From 5d66cf017cc4ea417ee68fa07a1a7d3359d8e276 Mon Sep 17 00:00:00 2001
From: Dmitri Smirnov <yuslepukhin@users.noreply.github.com>
Date: Tue, 3 Nov 2020 07:07:28 -0800
Subject: [PATCH] Register Clip for OpSet13 (#5671)

Signed-off-by: Dmitri Smirnov <dmitrism@microsoft.com>
---
 .../providers/cuda/cuda_execution_provider.cc    |  6 ++++--
 onnxruntime/core/providers/cuda/math/clip.cc     | 16 ++++++++++++++--
 2 files changed, 18 insertions(+), 4 deletions(-)

diff --git a/onnxruntime/core/providers/cuda/cuda_execution_provider.cc b/onnxruntime/core/providers/cuda/cuda_execution_provider.cc
index 650ae97874..bff38534a8 100644
--- a/onnxruntime/core/providers/cuda/cuda_execution_provider.cc
+++ b/onnxruntime/core/providers/cuda/cuda_execution_provider.cc
@@ -780,7 +780,7 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain,
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, int32_t_MLFloat16_int32_t, OneHot);
 
 // OpSet 12
-class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 12, Clip);
+class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 12, 12, Clip);
 
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 12, float, MaxPool);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 12, double, MaxPool);
@@ -818,6 +818,7 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain,
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, float, Add);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, double, Add);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, MLFloat16, Add);
+class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, Clip);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, int32_t, Sub);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, int64_t, Sub);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, uint32_t, Sub);
@@ -1467,7 +1468,7 @@ static Status RegisterCudaKernels(KernelRegistry& kernel_registry) {
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 11, int32_t_MLFloat16_int32_t, OneHot)>,
 
       // OpSet 12
-      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 12, Clip)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 12, 12, Clip)>,
 
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 12, float, MaxPool)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 12, double, MaxPool)>,
@@ -1504,6 +1505,7 @@ static Status RegisterCudaKernels(KernelRegistry& kernel_registry) {
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, uint64_t, Add)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, float, Add)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, double, Add)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, Clip)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, MLFloat16, Add)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, int32_t, Sub)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 13, int64_t, Sub)>,
diff --git a/onnxruntime/core/providers/cuda/math/clip.cc b/onnxruntime/core/providers/cuda/math/clip.cc
index 55176624f7..6189f703f2 100644
--- a/onnxruntime/core/providers/cuda/math/clip.cc
+++ b/onnxruntime/core/providers/cuda/math/clip.cc
@@ -30,10 +30,10 @@ ONNX_OPERATOR_VERSIONED_KERNEL_EX(
         .TypeConstraint("T", DataTypeImpl::GetTensorType<float>()),
     Clip);
 
-ONNX_OPERATOR_KERNEL_EX(
+ONNX_OPERATOR_VERSIONED_KERNEL_EX(
     Clip,
     kOnnxDomain,
-    12,
+    12, 12,
     kCudaExecutionProvider,
     KernelDefBuilder()
         .InputMemoryType<OrtMemTypeCPUInput>(1)
@@ -41,6 +41,18 @@ ONNX_OPERATOR_KERNEL_EX(
         .TypeConstraint("T", BuildKernelDefConstraints<float, double, int8_t, uint8_t, int64_t, uint64_t>()),
     Clip);
 
+ONNX_OPERATOR_KERNEL_EX(
+    Clip,
+    kOnnxDomain,
+    13,
+    kCudaExecutionProvider,
+    KernelDefBuilder()
+        .InputMemoryType<OrtMemTypeCPUInput>(1)
+        .InputMemoryType<OrtMemTypeCPUInput>(2)
+        .TypeConstraint("T", BuildKernelDefConstraints<float, double, int8_t, uint8_t, int64_t, uint64_t>()),
+    Clip);
+
+
 template <typename T>
 Status Clip_6<T>::ComputeInternal(OpKernelContext* ctx) const {
   const Tensor& X = *ctx->Input<Tensor>(0);