aarch64: vp9: Add NEON itxfm routines

This work is sponsored by, and copyright, Google.

These are ported from the ARM version; thanks to the larger
amount of registers available, we can do the 16x16 and 32x32
transforms in slices 8 pixels wide instead of 4. This gives
a speedup of around 1.4x compared to the 32 bit version.

The fact that aarch64 doesn't have the same d/q register
aliasing makes some of the macros quite a bit simpler as well.

Examples of runtimes vs the 32 bit version, on a Cortex A53:
                                       ARM  AArch64
vp9_inv_adst_adst_4x4_add_neon:       90.0     87.7
vp9_inv_adst_adst_8x8_add_neon:      400.0    354.7
vp9_inv_adst_adst_16x16_add_neon:   2526.5   1827.2
vp9_inv_dct_dct_4x4_add_neon:         74.0     72.7
vp9_inv_dct_dct_8x8_add_neon:        271.0    256.7
vp9_inv_dct_dct_16x16_add_neon:     1960.7   1372.7
vp9_inv_dct_dct_32x32_add_neon:    11988.9   8088.3
vp9_inv_wht_wht_4x4_add_neon:         63.0     57.7

The speedup vs C code (2-4x) is smaller than in the 32 bit case,
mostly because the C code ends up significantly faster (around
1.6x faster, with GCC 5.4) when built for aarch64.

Examples of runtimes vs C on a Cortex A57 (for a slightly older version
of the patch):
                                A57 gcc-5.3   neon
vp9_inv_adst_adst_4x4_add_neon:       152.2   60.0
vp9_inv_adst_adst_8x8_add_neon:       948.2  288.0
vp9_inv_adst_adst_16x16_add_neon:    4830.4 1380.5
vp9_inv_dct_dct_4x4_add_neon:         153.0   58.6
vp9_inv_dct_dct_8x8_add_neon:         789.2  180.2
vp9_inv_dct_dct_16x16_add_neon:      3639.6  917.1
vp9_inv_dct_dct_32x32_add_neon:     20462.1 4985.0
vp9_inv_wht_wht_4x4_add_neon:          91.0   49.8

The asm is around factor 3-4 faster than C on the cortex-a57 and the asm
is around 30-50% faster on the a57 compared to the a53.

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

libavcodec/aarch64/Makefile

@@ -44,4 +44,5 @@ NEON-OBJS-$(CONFIG_MPEGAUDIODSP)        += aarch64/mpegaudiodsp_neon.o
 NEON-OBJS-$(CONFIG_DCA_DECODER)         += aarch64/dcadsp_neon.o               \
                                            aarch64/synth_filter_neon.o
 NEON-OBJS-$(CONFIG_VORBIS_DECODER)      += aarch64/vorbisdsp_neon.o
-NEON-OBJS-$(CONFIG_VP9_DECODER)         += aarch64/vp9mc_neon.o
+NEON-OBJS-$(CONFIG_VP9_DECODER)         += aarch64/vp9itxfm_neon.o             \
+                                           aarch64/vp9mc_neon.o

libavcodec/aarch64/vp9dsp_init_aarch64.c

@@ -96,7 +96,7 @@ define_8tap_2d_funcs(16)
 define_8tap_2d_funcs(8)
 define_8tap_2d_funcs(4)
 
-av_cold void ff_vp9dsp_init_aarch64(VP9DSPContext *dsp)
+static av_cold void vp9dsp_mc_init_aarch64(VP9DSPContext *dsp)
 {
     int cpu_flags = av_get_cpu_flags();
 
@@ -151,3 +151,52 @@ av_cold void ff_vp9dsp_init_aarch64(VP9DSPContext *dsp)
         init_mc_funcs_dirs(4, 4);
     }
 }
+
+#define define_itxfm(type_a, type_b, sz)                                   \
+void ff_vp9_##type_a##_##type_b##_##sz##x##sz##_add_neon(uint8_t *_dst,    \
+                                                         ptrdiff_t stride, \
+                                                         int16_t *_block, int eob)
+
+#define define_itxfm_funcs(sz)      \
+    define_itxfm(idct,  idct,  sz); \
+    define_itxfm(iadst, idct,  sz); \
+    define_itxfm(idct,  iadst, sz); \
+    define_itxfm(iadst, iadst, sz)
+
+define_itxfm_funcs(4);
+define_itxfm_funcs(8);
+define_itxfm_funcs(16);
+define_itxfm(idct, idct, 32);
+define_itxfm(iwht, iwht, 4);
+
+
+static av_cold void vp9dsp_itxfm_init_aarch64(VP9DSPContext *dsp)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_neon(cpu_flags)) {
+#define init_itxfm(tx, sz)                                             \
+    dsp->itxfm_add[tx][DCT_DCT]   = ff_vp9_idct_idct_##sz##_add_neon;  \
+    dsp->itxfm_add[tx][DCT_ADST]  = ff_vp9_iadst_idct_##sz##_add_neon; \
+    dsp->itxfm_add[tx][ADST_DCT]  = ff_vp9_idct_iadst_##sz##_add_neon; \
+    dsp->itxfm_add[tx][ADST_ADST] = ff_vp9_iadst_iadst_##sz##_add_neon
+
+#define init_idct(tx, nm)           \
+    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_neon
+
+        init_itxfm(TX_4X4, 4x4);
+        init_itxfm(TX_8X8, 8x8);
+        init_itxfm(TX_16X16, 16x16);
+        init_idct(TX_32X32, idct_idct_32x32);
+        init_idct(4, iwht_iwht_4x4);
+    }
+}
+
+av_cold void ff_vp9dsp_init_aarch64(VP9DSPContext *dsp)
+{
+    vp9dsp_mc_init_aarch64(dsp);
+    vp9dsp_itxfm_init_aarch64(dsp);
+}