Merge branch 'develop' into amd-develop

test/gemm_universal/test_gemm_universal_xdl.cpp

@@ -7,9 +7,10 @@
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "test_gemm_universal_util.hpp"
 
-using F8  = ck::f8_t;
-using F16 = ck::half_t;
-using F32 = float;
+using F8   = ck::f8_t;
+using F16  = ck::half_t;
+using BF16 = ck::bhalf_t;
+using F32  = float;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -44,7 +45,8 @@ using KernelTypes = ::testing::Types<
     //         ADataType, BDataType, CDataType
     std::tuple<      F16,       F16,       F16>,
     std::tuple<      F16,        F8,       F16>,
-    std::tuple<      F8,        F16,       F16>
+    std::tuple<      F8,        F16,       F16>,
+    std::tuple<     BF16,      BF16,      BF16>
     >;
 // clang-format on
 

test/normalization_bwd_data/CMakeLists.txt

 add_custom_target(test_normalization_bwd_data)
+
 add_gtest_executable(test_layernorm2d_bwd_data_fp32 test_layernorm2d_bwd_data_fp32.cpp)
-target_link_libraries(test_layernorm2d_bwd_data_fp32 PRIVATE utility device_normalization_bwd_data_instance)
-add_dependencies(test_normalization_bwd_data test_layernorm2d_bwd_data_fp32)
+if (result EQUAL 0)
+    target_link_libraries(test_layernorm2d_bwd_data_fp32 PRIVATE utility device_normalization_bwd_data_instance)
+    add_dependencies(test_normalization_bwd_data test_layernorm2d_bwd_data_fp32)
+endif()
 
 add_gtest_executable(test_groupnorm_bwd_data_fp32 test_groupnorm_bwd_data_fp32.cpp)
-target_link_libraries(test_groupnorm_bwd_data_fp32 PRIVATE utility device_normalization_bwd_data_instance)
-add_dependencies(test_normalization_bwd_data test_groupnorm_bwd_data_fp32)
+if (result EQUAL 0)
+    target_link_libraries(test_groupnorm_bwd_data_fp32 PRIVATE utility device_normalization_bwd_data_instance)
+    add_dependencies(test_normalization_bwd_data test_groupnorm_bwd_data_fp32)
+endif()

test/normalization_bwd_gamma_beta/CMakeLists.txt

 add_custom_target(test_normalization_bwd_gamma_beta)
 add_gtest_executable(test_layernorm2d_bwd_gamma_beta_fp32 test_layernorm2d_bwd_gamma_beta_fp32.cpp)
-target_link_libraries(test_layernorm2d_bwd_gamma_beta_fp32 PRIVATE utility device_normalization_bwd_gamma_beta_instance)
-add_dependencies(test_normalization_bwd_gamma_beta test_layernorm2d_bwd_gamma_beta_fp32)
-
+if (result EQUAL 0)
+    target_link_libraries(test_layernorm2d_bwd_gamma_beta_fp32 PRIVATE utility device_normalization_bwd_gamma_beta_instance)
+    add_dependencies(test_normalization_bwd_gamma_beta test_layernorm2d_bwd_gamma_beta_fp32)
+endif()
 add_gtest_executable(test_groupnorm_bwd_gamma_beta_fp32 test_groupnorm_bwd_gamma_beta_fp32.cpp)
-target_link_libraries(test_groupnorm_bwd_gamma_beta_fp32 PRIVATE utility device_normalization_bwd_gamma_beta_instance)
-add_dependencies(test_normalization_bwd_gamma_beta test_groupnorm_bwd_gamma_beta_fp32)
+if (result EQUAL 0)
+    target_link_libraries(test_groupnorm_bwd_gamma_beta_fp32 PRIVATE utility device_normalization_bwd_gamma_beta_instance)
+    add_dependencies(test_normalization_bwd_gamma_beta test_groupnorm_bwd_gamma_beta_fp32)
+endif()
\ No newline at end of file