Fix CUDA_MS_OP_THREE_TYPED_CLASS_NAME

onnxruntime/contrib_ops/cuda/cuda_contrib_kernels.cc

@@ -17,6 +17,8 @@ using namespace onnxruntime::common;
   ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kMSDomain, start_ver, end_ver, name)
 #define CUDA_MS_OP_TWO_TYPED_CLASS_NAME(ver, type1, type2, name) \
   ONNX_OPERATOR_TWO_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kMSDomain, ver, type1, type2, name)
+#define CUDA_MS_OP_THREE_TYPED_CLASS_NAME(ver, type1, type2, type3, name) \
+  ONNX_OPERATOR_TWO_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kMSDomain, ver, type1, type2, type3, name)
 
 #define CUDA_ONNX_OP_TYPED_CLASS_NAME(ver, type, name) \
   ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, ver, type, name)
@@ -188,12 +190,24 @@ class CUDA_MS_OP_CLASS_NAME(1, GemmFloat8);
 class CUDA_MS_OP_TYPED_CLASS_NAME(1, MLFloat16, SparseAttention);
 class CUDA_MS_OP_TYPED_CLASS_NAME(1, BFloat16, SparseAttention);
 
-class CUDA_MS_OP_TWO_TYPED_CLASS_NAME(1, uint8_t, int32_t, GatherBlockQuantized);
-class CUDA_MS_OP_TWO_TYPED_CLASS_NAME(1, uint8_t, int64_t, GatherBlockQuantized);
-class CUDA_MS_OP_TWO_TYPED_CLASS_NAME(1, UInt4x2, int32_t, GatherBlockQuantized);
-class CUDA_MS_OP_TWO_TYPED_CLASS_NAME(1, UInt4x2, int64_t, GatherBlockQuantized);
-class CUDA_MS_OP_TWO_TYPED_CLASS_NAME(1, Int4x2, int32_t, GatherBlockQuantized);
-class CUDA_MS_OP_TWO_TYPED_CLASS_NAME(1, Int4x2, int64_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, uint8_t, float, int32_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, uint8_t, MLFloat16, int32_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, uint8_t, BFloat16, int32_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, uint8_t, float, int64_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, uint8_t, MLFloat16, int64_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, uint8_t, BFloat16, int64_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, UInt4x2, float, int32_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, UInt4x2, MLFloat16, int32_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, UInt4x2, BFloat16, int32_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, UInt4x2, float, int64_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, UInt4x2, MLFloat16, int64_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, UInt4x2, BFloat16, int64_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, Int4x2, float, int32_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, Int4x2, MLFloat16, int32_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, Int4x2, BFloat16, int32_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, Int4x2, float, int64_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, Int4x2, MLFloat16, int64_t, GatherBlockQuantized);
+class CUDA_MS_OP_THREE_TYPED_CLASS_NAME(1, Int4x2, BFloat16, int64_t, GatherBlockQuantized);
 
 #ifdef ENABLE_ATEN
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kPytorchAtenDomain, 1, ATen);