[HotFix] add config and version files to pass on build info (#856)

* experiment with config file

* experiment with version.h config

* add more info to version.h

* minor updates

* minor updates

* fix case where DTYPE is not used

* large amount of files but minor changes

* remove white space

* minor changes to add more MACROs

* fix cmakedefine01

* fix issue with CK internal conflict

* fix define and define value

* fix clang-format

* fix formatting issue

* experiment with cmake

* clang format v12 to be consistent with miopen

* avoid clang-format for config file

library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance(
     std::vector<std::unique_ptr<DeviceBatchedGemmGemm<Row,
                                                       Col,

library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp

@@ -19,7 +19,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
                                                         Row,
@@ -124,7 +124,7 @@ void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instan
                                                         PassThrough,
                                                         PassThrough>>>& instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
                                                         Row,
@@ -263,7 +263,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                      is_same_v<EDataType, half_t>)
         {
@@ -297,7 +297,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
             }
         }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
         else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
                           is_same_v<EDataType, int8_t>)
         {

library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp

@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {

library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_batched_gemm_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -59,7 +59,7 @@ void add_device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_g
                                                             MaskingSpecialization::MaskDisabled>>>&
         instances);
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 void add_device_batched_gemm_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -148,7 +148,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<B0DataType, half_t> &&
                      is_same_v<B1DataType, half_t> && is_same_v<CDataType, half_t>)
         {
@@ -164,7 +164,7 @@ struct DeviceOperationInstanceFactory<
             }
         }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
         else if constexpr(is_same_v<ADataType, BF16> && is_same_v<B0DataType, BF16> &&
                           is_same_v<B1DataType, BF16> && is_same_v<CDataType, BF16>)
         {

library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 // float
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -66,7 +66,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn
                                                            PassThrough,
                                                            Bilinear>>>& instances);
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
 // double
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -150,7 +150,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<DDataType, float> && is_same_v<EDataType, float>)
         {
@@ -167,7 +167,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
             }
         }
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
                      is_same_v<DDataType, double> && is_same_v<EDataType, double>)
         {

library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 // float
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -66,7 +66,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instanc
                                                            PassThrough,
                                                            Scale>>>& instances);
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
 // double
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -149,7 +149,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<EDataType, float>)
         {
@@ -166,7 +166,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
             }
         }
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
                      is_same_v<EDataType, double>)
         {

library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 // conv1d backward data
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<1,
@@ -30,19 +30,19 @@ void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(
     std::vector<std::unique_ptr<
         DeviceConvBwdData<1, NWC, KXC, NWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(
     std::vector<std::unique_ptr<
         DeviceConvBwdData<1, NWC, KXC, NWK, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<1,
                                                   NWC,
@@ -55,7 +55,7 @@ void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 // conv2d backward data
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -69,7 +69,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -82,7 +82,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -95,7 +95,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -109,7 +109,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
                                                   PassThrough>>>& instances);
 #endif
 #ifdef DL_KERNELS
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 // conv2d dl
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -123,7 +123,7 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -136,7 +136,7 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -150,7 +150,7 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
                                                   PassThrough>>>& instances);
 #endif
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 // conv3d backward data
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
@@ -164,7 +164,7 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
                                                   NDHWC,
@@ -177,7 +177,7 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
                                                   NDHWC,
@@ -190,7 +190,7 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
                                                   NDHWC,
@@ -245,21 +245,21 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(op_ptrs);
             }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(op_ptrs);
             }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(op_ptrs);
             }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
@@ -278,7 +278,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
                 add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
 #endif
             }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
@@ -288,14 +288,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
 #endif
             }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(op_ptrs);
             }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
@@ -314,21 +314,21 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(op_ptrs);
             }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(op_ptrs);
             }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(op_ptrs);
             }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {

library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp

@@ -18,7 +18,7 @@ namespace device {
 namespace instance {
 
 // conv2d forward
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<
         DeviceConvFwd<2, NHWC, KYXC, NHWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -28,7 +28,7 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(
         DeviceConvFwd<2, NHWC, KYXC, NHWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvFwd<2,
                                               NHWC,
@@ -41,13 +41,13 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(
                                               PassThrough,
                                               PassThrough>>>& instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<std::unique_ptr<
         DeviceConvFwd<2, NHWC, KYXC, NHWK, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvFwd<2,
                                               NHWC,
@@ -103,7 +103,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
             }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                               is_same_v<OutDataType, half_t>)
             {
@@ -111,7 +111,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(op_ptrs);
             }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
             else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                               is_same_v<WeiDataType, ck::bhalf_t> &&
                               is_same_v<OutDataType, ck::bhalf_t>)
@@ -119,7 +119,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(op_ptrs);
             }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
             else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                               is_same_v<OutDataType, int8_t>)
             {

library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp

@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {

library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp

@@ -17,7 +17,7 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-#if defined(__fp16__) && defined(DL_KERNELS)
+#if defined(CK_ENABLE_FP16) && defined(DL_KERNELS)
 void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -78,7 +78,7 @@ void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#if defined(__fp32__) && defined(DL_KERNELS)
+#if defined(CK_ENABLE_FP32) && defined(DL_KERNELS)
 void add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -99,7 +99,7 @@ void add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#if defined(__int8__) && defined(DL_KERNELS)
+#if defined(CK_ENABLE_INT8) && defined(DL_KERNELS)
 void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
@@ -140,7 +140,7 @@ void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
@@ -161,7 +161,7 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -208,7 +208,7 @@ void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
         instances);
 
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
@@ -229,7 +229,7 @@ void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -270,7 +270,7 @@ void add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
 void add_device_gemm_xdl_f64_f64_f64_km_kn_mn_instances(
     std::vector<std::unique_ptr<
 
@@ -363,7 +363,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(op_ptrs);
             }
         }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                           is_same_v<CDataType, half_t>)
         {
@@ -414,7 +414,7 @@ struct DeviceOperationInstanceFactory<
             }
         }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
         else if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
                           is_same_v<CDataType, ck::bhalf_t>)
         {
@@ -440,7 +440,7 @@ struct DeviceOperationInstanceFactory<
             }
         }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
         else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
                           is_same_v<CDataType, int8_t>)
         {

library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp

@@ -9,7 +9,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {

library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp

@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {

library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp

@@ -10,7 +10,7 @@
 #include "ck/ck.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {

library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmStreamK<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&

library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp

@@ -10,7 +10,7 @@
 #include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {

library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp

@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 // FP16
 void add_device_normalization_rank_2_1_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 2, 1>>>&);
@@ -27,7 +27,7 @@ void add_device_normalization_rank_4_3_f16_instances(
 void add_device_normalization_rank_5_3_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 5, 3>>>&);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 // FP32
 void add_device_normalization_rank_2_1_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, PassThrough, 2, 1>>>&);
@@ -66,7 +66,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F16> &&
                      is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16>)
         {
@@ -84,7 +84,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
             }
         }
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<XDataType, F32> && is_same_v<GammaDataType, F32> &&
                      is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32>)
         {

library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp

@@ -22,7 +22,7 @@ static constexpr auto WindowRank = 3;
 
 static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
 static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 // FP16
 void add_device_pool3d_fwd_ndhwc_f16_instances(
     std::vector<std::unique_ptr<
@@ -37,7 +37,7 @@ void add_device_pool3d_fwd_ndhwc_index_f16_instances(
     std::vector<std::unique_ptr<
         DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, NDHWC, NDHWC, MaxOp, true>>>&);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 // FP32
 void add_device_pool3d_fwd_ndhwc_f32_instances(
     std::vector<std::unique_ptr<
@@ -84,7 +84,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
         if constexpr(is_same_v<InLayout, NDHWC> && is_same_v<OutLayout, NDHWC>)
         {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
                          is_same_v<IndexDataType, I32>)
             {
@@ -98,7 +98,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                 }
             }
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
                          is_same_v<IndexDataType, I32>)
             {

library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp

@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp

@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {

library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp

@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {