| // Copyright 2015 Google Inc. All Rights Reserved. |
| // |
| // Licensed under the Apache License, Version 2.0 (the "License"); |
| // you may not use this file except in compliance with the License. |
| // You may obtain a copy of the License at |
| // |
| // http://www.apache.org/licenses/LICENSE-2.0 |
| // |
| // Unless required by applicable law or agreed to in writing, software |
| // distributed under the License is distributed on an "AS IS" BASIS, |
| // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| // See the License for the specific language governing permissions and |
| // limitations under the License. |
| |
| // pack_neon.h: optimized NEON specializations of the templates in pack.h. |
| |
| #ifndef GEMMLOWP_INTERNAL_PACK_NEON_H_ |
| #define GEMMLOWP_INTERNAL_PACK_NEON_H_ |
| |
| #include "pack.h" |
| |
| #include <arm_neon.h> |
| |
| namespace gemmlowp { |
| |
| // Variant of PseudoRandomNonzeroBytesGenerator that produces |
| // random NEON 128-bit vectors. |
| class NEONPseudoRandomNonzeroBytesGenerator { |
| public: |
| NEONPseudoRandomNonzeroBytesGenerator() { |
| uint8_t s = 1; |
| std::uint8_t a[16]; |
| for (int i = 0; i < 16; i++) { |
| a[i] = s; |
| // Xorshift8(7,7,1). Very important to choose a different |
| // xorshift than we do in get(), otherwise lanes would contain |
| // the same values! |
| s ^= s << 7; |
| s ^= s >> 7; |
| s ^= s << 1; |
| } |
| x_ = vld1q_u8(a); |
| } |
| |
| uint8x16_t get() { |
| // Xorshift8(7,5,3) |
| x_ = veorq_u8(x_, vshlq_n_u8(x_, 7)); |
| x_ = veorq_u8(x_, vshrq_n_u8(x_, 5)); |
| x_ = veorq_u8(x_, vshlq_n_u8(x_, 3)); |
| return x_; |
| } |
| |
| private: |
| // State |
| uint8x16_t x_; |
| }; |
| |
| // Requantizes source uint8 values in [0..255] range |
| // to the range specified by BitDepth, [0..((2^bits)-1)]. |
| // Bias must be avoided. Currently this is achieved |
| // by probabilistic rounding. |
| template <typename BitDepth, RoundingMode Rounding> |
| uint8x16_t Requantize(uint8x16_t raw_src_data, |
| NEONPseudoRandomNonzeroBytesGenerator* prng) { |
| static const int kBits = BitDepth::kBits; |
| static const std::uint8_t kMaxVal = (1 << kBits) - 1; |
| |
| if (kBits == 8) { |
| return raw_src_data; |
| } |
| |
| // We will need to temporarily work in 16 bit precision. |
| uint16x8_t x[2]; |
| |
| // Multiply source values by 2^kBits. |
| x[0] = vshll_n_u8(vget_low_u8(raw_src_data), kBits); |
| x[1] = vshll_n_u8(vget_high_u8(raw_src_data), kBits); |
| |
| // Subtract source values, so we effectively have them |
| // multiplied by (2^kBits) - 1. |
| x[0] = vsubw_u8(x[0], vget_low_u8(raw_src_data)); |
| x[1] = vsubw_u8(x[1], vget_high_u8(raw_src_data)); |
| |
| // Compute the rounding offset. |
| uint8x16_t rounding_offset; |
| switch (Rounding) { |
| case RoundingMode::Nearest: |
| // 128 is the midpoint in [1..255], and [1..255] is the interval |
| // that we use for offsets here, see the comment below on |
| // the Probabilistic case. |
| rounding_offset = vdupq_n_u8(128); |
| break; |
| case RoundingMode::Probabilistic: |
| // Take nonzero bytes in [1..255]. |
| // In principle we want a value in [0..254], but: |
| // 1) Below we will be multiplying by 257/256 instead of 256/255, |
| // which is slightly too low, and this helps compensate for that. |
| // (One checks this on paper and this also gives better results |
| // on TestWithRealData). |
| // 1 bis) Note also that this 257/256 != 256/255 helps ensure |
| // that no overflow can happen, even with offset=255. |
| // 2) Our PRNG, xorshift, inherently wants to generate values |
| // in [1..255] so this saves a couple of instructions. |
| rounding_offset = prng->get(); |
| break; |
| default: |
| assert(false); |
| rounding_offset = vdupq_n_u8(0); |
| } |
| |
| // Add the rounding offset. |
| x[0] = vaddw_u8(x[0], vget_low_u8(rounding_offset)); |
| x[1] = vaddw_u8(x[1], vget_high_u8(rounding_offset)); |
| |
| // Multiply by 257/256, which is close enough to 256/255. |
| x[0] = vaddq_u16(x[0], vshrq_n_u16(x[0], 8)); |
| x[1] = vaddq_u16(x[1], vshrq_n_u16(x[1], 8)); |
| |
| uint8x8_t y[2]; |
| // Divide again by 256. |
| y[0] = vshrn_n_u16(x[0], 8); |
| y[1] = vshrn_n_u16(x[1], 8); |
| |
| return vcombine_u8(y[0], y[1]); |
| } |
| |
| typedef SideMap<const std::uint8_t, SideMapOrder::WidthMajor> |
| WidthMajorUint8SideMap; |
| |
| template <int Cells> |
| using DepthMajorSideFormatNCells4x2 = KernelSideFormat<CellFormat<4, 2>, Cells>; |
| |
| template <int Cells> |
| class PackingRegisterBlock< |
| WidthMajorUint8SideMap, |
| PackedSideBlock<DepthMajorSideFormatNCells4x2<Cells> > > |
| : public PackingRegisterBlockBase< |
| WidthMajorUint8SideMap, |
| PackedSideBlock<DepthMajorSideFormatNCells4x2<Cells> > > { |
| public: |
| typedef DepthMajorSideFormatNCells4x2<Cells> KernelSideFormat; |
| typedef typename KernelSideFormat::Cell CellFormat; |
| static const int kCells = KernelSideFormat::kCells; |
| static const int kCellWidth = CellFormat::kWidth; |
| static const int kKernelWidth = CellFormat::kWidth * kCells; |
| static const int kCellDepth = CellFormat::kDepth; |
| static const int kCellSize = CellFormat::kSize; |
| |
| typedef NEONPseudoRandomNonzeroBytesGenerator |
| PseudoRandomNonzeroBytesGenerator; |
| |
| template <typename BitDepth, RoundingMode Rounding> |
| void Pack(PackedSideBlock<KernelSideFormat>* dst, int start_width, |
| PseudoRandomNonzeroBytesGenerator* prng) { |
| static const int kBits = BitDepth::kBits; |
| static const std::uint16_t kMaxVal = (1 << kBits) - 1; |
| std::uint8_t* dst_ptr = dst->current_data(); |
| const std::uint8_t* const src_ptr = this->complete_src_.data(); |
| const int stride = this->complete_src_.stride(); |
| // Load and requantize source WidthMajor data |
| uint8x16_t src_lines[4 * kCells]; |
| for (int i = 0; i < 4 * kCells; i++) { |
| src_lines[i] = Requantize<BitDepth, Rounding>( |
| vld1q_u8(src_ptr + i * stride), prng); |
| } |
| // Reorder the data within registers to make DepthMajor 4x2 cells |
| uint8x16x2_t src_lines_intertwined_2x[2 * kCells]; |
| for (int i = 0; i < kCells; i++) { |
| src_lines_intertwined_2x[2 * i] = |
| vzipq_u8(src_lines[4 * i], src_lines[4 * i + 2]); |
| src_lines_intertwined_2x[2 * i + 1] = |
| vzipq_u8(src_lines[4 * i + 1], src_lines[4 * i + 3]); |
| } |
| uint8x16x2_t src_lines_intertwined_4x[2 * kCells]; |
| for (int i = 0; i < kCells; i++) { |
| src_lines_intertwined_4x[2 * i] = |
| vzipq_u8(src_lines_intertwined_2x[2 * i].val[0], |
| src_lines_intertwined_2x[2 * i + 1].val[0]); |
| src_lines_intertwined_4x[2 * i + 1] = |
| vzipq_u8(src_lines_intertwined_2x[2 * i].val[1], |
| src_lines_intertwined_2x[2 * i + 1].val[1]); |
| } |
| // Store the resulting DepthMajor 4x2 cells in the destination packed block |
| for (int outer = 0; outer < 2; outer++) { |
| for (int inner = 0; inner < 2; inner++) { |
| for (int cell = 0; cell < kCells; cell++) { |
| uint8x8_t value = vget_low_u8( |
| src_lines_intertwined_4x[2 * cell + outer].val[inner]); |
| vst1_u8(dst_ptr, value); |
| dst_ptr += 8; |
| } |
| for (int cell = 0; cell < kCells; cell++) { |
| uint8x8_t value = vget_high_u8( |
| src_lines_intertwined_4x[2 * cell + outer].val[inner]); |
| vst1_u8(dst_ptr, value); |
| dst_ptr += 8; |
| } |
| } |
| } |
| // Compute sums across the depth dimension |
| uint16x8_t sums_of_2_cells[kCells][4]; |
| for (int outer = 0; outer < 2; outer++) { |
| for (int inner = 0; inner < 2; inner++) { |
| int i = 2 * outer + inner; |
| for (int cell = 0; cell < kCells; cell++) { |
| sums_of_2_cells[cell][i] = vaddl_u8( |
| vget_low_u8( |
| src_lines_intertwined_4x[2 * cell + outer].val[inner]), |
| vget_high_u8( |
| src_lines_intertwined_4x[2 * cell + outer].val[inner])); |
| } |
| } |
| } |
| int32x4_t sums_of_4_cells[kCells][4]; |
| for (int i = 0; i < 4; i++) { |
| for (int cell = 0; cell < kCells; cell++) { |
| sums_of_4_cells[cell][i] = vreinterpretq_s32_u32( |
| vaddl_u16(vget_low_u16(sums_of_2_cells[cell][i]), |
| vget_high_u16(sums_of_2_cells[cell][i]))); |
| } |
| } |
| // Update the rank_one_update vector |
| for (int cell = 0; cell < kCells; cell++) { |
| int32x4_t s01 = |
| vaddq_s32(sums_of_4_cells[cell][0], sums_of_4_cells[cell][1]); |
| int32x4_t s23 = |
| vaddq_s32(sums_of_4_cells[cell][2], sums_of_4_cells[cell][3]); |
| int32x4_t s = vaddq_s32(s01, s23); |
| int32x4_t u = vmulq_n_s32(s, dst->rank_one_update_multiplier()); |
| std::int32_t* rank_one_update_ptr = |
| dst->rank_one_update() + start_width + 4 * cell; |
| vst1q_s32(rank_one_update_ptr, |
| vaddq_s32(u, vld1q_s32(rank_one_update_ptr))); |
| } |
| dst->seek_forward_n_cells(kCells * kRegisterSize / kCellDepth); |
| } |
| }; |
| |
| template <int Cells> |
| using WidthMajorSideFormatNCells4x2 = |
| KernelSideFormat<CellFormat<4, 2, CellOrder::WidthMajor>, Cells>; |
| |
| template <int Cells> |
| class PackingRegisterBlock< |
| WidthMajorUint8SideMap, |
| PackedSideBlock<WidthMajorSideFormatNCells4x2<Cells> > > |
| : public PackingRegisterBlockBase< |
| WidthMajorUint8SideMap, |
| PackedSideBlock<WidthMajorSideFormatNCells4x2<Cells> > > { |
| public: |
| typedef WidthMajorSideFormatNCells4x2<Cells> KernelSideFormat; |
| typedef typename KernelSideFormat::Cell CellFormat; |
| static const int kCells = KernelSideFormat::kCells; |
| static const int kCellWidth = CellFormat::kWidth; |
| static const int kKernelWidth = CellFormat::kWidth * kCells; |
| static const int kCellDepth = CellFormat::kDepth; |
| static const int kCellSize = CellFormat::kSize; |
| |
| typedef NEONPseudoRandomNonzeroBytesGenerator |
| PseudoRandomNonzeroBytesGenerator; |
| |
| template <typename BitDepth, RoundingMode Rounding> |
| void Pack(PackedSideBlock<KernelSideFormat>* dst, int start_width, |
| PseudoRandomNonzeroBytesGenerator* prng) { |
| static const int kBits = BitDepth::kBits; |
| static const std::uint16_t kMaxVal = (1 << kBits) - 1; |
| std::uint8_t* dst_ptr = dst->current_data(); |
| const std::uint8_t* src_ptr = this->complete_src_.data(); |
| const int stride = this->complete_src_.stride(); |
| // Load and requantize source WidthMajor data |
| uint16x8_t src_lines[kCells * 4]; |
| for (int i = 0; i < kCells; i++) { |
| // This packing path is used with our current |
| // less-than-8-bit kernel, and the partial unrolling of this loop |
| // results in substantially faster code (thanks to better |
| // register allocation) on Nexus 5. |
| |
| #define GEMMLOWP_UNROLLED_LOOP_ITER(k) \ |
| src_lines[4 * i + k] = \ |
| vreinterpretq_u16_u8(Requantize<BitDepth, Rounding>( \ |
| vld1q_u8(src_ptr), prng)); \ |
| src_ptr += stride; |
| |
| GEMMLOWP_UNROLLED_LOOP_ITER(0) |
| GEMMLOWP_UNROLLED_LOOP_ITER(1) |
| GEMMLOWP_UNROLLED_LOOP_ITER(2) |
| GEMMLOWP_UNROLLED_LOOP_ITER(3) |
| |
| #undef GEMMLOWP_UNROLLED_LOOP_ITER |
| } |
| // Reorder the data within registers to make WidthMajor 4x2 cells |
| uint16x8x2_t src_lines_intertwined_2x[2 * kCells]; |
| for (int i = 0; i < kCells; i++) { |
| src_lines_intertwined_2x[2 * i] = |
| vzipq_u16(src_lines[4 * i], src_lines[4 * i + 2]); |
| src_lines_intertwined_2x[2 * i + 1] = |
| vzipq_u16(src_lines[4 * i + 1], src_lines[4 * i + 3]); |
| } |
| uint16x8x2_t src_lines_intertwined_4x[2 * kCells]; |
| for (int i = 0; i < kCells; i++) { |
| src_lines_intertwined_4x[2 * i] = |
| vzipq_u16(src_lines_intertwined_2x[2 * i].val[0], |
| src_lines_intertwined_2x[2 * i + 1].val[0]); |
| src_lines_intertwined_4x[2 * i + 1] = |
| vzipq_u16(src_lines_intertwined_2x[2 * i].val[1], |
| src_lines_intertwined_2x[2 * i + 1].val[1]); |
| } |
| // Store the resulting WidthMajor 4x2 cells in the destination packed block |
| for (int outer = 0; outer < 2; outer++) { |
| for (int inner = 0; inner < 2; inner++) { |
| for (int cell = 0; cell < kCells; cell++) { |
| uint8x8_t value = vreinterpret_u8_u16(vget_low_u16( |
| src_lines_intertwined_4x[2 * cell + outer].val[inner])); |
| vst1_u8(dst_ptr, value); |
| dst_ptr += 8; |
| } |
| for (int cell = 0; cell < kCells; cell++) { |
| uint8x8_t value = vreinterpret_u8_u16(vget_high_u16( |
| src_lines_intertwined_4x[2 * cell + outer].val[inner])); |
| vst1_u8(dst_ptr, value); |
| dst_ptr += 8; |
| } |
| } |
| } |
| // Compute sums across the depth dimension |
| uint16x8_t sums_of_2[kCells][4]; |
| for (int outer = 0; outer < 2; outer++) { |
| for (int inner = 0; inner < 2; inner++) { |
| int i = 2 * outer + inner; |
| for (int cell = 0; cell < kCells; cell++) { |
| sums_of_2[cell][i] = vpaddlq_u8(vreinterpretq_u8_u16( |
| src_lines_intertwined_4x[2 * cell + outer].val[inner])); |
| } |
| } |
| } |
| uint16x8_t sums_of_4[kCells][2]; |
| for (int i = 0; i < 2; i++) { |
| for (int cell = 0; cell < kCells; cell++) { |
| sums_of_4[cell][i] = |
| vaddq_u16(sums_of_2[cell][2 * i], sums_of_2[cell][2 * i + 1]); |
| } |
| } |
| uint16x8_t sums_of_8[kCells]; |
| for (int cell = 0; cell < kCells; cell++) { |
| sums_of_8[cell] = vaddq_u16(sums_of_4[cell][0], sums_of_4[cell][1]); |
| } |
| |
| uint16x4_t sums_of_16[kCells]; |
| for (int cell = 0; cell < kCells; cell++) { |
| sums_of_16[cell] = vadd_u16(vget_low_u16(sums_of_8[cell]), |
| vget_high_u16(sums_of_8[cell])); |
| } |
| // Update the rank_one_update vector |
| for (int cell = 0; cell < kCells; cell++) { |
| int32x4_t s = vreinterpretq_s32_u32(vmovl_u16(sums_of_16[cell])); |
| int32x4_t u = vmulq_n_s32(s, dst->rank_one_update_multiplier()); |
| std::int32_t* rank_one_update_ptr = |
| dst->rank_one_update() + start_width + 4 * cell; |
| vst1q_s32(rank_one_update_ptr, |
| vaddq_s32(u, vld1q_s32(rank_one_update_ptr))); |
| } |
| dst->seek_forward_n_cells(kCells * kRegisterSize / kCellDepth); |
| } |
| }; |
| |
| } // namespace gemmlowp |
| |
| #endif // GEMMLOWP_INTERNAL_PACK_NEON_H_ |