Refactoring cmake files to build data types separately. (#932)

* refactor cmake files for the tests

* refactor cmake files for examples

* fix cmake for gemm example

* fix the cmake file for all examples

* add splitting by data types in gemm_splitk instance header

* rename test to reflect only dl instances are used

* clean up CI workspace, update cmake for instances

* change the jenkinsfile syntax

* build all instances except DL on gfx11

* move workspace cleanup after stages

* clean up workspace after every stage

* isolate data types in grouped_conv_fwd header

* isolate dl instances for grouped_conv2d_fwd

* fix syntax

* fix cmake and batchnorm instances

* fix typo

* fix reduction instances

* fix grouped_conv headers

* fix syntax

* replace parsing logic for instances, replace bfp16 with bf16

* fix the client examples build

* clean up DTYPES from instances cmake files

* update the parsing logic in cmake files

* make an exception for reduction kernels

* update few remaining cmake files to handle DTYPES

* fix syntax

* fix cmake conflicts

* replace f8 with fp8 test name

* resolve conflicts for dpp instances

CMakeLists.txt

@@ -1,4 +1,5 @@
 cmake_minimum_required(VERSION 3.14)
+cmake_policy(SET CMP0140 NEW)
 
 # This has to be initialized before the project() command appears
 # Set the default of CMAKE_BUILD_TYPE to be release, unless user specifies with -D.  MSVC_IDE does not use CMAKE_BUILD_TYPE
@@ -383,31 +384,31 @@ IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu
     set(cmake_instance)
     file(READ "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}/CMakeLists.txt" cmake_instance)
     set(add_inst 0)
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp8\" " AND DTYPES MATCHES "fp8")
+    if(("${cmake_instance}" MATCHES "fp8" OR "${cmake_instance}" MATCHES "_f8") AND DTYPES MATCHES "fp8")
         #message("fp8 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf8\" " AND DTYPES MATCHES "bf8")
+    if(("${cmake_instance}" MATCHES "bf8" OR "${cmake_instance}" MATCHES "_b8") AND DTYPES MATCHES "bf8")
         #message("bf8 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp16\"" AND DTYPES MATCHES "fp16")
+    if(("${cmake_instance}" MATCHES "fp16" OR "${cmake_instance}" MATCHES "_f16") AND DTYPES MATCHES "fp16")
         #message("fp16 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp32\"" AND DTYPES MATCHES "fp32")
+    if(("${cmake_instance}" MATCHES "fp32" OR "${cmake_instance}" MATCHES "_f32") AND DTYPES MATCHES "fp32")
         #message("fp32 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp64\"" AND DTYPES MATCHES "fp64")
+    if(("${cmake_instance}" MATCHES "fp64" OR "${cmake_instance}" MATCHES "_f64") AND DTYPES MATCHES "fp64")
         #message("fp64 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf16\"" AND DTYPES MATCHES "bf16")
+    if(("${cmake_instance}" MATCHES "bf16" OR "${cmake_instance}" MATCHES "_b16") AND DTYPES MATCHES "bf16")
         #message("bf16 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"int8\"" AND DTYPES MATCHES "int8")
+    if(("${cmake_instance}" MATCHES "int8" OR "${cmake_instance}" MATCHES "_i8") AND DTYPES MATCHES "int8")
         #message("int8 instance found!")
         set(add_inst 1)
     endif()

example/01_gemm/CMakeLists.txt

@@ -1,51 +1,54 @@
-if(DL_KERNELS)
-  add_custom_target(example_gemm_dl)
+add_custom_target(example_gemm_dl)
 
-  add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
-  add_dependencies(example_gemm_dl example_gemm_dl_fp32)
-  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
+add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_gemm_dl example_gemm_dl_fp32)
+endif()
+add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_dl example_gemm_dl_fp16)
-    add_example_executable(example_gemm_dpp_fp16 gemm_dpp_fp16.cpp)
-  endif()
-  if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-    add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
+endif()
+add_example_executable(example_gemm_dpp_fp16 gemm_dpp_fp16.cpp)
+add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_dl example_gemm_dl_int8)
-  endif()
-
-  if(USE_BITINT_EXTENSION_INT4)
+endif()
+if(USE_BITINT_EXTENSION_INT4)
     add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
     add_dependencies(example_gemm_dl example_gemm_dl_int4)
-  endif(USE_BITINT_EXTENSION_INT4)
-endif()
+endif(USE_BITINT_EXTENSION_INT4)
 
 add_custom_target(example_gemm_xdl)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
-  add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
-  add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
-
-  if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
+add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
+endif()
+add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
+endif()
+add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
+endif()
+if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
     add_custom_target(example_gemm_wmma)
     add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
-    add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
-  endif()
-
+    if(result EQUAL 0)
+      add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
+    endif()
 endif()
 
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
+add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
+if(result EQUAL 0)
   add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
 
   add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp)
   add_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn)
 endif()
 
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-  add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
+add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_gemm_xdl example_gemm_xdl_int8)
 endif()
 
@@ -54,22 +57,23 @@ if(USE_BITINT_EXTENSION_INT4)
   add_dependencies(example_gemm_xdl example_gemm_xdl_int4)
 endif(USE_BITINT_EXTENSION_INT4)
 
-if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
-  # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
-  add_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp)
+# FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
+add_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp)
+if(result EQUAL 0)
   add_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
 endif()
 
 add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
 
-if(DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
-  if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
-    add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
+
+if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
+  add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
   endif()
 endif()
 
-if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
-  add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp)
+add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_f8)
 endif()

example/02_gemm_bilinear/CMakeLists.txt

@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list1 gfx1100 gfx1101 gfx1102)
 list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
@@ -19,4 +18,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-endif()

example/03_gemm_bias_relu/CMakeLists.txt

@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -7,4 +6,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-endif()

example/04_gemm_add_add_fastgelu/CMakeLists.txt

@@ -3,24 +3,24 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
     add_custom_target(example_gemm_add_add_fastgelu_xdl)
-    if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
+    if(result EQUAL 0)
       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
     endif()
-    if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
+    if(result EQUAL 0)
       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
     endif()
-    if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-      add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
+    if(result EQUAL 0)
       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
     endif()
     if(USE_BITINT_EXTENSION_INT4)
        add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
        add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
     endif(USE_BITINT_EXTENSION_INT4)
-    if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-      add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
+    if(result EQUAL 0)
       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
     endif()
    set(target 1)

example/09_convnd_fwd/CMakeLists.txt

@@ -2,34 +2,16 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-    if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
       add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
-    endif()
-    if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
       add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
-    endif()
-    if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
       add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
-    endif()
-    if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
       add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
-    endif()
-    # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
-    if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
+      # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
       add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
-    endif()
    set(target 1)
  endif()
 endforeach()
 
-if(DL_KERNELS)
-  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
-  endif()
-  if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
-  endif()
-  if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
-  endif()
-endif()
+add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
+add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
+add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)

example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt

@@ -2,27 +2,27 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_custom_target(example_convnd_fwd_reduce_xdl)
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
+  add_custom_target(example_convnd_fwd_reduce_xdl)
+  add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
-   endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
+  endif()
+  add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
-   endif()
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
+  endif()
+  add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
-   endif()
-   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
+  endif()
+  add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
-   endif()
-   if(USE_BITINT_EXTENSION_INT4)
+  endif()
+  if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
       add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
-   endif(USE_BITINT_EXTENSION_INT4)
-   set(target 1)
+  endif(USE_BITINT_EXTENSION_INT4)
+  set(target 1)
  endif()
 endforeach()

example/13_pool2d_fwd/CMakeLists.txt

@@ -1,6 +1,2 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_pool2d_fwd_fp16 pool2d_fwd_fp16.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_pool2d_fwd_fp32 pool2d_fwd_fp32.cpp)
-endif()
+add_example_executable(example_pool2d_fwd_fp16 pool2d_fwd_fp16.cpp)
+add_example_executable(example_pool2d_fwd_fp32 pool2d_fwd_fp32.cpp)

example/14_gemm_quantization/CMakeLists.txt

@@ -1,9 +1,5 @@
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 # dlops
-if(DL_KERNELS)
-  add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
-endif()
-
+add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
 # xdlops
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
@@ -14,4 +10,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-endif()

example/15_grouped_gemm/CMakeLists.txt

@@ -1,36 +1,44 @@
 add_custom_target(example_grouped_gemm_xdl)
-
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
+add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
+if(result EQUAL 0)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
 endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
-  add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
-  add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
-  add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp16 grouped_gemm_xdl_fixed_nk_fp16.cpp)
-  add_example_executable(example_grouped_gemm_xdl_fixed_nk_bias_fp16 grouped_gemm_xdl_fixed_nk_bias_fp16.cpp)
-  add_dependencies(example_grouped_gemm_xdl 
-                   example_grouped_gemm_xdl_fp16
-                   example_grouped_gemm_multiple_d_dl_fp16
-                   example_grouped_gemm_xdl_splitk_fp16
-                   example_grouped_gemm_xdl_fixed_nk_fp16
-                   example_grouped_gemm_xdl_fixed_nk_bias_fp16)
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_bfp16 grouped_gemm_xdl_bfp16.cpp)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bfp16)
-endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
+add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp16)
+endif()
+add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_multiple_d_dl_fp16)
+endif()
+add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_splitk_fp16)
+endif()
+add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp16 grouped_gemm_xdl_fixed_nk_fp16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp16)
+endif()
+add_example_executable(example_grouped_gemm_xdl_fixed_nk_bias_fp16 grouped_gemm_xdl_fixed_nk_bias_fp16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_bias_fp16)
+endif()
+add_example_executable(example_grouped_gemm_xdl_bf16 grouped_gemm_xdl_bf16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bf16)
+endif()
+add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8)
 endif()
-if(DTYPES MATCHES "f8" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp8 grouped_gemm_xdl_fixed_nk_fp8.cpp)
+add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp8 grouped_gemm_xdl_fixed_nk_fp8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp8)
 endif()
 
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
+  if(result EQUAL 0)
+    add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
+  endif()
 endif()