/* * Copyright (c) 2017-2020 ARM Limited. * * SPDX-License-Identifier: MIT * * Permission is hereby granted, free of charge, to any person obtaining a copy * of this software and associated documentation files (the "Software"), to * deal in the Software without restriction, including without limitation the * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or * sell copies of the Software, and to permit persons to whom the Software is * furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice shall be included in all * copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ #ifndef ARM_COMPUTE_NEDIRECTCONVOLUTIONDETAIL_H #define ARM_COMPUTE_NEDIRECTCONVOLUTIONDETAIL_H #include "arm_compute/core/AccessWindowStatic.h" #include "arm_compute/core/NEON/NEFixedPoint.h" #include "arm_compute/core/NEON/wrapper/wrapper.h" #include "arm_compute/core/utils/misc/Requires.h" #include namespace arm_compute { namespace detail { /** Loads a 3x3 matrix as a row (float). * * @param[in] ptr Pointer to a float 3x3 matrix. * @param[in] weights_offset (Optional) Weights quantization offset. * * @return The loaded matrix. */ inline float32x4x3_t load_matrix_row(const float *ptr, int weights_offset = 0) { ARM_COMPUTE_UNUSED(weights_offset); const float32x4x3_t r = { { vld1q_dup_f32(ptr), vld1q_dup_f32(1 + ptr), vld1q_dup_f32(2 + ptr) } }; return r; } /** Loads a 3x3 matrix as a row (uint8_t/int8_t). * * @param[in] ptr Pointer to a uint8_t/int8_t 3x3 matrix. * @param[in] weights_offset (Optional) Weights quantization offset. * * @return The loaded matrix. */ template < typename T, REQUIRES_TA(std::is_same::value || std::is_same::value) > inline int32x4x3_t load_matrix_row(const T *ptr, int weights_offset = 0) { const int32x4_t v_weights_offset = vdupq_n_s32(weights_offset); /* ptr is a pointer to a row in a 3x3 matrix, the function returns 3 vectors holding exactly the same value in all lanes: r.val[0] contains the first element, r.val[1] the second element and r.val[2] the third element (in all lanes) */ int32x4x3_t r = { { vaddq_s32(v_weights_offset, vdupq_n_s32(*ptr)), vaddq_s32(v_weights_offset, vdupq_n_s32(*(ptr + 1))), vaddq_s32(v_weights_offset, vdupq_n_s32(*(ptr + 2))) } }; return r; } /** Stores a float32x4x2_t array into a memory location. * * @param[in] buffer Pointer to the memory location where the values will be stored. * @param[in] values Values that will be stored. * */ template void store_results(float *buffer, const float32x4x2_t &values); template <> inline void store_results<1>(float *buffer, const float32x4x2_t &values) { vst1q_f32(buffer, values.val[0]); vst1q_f32(buffer + 4, values.val[1]); } template <> inline void store_results<2>(float *buffer, const float32x4x2_t &values) { vst1q_f32(buffer, values.val[0]); } template <> inline void store_results<3>(float *buffer, const float32x4x2_t &values) { vst1_f32(buffer, vget_low_f32(values.val[0])); } /** Stores a uint32_t array into a memory location. * * @param[in] buffer Pointer to the memory location where the values will be stored. * @param[in] values Values that will be stored. * */ template void store_results(int32_t *buffer, const int32x4x2_t &values); template <> inline void store_results<1>(int32_t *buffer, const int32x4x2_t &values) { vst1q_s32(buffer, values.val[0]); vst1q_s32(buffer + 4, values.val[1]); } template <> inline void store_results<2>(int32_t *buffer, const int32x4x2_t &values) { vst1q_s32(buffer, values.val[0]); } template <> inline void store_results<3>(int32_t *buffer, const int32x4x2_t &values) { vst1_s32(buffer, vget_low_s32(values.val[0])); } template inline void accumulate_results(float *buffer, const float32x4x2_t &values); template <> inline void accumulate_results<1>(float *buffer, const float32x4x2_t &values) { vst1q_f32(buffer, vaddq_f32(vld1q_f32(buffer), values.val[0])); vst1q_f32(buffer + 4, vaddq_f32(vld1q_f32(buffer + 4), values.val[1])); } template <> inline void accumulate_results<2>(float *buffer, const float32x4x2_t &values) { vst1q_f32(buffer, vaddq_f32(vld1q_f32(buffer), values.val[0])); } template <> inline void accumulate_results<3>(float *buffer, const float32x4x2_t &values) { vst1_f32(buffer, vadd_f32(vld1_f32(buffer), vget_low_f32(values.val[0]))); } template void accumulate_results(int32_t *buffer, const int32x4x2_t &values); template <> inline void accumulate_results<1>(int32_t *buffer, const int32x4x2_t &values) { vst1q_s32(buffer, vaddq_s32(vld1q_s32(buffer), values.val[0])); vst1q_s32(buffer + 4, vaddq_s32(vld1q_s32(buffer + 4), values.val[1])); } template <> inline void accumulate_results<2>(int32_t *buffer, const int32x4x2_t &values) { vst1q_s32(buffer, vaddq_s32(vld1q_s32(buffer), values.val[0])); } template <> inline void accumulate_results<3>(int32_t *buffer, const int32x4x2_t &values) { vst1_s32(buffer, vadd_s32(vld1_s32(buffer), vget_low_s32(values.val[0]))); } #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC /** Stores a float16x8x2_t array into a memory location. * * @param[in] buffer Pointer to the memory location where the values will be stored. * @param[in] values Values that will be stored. * */ template void store_results(float16_t *buffer, const float16x8x2_t &values); template <> inline void store_results<1>(float16_t *buffer, const float16x8x2_t &values) { vst1q_f16(buffer, values.val[0]); vst1q_f16(buffer + 8, values.val[1]); } template <> inline void store_results<2>(float16_t *buffer, const float16x8x2_t &values) { vst1q_f16(buffer, values.val[0]); } template <> inline void store_results<3>(float16_t *buffer, const float16x8x2_t &values) { vst1_f16(buffer, vget_low_f16(values.val[0])); } template inline void accumulate_results(float16_t *buffer, const float16x8x2_t &values); template <> inline void accumulate_results<1>(float16_t *buffer, const float16x8x2_t &values) { vst1q_f16(buffer, vaddq_f16(vld1q_f16(buffer), values.val[0])); vst1q_f16(buffer + 8, vaddq_f16(vld1q_f16(buffer + 8), values.val[1])); } template <> inline void accumulate_results<2>(float16_t *buffer, const float16x8x2_t &values) { vst1q_f16(buffer, vaddq_f16(vld1q_f16(buffer), values.val[0])); } template <> inline void accumulate_results<3>(float16_t *buffer, const float16x8x2_t &values) { vst1_f16(buffer, vadd_f16(vld1_f16(buffer), vget_low_f16(values.val[0]))); } #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ /** Perform a 3x3 convolution for 4 consecutive elements on float32 when dilation.x() or dilation.y() is not 1. * * @param[in] in_top Pointer to the first row of the input. * @param[in] in_mid Pointer to the second row of the input. * @param[in] in_low Pointer to the third row of the input. * @param[in] m0 First row of the filter. * @param[in] m1 Second row of the filter. * @param[in] m2 Third row of the filter. * @param[in] dilation_x Dilation, in elements across x. * @param[in] input_offset (Optional) Input quantization offset. * */ inline float32x4_t single_convolve_3x3_dilation(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2, const size_t dilation_x, int input_offset) { ARM_COMPUTE_UNUSED(input_offset); const float32x4x3_t vtop = { { vld1q_f32(in_top), vld1q_f32(in_top + dilation_x), vld1q_f32(in_top + 2 * dilation_x) } }; const float32x4x3_t vmid = { { vld1q_f32(in_mid), vld1q_f32(in_mid + dilation_x), vld1q_f32(in_mid + 2 * dilation_x) } }; const float32x4x3_t vlow = { { vld1q_f32(in_low), vld1q_f32(in_low + dilation_x), vld1q_f32(in_low + 2 * dilation_x) } }; float32x4_t out = vmulq_f32(vtop.val[0], m0.val[0]); out = vmlaq_f32(out, vtop.val[1], m0.val[1]); out = vmlaq_f32(out, vtop.val[2], m0.val[2]); out = vmlaq_f32(out, vmid.val[0], m1.val[0]); out = vmlaq_f32(out, vmid.val[1], m1.val[1]); out = vmlaq_f32(out, vmid.val[2], m1.val[2]); out = vmlaq_f32(out, vlow.val[0], m2.val[0]); out = vmlaq_f32(out, vlow.val[1], m2.val[1]); out = vmlaq_f32(out, vlow.val[2], m2.val[2]); return out; } /** Perform a 3x3 convolution for 8 consecutive elements on float32 when dilation.x() or dilation.y() is not 1. * * @param[in] in_top Pointer to the first row of the input. * @param[in] in_mid Pointer to the second row of the input. * @param[in] in_low Pointer to the third row of the input. * @param[in] m0 First row of the filter. * @param[in] m1 Second row of the filter. * @param[in] m2 Third row of the filter. * @param[in] dilation_x Dilation, in elements across x. * @param[in] stridex Stride value in elements across x. * @param[in] input_offset (Optional) Input quantization offset. * */ inline float32x4x2_t convolve_3x3_dilation(const float *in_top, const float *in_mid, const float *in_low, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2, const size_t dilation_x, unsigned int stridex, int input_offset = 0) { ARM_COMPUTE_ERROR_ON(stridex > 3); float32x4x2_t out = { { single_convolve_3x3_dilation(in_top, in_mid, in_low, m0, m1, m2, dilation_x, input_offset), single_convolve_3x3_dilation(in_top + 4, in_mid + 4, in_low + 4, m0, m1, m2, dilation_x, input_offset) } }; if(stridex == 2) { out.val[0] = vsetq_lane_f32(vgetq_lane_f32(out.val[0], 2), out.val[0], 1); out.val[0] = vsetq_lane_f32(vgetq_lane_f32(out.val[1], 0), out.val[0], 2); out.val[0] = vsetq_lane_f32(vgetq_lane_f32(out.val[1], 2), out.val[0], 3); } else if(stridex == 3) { out.val[0] = vsetq_lane_f32(vgetq_lane_f32(out.val[0], 3), out.val[0], 1); } return out; } /** Perform a convolve3x3 on float32. * * @param[in] in_top Pointer to the first row of the input. * @param[in] in_mid Pointer to the second row of the input. * @param[in] in_low Pointer to the third row of the input. * @param[out] out_ptr Pointer to the output. * @param[in] m0 First row of the filter. * @param[in] m1 Second row of the filter. * @param[in] m2 Third row of the filter. * @param[in] stridex Stride value in elements across x. * @param[in] input_offset (Optional) Input quantization offset. * */ template void convolve_3x3(const float *in_top, const float *in_mid, const float *in_low, float *out_ptr, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2, unsigned int stridex, int input_offset = 0); template inline void convolve_3x3(const float *in_top, const float *in_mid, const float *in_low, float *out_ptr, const float32x4x3_t &m0, const float32x4x3_t &m1, const float32x4x3_t &m2, unsigned int stridex, int input_offset) { ARM_COMPUTE_UNUSED(input_offset); ARM_COMPUTE_ERROR_ON(stridex > 3); float32x4x2_t out = { { vdupq_n_f32(0.f), vdupq_n_f32(0.f) } }; if(stridex == 2) { const float32x4x2_t vtop = vld2q_f32(in_top); const float32x4x2_t vmid = vld2q_f32(in_mid); const float32x4x2_t vlow = vld2q_f32(in_low); const float32x4_t vtop_end = vld1q_f32(in_top + 8); const float32x4_t vmid_end = vld1q_f32(in_mid + 8); const float32x4_t vlow_end = vld1q_f32(in_low + 8); out.val[0] = vmulq_f32(vtop.val[0], m0.val[0]); out.val[0] = vmlaq_f32(out.val[0], vtop.val[1], m0.val[1]); out.val[0] = vmlaq_f32(out.val[0], vextq_f32(vtop.val[0], vtop_end, 1), m0.val[2]); out.val[0] = vmlaq_f32(out.val[0], vmid.val[0], m1.val[0]); out.val[0] = vmlaq_f32(out.val[0], vmid.val[1], m1.val[1]); out.val[0] = vmlaq_f32(out.val[0], vextq_f32(vmid.val[0], vmid_end, 1), m1.val[2]); out.val[0] = vmlaq_f32(out.val[0], vlow.val[0], m2.val[0]); out.val[0] = vmlaq_f32(out.val[0], vlow.val[1], m2.val[1]); out.val[0] = vmlaq_f32(out.val[0], vextq_f32(vlow.val[0], vlow_end, 1), m2.val[2]); accumulate ? accumulate_results<2>(out_ptr, out) : store_results<2>(out_ptr, out); } else { const float32x4x3_t vtop = { { vld1q_f32(in_top), vld1q_f32(in_top + 4), vld1q_f32(in_top + 8) } }; const float32x4x3_t vmid = { { vld1q_f32(in_mid), vld1q_f32(in_mid + 4), vld1q_f32(in_mid + 8) } }; const float32x4x3_t vlow = { { vld1q_f32(in_low), vld1q_f32(in_low + 4), vld1q_f32(in_low + 8) } }; out.val[0] = vmulq_f32(vtop.val[0], m0.val[0]); out.val[1] = vmulq_f32(vtop.val[1], m0.val[0]); out.val[0] = vmlaq_f32(out.val[0], vextq_f32(vtop.val[0], vtop.val[1], 1), m0.val[1]); out.val[0] = vmlaq_f32(out.val[0], vextq_f32(vtop.val[0], vtop.val[1], 2), m0.val[2]); out.val[0] = vmlaq_f32(out.val[0], vmid.val[0], m1.val[0]); out.val[0] = vmlaq_f32(out.val[0], vextq_f32(vmid.val[0], vmid.val[1], 1), m1.val[1]); out.val[0] = vmlaq_f32(out.val[0], vextq_f32(vmid.val[0], vmid.val[1], 2), m1.val[2]); out.val[0] = vmlaq_f32(out.val[0], vlow.val[0], m2.val[0]); out.val[0] = vmlaq_f32(out.val[0], vextq_f32(vlow.val[0], vlow.val[1], 1), m2.val[1]); out.val[0] = vmlaq_f32(out.val[0], vextq_f32(vlow.val[0], vlow.val[1], 2), m2.val[2]); out.val[1] = vmlaq_f32(out.val[1], vextq_f32(vtop.val[1], vtop.val[2], 1), m0.val[1]); out.val[1] = vmlaq_f32(out.val[1], vextq_f32(vtop.val[1], vtop.val[2], 2), m0.val[2]); out.val[1] = vmlaq_f32(out.val[1], vmid.val[1], m1.val[0]); out.val[1] = vmlaq_f32(out.val[1], vextq_f32(vmid.val[1], vmid.val[2], 1), m1.val[1]); out.val[1] = vmlaq_f32(out.val[1], vextq_f32(vmid.val[1], vmid.val[2], 2), m1.val[2]); out.val[1] = vmlaq_f32(out.val[1], vlow.val[1], m2.val[0]); out.val[1] = vmlaq_f32(out.val[1], vextq_f32(vlow.val[1], vlow.val[2], 1), m2.val[1]); out.val[1] = vmlaq_f32(out.val[1], vextq_f32(vlow.val[1], vlow.val[2], 2), m2.val[2]); if(stridex == 3) { out.val[0] = vsetq_lane_f32(vgetq_lane_f32(out.val[0], 3), out.val[0], 1); accumulate ? accumulate_results<3>(out_ptr, out) : store_results<3>(out_ptr, out); } else { accumulate ? accumulate_results<1>(out_ptr, out) : store_results<1>(out_ptr, out); } } } /** Perform a 3x3 convolution for 4 consecutive 8-bit elements when dilation.x() or dilation.y() is not 1. * * @param[in] in_top Pointer to the first row of the input. * @param[in] in_mid Pointer to the second row of the input. * @param[in] in_low Pointer to the third row of the input. * @param[in] m0 First row of the filter. * @param[in] m1 Second row of the filter. * @param[in] m2 Third row of the filter. * @param[in] dilation_x Dilation, in elements across x. * @param[in] input_offset Input quantization offset. * */ template < typename T, REQUIRES_TA(std::is_same::value || std::is_same::value) > inline int32x4_t single_convolve_3x3_dilation(const T *in_top, const T *in_mid, const T *in_low, const int32x4x3_t &m0, const int32x4x3_t &m1, const int32x4x3_t &m2, size_t dilation_x, int32_t input_offset) { using VectorType = typename std::conditional::value, uint8x8x3_t, int8x8x3_t>::type; using OutputTagType = typename wrapper::traits::neon_bitvector_tag_t; const int32x4_t v_input_offset = wrapper::vdup_n(input_offset, OutputTagType{}); const VectorType vtop = { { wrapper::vload(in_top), wrapper::vload(in_top + dilation_x), wrapper::vload(in_top + 2 * dilation_x) } }; const VectorType vmid = { { wrapper::vload(in_mid), wrapper::vload(in_mid + dilation_x), wrapper::vload(in_mid + 2 * dilation_x) } }; const VectorType vlow = { { wrapper::vload(in_low), wrapper::vload(in_low + dilation_x), wrapper::vload(in_low + 2 * dilation_x) } }; const int32x4x3_t vtop_s32 = { { wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vtop.val[0])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vtop.val[1])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vtop.val[2])))), } }; const int32x4x3_t vmid_s32 = { { wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vmid.val[0])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vmid.val[1])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vmid.val[2])))), } }; const int32x4x3_t vlow_s32 = { { wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vlow.val[0])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vlow.val[1])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vlow.val[2])))), } }; int32x4_t out = wrapper::vmul(vtop_s32.val[0], m0.val[0]); out = wrapper::vmla(out, vtop_s32.val[1], m0.val[1]); out = wrapper::vmla(out, vtop_s32.val[2], m0.val[2]); out = wrapper::vmla(out, vmid_s32.val[0], m1.val[0]); out = wrapper::vmla(out, vmid_s32.val[1], m1.val[1]); out = wrapper::vmla(out, vmid_s32.val[2], m1.val[2]); out = wrapper::vmla(out, vlow_s32.val[0], m2.val[0]); out = wrapper::vmla(out, vlow_s32.val[1], m2.val[1]); out = wrapper::vmla(out, vlow_s32.val[2], m2.val[2]); return out; } /** Perform a 3x3 convolution for 4 consecutive 8-bit elements when dilation.x() or dilation.y() is not 1. * * @param[in] in_top Pointer to the first row of the input. * @param[in] in_mid Pointer to the second row of the input. * @param[in] in_low Pointer to the third row of the input. * @param[in] m0 First row of the filter. * @param[in] m1 Second row of the filter. * @param[in] m2 Third row of the filter. * @param[in] dilation_x Dilation, in elements across x. * @param[in] stridex Stride value in elements across x. * @param[in] input_offset Input quantization offset. * */ template < typename T, REQUIRES_TA(std::is_same::value || std::is_same::value) > inline int32x4x2_t convolve_3x3_dilation(const T *in_top, const T *in_mid, const T *in_low, const int32x4x3_t &m0, const int32x4x3_t &m1, const int32x4x3_t &m2, const size_t dilation_x, unsigned int stridex, int input_offset) { ARM_COMPUTE_ERROR_ON(stridex > 3); int32x4x2_t out = { { single_convolve_3x3_dilation(in_top, in_mid, in_low, m0, m1, m2, dilation_x, input_offset), single_convolve_3x3_dilation(in_top + 4, in_mid + 4, in_low + 4, m0, m1, m2, dilation_x, input_offset) } }; if(stridex == 2) { out.val[0] = wrapper::vsetlane(wrapper::vgetlane(out.val[0], 2), out.val[0], 1); out.val[0] = wrapper::vsetlane(wrapper::vgetlane(out.val[1], 0), out.val[0], 2); out.val[0] = wrapper::vsetlane(wrapper::vgetlane(out.val[1], 2), out.val[0], 3); } else if(stridex == 3) { out.val[0] = wrapper::vsetlane(wrapper::vgetlane(out.val[0], 3), out.val[0], 1); } return out; } /** Perform a convolve3x3 on 8-bit elements * * @param[in] in_top Pointer to the first row of the input. * @param[in] in_mid Pointer to the second row of the input. * @param[in] in_low Pointer to the third row of the input. * @param[out] out_ptr Pointer to the output. * @param[in] m0 First row of the filter. * @param[in] m1 Second row of the filter. * @param[in] m2 Third row of the filter. * @param[in] stridex Stride value in elements across x. * @param[in] input_offset Input quantization offset. * */ template < bool accumulate, typename T1, typename T2, REQUIRES_TA(std::is_same::value || std::is_same::value) > void convolve_3x3(const T1 *in_top, const T1 *in_mid, const T1 *in_low, T2 *out_ptr, const int32x4x3_t &m0, const int32x4x3_t &m1, const int32x4x3_t &m2, unsigned int stridex, int32_t input_offset) { ARM_COMPUTE_ERROR_ON(stridex > 3); using VectorType = typename std::conditional::value, uint8x8x2_t, int8x8x2_t>::type; using OutputTagType = typename wrapper::traits::neon_bitvector_tag_t; const int32x4_t v_input_offset = wrapper::vdup_n(input_offset, OutputTagType{}); const VectorType vtop = { { wrapper::vload(in_top), wrapper::vload(in_top + 8) } }; const VectorType vmid = { { wrapper::vload(in_mid), wrapper::vload(in_mid + 8) } }; const VectorType vlow = { { wrapper::vload(in_low), wrapper::vload(in_low + 8) } }; const int32x4x3_t vtop_s32 = { { wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vtop.val[0])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgethigh(wrapper::vmovl(vtop.val[0])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vtop.val[1])))), } }; const int32x4x3_t vmid_s32 = { { wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vmid.val[0])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgethigh(wrapper::vmovl(vmid.val[0])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vmid.val[1])))), } }; const int32x4x3_t vlow_s32 = { { wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vlow.val[0])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgethigh(wrapper::vmovl(vlow.val[0])))), wrapper::vaddw(v_input_offset, wrapper::vreinterpret(wrapper::vgetlow(wrapper::vmovl(vlow.val[1])))), } }; int32x4x2_t out { { wrapper::vdup_n(static_cast(0), OutputTagType{}), wrapper::vdup_n(static_cast(0), OutputTagType{}), } }; // 0 out.val[0] = wrapper::vmla(out.val[0], vtop_s32.val[0], m0.val[0]); out.val[0] = wrapper::vmla(out.val[0], wrapper::vext_1(vtop_s32.val[0], vtop_s32.val[1]), m0.val[1]); out.val[0] = wrapper::vmla(out.val[0], wrapper::vext_2(vtop_s32.val[0], vtop_s32.val[1]), m0.val[2]); out.val[0] = wrapper::vmla(out.val[0], vmid_s32.val[0], m1.val[0]); out.val[0] = wrapper::vmla(out.val[0], wrapper::vext_1(vmid_s32.val[0], vmid_s32.val[1]), m1.val[1]); out.val[0] = wrapper::vmla(out.val[0], wrapper::vext_2(vmid_s32.val[0], vmid_s32.val[1]), m1.val[2]); out.val[0] = wrapper::vmla(out.val[0], vlow_s32.val[0], m2.val[0]); out.val[0] = wrapper::vmla(out.val[0], wrapper::vext_1(vlow_s32.val[0], vlow_s32.val[1]), m2.val[1]); out.val[0] = wrapper::vmla(out.val[0], wrapper::vext_2(vlow_s32.val[0], vlow_s32.val[1]), m2.val[2]); // 1 out.val[1] = wrapper::vmla(out.val[1], vtop_s32.val[1], m0.val[0]); out.val[1] = wrapper::vmla(out.val[1], wrapper::vext_1(vtop_s32.val[1], vtop_s32.val[2]), m0.val[1]); out.val[1] = wrapper::vmla(out.val[1], wrapper::vext_2(vtop_s32.val[1], vtop_s32.val[2]), m0.val[2]); out.val[1] = wrapper::vmla(out.val[1], vmid_s32.val[1], m1.val[0]); out.val[1] = wrapper::vmla(out.val[1], wrapper::vext_1(vmid_s32.val[1], vmid_s32.val[2]), m1.val[1]); out.val[1] = wrapper::vmla(out.val[1], wrapper::vext_2(vmid_s32.val[1], vmid_s32.val[2]), m1.val[2]); out.val[1] = wrapper::vmla(out.val[1], vlow_s32.val[1], m2.val[0]); out.val[1] = wrapper::vmla(out.val[1], wrapper::vext_1(vlow_s32.val[1], vlow_s32.val[2]), m2.val[1]); out.val[1] = wrapper::vmla(out.val[1], wrapper::vext_2(vlow_s32.val[1], vlow_s32.val[2]), m2.val[2]); if(stridex == 1) { accumulate ? accumulate_results<1>(out_ptr, out) : store_results<1>(out_ptr, out); } else if(stridex == 2) { out.val[0] = wrapper::vsetlane(wrapper::vgetlane(out.val[0], 2), out.val[0], 1); out.val[0] = wrapper::vsetlane(wrapper::vgetlane(out.val[1], 0), out.val[0], 2); out.val[0] = wrapper::vsetlane(wrapper::vgetlane(out.val[1], 2), out.val[0], 3); accumulate ? accumulate_results<2>(out_ptr, out) : store_results<2>(out_ptr, out); } else if(stridex == 3) { out.val[0] = wrapper::vsetlane(wrapper::vgetlane(out.val[0], 3), out.val[0], 1); accumulate ? accumulate_results<3>(out_ptr, out) : store_results<3>(out_ptr, out); } } #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC /** Loads a 3x3 matrix as a row (float16_t). * * @param[in] ptr Pointer to a float 3x3 matrix. * * @return The loaded matrix. */ inline float16x8x3_t load_matrix_row(const float16_t *ptr, int weights_offset = 0) { ARM_COMPUTE_UNUSED(weights_offset); /* ptr is a pointer to a row in a 3x3 matrix, the function returns 3 vectors holding exactly the same value in all lanes: r.val[0] contains the first element, r.val[1] the second element and r.val[2] the third element (in all lanes) */ const float16x8x3_t r = { { vld1q_dup_f16(ptr), vld1q_dup_f16(1 + ptr), vld1q_dup_f16(2 + ptr) } }; return r; } /** Perform a 3x3 convolution for 8 consecutive elements on float16 when dilation.x() or dilation.y() is not 1. * * @param[in] in_top Pointer to the first row of the input. * @param[in] in_mid Pointer to the second row of the input. * @param[in] in_low Pointer to the third row of the input. * @param[in] m0 First row of the filter. * @param[in] m1 Second row of the filter. * @param[in] m2 Third row of the filter. * @param[in] dilation_x Dilation, in elements across x. * @param[in] input_offset (Optional)Input quantization offset. * */ inline float16x8_t single_convolve_3x3_dilation(const float16_t *in_top, const float16_t *in_mid, const float16_t *in_low, const float16x8x3_t &m0, const float16x8x3_t &m1, const float16x8x3_t &m2, const size_t dilation_x, int input_offset = 0) { ARM_COMPUTE_UNUSED(input_offset); const float16x8x3_t vtop = { { vld1q_f16(in_top), vld1q_f16(in_top + dilation_x), vld1q_f16(in_top + 2 * dilation_x) } }; const float16x8x3_t vmid = { { vld1q_f16(in_mid), vld1q_f16(in_mid + dilation_x), vld1q_f16(in_mid + 2 * dilation_x) } }; const float16x8x3_t vlow = { { vld1q_f16(in_low), vld1q_f16(in_low + dilation_x), vld1q_f16(in_low + 2 * dilation_x) } }; float16x8_t out = vmulq_f16(vtop.val[0], m0.val[0]); out = vaddq_f16(out, vmulq_f16(vtop.val[1], m0.val[1])); out = vaddq_f16(out, vmulq_f16(vtop.val[2], m0.val[2])); out = vaddq_f16(out, vmulq_f16(vmid.val[0], m1.val[0])); out = vaddq_f16(out, vmulq_f16(vmid.val[1], m1.val[1])); out = vaddq_f16(out, vmulq_f16(vmid.val[2], m1.val[2])); out = vaddq_f16(out, vmulq_f16(vlow.val[0], m2.val[0])); out = vaddq_f16(out, vmulq_f16(vlow.val[1], m2.val[1])); out = vaddq_f16(out, vmulq_f16(vlow.val[2], m2.val[2])); return out; } /** Perform a 3x3 convolution for 16 consecutive elements on float16 when dilation.x() or dilation.y() is not 1. * * @param[in] in_top Pointer to the first row of the input. * @param[in] in_mid Pointer to the second row of the input. * @param[in] in_low Pointer to the third row of the input. * @param[in] m0 First row of the filter. * @param[in] m1 Second row of the filter. * @param[in] m2 Third row of the filter. * @param[in] dilation_x Dilation, in elements across x. * @param[in] stridex Stride value in elements across x. * @param[in] input_offset (Optional) Input quantization offset. * */ inline float16x8x2_t convolve_3x3_dilation(const float16_t *in_top, const float16_t *in_mid, const float16_t *in_low, const float16x8x3_t &m0, const float16x8x3_t &m1, const float16x8x3_t &m2, const size_t dilation_x, unsigned int stridex, int input_offset = 0) { float16x8x2_t out = { { single_convolve_3x3_dilation(in_top, in_mid, in_low, m0, m1, m2, dilation_x, input_offset), single_convolve_3x3_dilation(in_top + 8, in_mid + 8, in_low + 8, m0, m1, m2, dilation_x, input_offset) } }; if(stridex == 2) { out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 2), out.val[0], 1); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 4), out.val[0], 2); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 6), out.val[0], 3); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 0), out.val[0], 4); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 2), out.val[0], 5); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 4), out.val[0], 6); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 6), out.val[0], 7); } else if(stridex == 3) { out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 3), out.val[0], 1); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 6), out.val[0], 2); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 1), out.val[0], 3); } return out; } /** Perform a convolve3x3 on float16. * * @param[in] in_top Pointer to the first row of the input. * @param[in] in_mid Pointer to the second row of the input. * @param[in] in_low Pointer to the third row of the input. * @param[out] out_ptr Pointer to the output. * @param[in] m0 First row of the filter. * @param[in] m1 Second row of the filter. * @param[in] m2 Third row of the filter. * @param[in] stridex Stride value in elements across x. * @param[in] input_offset (Optional) Input quantization offset. * */ template inline void convolve_3x3(const float16_t *in_top, const float16_t *in_mid, const float16_t *in_low, float16_t *out_ptr, const float16x8x3_t &m0, const float16x8x3_t &m1, const float16x8x3_t &m2, unsigned int stridex, int input_offset = 0) { ARM_COMPUTE_UNUSED(input_offset); float16x8x2_t out = { { vdupq_n_f16(0), vdupq_n_f16(0) } }; if(stridex == 2) { const float16x8x2_t vtop = vld2q_f16(in_top); const float16x8x2_t vmid = vld2q_f16(in_mid); const float16x8x2_t vlow = vld2q_f16(in_low); const float16x8_t vtop_end = vld1q_f16(in_top + 16); const float16x8_t vmid_end = vld1q_f16(in_mid + 16); const float16x8_t vlow_end = vld1q_f16(in_low + 16); out.val[0] = vmulq_f16(vtop.val[0], m0.val[0]); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vtop.val[1], m0.val[1])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vextq_f16(vtop.val[0], vtop_end, 1), m0.val[2])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vmid.val[0], m1.val[0])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vmid.val[1], m1.val[1])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vextq_f16(vmid.val[0], vmid_end, 1), m1.val[2])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vlow.val[0], m2.val[0])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vlow.val[1], m2.val[1])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vextq_f16(vlow.val[0], vlow_end, 1), m2.val[2])); accumulate ? accumulate_results<2>(out_ptr, out) : store_results<2>(out_ptr, out); } else { const float16x8x3_t vtop = { { vld1q_f16(in_top), vld1q_f16(in_top + 8), vld1q_f16(in_top + 16) } }; const float16x8x3_t vmid = { { vld1q_f16(in_mid), vld1q_f16(in_mid + 8), vld1q_f16(in_mid + 16) } }; const float16x8x3_t vlow = { { vld1q_f16(in_low), vld1q_f16(in_low + 8), vld1q_f16(in_low + 16) } }; out.val[0] = vmulq_f16(vtop.val[0], m0.val[0]); out.val[1] = vmulq_f16(vtop.val[1], m0.val[0]); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vextq_f16(vtop.val[0], vtop.val[1], 1), m0.val[1])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vextq_f16(vtop.val[0], vtop.val[1], 2), m0.val[2])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vmid.val[0], m1.val[0])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vextq_f16(vmid.val[0], vmid.val[1], 1), m1.val[1])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vextq_f16(vmid.val[0], vmid.val[1], 2), m1.val[2])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vlow.val[0], m2.val[0])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vextq_f16(vlow.val[0], vlow.val[1], 1), m2.val[1])); out.val[0] = vaddq_f16(out.val[0], vmulq_f16(vextq_f16(vlow.val[0], vlow.val[1], 2), m2.val[2])); out.val[1] = vaddq_f16(out.val[1], vmulq_f16(vextq_f16(vtop.val[1], vtop.val[2], 1), m0.val[1])); out.val[1] = vaddq_f16(out.val[1], vmulq_f16(vextq_f16(vtop.val[1], vtop.val[2], 2), m0.val[2])); out.val[1] = vaddq_f16(out.val[1], vmulq_f16(vmid.val[1], m1.val[0])); out.val[1] = vaddq_f16(out.val[1], vmulq_f16(vextq_f16(vmid.val[1], vmid.val[2], 1), m1.val[1])); out.val[1] = vaddq_f16(out.val[1], vmulq_f16(vextq_f16(vmid.val[1], vmid.val[2], 2), m1.val[2])); out.val[1] = vaddq_f16(out.val[1], vmulq_f16(vlow.val[1], m2.val[0])); out.val[1] = vaddq_f16(out.val[1], vmulq_f16(vextq_f16(vlow.val[1], vlow.val[2], 1), m2.val[1])); out.val[1] = vaddq_f16(out.val[1], vmulq_f16(vextq_f16(vlow.val[1], vlow.val[2], 2), m2.val[2])); if(stridex == 3) { out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 3), out.val[0], 1); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[0], 6), out.val[0], 2); out.val[0] = vsetq_lane_f16(vgetq_lane_f16(out.val[1], 1), out.val[0], 3); accumulate ? accumulate_results<3>(out_ptr, out) : store_results<3>(out_ptr, out); } else { accumulate ? accumulate_results<1>(out_ptr, out) : store_results<1>(out_ptr, out); } } } #endif /** __ARM_FEATURE_FP16_VECTOR_ARITHMETIC **/ /** Get the number of elements processed on 3x3 convolution. * * @param[in] num_elems_written_per_iteration Number of elements written per iteration on 3x3 convolution. * @param[in] stridex Stride value in elements across x. * * @return The number of elements processed. */ inline int get_input_num_elems_processed(unsigned int num_elems_written_per_iteration, unsigned int stridex) { switch(stridex) { case 1: return num_elems_written_per_iteration; case 2: return num_elems_written_per_iteration << 1; case 3: return num_elems_written_per_iteration * 3; default: ARM_COMPUTE_ERROR("stridex not supported"); return 0; } } } } // namespace arm_compute #endif /* ARM_COMPUTE_NEDIRECTCONVOLUTIONDETAIL_H */