feat(dnn): add compile for riscv64

GitOrigin-RevId: fa0c163527

CMakeLists.txt

@@ -117,6 +117,8 @@ if(CMAKE_TOOLCHAIN_FILE)
         else()
             message(FATAL_ERROR "Unsupported IOS_ARCH.")
         endif()
+    elseif(RISCV_TOOLCHAIN_ROOT)
+        set(MGE_ARCH "riscv64")
     elseif(NOT "${ARM_CROSS_BUILD_ARCH}" STREQUAL "")
         set(MGE_ARCH ${ARM_CROSS_BUILD_ARCH})
     else()
@@ -664,6 +666,11 @@ if(MGE_ARCH STREQUAL "aarch64")
 
 endif()
 
+if(MGE_ARCH STREQUAL "riscv64")
+    set(MEGDNN_RISCV64 1)
+    set(MEGDNN_64_BIT 1)
+endif()
+
 set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${MARCH}")
 
 set(MGB_ENABLE_IMPERATIVE ${MGE_BUILD_IMPERATIVE_RT})

dnn/src/common/postprocess.h

@@ -0,0 +1,21 @@
+/**
+ * \file dnn/src/common/postprocess.h
+ * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
+ *
+ * Copyright (c) 2014-2020 Megvii Inc. All rights reserved.
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
+ * implied.
+ */
+#pragma once
+namespace megdnn {
+enum class PostprocessMode : uint8_t {
+    FLOAT = 0,   ///< support all biasmode and no_nonlinemode
+    NO_PROCESS,  ///< support  non bias and identity
+    QUANTIZED,   ///< support  NOBIAS ,BROADCAST_CHANNEL_BIAS and relu hswish
+                 ///< identify nonline mode
+    ADD_BIAS,    ///< only add bias
+};
+}

dnn/src/common/postprocess_helper.h

@@ -0,0 +1,80 @@
+/**
+ * \file dnn/src/common/postprocess_helper.h
+ * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
+ *
+ * Copyright (c) 2014-2020 Megvii Inc. All rights reserved.
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
+ * implied.
+ */
+
+#pragma once
+
+#include "megdnn/basic_types.h"
+#include "midout.h"
+#include "src/common/postprocess.h"
+
+namespace {
+#define POST_PROCESS_UNUSED_VAR()       \
+    MEGDNN_MARK_USED_VAR(conv_dst_ptr); \
+    MEGDNN_MARK_USED_VAR(bias_ptr);     \
+    MEGDNN_MARK_USED_VAR(dst_ptr);      \
+    MEGDNN_MARK_USED_VAR(bias_mode);    \
+    MEGDNN_MARK_USED_VAR(nonlineMode);  \
+    MEGDNN_MARK_USED_VAR(bias_type);    \
+    MEGDNN_MARK_USED_VAR(dst_type);     \
+    MEGDNN_MARK_USED_VAR(N);            \
+    MEGDNN_MARK_USED_VAR(OC);           \
+    MEGDNN_MARK_USED_VAR(OH);           \
+    MEGDNN_MARK_USED_VAR(OW);           \
+    MEGDNN_MARK_USED_VAR(pack_oc_size)
+
+template <typename ctype, typename dtype = ctype,
+          megdnn::PostprocessMode postprocess_mode =
+                  megdnn::PostprocessMode::FLOAT>
+struct PostProcess {
+    static void run(void* conv_dst_ptr, const void* bias_ptr, void* dst_ptr,
+                    megdnn::BiasMode bias_mode, megdnn::NonlineMode nonlineMode,
+                    megdnn::DType bias_type, megdnn::DType dst_type, size_t N,
+                    size_t OC, size_t OH, size_t OW, size_t pack_oc_size = 1) {
+        POST_PROCESS_UNUSED_VAR();
+        megdnn_throw("not impl PostProcess");
+    }
+};
+
+template <typename ctype, typename dtype>
+struct PostProcess<ctype, dtype, megdnn::PostprocessMode::NO_PROCESS> {
+    static void run(void* conv_dst_ptr, void* bias_ptr, void* dst_ptr,
+                    megdnn::BiasMode bias_mode, megdnn::NonlineMode nonlineMode,
+                    megdnn::DType bias_type, megdnn::DType dst_type, size_t N,
+                    size_t OC, size_t OH, size_t OW, size_t pack_oc_size = 1) {
+        POST_PROCESS_UNUSED_VAR();
+        megdnn_throw("not impl PostProcess");
+    }
+};
+
+template <typename opctype, typename opdtype>
+struct PostProcess<opctype, opdtype, megdnn::PostprocessMode::QUANTIZED> {
+    static void run(void* conv_dst_ptr, const void* bias_ptr, void* dst_ptr,
+                    megdnn::BiasMode bias_mode, megdnn::NonlineMode nonlineMode,
+                    megdnn::DType bias_type, megdnn::DType dst_type, size_t N,
+                    size_t OC, size_t OH, size_t OW, size_t pack_oc_size = 1) {
+        POST_PROCESS_UNUSED_VAR();
+        megdnn_throw("not impl PostProcess");
+    }
+};
+
+template <typename ctype, typename dtype>
+struct PostProcess<ctype, dtype, megdnn::PostprocessMode::ADD_BIAS> {
+    static void run(void* conv_dst_ptr, void* bias_ptr, void* dst_ptr,
+                    megdnn::BiasMode bias_mode, megdnn::NonlineMode nonlineMode,
+                    megdnn::DType bias_type, megdnn::DType dst_type, size_t N,
+                    size_t OC, size_t OH, size_t OW, size_t pack_oc_size = 1) {
+        POST_PROCESS_UNUSED_VAR();
+        megdnn_throw("not impl PostProcess");
+    }
+};
+
+}  // namespace

dnn/src/common/relayout_helper.h

@@ -6,7 +6,8 @@
  *
  * Unless required by applicable law or agreed to in writing,
  * software distributed under the License is distributed on an
- * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
+ * implied.
  */
 #pragma once
 
@@ -42,8 +43,12 @@ namespace transpose_fallback {
 
 #if MEGDNN_X86
 constexpr size_t BLOCK_LINE_SIZE_BYTES = 64;
-#elif MEGDNN_AARCH64 || MEGDNN_ARMV7 
+#elif MEGDNN_AARCH64 || MEGDNN_ARMV7 /*BEGIN-INLINE-INTERNAL*/ || \
+        MEGDNN_MIPS /*END-INLINE-INTERNAL*/
 constexpr size_t BLOCK_LINE_SIZE_BYTES = 32;
+#elif MEGDNN_RISCV64
+//! ref U54-MC arch
+constexpr size_t BLOCK_LINE_SIZE_BYTES = 64;
 #else
 #error "unknown megdnn arch"
 #endif

dnn/src/fallback/conv_bias/common.h

@@ -6,12 +6,14 @@
  *
  * Unless required by applicable law or agreed to in writing,
  * software distributed under the License is distributed on an
- * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
+ * implied.
  */
 #pragma once
 
 #include <stdint.h>
 #include "megdnn/oprs.h"
+#include "src/common/postprocess.h"
 #include "src/common/utils.h"
 
 namespace megdnn {
@@ -157,13 +159,6 @@ private:                                                                       \
     mutable std::string m_name;                                                \
     uint32_t m_tile_size;
 
-enum class PostprocessMode : uint8_t {
-    FLOAT = 0,   ///< support all biasmode and no_nonlinemode
-    NO_PROCESS,  ///< support  non bias and identity
-    QUANTIZED,   ///< support  NOBIAS ,BROADCAST_CHANNEL_BIAS and relu hswish
-                 ///< identify nonline mode
-    ADD_BIAS,    ///< only add bias
-};
 }  // namespace megdnn
 
 // vim: syntax=cpp.doxygen

dnn/src/fallback/conv_bias/conv1x1/algos.cpp

@@ -24,6 +24,8 @@
 #include "src/x86/conv_bias/postprocess_helper.h"
 #elif (MEGDNN_ARMV7 || MEGDNN_AARCH64)
 #include "src/arm_common/conv_bias/postprocess_helper.h"
+#else
+#include "src/common/postprocess_helper.h"
 #endif
 
 #include "midout.h"
@@ -106,7 +108,7 @@ ConvBiasImpl::AlgoConv1x1::get_kerns_according_packmode(
 
     WorkspaceBundle whole_bundle = get_bundle_according_packmode(param);
     //! NO_PACK not implement get_bundle
-    WorkspaceBundle matmul_bundle ={nullptr,{}};
+    WorkspaceBundle matmul_bundle = {nullptr, {}};
     if (pack_mode == MatrixMulImpl::AlgoBase::PackMode::NO_PACK) {
         matmul_bundle = {nullptr,
                          {0, 0, m_matmul_algo->get_workspace(matmul_param)}};
@@ -281,7 +283,6 @@ bool ConvBiasImpl::AlgoConv1x1::usable(const NCBKernSizeParam& param,
     return false;
 }
 
-
 bool ConvBiasImpl::AlgoConv1x1::is_preferred(
         const NCBKernSizeParam& param) const {
     size_t OH = param.osz[0];

dnn/src/fallback/conv_bias/conv1x1/algos_conv1x1_gemv.cpp

@@ -25,9 +25,11 @@
 #include "src/x86/conv_bias/postprocess_helper.h"
 #elif (MEGDNN_ARMV7 || MEGDNN_AARCH64)
 #include "src/arm_common/conv_bias/postprocess_helper.h"
-#include "src/arm_common/matrix_mul/fp32/exec_sgemv.h"
 #include "src/arm_common/matrix_mul/fp16/hgemv.h"
+#include "src/arm_common/matrix_mul/fp32/exec_sgemv.h"
 #include "src/arm_common/matrix_mul/int8/gemv.h"
+#else
+#include "src/common/postprocess_helper.h"
 #endif
 
 #include "midout.h"
@@ -249,7 +251,7 @@ size_t ConvBiasImpl::AlgoConv1x1Gemv::get_oc_tile_size_heuristic(
 }
 
 size_t ConvBiasImpl::AlgoConv1x1Gemv::get_workspace(
-         const NCBKernSizeParam& param) const {
+        const NCBKernSizeParam& param) const {
     MIDOUT_BEGIN(megdnn_fallback_conv1x1_gemv,
                  midout_iv("AlgoConv1x1Gemv::get_workspace"_hash)) {
         size_t compt_oc_block_size = get_oc_tile_size_heuristic(param);
@@ -335,7 +337,8 @@ ConvBiasImpl::AlgoConv1x1Gemv::dispatch_kerns(
 #else
 #if !MEGDNN_DISABLE_FLOAT16
             cb1(param::ConvBias::Format::NCHW, dt_float16, dt_float16,
-                PostprocessMode::NO_PROCESS, "NCHW::GEMV::FLOAT16_FLOAT16"_hash);
+                PostprocessMode::NO_PROCESS,
+                "NCHW::GEMV::FLOAT16_FLOAT16"_hash);
 #endif
 #endif
             cb3(param::ConvBias::Format::NCHW, dt_int8, dt_int32, dt_int32,
@@ -361,7 +364,7 @@ ConvBiasImpl::AlgoConv1x1Gemv::dispatch_kerns(
                 dt_uint8, PostprocessMode::QUANTIZED,
                 "NCHW::GEMV::QUINT8x8x32_QUINT8"_hash);
             break;
-        //!no support nchw44 8x8x16
+        //! no support nchw44 8x8x16
         case param::ConvBias::Format::NCHW44:
             cb1(param::ConvBias::Format::NCHW44, dt_float32, dt_float32,
                 PostprocessMode::FLOAT, "NCHW44::GEMV::FLOAT"_hash);
@@ -377,7 +380,7 @@ ConvBiasImpl::AlgoConv1x1Gemv::dispatch_kerns(
                 dt_int8, PostprocessMode::QUANTIZED,
                 "NCHW44::GEMV::QINT8x8x32_QINT8"_hash);
             break;
-        //!no support nchw44-dot 8x8x16
+        //! no support nchw44-dot 8x8x16
         case param::ConvBias::Format::NCHW44_DOT:
             cb3(param::ConvBias::Format::NCHW44_DOT, dt_int8, dt_int32,
                 dt_int32, dt_int8, dt_int32, dt_int32,

dnn/src/fallback/conv_bias/conv1x1/conv1x1_strategy.h

@@ -19,6 +19,8 @@
 #include "src/x86/conv_bias/postprocess_helper.h"
 #elif (MEGDNN_ARMV7 || MEGDNN_AARCH64)
 #include "src/arm_common/conv_bias/postprocess_helper.h"
+#else
+#include "src/common/postprocess_helper.h"
 #endif
 
 namespace megdnn {

dnn/src/fallback/conv_bias/im2col/strategy_base.h

@@ -16,6 +16,8 @@
 #include "src/x86/conv_bias/postprocess_helper.h"
 #elif (MEGDNN_ARMV7 || MEGDNN_AARCH64)
 #include "src/arm_common/conv_bias/postprocess_helper.h"
+#else
+#include "src/common/postprocess_helper.h"
 #endif
 using namespace megdnn;
 #if MEGDNN_X86

dnn/src/fallback/conv_bias/im2col/strategy_default_nchw44.cpp

@@ -12,10 +12,10 @@
 #include "src/fallback/convolution/img2col_helper.h"
 #if MEGDNN_X86
 #include "src/x86/conv_bias/postprocess_helper.h"
-#endif
-
-#if (MEGDNN_ARMV7 || MEGDNN_AARCH64)
+#elif (MEGDNN_ARMV7 || MEGDNN_AARCH64)
 #include "src/arm_common/conv_bias/postprocess_helper.h"
+#else
+#include "src/common/postprocess_helper.h"
 #endif
 
 using namespace megdnn;

dnn/src/fallback/conv_bias/opr_impl.cpp

@@ -74,6 +74,10 @@ public:
             }
 #endif
 
+//! As we haven't riscv64 postprocess yet, im2col and conv1x1 can not pass ci
+//! test. so we just disable all im2col and conv1x1 in riscv64
+//! FIXME: remove it when impl postprocess for riscv64
+#if !MEGDNN_RISCV64
             for (size_t ohw_tile_size : {192, 384, 96, 48, 24}) {
                 refhold.emplace_back(new AlgoIm2col(
                         static_cast<MatrixMulImpl::AlgoBase*>(algo),
@@ -86,6 +90,8 @@ public:
                         oc_tile_size));
                 all_algos.emplace_back(refhold.back().get());
             }
+#endif
+
 #if 0
         //! As these algos maybe very slow, it will make fastrun search slow, so
         //! we disable it, but for the test of strategyhelper, we just keep it.

dnn/src/fallback/conv_bias/opr_impl.h

@@ -50,6 +50,7 @@ public:
               _megdnn_tensor_in bias, _megdnn_tensor_in z,
               _megdnn_tensor_out dst, const PreprocessedFilter*,
               _megdnn_workspace workspace) override;
+    bool is_thread_safe() const override { return true; }
 
     void exec_preprocess(const TensorLayout& src_layout,
                          _megdnn_tensor_in filter,

dnn/test/common/mask_conv.h

@@ -74,7 +74,7 @@ void mask_conv_test(Handle* handle) {
             arg[8], arg[9], arg[10], arg[11], arg[12]);
     }
 }
-
+#if MEGDNN_WITH_BENCHMARK
 void mask_conv_benchmark(Handle* handle) {
     auto benchmark = [&](size_t N, size_t IC, size_t OC, size_t IH, size_t IW,
                          size_t FH, size_t FW, size_t SH, size_t SW, size_t PH,
@@ -113,5 +113,6 @@ void mask_conv_benchmark(Handle* handle) {
                   arg[7], arg[8], arg[9], arg[10], arg[11], arg[12]);
     }
 }
+#endif
 
 }  // namespace

dnn/test/cpu/mask_conv.cpp

@@ -25,9 +25,11 @@ TEST_F(CPU, MASK_CONV) {
     mask_conv_test(handle());
 }
 
+#if MEGDNN_WITH_BENCHMARK
 TEST_F(CPU, MASK_CONV_BENCHMARK) {
     mask_conv_benchmark(handle());
 }
+#endif
 
 TEST_F(CPU, MASK_PROPAGATE) {
     param::MaskPropagate mask_param;

dnn/test/cpu/matrix_mul.cpp

@@ -17,6 +17,7 @@
 
 using namespace megdnn;
 using namespace test;
+#if MEGDNN_WITH_BENCHMARK
 namespace {
 
 void sgemm_sgemv_like(const float* __restrict A, const float* __restrict B,
@@ -70,6 +71,7 @@ TEST_F(CPU, BENCHMARK_MATRIX_MUL) {
             run(m, nk, nk);
         }
 }
+#endif
 
 TEST_F(CPU, MATRIX_MUL) {
     matrix_mul::check_matrix_mul(dtype::Float32{}, dtype::Float32{},

dnn/test/cpu/relayout.cpp

@@ -31,6 +31,7 @@ TYPED_TEST(CPU_RELAYOUT, run) {
 }
 }
 
+#if MEGDNN_WITH_BENCHMARK
 TEST_F(CPU, BENCHMARK_RELAYOUT_CV) {
     relayout::run_cv_benchmark(handle());
 }
@@ -55,6 +56,6 @@ TEST_F(CPU, BENCHMARK_RELAYOUT) {
         ASSERT_LE(cpu_time * 5, naive_time);
     }
 }
-
+#endif
 
 // vim: syntax=cpp.doxygen

dnn/test/cuda/mask_conv.cpp

@@ -22,10 +22,11 @@ using namespace test;
 TEST_F(CUDA, MASK_CONV) {
     mask_conv_test(handle_cuda());
 }
-
+#if MEGDNN_WITH_BENCHMARK
 TEST_F(CUDA, MASK_CONV_BENCHMARK) {
     mask_conv_benchmark(handle_cuda());
 }
+#endif
 
 TEST_F(CUDA, MASK_PROPAGATE) {
     Checker<MaskPropagate> checker(handle_cuda());

dnn/test/fallback/elemwise.cpp

@@ -27,7 +27,7 @@ TYPED_TEST_CASE(FALLBACK_ELEMWISE, elemwise::test_types);
 TYPED_TEST(FALLBACK_ELEMWISE, run) {
     elemwise::run_test<TypeParam>(this->handle());
 }
-
+#if MEGDNN_WITH_BENCHMARK
 TEST_F(FALLBACK, BENCHMARK_ELEMWISE) {
     auto naive_handle = create_cpu_handle(2);
     auto run = [&](const TensorShape &shp0, const TensorShape &shp1) {
@@ -72,6 +72,7 @@ TEST_F(FALLBACK, BENCHMARK_ELEMWISE) {
     // non-contig, fallback to naive
     run({1024, 1024, 32}, {1024, 1, 32});
 }
+#endif
 
 // vim: syntax=cpp.doxygen
 

dnn/test/fallback/elemwise_multi_type.cpp

@@ -25,7 +25,7 @@ TYPED_TEST_CASE(FALLBACK_ELEMWISE_MULTI_TYPE, elemwise_multi_type::test_types);
 TYPED_TEST(FALLBACK_ELEMWISE_MULTI_TYPE, run) {
     elemwise_multi_type::run_test<TypeParam>(this->handle());
 }
-
+#if MEGDNN_WITH_BENCHMARK
 TEST_F(FALLBACK, ELEMWISE_MULTI_TYPE_BENCHMARK_FMA3_INT16x32x32x32) {
     Benchmarker<ElemwiseMultiType> bench{handle()};
     bench.set_param({ElemwiseMultiType::Mode::FUSE_MUL_ADD3_INT16x32x32x32});
@@ -64,5 +64,5 @@ TEST_F(FALLBACK, ELEMWISE_MULTI_TYPE_BENCHMARK_FMA3_IXxf32xf32xI8) {
                        (1024.0 * 1024.0 * 1024.0));
     }
 }
-
+#endif
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}

dnn/test/fallback/relayout.cpp

@@ -31,7 +31,7 @@ TYPED_TEST(FALLBACK_RELAYOUT, run) {
     relayout::run_test<TypeParam>(this->handle());
 }
 }
-
+#if MEGDNN_WITH_BENCHMARK
 TEST_F(FALLBACK, BENCHMARK_RELAYOUT_CV) {
     relayout::run_cv_benchmark(handle());
 }
@@ -160,5 +160,6 @@ TEST_F(FALLBACK, BENCHMARK_RELAYOUT) {
         }
     }
 }
+#endif
 
 // vim: syntax=cpp.doxygen

dnn/test/fallback/roi_copy.cpp

@@ -34,7 +34,7 @@ TEST_F(FALLBACK, ROICOPY) {
     }
 
 }
-
+#if MEGDNN_WITH_BENCHMARK
 TEST_F(FALLBACK, BENCHMARK_ROICOPY) {
     auto run = [&](const TensorShapeArray& shapes) {
         Benchmarker<ROICopy> benchmarker(handle());
@@ -62,6 +62,7 @@ TEST_F(FALLBACK, BENCHMARK_ROICOPY) {
 
     run(shapes);
 }
+#endif
 
 
 }  // namespace test

toolchains/riscv64-linux-gnu.toolchain.cmake

@@ -0,0 +1,18 @@
+set(CMAKE_SYSTEM_NAME Linux)
+set(CMAKE_SYSTEM_PROCESSOR riscv64)
+set(RISCV_CROSS_BUILD_ARCH riscv64)
+
+if(DEFINED ENV{RISCV_TOOLCHAIN_ROOT})
+    file(TO_CMAKE_PATH $ENV{RISCV_TOOLCHAIN_ROOT} RISCV_TOOLCHAIN_ROOT)
+else()
+    message(FATAL_ERROR "RISCV_TOOLCHAIN_ROOT env must be defined")
+endif()
+
+set(RISCV_TOOLCHAIN_ROOT ${RISCV_TOOLCHAIN_ROOT} CACHE STRING "root path to riscv toolchain")
+
+set(CMAKE_C_COMPILER "${RISCV_TOOLCHAIN_ROOT}/bin/riscv64-unknown-linux-gnu-gcc")
+set(CMAKE_CXX_COMPILER "${RISCV_TOOLCHAIN_ROOT}/bin/riscv64-unknown-linux-gnu-g++")
+set(CMAKE_FIND_ROOT_PATH "${RISCV_TOOLCHAIN_ROOT}/riscv64-unknown-linux-gnu")
+set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
+set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)