Add init impl for riscv64 matmul.

Author: zhangyang2057

CMakeLists.txt

@@ -51,7 +51,6 @@ option(BUILD_BENCHMARK "Build benchmark programs" ON)
 option(BUILD_TESTING "Build test programs" OFF)
 option(ENABLE_OP_PROFILE "Profile ops cast time" OFF)
 option(ENABLE_DUMP_MANAGER "Enable dump manager" OFF)
-option(ENABLE_RVV "Some kernel impl by rvv" OFF)
 option(ENABLE_DUMP_MEM "Dump mem usage" OFF)
 
 if (BUILDING_RUNTIME)

src/Native/src/kernels/CMakeLists.txt

@@ -3,12 +3,6 @@
 set(SRCS kernel_context.cpp)
 
 if (BUILDING_RUNTIME)
-    # used for rvv
-    if(ENABLE_RVV)
-        add_definitions(-D__riscv_vector=1)
-        add_compile_options(-march=rv64imafdcv)
-    endif()
-
     add_library(kernels OBJECT ${SRCS})
     target_include_directories(kernels PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
     target_link_libraries(kernels PUBLIC gsl::gsl-lite)
@@ -25,9 +19,9 @@ else()
     #     if(os_name STREQUAL "linux")
     #         target_link_libraries(kernels PRIVATE -lpthread)
     #     endif()
-    #     target_compile_definitions(kernels PRIVATE "-DNNCASE_HALIDE")        
+    #     target_compile_definitions(kernels PRIVATE "-DNNCASE_HALIDE")
     # endif()
-    
+
     target_compile_definitions(kernels PUBLIC -DNNCASE_DLL -DNNCASE_SIMULATOR)
     set_property(TARGET kernels PROPERTY POSITION_INDEPENDENT_CODE ON)
 endif()

src/Native/src/kernels/stackvm/optimized/CMakeLists.txt

@@ -45,7 +45,7 @@ _TARGET_ARCH_FILES(TARGET kernels
                    FILES
                    binary.cpp
                    layer_norm.cpp
-                   #matmul.cpp
+                   matmul.cpp
                    sigmoid.cpp
                    softmax.cpp
                    unary.cpp

src/Native/src/kernels/stackvm/optimized/matmul.cpp

@@ -0,0 +1,33 @@
+/* Copyright 2019-2021 Canaan Inc.
+ *
+ * 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.
+ */
+#include "../reference/ref_ops.h"
+#include "opt_ops.h"
+#include <iostream>
+#include <nncase/kernels/kernel_utils.h>
+#include <nncase/runtime/runtime_op_utility.h>
+
+using namespace nncase;
+using namespace nncase::runtime;
+using namespace nncase::kernels;
+using namespace nncase::kernels::stackvm;
+using namespace nncase::kernels::stackvm::optimized;
+
+result<void> optimized::matmul(typecode_t typecode, const gsl::byte *input_a, const gsl::byte *input_b,
+                                gsl::byte *output,
+                                gsl::span<const size_t> in_a_shape,
+                                gsl::span<const size_t> in_b_shape,
+                                [[maybe_unused]] kernel_context &context) noexcept {
+    return stackvm::reference::matmul(typecode, input_a, input_b, output, in_a_shape, in_b_shape, context);
+}

src/Native/src/kernels/stackvm/optimized/opt_ops.h

@@ -139,15 +139,11 @@ unary(typecode_t dtype, runtime::stackvm::unary_op_t op, const gsl::byte *in,
       gsl::span<const size_t> out_strides,
       kernel_context &context = default_kernel_context()) noexcept;
 
-// template <typename T>
-// NNCASE_API result<void> matmul(const T *input_a, const T *input_b, const T
-// *bias, T *output,
-//                               gsl::span<const size_t> in_a_shape, const
-//                               dims_t &in_a_strides, gsl::span<const size_t>
-//                               in_b_shape, const dims_t &in_b_strides,
-//                               gsl::span<const size_t> out_shape,
-//                               gsl::span<const size_t> out_strides,
-//                               value_range<float> fused_activation) noexcept;
+NNCASE_API result<void>
+matmul(typecode_t typecode, const gsl::byte *input_a, const gsl::byte *input_b,
+    gsl::byte *output, gsl::span<const size_t> in_a_shape,
+    gsl::span<const size_t> in_b_shape,
+    [[maybe_unused]] kernel_context &context) noexcept;
 
 // template <typename T>
 NNCASE_API result<void>