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

1 files changed, 43 insertions, 154 deletions

diff --git a/libavcodec/x86/vp9dsp_init.c b/libavcodec/x86/vp9dsp_init.c
index 244f5f0f40..cd4af992d0 100644
--- a/libavcodec/x86/vp9dsp_init.c
+++ b/libavcodec/x86/vp9dsp_init.c
@@ -44,144 +44,52 @@ decl_fpel_func(put, 64,   , avx);
 decl_fpel_func(avg, 32, _8, avx2);
 decl_fpel_func(avg, 64, _8, avx2);
 
-#define mc_func(avg, sz, dir, opt, type, f_sz) \
-void ff_vp9_##avg##_8tap_1d_##dir##_##sz##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
-                                                 const uint8_t *src, ptrdiff_t src_stride, \
-                                                 int h, const type (*filter)[f_sz])
-#define mc_funcs(sz, opt, type, fsz) \
-mc_func(put, sz, h, opt, type, fsz); \
-mc_func(avg, sz, h, opt, type, fsz); \
-mc_func(put, sz, v, opt, type, fsz); \
-mc_func(avg, sz, v, opt, type, fsz)
-
-mc_funcs(4, mmxext, int16_t, 8);
-mc_funcs(8, sse2, int16_t, 8);
-mc_funcs(4, ssse3, int8_t, 32);
-mc_funcs(8, ssse3, int8_t, 32);
+decl_mc_funcs(4, mmxext, int16_t, 8, 8);
+decl_mc_funcs(8, sse2, int16_t,  8, 8);
+decl_mc_funcs(4, ssse3, int8_t, 32, 8);
+decl_mc_funcs(8, ssse3, int8_t, 32, 8);
 #if ARCH_X86_64
-mc_funcs(16, ssse3, int8_t, 32);
-mc_funcs(32, avx2, int8_t, 32);
+decl_mc_funcs(16, ssse3, int8_t, 32, 8);
+decl_mc_funcs(32, avx2, int8_t, 32, 8);
 #endif
 
-#undef mc_funcs
-#undef mc_func
-
-#define mc_rep_func(avg, sz, hsz, dir, opt, type, f_sz) \
-static av_always_inline void \
-ff_vp9_##avg##_8tap_1d_##dir##_##sz##_##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##_##opt(dst,       dst_stride, src, \
-                                                 src_stride, h, filter); \
-    ff_vp9_##avg##_8tap_1d_##dir##_##hsz##_##opt(dst + hsz, dst_stride, src + hsz, \
-                                                 src_stride, h, filter); \
-}
-
-#define mc_rep_funcs(sz, hsz, opt, type, fsz) \
-mc_rep_func(put, sz, hsz, h, opt, type, fsz); \
-mc_rep_func(avg, sz, hsz, h, opt, type, fsz); \
-mc_rep_func(put, sz, hsz, v, opt, type, fsz); \
-mc_rep_func(avg, sz, hsz, v, opt, type, fsz)
-
-mc_rep_funcs(16, 8, sse2, int16_t, 8);
+mc_rep_funcs(16,  8,  8,  sse2, int16_t,  8, 8);
 #if ARCH_X86_32
-mc_rep_funcs(16, 8, ssse3, int8_t, 32);
+mc_rep_funcs(16,  8,  8, ssse3, int8_t,  32, 8);
 #endif
-mc_rep_funcs(32, 16, sse2, int16_t, 8);
-mc_rep_funcs(32, 16, ssse3, int8_t, 32);
-mc_rep_funcs(64, 32, sse2, int16_t, 8);
-mc_rep_funcs(64, 32, ssse3, int8_t, 32);
+mc_rep_funcs(32, 16, 16, sse2,  int16_t,  8, 8);
+mc_rep_funcs(32, 16, 16, ssse3, int8_t,  32, 8);
+mc_rep_funcs(64, 32, 32, sse2,  int16_t,  8, 8);
+mc_rep_funcs(64, 32, 32, ssse3, int8_t,  32, 8);
 #if ARCH_X86_64 && HAVE_AVX2_EXTERNAL
-mc_rep_funcs(64, 32, avx2, int8_t, 32);
+mc_rep_funcs(64, 32, 32, avx2,  int8_t,  32, 8);
 #endif
 
-#undef mc_rep_funcs
-#undef mc_rep_func
-
 extern const int8_t ff_filters_ssse3[3][15][4][32];
 extern const int16_t ff_filters_sse2[3][15][8][8];
 
-#define filter_8tap_2d_fn(op, sz, f, f_opt, fname, align, opt) \
-static void op##_8tap_##fname##_##sz##hv_##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]); \
-    ff_vp9_put_8tap_1d_h_##sz##_##opt(temp, 64, src - 3 * src_stride, src_stride, \
-                                      h + 7, ff_filters_##f_opt[f][mx - 1]); \
-    ff_vp9_##op##_8tap_1d_v_##sz##_##opt(dst, dst_stride, temp + 3 * 64, 64, \
-                                         h, ff_filters_##f_opt[f][my - 1]); \
-}
-
-#define filters_8tap_2d_fn(op, sz, align, opt, f_opt) \
-filter_8tap_2d_fn(op, sz, FILTER_8TAP_REGULAR, f_opt, regular, align, opt) \
-filter_8tap_2d_fn(op, sz, FILTER_8TAP_SHARP,   f_opt, sharp, align, opt) \
-filter_8tap_2d_fn(op, sz, FILTER_8TAP_SMOOTH,  f_opt, smooth, align, opt)
-
-#define filters_8tap_2d_fn2(op, align, opt4, opt8, f_opt) \
-filters_8tap_2d_fn(op, 64, align, opt8, f_opt) \
-filters_8tap_2d_fn(op, 32, align, opt8, f_opt) \
-filters_8tap_2d_fn(op, 16, align, opt8, f_opt) \
-filters_8tap_2d_fn(op, 8, align, opt8, f_opt) \
-filters_8tap_2d_fn(op, 4, align, opt4, f_opt)
-
-filters_8tap_2d_fn2(put, 16, mmxext, sse2, sse2)
-filters_8tap_2d_fn2(avg, 16, mmxext, sse2, sse2)
-filters_8tap_2d_fn2(put, 16, ssse3, ssse3, ssse3)
-filters_8tap_2d_fn2(avg, 16, ssse3, ssse3, ssse3)
+filters_8tap_2d_fn2(put, 16, 8, 1, mmxext, sse2, sse2)
+filters_8tap_2d_fn2(avg, 16, 8, 1, mmxext, sse2, sse2)
+filters_8tap_2d_fn2(put, 16, 8, 1, ssse3, ssse3, ssse3)
+filters_8tap_2d_fn2(avg, 16, 8, 1, ssse3, ssse3, ssse3)
 #if ARCH_X86_64 && HAVE_AVX2_EXTERNAL
-filters_8tap_2d_fn(put, 64, 32, avx2, ssse3)
-filters_8tap_2d_fn(put, 32, 32, avx2, ssse3)
-filters_8tap_2d_fn(avg, 64, 32, avx2, ssse3)
-filters_8tap_2d_fn(avg, 32, 32, avx2, ssse3)
+filters_8tap_2d_fn(put, 64, 32, 8, 1, avx2, ssse3)
+filters_8tap_2d_fn(put, 32, 32, 8, 1, avx2, ssse3)
+filters_8tap_2d_fn(avg, 64, 32, 8, 1, avx2, ssse3)
+filters_8tap_2d_fn(avg, 32, 32, 8, 1, avx2, ssse3)
 #endif
 
-#undef filters_8tap_2d_fn2
-#undef filters_8tap_2d_fn
-#undef filter_8tap_2d_fn
-
-#define filter_8tap_1d_fn(op, sz, f, f_opt, fname, dir, dvar, opt) \
-static void op##_8tap_##fname##_##sz##dir##_##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##_##opt(dst, dst_stride, src, src_stride, \
-                                               h, ff_filters_##f_opt[f][dvar - 1]); \
-}
-
-#define filters_8tap_1d_fn(op, sz, dir, dvar, opt, f_opt) \
-filter_8tap_1d_fn(op, sz, FILTER_8TAP_REGULAR, f_opt, regular, dir, dvar, opt) \
-filter_8tap_1d_fn(op, sz, FILTER_8TAP_SHARP,   f_opt, sharp,   dir, dvar, opt) \
-filter_8tap_1d_fn(op, sz, FILTER_8TAP_SMOOTH,  f_opt, smooth,  dir, dvar, opt)
-
-#define filters_8tap_1d_fn2(op, sz, opt, f_opt) \
-filters_8tap_1d_fn(op, sz, h, mx, opt, f_opt) \
-filters_8tap_1d_fn(op, sz, v, my, opt, f_opt)
-
-#define filters_8tap_1d_fn3(op, opt4, opt8, f_opt) \
-filters_8tap_1d_fn2(op, 64, opt8, f_opt) \
-filters_8tap_1d_fn2(op, 32, opt8, f_opt) \
-filters_8tap_1d_fn2(op, 16, opt8, f_opt) \
-filters_8tap_1d_fn2(op, 8, opt8, f_opt) \
-filters_8tap_1d_fn2(op, 4, opt4, f_opt)
-
-filters_8tap_1d_fn3(put, mmxext, sse2, sse2)
-filters_8tap_1d_fn3(avg, mmxext, sse2, sse2)
-filters_8tap_1d_fn3(put, ssse3, ssse3, ssse3)
-filters_8tap_1d_fn3(avg, ssse3, ssse3, ssse3)
+filters_8tap_1d_fn3(put, 8, mmxext, sse2, sse2)
+filters_8tap_1d_fn3(avg, 8, mmxext, sse2, sse2)
+filters_8tap_1d_fn3(put, 8, ssse3, ssse3, ssse3)
+filters_8tap_1d_fn3(avg, 8, ssse3, ssse3, ssse3)
 #if ARCH_X86_64 && HAVE_AVX2_EXTERNAL
-filters_8tap_1d_fn2(put, 64, avx2, ssse3)
-filters_8tap_1d_fn2(put, 32, avx2, ssse3)
-filters_8tap_1d_fn2(avg, 64, avx2, ssse3)
-filters_8tap_1d_fn2(avg, 32, avx2, ssse3)
+filters_8tap_1d_fn2(put, 64, 8, avx2, ssse3)
+filters_8tap_1d_fn2(put, 32, 8, avx2, ssse3)
+filters_8tap_1d_fn2(avg, 64, 8, avx2, ssse3)
+filters_8tap_1d_fn2(avg, 32, 8, avx2, ssse3)
 #endif
 
-#undef filters_8tap_1d_fn
-#undef filters_8tap_1d_fn2
-#undef filters_8tap_1d_fn3
-#undef filter_8tap_1d_fn
-
 #define itxfm_func(typea, typeb, size, opt) \
 void ff_vp9_##typea##_##typeb##_##size##x##size##_add_##opt(uint8_t *dst, ptrdiff_t stride, \
                                                             int16_t *block, int eob)
@@ -306,36 +214,17 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact)
 {
 #if HAVE_YASM
     int cpu_flags;
-    if (bpp != 8) {
-        ff_vp9dsp_init_16bpp_x86(dsp, bpp);
+
+    if (bpp == 10) {
+        ff_vp9dsp_init_10bpp_x86(dsp);
+        return;
+    } else if (bpp == 12) {
+        ff_vp9dsp_init_12bpp_x86(dsp);
         return;
     }
 
     cpu_flags = av_get_cpu_flags();
 
-#define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type, opt) \
-    dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][idxh][idxv] = type##_8tap_smooth_##sz##dir##_##opt; \
-    dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = type##_8tap_regular_##sz##dir##_##opt; \
-    dsp->mc[idx1][FILTER_8TAP_SHARP  ][idx2][idxh][idxv] = type##_8tap_sharp_##sz##dir##_##opt
-
-#define init_subpel2(idx1, idx2, sz, type, opt) \
-    init_subpel1(idx1, idx2, 1, 1, sz, hv, type, opt); \
-    init_subpel1(idx1, idx2, 0, 1, sz, v,  type, opt); \
-    init_subpel1(idx1, idx2, 1, 0, sz, h,  type, opt)
-
-#define init_subpel3_32_64(idx, type, opt) \
-    init_subpel2(0, idx, 64, type, opt); \
-    init_subpel2(1, idx, 32, type, opt)
-
-#define init_subpel3_8to64(idx, type, opt) \
-    init_subpel3_32_64(idx, type, opt); \
-    init_subpel2(2, idx, 16, type, opt); \
-    init_subpel2(3, idx,  8, type, opt)
-
-#define init_subpel3(idx, type, opt) \
-    init_subpel3_8to64(idx, type, opt); \
-    init_subpel2(4, idx,  4, type, opt)
-
 #define init_lpf(opt) do { \
     dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_##opt; \
     dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_##opt; \
@@ -390,8 +279,8 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact)
     }
 
     if (EXTERNAL_MMXEXT(cpu_flags)) {
-        init_subpel2(4, 0, 4, put, mmxext);
-        init_subpel2(4, 1, 4, avg, mmxext);
+        init_subpel2(4, 0, 4, put, 8, mmxext);
+        init_subpel2(4, 1, 4, avg, 8, mmxext);
         init_fpel_func(4, 1,  4, avg, _8, mmxext);
         init_fpel_func(3, 1,  8, avg, _8, mmxext);
         dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_mmxext;
@@ -409,8 +298,8 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact)
     }
 
     if (EXTERNAL_SSE2(cpu_flags)) {
-        init_subpel3_8to64(0, put, sse2);
-        init_subpel3_8to64(1, avg, sse2);
+        init_subpel3_8to64(0, put, 8, sse2);
+        init_subpel3_8to64(1, avg, 8, sse2);
         init_fpel_func(2, 1, 16, avg,  _8, sse2);
         init_fpel_func(1, 1, 32, avg,  _8, sse2);
         init_fpel_func(0, 1, 64, avg,  _8, sse2);
@@ -439,8 +328,8 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact)
     }
 
     if (EXTERNAL_SSSE3(cpu_flags)) {
-        init_subpel3(0, put, ssse3);
-        init_subpel3(1, avg, ssse3);
+        init_subpel3(0, put, 8, ssse3);
+        init_subpel3(1, avg, 8, ssse3);
         dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_ssse3;
         dsp->itxfm_add[TX_4X4][ADST_DCT]  = ff_vp9_idct_iadst_4x4_add_ssse3;
         dsp->itxfm_add[TX_4X4][DCT_ADST]  = ff_vp9_iadst_idct_4x4_add_ssse3;
@@ -493,8 +382,8 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact)
         init_fpel_func(0, 1, 64, avg, _8, avx2);
         if (ARCH_X86_64) {
 #if ARCH_X86_64 && HAVE_AVX2_EXTERNAL
-            init_subpel3_32_64(0, put, avx2);
-            init_subpel3_32_64(1, avg, avx2);
+            init_subpel3_32_64(0, put, 8, avx2);
+            init_subpel3_32_64(1, avg, 8, avx2);
 #endif
         }
         init_dc_ipred(32, avx2);