Allow building CK for specific data types and split off last remaining DL instances. (#830)

* properly split conv_nd_bwd_data instances

* split conv2d_fwd instance data types

* split the gemm, conv2d_fwd and batched_gemm_softamx_gemm

* split the tests by data types where possible

* filter examples by DTYPES

* split few remaining examples by DTYPES

* filter most instances by DTYPES

* add new lines at end of headers, fix grouped_gemm profiler

* fix syntax

* split the ckprofiler instances by DTYPES

* split the conv2d and quantization DL and XDL instances

* fix the splitting of conv2d DL instances

* split softmax and pool_fwd tests for fp16 and fp32 types

* fix syntax

* fix the dl_int8 quantization instances isolation

library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp

@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __bf16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -126,3 +126,4 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif

library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp

@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -118,3 +118,4 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif

library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp

@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __fp32__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -117,3 +117,4 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif

library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp

@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -123,3 +123,4 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif

library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_elementwise_normalization_instance
     device_elementwise_normalization_f16_instance.cpp
 )
+endif()

library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_add_add_fastgelu_instance
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
 )
+endif()

library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_add_fastgelu_instance
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
 )
+endif()

library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_add_relu_add_layernorm_instance
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
 )
+endif()

library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_bilinear_instance
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
 )
+endif()

library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_fastgelu_instance
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
 )
+endif()

library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_streamk_instance
    # device_gemm_xdl_streamk_f32_f32_f32_mk_kn_mn_instance.cpp
    # device_gemm_xdl_streamk_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -8,3 +9,4 @@ add_instance_library(device_gemm_streamk_instance
    # device_gemm_xdl_streamk_f16_f16_f16_km_kn_mn_instance.cpp
    # device_gemm_xdl_streamk_f16_f16_f16_km_nk_mn_instance.cpp
 )
+endif()

library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_grouped_gemm_instance
    device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
    device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -8,3 +9,4 @@ add_instance_library(device_grouped_gemm_instance
    device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
    device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
 )
+endif()

library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt

+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_grouped_gemm_fastgelu_instance
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
 )
+endif()

library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt

-add_instance_library(device_normalization_instance
-    device_layernorm2d_f16_instance.cpp
-    device_layernorm2d_f32_instance.cpp
+set(DEVICE_NORMALIZATION_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_NORMALIZATION_INSTANCES device_layernorm2d_f16_instance.cpp
     device_layernorm4d_f16_instance.cpp
-    device_layernorm4d_f32_instance.cpp
     device_groupnorm_f16_instance.cpp
-    device_groupnorm_f32_instance.cpp
     device_groupnorm_swish_f16_instance.cpp
-    device_groupnorm_swish_f32_instance.cpp
-    device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
-)
+    device_groupnorm_swish_f16_f32_f32_f16_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_NORMALIZATION_INSTANCES device_layernorm2d_f32_instance.cpp
+    device_layernorm4d_f32_instance.cpp
+    device_groupnorm_f32_instance.cpp
+    device_groupnorm_swish_f32_instance.cpp)
+endif()
+add_instance_library(device_normalization_instance ${DEVICE_NORMALIZATION_INSTANCES})

library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt

-add_instance_library(device_pool_fwd_instance
-    device_avg_pool2d_fwd_nhwc_f16_instance.cpp
-    device_avg_pool2d_fwd_nhwc_f32_instance.cpp
-    device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
-    device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
-    device_max_pool2d_fwd_nhwc_f16_instance.cpp
-    device_max_pool2d_fwd_nhwc_f32_instance.cpp
-    device_max_pool3d_fwd_ndhwc_f16_instance.cpp
-    device_max_pool3d_fwd_ndhwc_f32_instance.cpp
-)
+set(DEVICE_POOL_FWD_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_POOL_FWD_INSTANCES device_avg_pool2d_fwd_nhwc_f16_instance.cpp
+                                          device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
+                                          device_max_pool2d_fwd_nhwc_f16_instance.cpp
+                                          device_max_pool3d_fwd_ndhwc_f16_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_POOL_FWD_INSTANCES device_avg_pool2d_fwd_nhwc_f32_instance.cpp
+                                          device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
+                                          device_max_pool2d_fwd_nhwc_f32_instance.cpp
+                                          device_max_pool3d_fwd_ndhwc_f32_instance.cpp)
+endif()
+add_instance_library(device_pool_fwd_instance ${DEVICE_POOL_FWD_INSTANCES})

library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt

 if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-set(CONV2D_PERLAYER_QUANT_SRC
-    conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
-    conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
-)
-
-set(CONV2D_PERCHANNEL_QUANT_SRC
-    conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
-    conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
-)
-
-set(CONV2D_BIAS_PERLAYER_QUANT_SRC
-    conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
-    conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
-)
-
-set(CONV2D_BIAS_PERCHANNEL_QUANT_SRC
-    conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
-    conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
-)
 
+set(CONV2D_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp)
+set(CONV2D_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp)
+set(CONV2D_BIAS_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp)
+set(CONV2D_BIAS_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp)
 set(GEMM_QUANT_SRC
-    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
-    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
-    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
-    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
 )
+if(DL_KERNELS)
+    list(APPEND CONV2D_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp)
+    list(APPEND CONV2D_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp)
+    list(APPEND CONV2D_BIAS_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp)
+    list(APPEND CONV2D_BIAS_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp)
+    list(APPEND GEMM_QUANT_SRC
+        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp)
+endif()
 
 add_instance_library(device_quantization_instance
     ${CONV2D_PERLAYER_QUANT_SRC}

library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt

-add_instance_library(device_softmax_instance
-    device_softmax_f16_f16_instance_rank3_reduce1.cpp
+set(DEVICE_SOFTMAX_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_SOFTMAX_INSTANCES device_softmax_f16_f16_instance_rank3_reduce1.cpp
     device_softmax_f16_f16_instance_rank3_reduce2.cpp
     device_softmax_f16_f16_instance_rank3_reduce3.cpp
     device_softmax_f16_f16_instance_rank4_reduce1.cpp
     device_softmax_f16_f16_instance_rank4_reduce2.cpp
     device_softmax_f16_f16_instance_rank4_reduce3.cpp
-    device_softmax_f16_f16_instance_rank4_reduce4.cpp
-    device_softmax_f32_f32_instance_rank3_reduce1.cpp
+    device_softmax_f16_f16_instance_rank4_reduce4.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_SOFTMAX_INSTANCES device_softmax_f32_f32_instance_rank3_reduce1.cpp
     device_softmax_f32_f32_instance_rank3_reduce2.cpp
     device_softmax_f32_f32_instance_rank3_reduce3.cpp
     device_softmax_f32_f32_instance_rank4_reduce1.cpp
     device_softmax_f32_f32_instance_rank4_reduce2.cpp
     device_softmax_f32_f32_instance_rank4_reduce3.cpp
-    device_softmax_f32_f32_instance_rank4_reduce4.cpp
-)
+    device_softmax_f32_f32_instance_rank4_reduce4.cpp)
+endif()
+add_instance_library(device_softmax_instance ${DEVICE_SOFTMAX_INSTANCES})

profiler/src/CMakeLists.txt

@@ -3,20 +3,11 @@ set(PROFILER_SOURCES
     profiler.cpp
     profile_gemm.cpp
     profile_gemm_splitk.cpp
-    profile_gemm_streamk.cpp
-    profile_gemm_bilinear.cpp
     profile_gemm_bias_add_reduce.cpp
-    profile_gemm_add_add_fastgelu.cpp
     profile_gemm_add_multiply.cpp
-    profile_gemm_add_fastgelu.cpp
-    profile_gemm_add_relu_add_layernorm.cpp
-    profile_gemm_fastgelu.cpp
     profile_gemm_reduce.cpp
     profile_batched_gemm.cpp
-    profile_batched_gemm_gemm.cpp
-    profile_batched_gemm_add_relu_gemm_add.cpp
     profile_batched_gemm_reduce.cpp
-    profile_grouped_gemm.cpp
     profile_conv_fwd.cpp
     profile_conv_fwd_bias_relu.cpp
     profile_conv_fwd_bias_relu_add.cpp
@@ -32,7 +23,6 @@ set(PROFILER_SOURCES
     profile_batchnorm_fwd.cpp
     profile_batchnorm_bwd.cpp
     profile_batchnorm_infer.cpp
-    profile_grouped_gemm_fastgelu.cpp
     profile_contraction_bilinear.cpp
     profile_contraction_scale.cpp
     profile_grouped_conv_bwd_data.cpp
@@ -40,6 +30,18 @@ set(PROFILER_SOURCES
 if(DL_KERNELS)
   list(APPEND PROFILER_SOURCES profile_batched_gemm_multi_d.cpp)
 endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  list(APPEND PROFILER_SOURCES profile_batched_gemm_gemm.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_fastgelu.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_streamk.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_bilinear.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_add_fastgelu.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_add_add_fastgelu.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_add_relu_add_layernorm.cpp)
+  list(APPEND PROFILER_SOURCES profile_batched_gemm_add_relu_gemm_add.cpp)
+  list(APPEND PROFILER_SOURCES profile_grouped_gemm.cpp)
+  list(APPEND PROFILER_SOURCES profile_grouped_gemm_fastgelu.cpp)
+endif()
 
 set(PROFILER_EXECUTABLE ckProfiler)
 
@@ -49,20 +51,11 @@ target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_streamk_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_reduce_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_fwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_fwd_instance)
@@ -79,13 +72,24 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_instan
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
 if(DL_KERNELS)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
 endif()
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_streamk_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
+endif()
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)

profiler/src/profile_grouped_gemm.cpp

@@ -88,7 +88,7 @@ int profile_grouped_gemm(int argc, char* argv[])
     const auto StrideBs = argToIntArray(argv[12]);
     const auto StrideCs = argToIntArray(argv[13]);
     const int kbatch    = argc == 15 ? std::stoi(argv[14]) : 1;
-
+#ifdef __fp16__
     if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         ck::profiler::profile_grouped_gemm_impl<ck::half_t,
@@ -173,7 +173,7 @@ int profile_grouped_gemm(int argc, char* argv[])
     {
         throw std::runtime_error("wrong! this GEMM data_type & layout is not implemented");
     }
-
+#endif
     return 0;
 }
 

test/CMakeLists.txt

@@ -60,6 +60,6 @@ add_subdirectory(contraction)
 add_subdirectory(pool_fwd)
 add_subdirectory(batched_gemm_multi_d)
 add_subdirectory(grouped_convnd_bwd_data)
-if(GPU_TARGETS MATCHES "gfx1100")
+if(GPU_TARGETS MATCHES "gfx11")
     add_subdirectory(wmma_op)
 endif()