aarch64: Add NEON optimizations for 10 and 12 bit vp9 itxfm

This work is sponsored by, and copyright, Google.

Compared to the arm version, on aarch64 we can keep the full 8x8
transform in registers, and for 16x16 and 32x32, we can process
it in slices of 4 pixels instead of 2.

Examples of runtimes vs the 32 bit version, on a Cortex A53:
                                                ARM  AArch64
vp9_inv_adst_adst_4x4_sub4_add_10_neon:       111.0    109.7
vp9_inv_adst_adst_8x8_sub8_add_10_neon:       914.0    733.5
vp9_inv_adst_adst_16x16_sub16_add_10_neon:   5184.0   3745.7
vp9_inv_dct_dct_4x4_sub1_add_10_neon:          65.0     65.7
vp9_inv_dct_dct_4x4_sub4_add_10_neon:         100.0     96.7
vp9_inv_dct_dct_8x8_sub1_add_10_neon:         111.0    119.7
vp9_inv_dct_dct_8x8_sub8_add_10_neon:         618.0    494.7
vp9_inv_dct_dct_16x16_sub1_add_10_neon:       295.1    284.6
vp9_inv_dct_dct_16x16_sub2_add_10_neon:      2303.2   1883.9
vp9_inv_dct_dct_16x16_sub8_add_10_neon:      2984.8   2189.3
vp9_inv_dct_dct_16x16_sub16_add_10_neon:     3890.0   2799.4
vp9_inv_dct_dct_32x32_sub1_add_10_neon:      1044.4   1012.7
vp9_inv_dct_dct_32x32_sub2_add_10_neon:     13333.7   9695.1
vp9_inv_dct_dct_32x32_sub16_add_10_neon:    18531.3  12459.8
vp9_inv_dct_dct_32x32_sub32_add_10_neon:    24470.7  16160.2
vp9_inv_wht_wht_4x4_sub4_add_10_neon:          83.0     79.7

The larger transforms are significantly faster than the corresponding
ARM versions.

The speedup vs C code is smaller than in 32 bit mode, probably
because the 64 bit intermediates in the C code can be expressed
more efficiently in aarch64.

Signed-off-by: Martin Storsjö <martin@martin.st>

libavcodec/aarch64/Makefile

@@ -42,7 +42,8 @@ NEON-OBJS-$(CONFIG_MPEGAUDIODSP)        += aarch64/mpegaudiodsp_neon.o
 # decoders/encoders
 NEON-OBJS-$(CONFIG_DCA_DECODER)         += aarch64/synth_filter_neon.o
 NEON-OBJS-$(CONFIG_VORBIS_DECODER)      += aarch64/vorbisdsp_neon.o
-NEON-OBJS-$(CONFIG_VP9_DECODER)         += aarch64/vp9itxfm_neon.o             \
+NEON-OBJS-$(CONFIG_VP9_DECODER)         += aarch64/vp9itxfm_16bpp_neon.o       \
+                                           aarch64/vp9itxfm_neon.o             \
                                            aarch64/vp9lpf_neon.o               \
                                            aarch64/vp9mc_16bpp_neon.o          \
                                            aarch64/vp9mc_neon.o

libavcodec/aarch64/vp9dsp_init_16bpp_aarch64_template.c

@@ -157,7 +157,54 @@ static av_cold void vp9dsp_mc_init_aarch64(VP9DSPContext *dsp)
     }
 }
 
+#define define_itxfm2(type_a, type_b, sz, bpp)                                     \
+void ff_vp9_##type_a##_##type_b##_##sz##x##sz##_add_##bpp##_neon(uint8_t *_dst,    \
+                                                                 ptrdiff_t stride, \
+                                                                 int16_t *_block, int eob)
+#define define_itxfm(type_a, type_b, sz, bpp) define_itxfm2(type_a, type_b, sz, bpp)
+
+#define define_itxfm_funcs(sz, bpp)      \
+    define_itxfm(idct,  idct,  sz, bpp); \
+    define_itxfm(iadst, idct,  sz, bpp); \
+    define_itxfm(idct,  iadst, sz, bpp); \
+    define_itxfm(iadst, iadst, sz, bpp)
+
+define_itxfm_funcs(4,  BPP);
+define_itxfm_funcs(8,  BPP);
+define_itxfm_funcs(16, BPP);
+define_itxfm(idct, idct, 32, BPP);
+define_itxfm(iwht, iwht, 4,  BPP);
+
+
+static av_cold void vp9dsp_itxfm_init_aarch64(VP9DSPContext *dsp)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_neon(cpu_flags)) {
+#define init_itxfm2(tx, sz, bpp)                                               \
+    dsp->itxfm_add[tx][DCT_DCT]   = ff_vp9_idct_idct_##sz##_add_##bpp##_neon;  \
+    dsp->itxfm_add[tx][DCT_ADST]  = ff_vp9_iadst_idct_##sz##_add_##bpp##_neon; \
+    dsp->itxfm_add[tx][ADST_DCT]  = ff_vp9_idct_iadst_##sz##_add_##bpp##_neon; \
+    dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_iadst_iadst_##sz##_add_##bpp##_neon
+#define init_itxfm(tx, sz, bpp) init_itxfm2(tx, sz, bpp)
+
+#define init_idct2(tx, nm, bpp)     \
+    dsp->itxfm_add[tx][DCT_DCT]   = \
+    dsp->itxfm_add[tx][ADST_DCT]  = \
+    dsp->itxfm_add[tx][DCT_ADST]  = \
+    dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_##nm##_add_##bpp##_neon
+#define init_idct(tx, nm, bpp) init_idct2(tx, nm, bpp)
+
+        init_itxfm(TX_4X4,   4x4,   BPP);
+        init_itxfm(TX_8X8,   8x8,   BPP);
+        init_itxfm(TX_16X16, 16x16, BPP);
+        init_idct(TX_32X32, idct_idct_32x32, BPP);
+        init_idct(4,        iwht_iwht_4x4,   BPP);
+    }
+}
+
 av_cold void INIT_FUNC(VP9DSPContext *dsp)
 {
     vp9dsp_mc_init_aarch64(dsp);
+    vp9dsp_itxfm_init_aarch64(dsp);
 }