Reference CGEMM + test stub

example/19_cgemm/CMakeLists.txt

@@ -0,0 +1 @@
+add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)

example/19_cgemm/cgemm_xdl_bf16.cpp

@@ -0,0 +1,23 @@
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+#include <stdlib.h>
+#include <half.hpp>
+
+#include "check_err.hpp"
+#include "config.hpp"
+#include "device.hpp"
+#include "host_tensor.hpp"
+#include "host_tensor_generator.hpp"
+#include "device_tensor.hpp"
+#include "device_gemm_xdl_cshuffle.hpp"
+#include "element_wise_operation.hpp"
+#include "reference_cgemm.hpp"
+#include "gemm_specialization.hpp"
+
+// stub only
+int main()
+{
+    return 0;
+}

example/CMakeLists.txt

@@ -38,7 +38,8 @@ add_subdirectory(11_conv2d_bwd_weight)
 add_subdirectory(12_reduce)
 add_subdirectory(13_pool2d_fwd)
 add_subdirectory(14_gemm_xdl_requant_relu_requant)
-add_subdirectory(17_convnd_bwd_data_xdl)
 add_subdirectory(15_grouped_gemm)
 add_subdirectory(16_gemm_reduce)
+add_subdirectory(17_convnd_bwd_data_xdl)
 add_subdirectory(18_batched_gemm_reduce)
+add_subdirectory(19_cgemm)