clean

profiler/CMakeLists.txt

@@ -24,21 +24,21 @@ include_directories(BEFORE
 # ck_profiler
 set(PROFILER_SOURCE
     src/profiler.cpp
-#   src/profile_gemm.cpp
-#   src/profile_gemm_bias_2d.cpp
-#   src/profile_gemm_bias_relu.cpp
-#   src/profile_gemm_bias_relu_add.cpp
-#   src/profile_gemm_reduce.cpp
-#   src/profile_batched_gemm.cpp
-#   src/profile_conv_fwd_bias_relu.cpp
-#   src/profile_conv_fwd_bias_relu_add.cpp
-#   src/profile_conv_fwd_bias_relu_atomic_add.cpp
-#   src/profile_convnd_fwd.cpp
-#   src/profile_convnd_bwd_data.cpp
-#   src/profile_reduce.cpp
-#   src/profile_grouped_gemm.cpp
-#   src/profile_conv_bwd_weight.cpp
-#   src/profile_batched_gemm_reduce.cpp
+    src/profile_gemm.cpp
+    src/profile_gemm_bias_2d.cpp
+    src/profile_gemm_bias_relu.cpp
+    src/profile_gemm_bias_relu_add.cpp
+    src/profile_gemm_reduce.cpp
+    src/profile_batched_gemm.cpp
+    src/profile_conv_fwd_bias_relu.cpp
+    src/profile_conv_fwd_bias_relu_add.cpp
+    src/profile_conv_fwd_bias_relu_atomic_add.cpp
+    src/profile_convnd_fwd.cpp
+    src/profile_convnd_bwd_data.cpp
+    src/profile_reduce.cpp
+    src/profile_grouped_gemm.cpp
+    src/profile_conv_bwd_weight.cpp
+    src/profile_batched_gemm_reduce.cpp
     src/profile_gemm_add_add_fastgelu.cpp
 )
 
@@ -46,21 +46,21 @@ add_executable(ckProfiler ${PROFILER_SOURCE})
 
 target_link_libraries(ckProfiler PRIVATE host_tensor)
 target_link_libraries(ckProfiler PRIVATE conv_util)
-#target_link_libraries(ckProfiler PRIVATE device_gemm_reduce_instance)
-#target_link_libraries(ckProfiler PRIVATE device_gemm_instance)
-#target_link_libraries(ckProfiler PRIVATE device_gemm_bias2d_instance)
-#target_link_libraries(ckProfiler PRIVATE device_gemm_bias_relu_instance)
-#target_link_libraries(ckProfiler PRIVATE device_gemm_bias_relu_add_instance)
-#target_link_libraries(ckProfiler PRIVATE device_batched_gemm_instance)
-#target_link_libraries(ckProfiler PRIVATE device_conv1d_fwd_instance)
-#target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_instance)
-#target_link_libraries(ckProfiler PRIVATE device_conv3d_fwd_instance)
-#target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_instance)
-#target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_add_instance)
-#target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_atomic_add_instance)
-#target_link_libraries(ckProfiler PRIVATE device_convnd_bwd_data_instance)
-#target_link_libraries(ckProfiler PRIVATE device_reduce_instance)
-#target_link_libraries(ckProfiler PRIVATE device_grouped_gemm_instance)
-#target_link_libraries(ckProfiler PRIVATE device_conv2d_bwd_weight_instance)
-#target_link_libraries(ckProfiler PRIVATE device_batched_gemm_reduce_instance)
+target_link_libraries(ckProfiler PRIVATE device_gemm_reduce_instance)
+target_link_libraries(ckProfiler PRIVATE device_gemm_instance)
+target_link_libraries(ckProfiler PRIVATE device_gemm_bias2d_instance)
+target_link_libraries(ckProfiler PRIVATE device_gemm_bias_relu_instance)
+target_link_libraries(ckProfiler PRIVATE device_gemm_bias_relu_add_instance)
+target_link_libraries(ckProfiler PRIVATE device_batched_gemm_instance)
+target_link_libraries(ckProfiler PRIVATE device_conv1d_fwd_instance)
+target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_instance)
+target_link_libraries(ckProfiler PRIVATE device_conv3d_fwd_instance)
+target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_instance)
+target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_add_instance)
+target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_atomic_add_instance)
+target_link_libraries(ckProfiler PRIVATE device_convnd_bwd_data_instance)
+target_link_libraries(ckProfiler PRIVATE device_reduce_instance)
+target_link_libraries(ckProfiler PRIVATE device_grouped_gemm_instance)
+target_link_libraries(ckProfiler PRIVATE device_conv2d_bwd_weight_instance)
+target_link_libraries(ckProfiler PRIVATE device_batched_gemm_reduce_instance)
 target_link_libraries(ckProfiler PRIVATE device_gemm_add_add_fastgelu_instance)

profiler/src/profiler.cpp

@@ -6,21 +6,21 @@
 
 #include "profile_convnd_fwd.hpp"
 
-// int profile_gemm(int, char*[]);
-// int profile_gemm_bias_2d(int, char*[]);
-// int profile_gemm_bias_relu(int, char*[]);
-// int profile_gemm_bias_relu_add(int, char*[]);
-// int profile_gemm_reduce(int, char*[]);
-// int profile_batched_gemm(int, char*[]);
-// int profile_grouped_gemm(int, char*[]);
-// int profile_conv_fwd(int, char*[]);
-// int profile_conv_fwd_bias_relu(int, char*[]);
-// int profile_conv_fwd_bias_relu_add(int, char*[]);
-// int profile_conv_fwd_bias_relu_atomic_add(int, char*[]);
-// int profile_convnd_bwd_data(int, char*[], int);
-// int profile_reduce(int, char*[]);
-// int profile_conv_bwd_weight(int, char*[]);
-// int profile_batched_gemm_reduce(int, char*[]);
+int profile_gemm(int, char*[]);
+int profile_gemm_bias_2d(int, char*[]);
+int profile_gemm_bias_relu(int, char*[]);
+int profile_gemm_bias_relu_add(int, char*[]);
+int profile_gemm_reduce(int, char*[]);
+int profile_batched_gemm(int, char*[]);
+int profile_grouped_gemm(int, char*[]);
+int profile_conv_fwd(int, char*[]);
+int profile_conv_fwd_bias_relu(int, char*[]);
+int profile_conv_fwd_bias_relu_add(int, char*[]);
+int profile_conv_fwd_bias_relu_atomic_add(int, char*[]);
+int profile_convnd_bwd_data(int, char*[], int);
+int profile_reduce(int, char*[]);
+int profile_conv_bwd_weight(int, char*[]);
+int profile_batched_gemm_reduce(int, char*[]);
 int profile_gemm_add_add_fastgelu(int, char*[]);
 
 static void print_helper_message()
@@ -54,7 +54,6 @@ int main(int argc, char* argv[])
         return 0;
     }
 
-#if 0
     if(strcmp(argv[1], "gemm") == 0)
     {
         return profile_gemm(argc, argv);
@@ -123,8 +122,7 @@ int main(int argc, char* argv[])
     {
         return profile_conv_bwd_weight(argc, argv);
     }
-#endif
-    if(strcmp(argv[1], "gemm_add_add_fastgelu") == 0)
+    else if(strcmp(argv[1], "gemm_add_add_fastgelu") == 0)
     {
         return profile_gemm_add_add_fastgelu(argc, argv);
     }