Merge remote-tracking branch 'origin/develop' into revert_sfc

Jenkinsfile

@@ -60,7 +60,8 @@ def cmake_build(Map conf=[:]){
             cd build
         """
     def setup_cmd = conf.get("setup_cmd", "${cmake_envs} cmake ${setup_args}   .. ")
-    def build_cmd = conf.get("build_cmd", "${build_envs} dumb-init make  -j\$(( \$(nproc) / 4 )) ${config_targets}")
+    // reduce parallelism when compiling, clang uses too much memory
+    def build_cmd = conf.get("build_cmd", "${build_envs} dumb-init make  -j\$(( \$(nproc) / 5 )) ${config_targets}")
     def execute_cmd = conf.get("execute_cmd", "")
 
     def cmd = conf.get("cmd", """

test/CMakeLists.txt

@@ -21,34 +21,16 @@ function(add_test_executeable TEST_NAME)
     target_link_libraries(${TEST_NAME} PRIVATE host_tensor)
     target_link_libraries(${TEST_NAME} PRIVATE device_gemm_instance)
     target_link_libraries(${TEST_NAME} PRIVATE device_conv2d_fwd_instance)
+    target_link_libraries(${TEST_NAME} PRIVATE device_conv2d_bwd_data_instance)
     add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}> )
     add_dependencies(tests ${TEST_NAME})
     add_dependencies(check ${TEST_NAME})
 endfunction(add_test_executeable TEST_NAME)
 
-
-file(GLOB TESTS *.cpp)
+file(GLOB TESTS */*.cpp)
 
 foreach(TEST ${TESTS})
     get_filename_component(BASE_NAME ${TEST} NAME_WE)
     message("adding test ${BASE_NAME}")
     add_test_executeable(test_${BASE_NAME} ${TEST})
 endforeach(TEST ${TESTS})
-
-# test_gemm_xdl_fp32
-set(GEMM_XDL_FP32_SOURCE gemm_xdl/test_gemm_fp32.cpp)
-add_executable(test_gemm_xdl_fp32 ${GEMM_XDL_FP32_SOURCE})
-target_link_libraries(test_gemm_xdl_fp32 PRIVATE host_tensor)
-target_link_libraries(test_gemm_xdl_fp32 PRIVATE device_gemm_instance)
-
-# test_gemm_xdl_bf16
-set(GEMM_XDL_BF16_SOURCE gemm_xdl/test_gemm_bf16.cpp)
-add_executable(test_gemm_xdl_bf16 ${GEMM_XDL_BF16_SOURCE})
-target_link_libraries(test_gemm_xdl_bf16 PRIVATE host_tensor)
-target_link_libraries(test_gemm_xdl_bf16 PRIVATE device_gemm_instance)
-
-# test_gemm_xdl_int8
-set(GEMM_XDL_INT8_SOURCE gemm_xdl/test_gemm_int8.cpp)
-add_executable(test_gemm_xdl_int8 ${GEMM_XDL_INT8_SOURCE})
-target_link_libraries(test_gemm_xdl_int8 PRIVATE host_tensor)
-target_link_libraries(test_gemm_xdl_int8 PRIVATE device_gemm_instance)

test/conv2d_bwd_data/main.cpp → test/conv2d_bwd_data/conv2d_bwd_data.cpp

@@ -11,8 +11,9 @@
 
 using F16  = ck::half_t;
 using F32  = float;
-using BF16 = ushort;
+using BF16 = ck::bhalf_t;
 using INT8 = int8_t;
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -22,6 +23,7 @@ using DeviceConvBwdDataNoOpPtr =
     DeviceConvBwdDataPtr<ck::tensor_operation::element_wise::PassThrough,
                          ck::tensor_operation::element_wise::PassThrough,
                          ck::tensor_operation::element_wise::PassThrough>;
+
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<DeviceConvBwdDataNoOpPtr>&);
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
@@ -30,6 +32,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
     std::vector<DeviceConvBwdDataNoOpPtr>&);
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<DeviceConvBwdDataNoOpPtr>&);
+
 } // namespace device_conv2d_bwd_data_instance
 } // namespace device
 } // namespace tensor_operation
@@ -78,7 +81,12 @@ int main(int argc, char* argv[])
     ck::index_t in_right_pad_h  = 1;
     ck::index_t in_right_pad_w  = 1;
 
-    if(argc == 3)
+    if(argc == 1)
+    {
+        data_type   = 1;
+        init_method = 1;
+    }
+    else if(argc == 3)
     {
         data_type   = std::stoi(argv[1]);
         init_method = std::stoi(argv[2]);
@@ -106,11 +114,9 @@ int main(int argc, char* argv[])
     }
     else
     {
-        printf("arg1: data type (0=fp32 )\n");
-        printf("arg2: verification (0=no, 1=yes)\n");
-        printf("arg3: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg4: run kernel # of times (>1)\n");
-        printf("arg5 to 19: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
+        printf("arg1: data type (0=fp32, 1=fp16, 2= bfp16, 3= int8_t )\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3 to 17: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
                "RightPx\n");
         exit(1);
     }
@@ -296,7 +302,7 @@ int main(int argc, char* argv[])
 
     if(data_type == 0)
     {
-        Run(float(), float(), F32());
+        Run(F32(), F32(), F32());
     }
     else if(data_type == 1)
     {

test/conv2d_fwd.cpp → test/conv2d_fwd/conv2d_fwd.cpp

@@ -77,8 +77,8 @@ int main(int argc, char* argv[])
     ck::index_t in_right_pad_w  = 1;
     if(argc == 1)
     {
+        data_type   = 1;
         init_method = 1;
-        data_type   = 0;
     }
     else if(argc == 3)
     {
@@ -108,10 +108,9 @@ int main(int argc, char* argv[])
     }
     else
     {
-        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg1: data type (0=fp32, 1=fp16, 2= bfp16, 3= int8_t )\n");
         printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: run kernel # of times (>1)\n");
-        printf("arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
+        printf("arg3 to 17: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
                "RightPx\n");
         exit(1);
     }