Resolve some data type issues and cmake policy. (#940)

* split the types in gemm_bilinear instances, add condition to cmake policy

* fix syntax

* split the data types in batchnorm examples

* fix the batchnorm_bwd test

* fix types in the batchnorm_bwd test

CMakeLists.txt

 cmake_minimum_required(VERSION 3.14)
-cmake_policy(SET CMP0140 NEW)
+if(POLICY CMP0140)
+  # policies CMP0140 not known to CMake until 3.25
+  cmake_policy(SET CMP0140 NEW)
+endif()
 
 # 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

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

@@ -11,12 +11,12 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef CK_ENABLE_FP16
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef CK_ENABLE_FP16
 void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
                                                     Row,
@@ -68,7 +68,8 @@ void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance
                                                     PassThrough,
                                                     PassThrough,
                                                     Bilinear>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_INT8
 void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
                                                     Row,
@@ -120,7 +121,7 @@ void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(
                                                     PassThrough,
                                                     PassThrough,
                                                     Bilinear>>>& instances);
-
+#endif
 // GEMM + Bilinear
 template <typename ALayout,
           typename BLayout,
@@ -158,7 +159,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                      is_same_v<DDataType, half_t> && is_same_v<EDataType, half_t>)
         {
@@ -187,8 +188,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                     op_ptrs);
             }
         }
-        else if constexpr(is_same_v<ADataType, std::int8_t> && is_same_v<BDataType, std::int8_t> &&
-                          is_same_v<DDataType, std::int8_t> && is_same_v<EDataType, std::int8_t>)
+#endif
+#ifdef CK_ENABLE_INT8
+        if constexpr(is_same_v<ADataType, std::int8_t> && is_same_v<BDataType, std::int8_t> &&
+                     is_same_v<DDataType, std::int8_t> && is_same_v<EDataType, std::int8_t>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
@@ -211,7 +214,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                 add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
@@ -220,4 +223,3 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
-#endif

test/batchnorm/batchnorm_bwd_rank_4.cpp

@@ -70,10 +70,23 @@ class TestBatchNormBwdRank4 : public ::testing::Test
     }
 };
 
-using KernelTypes = ::testing::Types<std::tuple<F16, F32, F32, F32, F16, F32, F32>,
-                                     std::tuple<F32, F32, F32, F32, F32, F32, F32>,
-                                     std::tuple<BF16, F32, F32, F32, BF16, F32, F32>,
-                                     std::tuple<F64, F64, F64, F64, F64, F64, F64>>;
+using KernelTypes = ::testing::Types<
+#ifdef CK_ENABLE_FP16
+    std::tuple<F16, F32, F32, F32, F16, F32, F32>
+#endif
+#ifdef CK_ENABLE_FP32
+    ,
+    std::tuple<F32, F32, F32, F32, F32, F32, F32>
+#endif
+#ifdef CK_ENABLE_BF16
+    ,
+    std::tuple<BF16, F32, F32, F32, BF16, F32, F32>
+#endif
+#ifdef CK_ENABLE_FP64
+    ,
+    std::tuple<F64, F64, F64, F64, F64, F64, F64>
+#endif
+    >;
 
 TYPED_TEST_SUITE(TestBatchNormBwdRank4, KernelTypes);
 

test/batchnorm/batchnorm_fwd_rank_4.cpp

@@ -87,10 +87,23 @@ class TestBatchNormFwdRank4 : public ::testing::Test
     }
 };
 
-using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, F16, F16, F32>,
-                                     std::tuple<F32, F32, F32, F32, F32, F32>,
-                                     std::tuple<BF16, BF16, F32, BF16, BF16, F32>,
-                                     std::tuple<F64, F64, F64, F64, F64, F64>>;
+using KernelTypes = ::testing::Types<
+#ifdef CK_ENABLE_FP16
+    std::tuple<F16, F16, F32, F16, F16, F32>
+#endif
+#ifdef CK_ENABLE_FP32
+    ,
+    std::tuple<F32, F32, F32, F32, F32, F32>
+#endif
+#ifdef CK_ENABLE_BF16
+    ,
+    std::tuple<BF16, BF16, F32, BF16, BF16, F32>
+#endif
+#ifdef CK_ENABLE_FP64
+    ,
+    std::tuple<F64, F64, F64, F64, F64, F64>
+#endif
+    >;
 
 TYPED_TEST_SUITE(TestBatchNormFwdRank4, KernelTypes);
 

test/batchnorm/batchnorm_infer_rank_4.cpp

@@ -67,10 +67,23 @@ class TestBatchNormInferRank4 : public ::testing::Test
     }
 };
 
-using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, F16, F16, F32>,
-                                     std::tuple<F32, F32, F32, F32, F32, F32>,
-                                     std::tuple<BF16, BF16, F32, BF16, BF16, F32>,
-                                     std::tuple<F64, F64, F64, F64, F64, F64>>;
+using KernelTypes = ::testing::Types<
+#ifdef CK_ENABLE_FP16
+    std::tuple<F16, F16, F32, F16, F16, F32>
+#endif
+#ifdef CK_ENABLE_FP32
+    ,
+    std::tuple<F32, F32, F32, F32, F32, F32>
+#endif
+#ifdef CK_ENABLE_BF16
+    ,
+    std::tuple<BF16, BF16, F32, BF16, BF16, F32>
+#endif
+#ifdef CK_ENABLE_FP64
+    ,
+    std::tuple<F64, F64, F64, F64, F64, F64>
+#endif
+    >;
 
 TYPED_TEST_SUITE(TestBatchNormInferRank4, KernelTypes);