Merge remote-tracking branch 'origin/develop' into aosewski/ggemm_multi_d2

library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp

@@ -360,6 +360,8 @@ struct ReferenceConvFwd : public device::BaseOperator
 
                 return 0;
             }
+            throw std::runtime_error("Conv_fwd: number of dimensions must be between 1 and 3.");
+            return 1;
         }
 
         float Run(const device::BaseArgument* p_arg,

library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp

@@ -63,12 +63,11 @@ struct ReferenceGemm : public device::BaseOperator
                 const int K = arg.a_m_k_.mDesc.GetLengths()[1];
 
                 AccDataType v_acc = 0;
+                ComputeTypeA v_a  = 0;
+                ComputeTypeB v_b  = 0;
 
                 for(int k = 0; k < K; ++k)
                 {
-                    ComputeTypeA v_a;
-                    ComputeTypeB v_b;
-
                     // use PassThrough instead of ConvertBF16RTN for reference calculation
                     if constexpr(is_same_v<AElementwiseOperation,
                                            ck::tensor_operation::element_wise::ConvertBF16RTN>)
@@ -94,7 +93,7 @@ struct ReferenceGemm : public device::BaseOperator
                         ck::type_convert<AccDataType>(v_a) * ck::type_convert<AccDataType>(v_b);
                 }
 
-                CDataType v_c;
+                CDataType v_c = 0;
 
                 arg.c_element_op_(v_c, v_acc);
 

library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -10,6 +10,7 @@
 #include "ck/tensor_operation/gpu/device/device_base.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/numeric.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -229,6 +230,8 @@ struct ReferenceImageToColumn : public device::BaseOperator
 
                 return 0;
             }
+            throw std::runtime_error("Img2Col: number of dimensions should be between 1 and 3.");
+            return 1;
         }
 
         float Run(const device::BaseArgument* p_arg,

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

@@ -106,9 +106,8 @@ struct DeviceOperationInstanceFactory<
         return op_ptrs;
     }
 };
-
+#endif
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
-#endif

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

@@ -114,9 +114,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmSt
         return op_ptrs;
     }
 };
-
+#endif
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
-#endif

library/include/ck/library/tensor_operation_instance/gpu/transpose/device_transpose_instance.hpp

@@ -21,20 +21,19 @@ template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 
 using device_transpose_f16_instances = std::tuple<
-    // FOR 16, 32, 16, 32, 16
     // clang-format off
-    DeviceElementwise3dImpl<ck::Tuple<F16>, ck::Tuple<F16>, PassThrough, 2, 2, 1, 8, 8, 8, ck::Sequence<1>, ck::Sequence<1>>,
-    DeviceElementwise3dImpl<ck::Tuple<F16>, ck::Tuple<F16>, PassThrough, 2, 2, 1, 8, 1, 1, ck::Sequence<1>, ck::Sequence<1>>,
-    DeviceElementwise3dImpl<ck::Tuple<F16>, ck::Tuple<F16>, PassThrough, 2, 2, 1, 8, 4, 4, ck::Sequence<1>, ck::Sequence<1>>
+    DeviceElementwise3dImpl<ck::Tuple<F16>, ck::Tuple<F16>, PassThrough, 2, 2, 1, 8, 8, 8, ck::Sequence<8>, ck::Sequence<8>>,
+    DeviceElementwise3dImpl<ck::Tuple<F16>, ck::Tuple<F16>, PassThrough, 2, 2, 1, 8, 8, 8, ck::Sequence<8>, ck::Sequence<4>>,
+    DeviceElementwise3dImpl<ck::Tuple<F16>, ck::Tuple<F16>, PassThrough, 2, 2, 1, 4, 4, 8, ck::Sequence<4>, ck::Sequence<4>>,
+    DeviceElementwise3dImpl<ck::Tuple<F16>, ck::Tuple<F16>, PassThrough, 2, 2, 1, 4, 4, 4, ck::Sequence<1>, ck::Sequence<1>>
     // clang-format on
     >;
 
 using device_transpose_f32_instances = std::tuple<
-    // for 16, 8, 16, 32, 8 -> test with instances for fp16
     // clang-format off
     DeviceElementwise3dImpl<ck::Tuple<F32>, ck::Tuple<F32>, PassThrough, 2, 2, 1, 4, 4, 4, ck::Sequence<1>, ck::Sequence<1>>,
-    DeviceElementwise3dImpl<ck::Tuple<F32>, ck::Tuple<F32>, PassThrough, 2, 2, 1, 4, 8, 4, ck::Sequence<1>, ck::Sequence<1>>,
-    DeviceElementwise3dImpl<ck::Tuple<F32>, ck::Tuple<F32>, PassThrough, 2, 2, 1, 4, 8, 8, ck::Sequence<1>, ck::Sequence<1>>
+    DeviceElementwise3dImpl<ck::Tuple<F32>, ck::Tuple<F32>, PassThrough, 2, 2, 1, 4, 4, 4, ck::Sequence<4>, ck::Sequence<1>>,
+    DeviceElementwise3dImpl<ck::Tuple<F32>, ck::Tuple<F32>, PassThrough, 2, 2, 1, 4, 4, 4, ck::Sequence<4>, ck::Sequence<4>>
     // clang-format on
     >;