From 400aa23e57ab4492227ea95f849cd472d17be0dc Mon Sep 17 00:00:00 2001 From: nihuini Date: Tue, 29 Jun 2021 19:12:19 +0800 Subject: [PATCH] riscv v optimization for convolution sgemm pack1 --- src/layer/riscv/convolution_1x1.h | 26 ++ src/layer/riscv/convolution_1x1_fp16s.h | 26 ++ src/layer/riscv/convolution_riscv.cpp | 52 +++ src/layer/riscv/convolution_sgemm.h | 540 ++++++++++++++++++++++ src/layer/riscv/convolution_sgemm_fp16s.h | 540 ++++++++++++++++++++++ 5 files changed, 1184 insertions(+) create mode 100644 src/layer/riscv/convolution_1x1.h create mode 100644 src/layer/riscv/convolution_1x1_fp16s.h create mode 100644 src/layer/riscv/convolution_sgemm.h create mode 100644 src/layer/riscv/convolution_sgemm_fp16s.h diff --git a/src/layer/riscv/convolution_1x1.h b/src/layer/riscv/convolution_1x1.h new file mode 100644 index 000000000..5566d5437 --- /dev/null +++ b/src/layer/riscv/convolution_1x1.h @@ -0,0 +1,26 @@ +// Tencent is pleased to support the open source community by making ncnn available. +// +// Copyright (C) 2021 THL A29 Limited, a Tencent company. All rights reserved. +// +// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +// in compliance with the License. You may obtain a copy of the License at +// +// https://opensource.org/licenses/BSD-3-Clause +// +// 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. + +static void conv1x1s1_sgemm_rvv(const Mat& bottom_blob, Mat& top_blob, const Mat& kernel, const Mat& _bias, const Option& opt) +{ + int w = bottom_blob.w; + int h = bottom_blob.h; + const int size = w * h; + + Mat bottom_im2col = bottom_blob; + bottom_im2col.w = size; + bottom_im2col.h = 1; + + im2col_sgemm_rvv(bottom_im2col, top_blob, kernel, _bias, opt); +} diff --git a/src/layer/riscv/convolution_1x1_fp16s.h b/src/layer/riscv/convolution_1x1_fp16s.h new file mode 100644 index 000000000..2b652eecb --- /dev/null +++ b/src/layer/riscv/convolution_1x1_fp16s.h @@ -0,0 +1,26 @@ +// Tencent is pleased to support the open source community by making ncnn available. +// +// Copyright (C) 2021 THL A29 Limited, a Tencent company. All rights reserved. +// +// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +// in compliance with the License. You may obtain a copy of the License at +// +// https://opensource.org/licenses/BSD-3-Clause +// +// 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. + +static void conv1x1s1_sgemm_fp16sa_rvv(const Mat& bottom_blob, Mat& top_blob, const Mat& kernel, const Mat& _bias, const Option& opt) +{ + int w = bottom_blob.w; + int h = bottom_blob.h; + const int size = w * h; + + Mat bottom_im2col = bottom_blob; + bottom_im2col.w = size; + bottom_im2col.h = 1; + + im2col_sgemm_fp16sa_rvv(bottom_im2col, top_blob, kernel, _bias, opt); +} diff --git a/src/layer/riscv/convolution_riscv.cpp b/src/layer/riscv/convolution_riscv.cpp index 338a914a8..791327cd6 100644 --- a/src/layer/riscv/convolution_riscv.cpp +++ b/src/layer/riscv/convolution_riscv.cpp @@ -33,6 +33,9 @@ namespace ncnn { +#include "convolution_sgemm.h" +#include "convolution_1x1.h" + #if __riscv_vector #include "convolution_packn.h" #include "convolution_pack1ton.h" @@ -48,6 +51,9 @@ namespace ncnn { #include "convolution_pack1ton_fp16s.h" #include "convolution_packnto1_fp16s.h" +#include "convolution_sgemm_fp16s.h" +#include "convolution_1x1_fp16s.h" + #include "convolution_sgemm_packn_fp16s.h" #include "convolution_1x1_packn_fp16s.h" #include "convolution_3x3_packn_fp16s.h" @@ -194,6 +200,10 @@ int Convolution_riscv::create_pipeline(const Option& opt) // pack1 if (elempack == 1 && out_elempack == 1) { + if (opt.use_sgemm_convolution) + { + convolution_im2col_sgemm_transform_kernel_rvv(weight_data, weight_data_packed, num_input, num_output, kernel_w, kernel_h); + } } return 0; @@ -385,6 +395,25 @@ int Convolution_riscv::forward(const Mat& bottom_blob, Mat& top_blob, const Opti if (elempack == 1 && out_elempack == 1) { + if (opt.use_sgemm_convolution && kernel_w == 1 && kernel_h == 1 && dilation_w == 1 && dilation_h == 1 && stride_w == 1 && stride_h == 1) + { + conv1x1s1_sgemm_rvv(bottom_blob_bordered, top_blob, weight_data_packed, bias_data, opt); + + if (activation) + { + activation->forward_inplace(top_blob, opt); + } + } + else if (opt.use_sgemm_convolution) + { + convolution_im2col_sgemm_rvv(bottom_blob_bordered, top_blob, weight_data_packed, bias_data, kernel_w, kernel_h, dilation_w, dilation_h, stride_w, stride_h, opt); + + if (activation) + { + activation->forward_inplace(top_blob, opt); + } + } + else { const int maxk = kernel_w * kernel_h; @@ -529,6 +558,10 @@ int Convolution_riscv::create_pipeline_fp16s(const Option& opt) // pack1 if (elempack == 1 && out_elempack == 1) { + if (opt.use_fp16_arithmetic && opt.use_sgemm_convolution) + { + convolution_im2col_sgemm_transform_kernel_fp16sa_rvv(weight_data, weight_data_fp16, num_input, num_output, kernel_w, kernel_h); + } } ncnn::cast_float32_to_float16(bias_data, bias_data_fp16, opt); @@ -700,6 +733,25 @@ int Convolution_riscv::forward_fp16sa(const Mat& bottom_blob, Mat& top_blob, con if (elempack == 1 && out_elempack == 1) { + if (opt.use_sgemm_convolution && kernel_w == 1 && kernel_h == 1 && dilation_w == 1 && dilation_h == 1 && stride_w == 1 && stride_h == 1) + { + conv1x1s1_sgemm_fp16sa_rvv(bottom_blob_bordered, top_blob, weight_data_fp16, bias_data_fp16, opt); + + if (activation) + { + activation->forward_inplace(top_blob, opt); + } + } + else if (opt.use_sgemm_convolution) + { + convolution_im2col_sgemm_fp16sa_rvv(bottom_blob_bordered, top_blob, weight_data_fp16, bias_data_fp16, kernel_w, kernel_h, dilation_w, dilation_h, stride_w, stride_h, opt); + + if (activation) + { + activation->forward_inplace(top_blob, opt); + } + } + else { convolution_fp16s(bottom_blob_bordered, top_blob, weight_data_fp16, bias_data, kernel_w, kernel_h, dilation_w, dilation_h, stride_w, stride_h, activation_type, activation_params, opt); } diff --git a/src/layer/riscv/convolution_sgemm.h b/src/layer/riscv/convolution_sgemm.h new file mode 100644 index 000000000..a65610af3 --- /dev/null +++ b/src/layer/riscv/convolution_sgemm.h @@ -0,0 +1,540 @@ +// Tencent is pleased to support the open source community by making ncnn available. +// +// Copyright (C) 2021 THL A29 Limited, a Tencent company. All rights reserved. +// +// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +// in compliance with the License. You may obtain a copy of the License at +// +// https://opensource.org/licenses/BSD-3-Clause +// +// 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. + +static void im2col_sgemm_rvv(const Mat& bottom_im2col, Mat& top_blob, const Mat& kernel, const Mat& _bias, const Option& opt) +{ +#if __riscv_vector + const int packn = csrr_vlenb() / 4; + const word_type vl = vsetvl_e32m1(packn); +#endif + + // Mat bottom_im2col(size, maxk, inch, 4u, 1, opt.workspace_allocator); + + const int size = bottom_im2col.w; + const int maxk = bottom_im2col.h; + const int inch = bottom_im2col.c; + + const int outch = top_blob.c; + + const float* bias = _bias; + + // permute + Mat tmp; +#if __riscv_vector + if (size >= packn) + tmp.create(packn * maxk, inch, size / packn + size % packn, 4u, 1, opt.workspace_allocator); + else + tmp.create(maxk, inch, size, 4u, 1, opt.workspace_allocator); + { + int nn_size = size / packn; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int ii = 0; ii < nn_size; ii++) + { + int i = ii * packn; + + float* tmpptr = tmp.channel(i / packn); + + for (int q = 0; q < inch; q++) + { + const float* img0 = (const float*)bottom_im2col.channel(q) + i; + + for (int k = 0; k < maxk; k++) + { + vse32_v_f32m1(tmpptr, vle32_v_f32m1(img0, vl), vl); + img0 += size; + tmpptr += packn; + } + } + } + + int remain_size_start = nn_size * packn; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int i = remain_size_start; i < size; i++) + { + float* tmpptr = tmp.channel(i / packn + i % packn); + + for (int q = 0; q < inch; q++) + { + const float* img0 = (const float*)bottom_im2col.channel(q) + i; + + for (int k = 0; k < maxk; k++) + { + tmpptr[0] = img0[0]; + img0 += size; + tmpptr += 1; + } + } + } + } +#else // __riscv_vector + tmp.create(maxk, inch, size, 4u, 1, opt.workspace_allocator); + { + #pragma omp parallel for num_threads(opt.num_threads) + for (int i = 0; i < size; i++) + { + float* tmpptr = tmp.channel(i); + + for (int q = 0; q < inch; q++) + { + const float* img0 = (const float*)bottom_im2col.channel(q) + i; + + for (int k = 0; k < maxk; k++) + { + tmpptr[0] = img0[0]; + img0 += size; + tmpptr += 1; + } + } + } + } +#endif // __riscv_vector + +#if __riscv_vector + int nn_outch = outch >> 3; + int remain_outch_start = nn_outch << 3; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int pp = 0; pp < nn_outch; pp++) + { + int p = pp * 8; + + float* outptr0 = top_blob.channel(p); + float* outptr1 = top_blob.channel(p + 1); + float* outptr2 = top_blob.channel(p + 2); + float* outptr3 = top_blob.channel(p + 3); + float* outptr4 = top_blob.channel(p + 4); + float* outptr5 = top_blob.channel(p + 5); + float* outptr6 = top_blob.channel(p + 6); + float* outptr7 = top_blob.channel(p + 7); + + const float zeros[8] = {0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f}; + const float* biasptr = bias ? bias + p : zeros; + + int i = 0; + for (; i + (packn - 1) < size; i += packn) + { + const float* tmpptr = tmp.channel(i / packn); + const float* kptr = kernel.channel(p / 8); + + int nn = inch * maxk; // inch always > 0 + + vfloat32m1_t _sum0 = vfmv_v_f_f32m1(biasptr[0], vl); + vfloat32m1_t _sum1 = vfmv_v_f_f32m1(biasptr[1], vl); + vfloat32m1_t _sum2 = vfmv_v_f_f32m1(biasptr[2], vl); + vfloat32m1_t _sum3 = vfmv_v_f_f32m1(biasptr[3], vl); + vfloat32m1_t _sum4 = vfmv_v_f_f32m1(biasptr[4], vl); + vfloat32m1_t _sum5 = vfmv_v_f_f32m1(biasptr[5], vl); + vfloat32m1_t _sum6 = vfmv_v_f_f32m1(biasptr[6], vl); + vfloat32m1_t _sum7 = vfmv_v_f_f32m1(biasptr[7], vl); + + for (int q = 0; q < nn; q++) + { + vfloat32m1_t _val = vle32_v_f32m1(tmpptr, vl); + _sum0 = vfmacc_vf_f32m1(_sum0, kptr[0], _val, vl); + _sum1 = vfmacc_vf_f32m1(_sum1, kptr[1], _val, vl); + _sum2 = vfmacc_vf_f32m1(_sum2, kptr[2], _val, vl); + _sum3 = vfmacc_vf_f32m1(_sum3, kptr[3], _val, vl); + _sum4 = vfmacc_vf_f32m1(_sum4, kptr[4], _val, vl); + _sum5 = vfmacc_vf_f32m1(_sum5, kptr[5], _val, vl); + _sum6 = vfmacc_vf_f32m1(_sum6, kptr[6], _val, vl); + _sum7 = vfmacc_vf_f32m1(_sum7, kptr[7], _val, vl); + tmpptr += packn; + kptr += 8; + } + + vse32_v_f32m1(outptr0, _sum0, vl); + vse32_v_f32m1(outptr1, _sum1, vl); + vse32_v_f32m1(outptr2, _sum2, vl); + vse32_v_f32m1(outptr3, _sum3, vl); + vse32_v_f32m1(outptr4, _sum4, vl); + vse32_v_f32m1(outptr5, _sum5, vl); + vse32_v_f32m1(outptr6, _sum6, vl); + vse32_v_f32m1(outptr7, _sum7, vl); + + outptr0 += packn; + outptr1 += packn; + outptr2 += packn; + outptr3 += packn; + outptr4 += packn; + outptr5 += packn; + outptr6 += packn; + outptr7 += packn; + } + for (; i < size; i++) + { + const float* tmpptr = tmp.channel(i / packn + i % packn); + const float* kptr = kernel.channel(p / 8); + + int nn = inch * maxk; // inch always > 0 + + float sum0 = biasptr[0]; + float sum1 = biasptr[1]; + float sum2 = biasptr[2]; + float sum3 = biasptr[3]; + float sum4 = biasptr[4]; + float sum5 = biasptr[5]; + float sum6 = biasptr[6]; + float sum7 = biasptr[7]; + + for (int q = 0; q < nn; q++) + { + sum0 += tmpptr[0] * kptr[0]; + sum1 += tmpptr[0] * kptr[1]; + sum2 += tmpptr[0] * kptr[2]; + sum3 += tmpptr[0] * kptr[3]; + sum4 += tmpptr[0] * kptr[4]; + sum5 += tmpptr[0] * kptr[5]; + sum6 += tmpptr[0] * kptr[6]; + sum7 += tmpptr[0] * kptr[7]; + tmpptr++; + kptr += 8; + } + + outptr0[0] = sum0; + outptr1[0] = sum1; + outptr2[0] = sum2; + outptr3[0] = sum3; + outptr4[0] = sum4; + outptr5[0] = sum5; + outptr6[0] = sum6; + outptr7[0] = sum7; + + outptr0++; + outptr1++; + outptr2++; + outptr3++; + outptr4++; + outptr5++; + outptr6++; + outptr7++; + } + } + + nn_outch = (outch - remain_outch_start) >> 2; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int pp = 0; pp < nn_outch; pp++) + { + int p = remain_outch_start + pp * 4; + + float* outptr0 = top_blob.channel(p); + float* outptr1 = top_blob.channel(p + 1); + float* outptr2 = top_blob.channel(p + 2); + float* outptr3 = top_blob.channel(p + 3); + + const float zeros[4] = {0.f, 0.f, 0.f, 0.f}; + const float* biasptr = bias ? bias + p : zeros; + + int i = 0; + for (; i + (packn - 1) < size; i += packn) + { + const float* tmpptr = tmp.channel(i / packn); + const float* kptr = kernel.channel(p / 8 + (p % 8) / 4); + + int nn = inch * maxk; // inch always > 0 + + vfloat32m1_t _sum0 = vfmv_v_f_f32m1(biasptr[0], vl); + vfloat32m1_t _sum1 = vfmv_v_f_f32m1(biasptr[1], vl); + vfloat32m1_t _sum2 = vfmv_v_f_f32m1(biasptr[2], vl); + vfloat32m1_t _sum3 = vfmv_v_f_f32m1(biasptr[3], vl); + + for (int q = 0; q < nn; q++) + { + vfloat32m1_t _val = vle32_v_f32m1(tmpptr, vl); + _sum0 = vfmacc_vf_f32m1(_sum0, kptr[0], _val, vl); + _sum1 = vfmacc_vf_f32m1(_sum1, kptr[1], _val, vl); + _sum2 = vfmacc_vf_f32m1(_sum2, kptr[2], _val, vl); + _sum3 = vfmacc_vf_f32m1(_sum3, kptr[3], _val, vl); + tmpptr += packn; + kptr += 4; + } + + vse32_v_f32m1(outptr0, _sum0, vl); + vse32_v_f32m1(outptr1, _sum1, vl); + vse32_v_f32m1(outptr2, _sum2, vl); + vse32_v_f32m1(outptr3, _sum3, vl); + + outptr0 += packn; + outptr1 += packn; + outptr2 += packn; + outptr3 += packn; + } + for (; i < size; i++) + { + const float* tmpptr = tmp.channel(i / packn + i % packn); + const float* kptr = kernel.channel(p / 8 + (p % 8) / 4); + + int nn = inch * maxk; // inch always > 0 + + float sum0 = biasptr[0]; + float sum1 = biasptr[1]; + float sum2 = biasptr[2]; + float sum3 = biasptr[3]; + + for (int q = 0; q < nn; q++) + { + sum0 += tmpptr[0] * kptr[0]; + sum1 += tmpptr[0] * kptr[1]; + sum2 += tmpptr[0] * kptr[2]; + sum3 += tmpptr[0] * kptr[3]; + tmpptr++; + kptr += 4; + } + + outptr0[0] = sum0; + outptr1[0] = sum1; + outptr2[0] = sum2; + outptr3[0] = sum3; + + outptr0++; + outptr1++; + outptr2++; + outptr3++; + } + } + + remain_outch_start += nn_outch << 2; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int p = remain_outch_start; p < outch; p++) + { + float* outptr0 = top_blob.channel(p); + + const float bias0 = bias ? bias[p] : 0.f; + + int i = 0; + for (; i + (packn - 1) < size; i += packn) + { + const float* tmpptr = tmp.channel(i / packn); + const float* kptr = kernel.channel(p / 8 + (p % 8) / 4 + p % 4); + + int nn = inch * maxk; // inch always > 0 + + vfloat32m1_t _sum0 = vfmv_v_f_f32m1(bias0, vl); + + for (int q = 0; q < nn; q++) + { + _sum0 = vfmacc_vf_f32m1(_sum0, kptr[0], vle32_v_f32m1(tmpptr, vl), vl); + tmpptr += packn; + kptr++; + } + + vse32_v_f32m1(outptr0, _sum0, vl); + + outptr0 += packn; + } + for (; i < size; i++) + { + const float* tmpptr = tmp.channel(i / packn + i % packn); + const float* kptr = kernel.channel(p / 8 + (p % 8) / 4 + p % 4); + + int nn = inch * maxk; // inch always > 0 + + float sum0 = bias0; + + for (int q = 0; q < nn; q++) + { + sum0 += tmpptr[0] * kptr[0]; + tmpptr++; + kptr++; + } + + outptr0[0] = sum0; + + outptr0++; + } + } +#else // __riscv_vector + #pragma omp parallel for num_threads(opt.num_threads) + for (int p = 0; p < outch; p++) + { + float* outptr0 = top_blob.channel(p); + + const float bias0 = bias ? bias[p] : 0.f; + + for (int i = 0; i < size; i++) + { + const float* tmpptr = tmp.channel(i); + const float* kptr = kernel.channel(p); + + int nn = inch * maxk; // inch always > 0 + + float sum0 = bias0; + + for (int q = 0; q < nn; q++) + { + sum0 += tmpptr[0] * kptr[0]; + tmpptr++; + kptr++; + } + + outptr0[0] = sum0; + + outptr0++; + } + } +#endif // __riscv_vector +} + +static void convolution_im2col_sgemm_transform_kernel_rvv(const Mat& _kernel, Mat& kernel_tm, int inch, int outch, int kernel_w, int kernel_h) +{ + const int maxk = kernel_w * kernel_h; + + // interleave + // src = maxk-inch-outch + // dst = 8b-maxk-inch-outch/8b + Mat kernel = _kernel.reshape(maxk, inch, outch); +#if __riscv_vector + kernel_tm.create(8 * maxk, inch, outch / 8 + (outch % 8) / 4 + outch % 4); + + int q = 0; + for (; q + 7 < outch; q += 8) + { + const Mat k0 = kernel.channel(q); + const Mat k1 = kernel.channel(q + 1); + const Mat k2 = kernel.channel(q + 2); + const Mat k3 = kernel.channel(q + 3); + const Mat k4 = kernel.channel(q + 4); + const Mat k5 = kernel.channel(q + 5); + const Mat k6 = kernel.channel(q + 6); + const Mat k7 = kernel.channel(q + 7); + + float* g00 = kernel_tm.channel(q / 8); + + for (int p = 0; p < inch; p++) + { + const float* k00 = k0.row(p); + const float* k10 = k1.row(p); + const float* k20 = k2.row(p); + const float* k30 = k3.row(p); + const float* k40 = k4.row(p); + const float* k50 = k5.row(p); + const float* k60 = k6.row(p); + const float* k70 = k7.row(p); + + for (int k = 0; k < maxk; k++) + { + g00[0] = k00[k]; + g00[1] = k10[k]; + g00[2] = k20[k]; + g00[3] = k30[k]; + g00[4] = k40[k]; + g00[5] = k50[k]; + g00[6] = k60[k]; + g00[7] = k70[k]; + + g00 += 8; + } + } + } + for (; q + 3 < outch; q += 4) + { + const Mat k0 = kernel.channel(q); + const Mat k1 = kernel.channel(q + 1); + const Mat k2 = kernel.channel(q + 2); + const Mat k3 = kernel.channel(q + 3); + + float* g00 = kernel_tm.channel(q / 8 + (q % 8) / 4); + + for (int p = 0; p < inch; p++) + { + const float* k00 = k0.row(p); + const float* k10 = k1.row(p); + const float* k20 = k2.row(p); + const float* k30 = k3.row(p); + + for (int k = 0; k < maxk; k++) + { + g00[0] = k00[k]; + g00[1] = k10[k]; + g00[2] = k20[k]; + g00[3] = k30[k]; + + g00 += 4; + } + } + } + for (; q < outch; q++) + { + const Mat k0 = kernel.channel(q); + + float* g00 = kernel_tm.channel(q / 8 + (q % 8) / 4 + q % 4); + + for (int p = 0; p < inch; p++) + { + const float* k00 = k0.row(p); + + for (int k = 0; k < maxk; k++) + { + g00[0] = k00[k]; + + g00 += 1; + } + } + } +#else + kernel_tm = kernel; +#endif // __riscv_vector +} + +static void convolution_im2col_sgemm_rvv(const Mat& bottom_blob, Mat& top_blob, const Mat& kernel, const Mat& _bias, int kernel_w, int kernel_h, int dilation_w, int dilation_h, int stride_w, int stride_h, const Option& opt) +{ + int w = bottom_blob.w; + int inch = bottom_blob.c; + + int outw = top_blob.w; + int outh = top_blob.h; + const int size = outw * outh; + + const int maxk = kernel_w * kernel_h; + + // im2col + Mat bottom_im2col(size, maxk, inch, 4u, 1, opt.workspace_allocator); + { + const int gap = w * stride_h - outw * stride_w; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int p = 0; p < inch; p++) + { + const Mat img = bottom_blob.channel(p); + float* ptr = bottom_im2col.channel(p); + + for (int u = 0; u < kernel_h; u++) + { + for (int v = 0; v < kernel_w; v++) + { + const float* sptr = img.row(dilation_h * u) + dilation_w * v; + + for (int i = 0; i < outh; i++) + { + int j = 0; + for (; j < outw; j++) + { + ptr[0] = sptr[0]; + + sptr += stride_w; + ptr += 1; + } + + sptr += gap; + } + } + } + } + } + + im2col_sgemm_rvv(bottom_im2col, top_blob, kernel, _bias, opt); +} diff --git a/src/layer/riscv/convolution_sgemm_fp16s.h b/src/layer/riscv/convolution_sgemm_fp16s.h new file mode 100644 index 000000000..bf2d4f2ad --- /dev/null +++ b/src/layer/riscv/convolution_sgemm_fp16s.h @@ -0,0 +1,540 @@ +// Tencent is pleased to support the open source community by making ncnn available. +// +// Copyright (C) 2021 THL A29 Limited, a Tencent company. All rights reserved. +// +// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except +// in compliance with the License. You may obtain a copy of the License at +// +// https://opensource.org/licenses/BSD-3-Clause +// +// 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. + +static void im2col_sgemm_fp16sa_rvv(const Mat& bottom_im2col, Mat& top_blob, const Mat& kernel, const Mat& _bias, const Option& opt) +{ +#if __riscv_vector + const int packn = csrr_vlenb() / 2; + const word_type vl = vsetvl_e16m1(packn); +#endif + + // Mat bottom_im2col(size, maxk, inch, 4u, 1, opt.workspace_allocator); + + const int size = bottom_im2col.w; + const int maxk = bottom_im2col.h; + const int inch = bottom_im2col.c; + + const int outch = top_blob.c; + + const __fp16* bias = _bias; + + // permute + Mat tmp; +#if __riscv_vector + if (size >= packn) + tmp.create(packn * maxk, inch, size / packn + size % packn, 4u, 1, opt.workspace_allocator); + else + tmp.create(maxk, inch, size, 4u, 1, opt.workspace_allocator); + { + int nn_size = size / packn; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int ii = 0; ii < nn_size; ii++) + { + int i = ii * packn; + + __fp16* tmpptr = tmp.channel(i / packn); + + for (int q = 0; q < inch; q++) + { + const __fp16* img0 = (const __fp16*)bottom_im2col.channel(q) + i; + + for (int k = 0; k < maxk; k++) + { + vse16_v_f16m1(tmpptr, vle16_v_f16m1(img0, vl), vl); + img0 += size; + tmpptr += packn; + } + } + } + + int remain_size_start = nn_size * packn; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int i = remain_size_start; i < size; i++) + { + __fp16* tmpptr = tmp.channel(i / packn + i % packn); + + for (int q = 0; q < inch; q++) + { + const __fp16* img0 = (const __fp16*)bottom_im2col.channel(q) + i; + + for (int k = 0; k < maxk; k++) + { + tmpptr[0] = img0[0]; + img0 += size; + tmpptr += 1; + } + } + } + } +#else // __riscv_vector + tmp.create(maxk, inch, size, 4u, 1, opt.workspace_allocator); + { + #pragma omp parallel for num_threads(opt.num_threads) + for (int i = 0; i < size; i++) + { + __fp16* tmpptr = tmp.channel(i); + + for (int q = 0; q < inch; q++) + { + const __fp16* img0 = (const __fp16*)bottom_im2col.channel(q) + i; + + for (int k = 0; k < maxk; k++) + { + tmpptr[0] = img0[0]; + img0 += size; + tmpptr += 1; + } + } + } + } +#endif // __riscv_vector + +#if __riscv_vector + int nn_outch = outch >> 3; + int remain_outch_start = nn_outch << 3; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int pp = 0; pp < nn_outch; pp++) + { + int p = pp * 8; + + __fp16* outptr0 = top_blob.channel(p); + __fp16* outptr1 = top_blob.channel(p + 1); + __fp16* outptr2 = top_blob.channel(p + 2); + __fp16* outptr3 = top_blob.channel(p + 3); + __fp16* outptr4 = top_blob.channel(p + 4); + __fp16* outptr5 = top_blob.channel(p + 5); + __fp16* outptr6 = top_blob.channel(p + 6); + __fp16* outptr7 = top_blob.channel(p + 7); + + const __fp16 zeros[8] = {0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f}; + const __fp16* biasptr = bias ? bias + p : zeros; + + int i = 0; + for (; i + (packn - 1) < size; i += packn) + { + const __fp16* tmpptr = tmp.channel(i / packn); + const __fp16* kptr = kernel.channel(p / 8); + + int nn = inch * maxk; // inch always > 0 + + vfloat16m1_t _sum0 = vfmv_v_f_f16m1(biasptr[0], vl); + vfloat16m1_t _sum1 = vfmv_v_f_f16m1(biasptr[1], vl); + vfloat16m1_t _sum2 = vfmv_v_f_f16m1(biasptr[2], vl); + vfloat16m1_t _sum3 = vfmv_v_f_f16m1(biasptr[3], vl); + vfloat16m1_t _sum4 = vfmv_v_f_f16m1(biasptr[4], vl); + vfloat16m1_t _sum5 = vfmv_v_f_f16m1(biasptr[5], vl); + vfloat16m1_t _sum6 = vfmv_v_f_f16m1(biasptr[6], vl); + vfloat16m1_t _sum7 = vfmv_v_f_f16m1(biasptr[7], vl); + + for (int q = 0; q < nn; q++) + { + vfloat16m1_t _val = vle16_v_f16m1(tmpptr, vl); + _sum0 = vfmacc_vf_f16m1(_sum0, kptr[0], _val, vl); + _sum1 = vfmacc_vf_f16m1(_sum1, kptr[1], _val, vl); + _sum2 = vfmacc_vf_f16m1(_sum2, kptr[2], _val, vl); + _sum3 = vfmacc_vf_f16m1(_sum3, kptr[3], _val, vl); + _sum4 = vfmacc_vf_f16m1(_sum4, kptr[4], _val, vl); + _sum5 = vfmacc_vf_f16m1(_sum5, kptr[5], _val, vl); + _sum6 = vfmacc_vf_f16m1(_sum6, kptr[6], _val, vl); + _sum7 = vfmacc_vf_f16m1(_sum7, kptr[7], _val, vl); + tmpptr += packn; + kptr += 8; + } + + vse16_v_f16m1(outptr0, _sum0, vl); + vse16_v_f16m1(outptr1, _sum1, vl); + vse16_v_f16m1(outptr2, _sum2, vl); + vse16_v_f16m1(outptr3, _sum3, vl); + vse16_v_f16m1(outptr4, _sum4, vl); + vse16_v_f16m1(outptr5, _sum5, vl); + vse16_v_f16m1(outptr6, _sum6, vl); + vse16_v_f16m1(outptr7, _sum7, vl); + + outptr0 += packn; + outptr1 += packn; + outptr2 += packn; + outptr3 += packn; + outptr4 += packn; + outptr5 += packn; + outptr6 += packn; + outptr7 += packn; + } + for (; i < size; i++) + { + const __fp16* tmpptr = tmp.channel(i / packn + i % packn); + const __fp16* kptr = kernel.channel(p / 8); + + int nn = inch * maxk; // inch always > 0 + + __fp16 sum0 = biasptr[0]; + __fp16 sum1 = biasptr[1]; + __fp16 sum2 = biasptr[2]; + __fp16 sum3 = biasptr[3]; + __fp16 sum4 = biasptr[4]; + __fp16 sum5 = biasptr[5]; + __fp16 sum6 = biasptr[6]; + __fp16 sum7 = biasptr[7]; + + for (int q = 0; q < nn; q++) + { + sum0 += tmpptr[0] * kptr[0]; + sum1 += tmpptr[0] * kptr[1]; + sum2 += tmpptr[0] * kptr[2]; + sum3 += tmpptr[0] * kptr[3]; + sum4 += tmpptr[0] * kptr[4]; + sum5 += tmpptr[0] * kptr[5]; + sum6 += tmpptr[0] * kptr[6]; + sum7 += tmpptr[0] * kptr[7]; + tmpptr++; + kptr += 8; + } + + outptr0[0] = sum0; + outptr1[0] = sum1; + outptr2[0] = sum2; + outptr3[0] = sum3; + outptr4[0] = sum4; + outptr5[0] = sum5; + outptr6[0] = sum6; + outptr7[0] = sum7; + + outptr0++; + outptr1++; + outptr2++; + outptr3++; + outptr4++; + outptr5++; + outptr6++; + outptr7++; + } + } + + nn_outch = (outch - remain_outch_start) >> 2; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int pp = 0; pp < nn_outch; pp++) + { + int p = remain_outch_start + pp * 4; + + __fp16* outptr0 = top_blob.channel(p); + __fp16* outptr1 = top_blob.channel(p + 1); + __fp16* outptr2 = top_blob.channel(p + 2); + __fp16* outptr3 = top_blob.channel(p + 3); + + const __fp16 zeros[4] = {0.f, 0.f, 0.f, 0.f}; + const __fp16* biasptr = bias ? bias + p : zeros; + + int i = 0; + for (; i + (packn - 1) < size; i += packn) + { + const __fp16* tmpptr = tmp.channel(i / packn); + const __fp16* kptr = kernel.channel(p / 8 + (p % 8) / 4); + + int nn = inch * maxk; // inch always > 0 + + vfloat16m1_t _sum0 = vfmv_v_f_f16m1(biasptr[0], vl); + vfloat16m1_t _sum1 = vfmv_v_f_f16m1(biasptr[1], vl); + vfloat16m1_t _sum2 = vfmv_v_f_f16m1(biasptr[2], vl); + vfloat16m1_t _sum3 = vfmv_v_f_f16m1(biasptr[3], vl); + + for (int q = 0; q < nn; q++) + { + vfloat16m1_t _val = vle16_v_f16m1(tmpptr, vl); + _sum0 = vfmacc_vf_f16m1(_sum0, kptr[0], _val, vl); + _sum1 = vfmacc_vf_f16m1(_sum1, kptr[1], _val, vl); + _sum2 = vfmacc_vf_f16m1(_sum2, kptr[2], _val, vl); + _sum3 = vfmacc_vf_f16m1(_sum3, kptr[3], _val, vl); + tmpptr += packn; + kptr += 4; + } + + vse16_v_f16m1(outptr0, _sum0, vl); + vse16_v_f16m1(outptr1, _sum1, vl); + vse16_v_f16m1(outptr2, _sum2, vl); + vse16_v_f16m1(outptr3, _sum3, vl); + + outptr0 += packn; + outptr1 += packn; + outptr2 += packn; + outptr3 += packn; + } + for (; i < size; i++) + { + const __fp16* tmpptr = tmp.channel(i / packn + i % packn); + const __fp16* kptr = kernel.channel(p / 8 + (p % 8) / 4); + + int nn = inch * maxk; // inch always > 0 + + __fp16 sum0 = biasptr[0]; + __fp16 sum1 = biasptr[1]; + __fp16 sum2 = biasptr[2]; + __fp16 sum3 = biasptr[3]; + + for (int q = 0; q < nn; q++) + { + sum0 += tmpptr[0] * kptr[0]; + sum1 += tmpptr[0] * kptr[1]; + sum2 += tmpptr[0] * kptr[2]; + sum3 += tmpptr[0] * kptr[3]; + tmpptr++; + kptr += 4; + } + + outptr0[0] = sum0; + outptr1[0] = sum1; + outptr2[0] = sum2; + outptr3[0] = sum3; + + outptr0++; + outptr1++; + outptr2++; + outptr3++; + } + } + + remain_outch_start += nn_outch << 2; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int p = remain_outch_start; p < outch; p++) + { + __fp16* outptr0 = top_blob.channel(p); + + const __fp16 bias0 = bias ? bias[p] : 0.f; + + int i = 0; + for (; i + (packn - 1) < size; i += packn) + { + const __fp16* tmpptr = tmp.channel(i / packn); + const __fp16* kptr = kernel.channel(p / 8 + (p % 8) / 4 + p % 4); + + int nn = inch * maxk; // inch always > 0 + + vfloat16m1_t _sum0 = vfmv_v_f_f16m1(bias0, vl); + + for (int q = 0; q < nn; q++) + { + _sum0 = vfmacc_vf_f16m1(_sum0, kptr[0], vle16_v_f16m1(tmpptr, vl), vl); + tmpptr += packn; + kptr++; + } + + vse16_v_f16m1(outptr0, _sum0, vl); + + outptr0 += packn; + } + for (; i < size; i++) + { + const __fp16* tmpptr = tmp.channel(i / packn + i % packn); + const __fp16* kptr = kernel.channel(p / 8 + (p % 8) / 4 + p % 4); + + int nn = inch * maxk; // inch always > 0 + + __fp16 sum0 = bias0; + + for (int q = 0; q < nn; q++) + { + sum0 += tmpptr[0] * kptr[0]; + tmpptr++; + kptr++; + } + + outptr0[0] = sum0; + + outptr0++; + } + } +#else // __riscv_vector + #pragma omp parallel for num_threads(opt.num_threads) + for (int p = 0; p < outch; p++) + { + __fp16* outptr0 = top_blob.channel(p); + + const __fp16 bias0 = bias ? bias[p] : 0.f; + + for (int i = 0; i < size; i++) + { + const __fp16* tmpptr = tmp.channel(i); + const __fp16* kptr = kernel.channel(p); + + int nn = inch * maxk; // inch always > 0 + + __fp16 sum0 = bias0; + + for (int q = 0; q < nn; q++) + { + sum0 += tmpptr[0] * kptr[0]; + tmpptr++; + kptr++; + } + + outptr0[0] = sum0; + + outptr0++; + } + } +#endif // __riscv_vector +} + +static void convolution_im2col_sgemm_transform_kernel_fp16sa_rvv(const Mat& _kernel, Mat& kernel_tm, int inch, int outch, int kernel_w, int kernel_h) +{ + const int maxk = kernel_w * kernel_h; + + // interleave + // src = maxk-inch-outch + // dst = 8b-maxk-inch-outch/8b + Mat kernel = _kernel.reshape(maxk, inch, outch); +#if __riscv_vector + kernel_tm.create(8 * maxk, inch, outch / 8 + (outch % 8) / 4 + outch % 4, 2u); + + int q = 0; + for (; q + 7 < outch; q += 8) + { + const Mat k0 = kernel.channel(q); + const Mat k1 = kernel.channel(q + 1); + const Mat k2 = kernel.channel(q + 2); + const Mat k3 = kernel.channel(q + 3); + const Mat k4 = kernel.channel(q + 4); + const Mat k5 = kernel.channel(q + 5); + const Mat k6 = kernel.channel(q + 6); + const Mat k7 = kernel.channel(q + 7); + + __fp16* g00 = kernel_tm.channel(q / 8); + + for (int p = 0; p < inch; p++) + { + const float* k00 = k0.row(p); + const float* k10 = k1.row(p); + const float* k20 = k2.row(p); + const float* k30 = k3.row(p); + const float* k40 = k4.row(p); + const float* k50 = k5.row(p); + const float* k60 = k6.row(p); + const float* k70 = k7.row(p); + + for (int k = 0; k < maxk; k++) + { + g00[0] = (__fp16)k00[k]; + g00[1] = (__fp16)k10[k]; + g00[2] = (__fp16)k20[k]; + g00[3] = (__fp16)k30[k]; + g00[4] = (__fp16)k40[k]; + g00[5] = (__fp16)k50[k]; + g00[6] = (__fp16)k60[k]; + g00[7] = (__fp16)k70[k]; + + g00 += 8; + } + } + } + for (; q + 3 < outch; q += 4) + { + const Mat k0 = kernel.channel(q); + const Mat k1 = kernel.channel(q + 1); + const Mat k2 = kernel.channel(q + 2); + const Mat k3 = kernel.channel(q + 3); + + __fp16* g00 = kernel_tm.channel(q / 8 + (q % 8) / 4); + + for (int p = 0; p < inch; p++) + { + const float* k00 = k0.row(p); + const float* k10 = k1.row(p); + const float* k20 = k2.row(p); + const float* k30 = k3.row(p); + + for (int k = 0; k < maxk; k++) + { + g00[0] = (__fp16)k00[k]; + g00[1] = (__fp16)k10[k]; + g00[2] = (__fp16)k20[k]; + g00[3] = (__fp16)k30[k]; + + g00 += 4; + } + } + } + for (; q < outch; q++) + { + const Mat k0 = kernel.channel(q); + + __fp16* g00 = kernel_tm.channel(q / 8 + (q % 8) / 4 + q % 4); + + for (int p = 0; p < inch; p++) + { + const float* k00 = k0.row(p); + + for (int k = 0; k < maxk; k++) + { + g00[0] = (__fp16)k00[k]; + + g00 += 1; + } + } + } +#else + kernel_tm = kernel; +#endif // __riscv_vector +} + +static void convolution_im2col_sgemm_fp16sa_rvv(const Mat& bottom_blob, Mat& top_blob, const Mat& kernel, const Mat& _bias, int kernel_w, int kernel_h, int dilation_w, int dilation_h, int stride_w, int stride_h, const Option& opt) +{ + int w = bottom_blob.w; + int inch = bottom_blob.c; + + int outw = top_blob.w; + int outh = top_blob.h; + const int size = outw * outh; + + const int maxk = kernel_w * kernel_h; + + // im2col + Mat bottom_im2col(size, maxk, inch, 4u, 1, opt.workspace_allocator); + { + const int gap = w * stride_h - outw * stride_w; + + #pragma omp parallel for num_threads(opt.num_threads) + for (int p = 0; p < inch; p++) + { + const Mat img = bottom_blob.channel(p); + __fp16* ptr = bottom_im2col.channel(p); + + for (int u = 0; u < kernel_h; u++) + { + for (int v = 0; v < kernel_w; v++) + { + const __fp16* sptr = img.row(dilation_h * u) + dilation_w * v; + + for (int i = 0; i < outh; i++) + { + int j = 0; + for (; j < outw; j++) + { + ptr[0] = sptr[0]; + + sptr += stride_w; + ptr += 1; + } + + sptr += gap; + } + } + } + } + } + + im2col_sgemm_fp16sa_rvv(bottom_im2col, top_blob, kernel, _bias, opt); +}