diff options
Diffstat (limited to 'src/dsp/arm/weight_mask_neon.cc')
| -rw-r--r-- | src/dsp/arm/weight_mask_neon.cc | 289 |
1 files changed, 207 insertions, 82 deletions
diff --git a/src/dsp/arm/weight_mask_neon.cc b/src/dsp/arm/weight_mask_neon.cc index 7e5bff0..5ad6b97 100644 --- a/src/dsp/arm/weight_mask_neon.cc +++ b/src/dsp/arm/weight_mask_neon.cc @@ -32,20 +32,51 @@ namespace libgav1 { namespace dsp { -namespace low_bitdepth { namespace { -constexpr int kRoundingBits8bpp = 4; +inline int16x8x2_t LoadPred(const int16_t* LIBGAV1_RESTRICT prediction_0, + const int16_t* LIBGAV1_RESTRICT prediction_1) { + const int16x8x2_t pred = {vld1q_s16(prediction_0), vld1q_s16(prediction_1)}; + return pred; +} + +#if LIBGAV1_MAX_BITDEPTH >= 10 +inline uint16x8x2_t LoadPred(const uint16_t* LIBGAV1_RESTRICT prediction_0, + const uint16_t* LIBGAV1_RESTRICT prediction_1) { + const uint16x8x2_t pred = {vld1q_u16(prediction_0), vld1q_u16(prediction_1)}; + return pred; +} +#endif // LIBGAV1_MAX_BITDEPTH >= 10 + +template <int bitdepth> +inline uint16x8_t AbsolutePredDifference(const int16x8x2_t pred) { + static_assert(bitdepth == 8, ""); + constexpr int rounding_bits = bitdepth - 8 + ((bitdepth == 12) ? 2 : 4); + return vrshrq_n_u16( + vreinterpretq_u16_s16(vabdq_s16(pred.val[0], pred.val[1])), + rounding_bits); +} -template <bool mask_is_inverse> -inline void WeightMask8_NEON(const int16_t* prediction_0, - const int16_t* prediction_1, uint8_t* mask) { - const int16x8_t pred_0 = vld1q_s16(prediction_0); - const int16x8_t pred_1 = vld1q_s16(prediction_1); +template <int bitdepth> +inline uint16x8_t AbsolutePredDifference(const uint16x8x2_t pred) { + constexpr int rounding_bits = bitdepth - 8 + ((bitdepth == 12) ? 2 : 4); + return vrshrq_n_u16(vabdq_u16(pred.val[0], pred.val[1]), rounding_bits); +} + +template <bool mask_is_inverse, int bitdepth> +inline void WeightMask8_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask) { + using PredType = + typename std::conditional<bitdepth == 8, int16_t, uint16_t>::type; + using PredTypeVecx2 = + typename std::conditional<bitdepth == 8, int16x8x2_t, uint16x8x2_t>::type; + const PredTypeVecx2 pred = + LoadPred(static_cast<const PredType*>(prediction_0), + static_cast<const PredType*>(prediction_1)); + const uint16x8_t difference = AbsolutePredDifference<bitdepth>(pred); const uint8x8_t difference_offset = vdup_n_u8(38); const uint8x8_t mask_ceiling = vdup_n_u8(64); - const uint16x8_t difference = vrshrq_n_u16( - vreinterpretq_u16_s16(vabdq_s16(pred_0, pred_1)), kRoundingBits8bpp); const uint8x8_t adjusted_difference = vqadd_u8(vqshrn_n_u16(difference, 4), difference_offset); const uint8x8_t mask_value = vmin_u8(adjusted_difference, mask_ceiling); @@ -58,7 +89,7 @@ inline void WeightMask8_NEON(const int16_t* prediction_0, } #define WEIGHT8_WITHOUT_STRIDE \ - WeightMask8_NEON<mask_is_inverse>(pred_0, pred_1, mask) + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0, pred_1, mask) #define WEIGHT8_AND_STRIDE \ WEIGHT8_WITHOUT_STRIDE; \ @@ -66,9 +97,12 @@ inline void WeightMask8_NEON(const int16_t* prediction_0, pred_1 += 8; \ mask += mask_stride -template <bool mask_is_inverse> -void WeightMask8x8_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +// |pred_0| and |pred_1| are cast as int16_t* for the sake of pointer math. They +// are uint16_t* for 10bpp and 12bpp, and this is handled in WeightMask8_NEON. +template <bool mask_is_inverse, int bitdepth> +void WeightMask8x8_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y = 0; @@ -78,9 +112,11 @@ void WeightMask8x8_NEON(const void* prediction_0, const void* prediction_1, WEIGHT8_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask8x16_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask8x16_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -92,9 +128,11 @@ void WeightMask8x16_NEON(const void* prediction_0, const void* prediction_1, WEIGHT8_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask8x32_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask8x32_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y5 = 0; @@ -109,9 +147,9 @@ void WeightMask8x32_NEON(const void* prediction_0, const void* prediction_1, WEIGHT8_WITHOUT_STRIDE; } -#define WEIGHT16_WITHOUT_STRIDE \ - WeightMask8_NEON<mask_is_inverse>(pred_0, pred_1, mask); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 8, pred_1 + 8, mask + 8) +#define WEIGHT16_WITHOUT_STRIDE \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0, pred_1, mask); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 8, pred_1 + 8, mask + 8) #define WEIGHT16_AND_STRIDE \ WEIGHT16_WITHOUT_STRIDE; \ @@ -119,9 +157,11 @@ void WeightMask8x32_NEON(const void* prediction_0, const void* prediction_1, pred_1 += 16; \ mask += mask_stride -template <bool mask_is_inverse> -void WeightMask16x8_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask16x8_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y = 0; @@ -131,9 +171,11 @@ void WeightMask16x8_NEON(const void* prediction_0, const void* prediction_1, WEIGHT16_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask16x16_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask16x16_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -145,9 +187,11 @@ void WeightMask16x16_NEON(const void* prediction_0, const void* prediction_1, WEIGHT16_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask16x32_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask16x32_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y5 = 0; @@ -162,9 +206,11 @@ void WeightMask16x32_NEON(const void* prediction_0, const void* prediction_1, WEIGHT16_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask16x64_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask16x64_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -176,11 +222,14 @@ void WeightMask16x64_NEON(const void* prediction_0, const void* prediction_1, WEIGHT16_WITHOUT_STRIDE; } -#define WEIGHT32_WITHOUT_STRIDE \ - WeightMask8_NEON<mask_is_inverse>(pred_0, pred_1, mask); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 8, pred_1 + 8, mask + 8); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 16, pred_1 + 16, mask + 16); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 24, pred_1 + 24, mask + 24) +#define WEIGHT32_WITHOUT_STRIDE \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0, pred_1, mask); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 8, pred_1 + 8, \ + mask + 8); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 16, pred_1 + 16, \ + mask + 16); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 24, pred_1 + 24, \ + mask + 24) #define WEIGHT32_AND_STRIDE \ WEIGHT32_WITHOUT_STRIDE; \ @@ -188,9 +237,11 @@ void WeightMask16x64_NEON(const void* prediction_0, const void* prediction_1, pred_1 += 32; \ mask += mask_stride -template <bool mask_is_inverse> -void WeightMask32x8_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask32x8_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); WEIGHT32_AND_STRIDE; @@ -203,9 +254,11 @@ void WeightMask32x8_NEON(const void* prediction_0, const void* prediction_1, WEIGHT32_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask32x16_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask32x16_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -217,9 +270,11 @@ void WeightMask32x16_NEON(const void* prediction_0, const void* prediction_1, WEIGHT32_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask32x32_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask32x32_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y5 = 0; @@ -234,9 +289,11 @@ void WeightMask32x32_NEON(const void* prediction_0, const void* prediction_1, WEIGHT32_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask32x64_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask32x64_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -248,15 +305,22 @@ void WeightMask32x64_NEON(const void* prediction_0, const void* prediction_1, WEIGHT32_WITHOUT_STRIDE; } -#define WEIGHT64_WITHOUT_STRIDE \ - WeightMask8_NEON<mask_is_inverse>(pred_0, pred_1, mask); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 8, pred_1 + 8, mask + 8); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 16, pred_1 + 16, mask + 16); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 24, pred_1 + 24, mask + 24); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 32, pred_1 + 32, mask + 32); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 40, pred_1 + 40, mask + 40); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 48, pred_1 + 48, mask + 48); \ - WeightMask8_NEON<mask_is_inverse>(pred_0 + 56, pred_1 + 56, mask + 56) +#define WEIGHT64_WITHOUT_STRIDE \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0, pred_1, mask); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 8, pred_1 + 8, \ + mask + 8); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 16, pred_1 + 16, \ + mask + 16); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 24, pred_1 + 24, \ + mask + 24); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 32, pred_1 + 32, \ + mask + 32); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 40, pred_1 + 40, \ + mask + 40); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 48, pred_1 + 48, \ + mask + 48); \ + WeightMask8_NEON<mask_is_inverse, bitdepth>(pred_0 + 56, pred_1 + 56, \ + mask + 56) #define WEIGHT64_AND_STRIDE \ WEIGHT64_WITHOUT_STRIDE; \ @@ -264,9 +328,11 @@ void WeightMask32x64_NEON(const void* prediction_0, const void* prediction_1, pred_1 += 64; \ mask += mask_stride -template <bool mask_is_inverse> -void WeightMask64x16_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask64x16_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -278,9 +344,11 @@ void WeightMask64x16_NEON(const void* prediction_0, const void* prediction_1, WEIGHT64_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask64x32_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask64x32_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y5 = 0; @@ -295,9 +363,11 @@ void WeightMask64x32_NEON(const void* prediction_0, const void* prediction_1, WEIGHT64_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask64x64_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask64x64_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -309,9 +379,11 @@ void WeightMask64x64_NEON(const void* prediction_0, const void* prediction_1, WEIGHT64_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask64x128_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask64x128_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -324,9 +396,11 @@ void WeightMask64x128_NEON(const void* prediction_0, const void* prediction_1, WEIGHT64_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask128x64_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask128x64_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -366,9 +440,11 @@ void WeightMask128x64_NEON(const void* prediction_0, const void* prediction_1, WEIGHT64_WITHOUT_STRIDE; } -template <bool mask_is_inverse> -void WeightMask128x128_NEON(const void* prediction_0, const void* prediction_1, - uint8_t* mask, ptrdiff_t mask_stride) { +template <bool mask_is_inverse, int bitdepth> +void WeightMask128x128_NEON(const void* LIBGAV1_RESTRICT prediction_0, + const void* LIBGAV1_RESTRICT prediction_1, + uint8_t* LIBGAV1_RESTRICT mask, + ptrdiff_t mask_stride) { const auto* pred_0 = static_cast<const int16_t*>(prediction_0); const auto* pred_1 = static_cast<const int16_t*>(prediction_1); int y3 = 0; @@ -416,11 +492,20 @@ void WeightMask128x128_NEON(const void* prediction_0, const void* prediction_1, mask += 64; WEIGHT64_WITHOUT_STRIDE; } +#undef WEIGHT8_WITHOUT_STRIDE +#undef WEIGHT8_AND_STRIDE +#undef WEIGHT16_WITHOUT_STRIDE +#undef WEIGHT16_AND_STRIDE +#undef WEIGHT32_WITHOUT_STRIDE +#undef WEIGHT32_AND_STRIDE +#undef WEIGHT64_WITHOUT_STRIDE +#undef WEIGHT64_AND_STRIDE #define INIT_WEIGHT_MASK_8BPP(width, height, w_index, h_index) \ dsp->weight_mask[w_index][h_index][0] = \ - WeightMask##width##x##height##_NEON<0>; \ - dsp->weight_mask[w_index][h_index][1] = WeightMask##width##x##height##_NEON<1> + WeightMask##width##x##height##_NEON<0, 8>; \ + dsp->weight_mask[w_index][h_index][1] = \ + WeightMask##width##x##height##_NEON<1, 8> void Init8bpp() { Dsp* const dsp = dsp_internal::GetWritableDspTable(kBitdepth8); assert(dsp != nullptr); @@ -442,11 +527,51 @@ void Init8bpp() { INIT_WEIGHT_MASK_8BPP(128, 64, 4, 3); INIT_WEIGHT_MASK_8BPP(128, 128, 4, 4); } +#undef INIT_WEIGHT_MASK_8BPP } // namespace -} // namespace low_bitdepth -void WeightMaskInit_NEON() { low_bitdepth::Init8bpp(); } +#if LIBGAV1_MAX_BITDEPTH >= 10 +namespace high_bitdepth { +namespace { + +#define INIT_WEIGHT_MASK_10BPP(width, height, w_index, h_index) \ + dsp->weight_mask[w_index][h_index][0] = \ + WeightMask##width##x##height##_NEON<0, 10>; \ + dsp->weight_mask[w_index][h_index][1] = \ + WeightMask##width##x##height##_NEON<1, 10> +void Init10bpp() { + Dsp* const dsp = dsp_internal::GetWritableDspTable(kBitdepth10); + assert(dsp != nullptr); + INIT_WEIGHT_MASK_10BPP(8, 8, 0, 0); + INIT_WEIGHT_MASK_10BPP(8, 16, 0, 1); + INIT_WEIGHT_MASK_10BPP(8, 32, 0, 2); + INIT_WEIGHT_MASK_10BPP(16, 8, 1, 0); + INIT_WEIGHT_MASK_10BPP(16, 16, 1, 1); + INIT_WEIGHT_MASK_10BPP(16, 32, 1, 2); + INIT_WEIGHT_MASK_10BPP(16, 64, 1, 3); + INIT_WEIGHT_MASK_10BPP(32, 8, 2, 0); + INIT_WEIGHT_MASK_10BPP(32, 16, 2, 1); + INIT_WEIGHT_MASK_10BPP(32, 32, 2, 2); + INIT_WEIGHT_MASK_10BPP(32, 64, 2, 3); + INIT_WEIGHT_MASK_10BPP(64, 16, 3, 1); + INIT_WEIGHT_MASK_10BPP(64, 32, 3, 2); + INIT_WEIGHT_MASK_10BPP(64, 64, 3, 3); + INIT_WEIGHT_MASK_10BPP(64, 128, 3, 4); + INIT_WEIGHT_MASK_10BPP(128, 64, 4, 3); + INIT_WEIGHT_MASK_10BPP(128, 128, 4, 4); +} +#undef INIT_WEIGHT_MASK_10BPP + +} // namespace +} // namespace high_bitdepth +#endif // LIBGAV1_MAX_BITDEPTH >= 10 +void WeightMaskInit_NEON() { + Init8bpp(); +#if LIBGAV1_MAX_BITDEPTH >= 10 + high_bitdepth::Init10bpp(); +#endif // LIBGAV1_MAX_BITDEPTH >= 10 +} } // namespace dsp } // namespace libgav1 |