \
+static av_always_inline void \
+ff_vp9_##avg##_8tap_1d_##dir##_##sz##_##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
+                                                    const uint8_t *src, ptrdiff_t src_stride, \
+                                                    int h, const type (*filter)[f_sz]) \
+{ \
+    ff_vp9_##avg##_8tap_1d_##dir##_##hsz##_##bpp##_##opt(dst,        dst_stride, src, \
+                                                         src_stride, h, filter); \
+    ff_vp9_##avg##_8tap_1d_##dir##_##hsz##_##bpp##_##opt(dst + hszb, dst_stride, src + hszb, \
+                                                         src_stride, h, filter); \
+}
+
+#define mc_rep_funcs(sz, hsz, hszb, opt, type, fsz, bpp) \
+mc_rep_func(put, sz, hsz, hszb, h, opt, type, fsz, bpp); \
+mc_rep_func(avg, sz, hsz, hszb, h, opt, type, fsz, bpp); \
+mc_rep_func(put, sz, hsz, hszb, v, opt, type, fsz, bpp); \
+mc_rep_func(avg, sz, hsz, hszb, v, opt, type, fsz, bpp)
+
+#define filter_8tap_1d_fn(op, sz, f, f_opt, fname, dir, dvar, bpp, opt) \
+static void op##_8tap_##fname##_##sz##dir##_##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
+                                                          const uint8_t *src, ptrdiff_t src_stride, \
+                                                          int h, int mx, int my) \
+{ \
+    ff_vp9_##op##_8tap_1d_##dir##_##sz##_##bpp##_##opt(dst, dst_stride, src, src_stride, \
+                                                       h, ff_filters_##f_opt[f][dvar - 1]); \
+}
+
+#define filters_8tap_1d_fn(op, sz, dir, dvar, bpp, opt, f_opt) \
+filter_8tap_1d_fn(op, sz, FILTER_8TAP_REGULAR, f_opt, regular, dir, dvar, bpp, opt) \
+filter_8tap_1d_fn(op, sz, FILTER_8TAP_SHARP,   f_opt, sharp,   dir, dvar, bpp, opt) \
+filter_8tap_1d_fn(op, sz, FILTER_8TAP_SMOOTH,  f_opt, smooth,  dir, dvar, bpp, opt)
+
+#define filters_8tap_1d_fn2(op, sz, bpp, opt, f_opt) \
+filters_8tap_1d_fn(op, sz, h, mx, bpp, opt, f_opt) \
+filters_8tap_1d_fn(op, sz, v, my, bpp, opt, f_opt)
+
+#define filters_8tap_1d_fn3(op, bpp, opt4, opt8, f_opt) \
+filters_8tap_1d_fn2(op, 64, bpp, opt8, f_opt) \
+filters_8tap_1d_fn2(op, 32, bpp, opt8, f_opt) \
+filters_8tap_1d_fn2(op, 16, bpp, opt8, f_opt) \
+filters_8tap_1d_fn2(op, 8, bpp, opt8, f_opt) \
+filters_8tap_1d_fn2(op, 4, bpp, opt4, f_opt)
+
+#define filter_8tap_2d_fn(op, sz, f, f_opt, fname, align, bpp, bytes, opt) \
+static void op##_8tap_##fname##_##sz##hv_##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
+                                                       const uint8_t *src, ptrdiff_t src_stride, \
+                                                       int h, int mx, int my) \
+{ \
+    LOCAL_ALIGNED_##align(uint8_t, temp, [71 * 64 * bytes]); \
+    ff_vp9_put_8tap_1d_h_##sz##_##bpp##_##opt(temp, 64 * bytes, src - 3 * src_stride, \
+                                              src_stride,  h + 7, \
+                                              ff_filters_##f_opt[f][mx - 1]); \
+    ff_vp9_##op##_8tap_1d_v_##sz##_##bpp##_##opt(dst, dst_stride, temp + 3 * bytes * 64, \
+                                                 64 * bytes, h, \
+                                                 ff_filters_##f_opt[f][my - 1]); \
+}
+
+#define filters_8tap_2d_fn(op, sz, align, bpp, bytes, opt, f_opt) \
+filter_8tap_2d_fn(op, sz, FILTER_8TAP_REGULAR, f_opt, regular, align, bpp, bytes, opt) \
+filter_8tap_2d_fn(op, sz, FILTER_8TAP_SHARP,   f_opt, sharp, align, bpp, bytes, opt) \
+filter_8tap_2d_fn(op, sz, FILTER_8TAP_SMOOTH,  f_opt, smooth, align, bpp, bytes, opt)
+
+#define filters_8tap_2d_fn2(op, align, bpp, bytes, opt4, opt8, f_opt) \
+filters_8tap_2d_fn(op, 64, align, bpp, bytes, opt8, f_opt) \
+filters_8tap_2d_fn(op, 32, align, bpp, bytes, opt8, f_opt) \
+filters_8tap_2d_fn(op, 16, align, bpp, bytes, opt8, f_opt) \
+filters_8tap_2d_fn(op, 8, align, bpp, bytes, opt8, f_opt) \
+filters_8tap_2d_fn(op, 4, align, bpp, bytes, opt4, f_opt)
+
 #define init_fpel_func(idx1, idx2, sz, type, bpp, opt) \
     dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
     dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
     dsp->mc[idx1][FILTER_8TAP_SHARP  ][idx2][0][0] = \
     dsp->mc[idx1][FILTER_BILINEAR    ][idx2][0][0] = ff_vp9_##type##sz##bpp##_##opt
 
-void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp, int bpp);
+#define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type, bpp, opt) \
+    dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][idxh][idxv] = \
+        type##_8tap_smooth_##sz##dir##_##bpp##_##opt; \
+    dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = \
+        type##_8tap_regular_##sz##dir##_##bpp##_##opt; \
+    dsp->mc[idx1][FILTER_8TAP_SHARP  ][idx2][idxh][idxv] = \
+        type##_8tap_sharp_##sz##dir##_##bpp##_##opt
+
+#define init_subpel2(idx1, idx2, sz, type, bpp, opt) \
+    init_subpel1(idx1, idx2, 1, 1, sz, hv, type, bpp, opt); \
+    init_subpel1(idx1, idx2, 0, 1, sz, v,  type, bpp, opt); \
+    init_subpel1(idx1, idx2, 1, 0, sz, h,  type, bpp, opt)
+
+#define init_subpel3_32_64(idx, type, bpp, opt) \
+    init_subpel2(0, idx, 64, type, bpp, opt); \
+    init_subpel2(1, idx, 32, type, bpp, opt)
+
+#define init_subpel3_8to64(idx, type, bpp, opt) \
+    init_subpel3_32_64(idx, type, bpp, opt); \
+    init_subpel2(2, idx, 16, type, bpp, opt); \
+    init_subpel2(3, idx,  8, type, bpp, opt)
+
+#define init_subpel3(idx, type, bpp, opt) \
+    init_subpel3_8to64(idx, type, bpp, opt); \
+    init_subpel2(4, idx,  4, type, bpp, opt)
+
+void ff_vp9dsp_init_10bpp_x86(VP9DSPContext *dsp);
+void ff_vp9dsp_init_12bpp_x86(VP9DSPContext *dsp);
+void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp);
 
 #endif /* AVCODEC_X86_VP9DSP_INIT_H */

libavcodec/x86/vp9dsp_init_10bpp.c

+/*
+ * VP9 SIMD optimizations
+ *
+ * Copyright (c) 2013 Ronald S. Bultje <rsbultje gmail com>
+ *
+ * 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
+ */
+
+#define BPC 10
+#define INIT_FUNC ff_vp9dsp_init_10bpp_x86
+#include "vp9dsp_init_16bpp_template.c"

libavcodec/x86/vp9dsp_init_12bpp.c

+/*
+ * VP9 SIMD optimizations
+ *
+ * Copyright (c) 2013 Ronald S. Bultje <rsbultje gmail com>
+ *
+ * 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
+ */
+
+#define BPC 12
+#define INIT_FUNC ff_vp9dsp_init_12bpp_x86
+#include "vp9dsp_init_16bpp_template.c"

libavcodec/x86/vp9dsp_init_16bpp.c

@@ -48,7 +48,7 @@ decl_fpel_func(avg, 128, _16, avx2);
 
 #endif /* HAVE_YASM */
 
-av_cold void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp, int bpp)
+av_cold void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp)
 {
 #if HAVE_YASM
     int cpu_flags = av_get_cpu_flags();

libavcodec/x86/vp9dsp_init_16bpp_template.c

+/*
+ * VP9 SIMD optimizations
+ *
+ * Copyright (c) 2013 Ronald S. Bultje <rsbultje gmail com>
+ *
+ * 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/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/mem.h"
+#include "libavutil/x86/cpu.h"
+#include "libavcodec/vp9dsp.h"
+#include "libavcodec/x86/vp9dsp_init.h"
+
+#if HAVE_YASM
+
+extern const int16_t ff_filters_16bpp[3][15][4][16];
+
+decl_mc_funcs(4, sse2, int16_t, 16, BPC);
+decl_mc_funcs(8, sse2, int16_t, 16, BPC);
+
+mc_rep_funcs(16,  8, 16, sse2, int16_t, 16, BPC);
+mc_rep_funcs(32, 16, 32, sse2, int16_t, 16, BPC);
+mc_rep_funcs(64, 32, 64, sse2, int16_t, 16, BPC);
+
+filters_8tap_2d_fn2(put, 16, BPC, 2, sse2, sse2, 16bpp)
+filters_8tap_2d_fn2(avg, 16, BPC, 2, sse2, sse2, 16bpp)
+
+filters_8tap_1d_fn3(put, BPC, sse2, sse2, 16bpp)
+filters_8tap_1d_fn3(avg, BPC, sse2, sse2, 16bpp)
+
+#endif /* HAVE_YASM */
+
+av_cold void INIT_FUNC(VP9DSPContext *dsp)
+{
+#if HAVE_YASM
+    int cpu_flags = av_get_cpu_flags();
+
+    if (EXTERNAL_SSE2(cpu_flags)) {
+        init_subpel3(0, put, BPC, sse2);
+        init_subpel3(1, avg, BPC, sse2);
+    }
+
+#endif /* HAVE_YASM */
+
+    ff_vp9dsp_init_16bpp_x86(dsp);
+}

libavcodec/x86/vp9mc.asm

@@ -45,6 +45,13 @@ times 8 dw %7
 times 8 dw %8
 %endmacro
 
+%macro F8_16BPP_TAPS 8
+times 8 dw %1, %2
+times 8 dw %3, %4
+times 8 dw %5, %6
+times 8 dw %7, %8
+%endmacro
+
 %macro FILTER 1
 const filters_%1 ; smooth
                     F8_TAPS -3, -1,  32,  64,  38,   1, -3,  0
@@ -102,12 +109,15 @@ FILTER ssse3
 %define F8_TAPS F8_SSE2_TAPS
 ; int16_t ff_filters_sse2[3][15][8][8]
 FILTER sse2
+%define F8_TAPS F8_16BPP_TAPS
+; int16_t ff_filters_16bpp[3][15][4][16]
+FILTER 16bpp
 
 SECTION .text
 
 %macro filter_sse2_h_fn 1
 %assign %%px mmsize/2
-cglobal vp9_%1_8tap_1d_h_ %+ %%px, 6, 6, 15, dst, dstride, src, sstride, h, filtery
+cglobal vp9_%1_8tap_1d_h_ %+ %%px %+ _8, 6, 6, 15, dst, dstride, src, sstride, h, filtery
     pxor        m5, m5
     mova        m6, [pw_64]
     mova        m7, [filteryq+  0]
@@ -192,7 +202,7 @@ filter_sse2_h_fn avg
 
 %macro filter_h_fn 1
 %assign %%px mmsize/2
-cglobal vp9_%1_8tap_1d_h_ %+ %%px, 6, 6, 11, dst, dstride, src, sstride, h, filtery
+cglobal vp9_%1_8tap_1d_h_ %+ %%px %+ _8, 6, 6, 11, dst, dstride, src, sstride, h, filtery
     mova        m6, [pw_256]
     mova        m7, [filteryq+ 0]
 %if ARCH_X86_64 && mmsize > 8
@@ -253,7 +263,7 @@ filter_h_fn avg
 %if ARCH_X86_64
 %macro filter_hx2_fn 1
 %assign %%px mmsize
-cglobal vp9_%1_8tap_1d_h_ %+ %%px, 6, 6, 14, dst, dstride, src, sstride, h, filtery
+cglobal vp9_%1_8tap_1d_h_ %+ %%px %+ _8, 6, 6, 14, dst, dstride, src, sstride, h, filtery
     mova       m13, [pw_256]
     mova        m8, [filteryq+ 0]
     mova        m9, [filteryq+32]
@@ -315,9 +325,9 @@ filter_hx2_fn avg
 %macro filter_sse2_v_fn 1
 %assign %%px mmsize/2
 %if ARCH_X86_64
-cglobal vp9_%1_8tap_1d_v_ %+ %%px, 6, 8, 15, dst, dstride, src, sstride, h, filtery, src4, sstride3
+cglobal vp9_%1_8tap_1d_v_ %+ %%px %+ _8, 6, 8, 15, dst, dstride, src, sstride, h, filtery, src4, sstride3
 %else
-cglobal vp9_%1_8tap_1d_v_ %+ %%px, 4, 7, 15, dst, dstride, src, sstride, filtery, src4, sstride3
+cglobal vp9_%1_8tap_1d_v_ %+ %%px %+ _8, 4, 7, 15, dst, dstride, src, sstride, filtery, src4, sstride3
     mov   filteryq, r5mp
 %define hd r4mp
 %endif
@@ -413,9 +423,9 @@ filter_sse2_v_fn avg
 %macro filter_v_fn 1
 %assign %%px mmsize/2
 %if ARCH_X86_64
-cglobal vp9_%1_8tap_1d_v_ %+ %%px, 6, 8, 11, dst, dstride, src, sstride, h, filtery, src4, sstride3
+cglobal vp9_%1_8tap_1d_v_ %+ %%px %+ _8, 6, 8, 11, dst, dstride, src, sstride, h, filtery, src4, sstride3
 %else
-cglobal vp9_%1_8tap_1d_v_ %+ %%px, 4, 7, 11, dst, dstride, src, sstride, filtery, src4, sstride3
+cglobal vp9_%1_8tap_1d_v_ %+ %%px %+ _8, 4, 7, 11, dst, dstride, src, sstride, filtery, src4, sstride3
     mov   filteryq, r5mp
 %define hd r4mp
 %endif
@@ -487,7 +497,7 @@ filter_v_fn avg
 
 %macro filter_vx2_fn 1
 %assign %%px mmsize
-cglobal vp9_%1_8tap_1d_v_ %+ %%px, 6, 8, 14, dst, dstride, src, sstride, h, filtery, src4, sstride3
+cglobal vp9_%1_8tap_1d_v_ %+ %%px %+ _8, 6, 8, 14, dst, dstride, src, sstride, h, filtery, src4, sstride3
     mova       m13, [pw_256]
     lea  sstride3q, [sstrideq*3]
     lea      src4q, [srcq+sstrideq]