avcodec/x86/vp9: Add AVX-512ICL for 16x16 and 32x32 10bpc inverse transforms

libavcodec/vp9dec.h

@@ -220,8 +220,8 @@ struct VP9TileData {
     DECLARE_ALIGNED(8, uint8_t, left_ref_ctx)[8];
     DECLARE_ALIGNED(8, uint8_t, left_filter_ctx)[8];
     // block reconstruction intermediates
-    DECLARE_ALIGNED(32, uint8_t, tmp_y)[64 * 64 * 2];
-    DECLARE_ALIGNED(32, uint8_t, tmp_uv)[2][64 * 64 * 2];
+    DECLARE_ALIGNED(64, uint8_t, tmp_y)[64 * 64 * 2];
+    DECLARE_ALIGNED(64, uint8_t, tmp_uv)[2][64 * 64 * 2];
     struct { int x, y; } min_mv, max_mv;
     int16_t *block_base, *block, *uvblock_base[2], *uvblock[2];
     uint8_t *eob_base, *uveob_base[2], *eob, *uveob[2];

libavcodec/x86/Makefile

@@ -186,6 +186,7 @@ X86ASM-OBJS-$(CONFIG_VP9_DECODER)      += x86/vp9intrapred.o            \
                                           x86/vp9itxfm.o                \
                                           x86/vp9itxfm_avx512.o         \
                                           x86/vp9itxfm_16bpp.o          \
+                                          x86/vp9itxfm_16bpp_avx512.o   \
                                           x86/vp9lpf.o                  \
                                           x86/vp9lpf_16bpp.o            \
                                           x86/vp9mc.o                   \

libavcodec/x86/vp9dsp_init_16bpp_template.c

@@ -127,6 +127,8 @@ decl_itxfm_func(iwht, iwht, 4, BPC, mmxext);
 #if BPC == 10
 decl_itxfm_func(idct,  idct,  4, BPC, mmxext);
 decl_itxfm_funcs(4, BPC, ssse3);
+decl_itxfm_funcs(16, BPC, avx512icl);
+decl_itxfm_func(idct,  idct, 32, BPC, avx512icl);
 #else
 decl_itxfm_func(idct,  idct,  4, BPC, sse2);
 #endif
@@ -233,6 +235,12 @@ av_cold void INIT_FUNC(VP9DSPContext *dsp, int bitexact)
 #endif
     }
 
+#if ARCH_X86_64 && BPC == 10
+    if (EXTERNAL_AVX512ICL(cpu_flags)) {
+        init_itx_funcs(TX_16X16, 16, BPC, avx512icl);
+        init_itx_func_one(TX_32X32, idct, idct, 32, BPC, avx512icl);
+    }
+#endif
 #endif /* HAVE_X86ASM */
 
     ff_vp9dsp_init_16bpp_x86(dsp);