From 49d197a8e6368802f3a0c86a52cabeaffb154cc7 Mon Sep 17 00:00:00 2001
From: Yi-Hong Lyu <yilyu@microsoft.com>
Date: Fri, 10 May 2024 16:07:42 -0700
Subject: [PATCH] Enable ClipQuantFusion exclusively on CPU EP (#20627)

### Motivation and Context

The Intel NPU does not support 16-bit int quantized operators.
Consequently, the execution provider removes the
QuantizeLinear/DeQuantizeLinear (Q/DQ) operators from node units and
executes the operation as FP16 in the backend. However, if a Clip
operator was fused into a Q operator in the node unit, the removal of
Q/DQ operators results in inaccuracies because the effect of the
original Clip operators is lost.

Consider the following example:
- FP32 model: -> Op_FP32 -> Clip ->
- QDQ model: -> (DQ-> Op_FP32 -> Q) -> (DQ' -> Clip -> Q') ->
- After ClipQuantFusion: -> (DQ-> Op_FP32 -> Q) -> (DQ' -> Q') ->
- Intel Execution Provider strips Q/DQ: -> Op_FP16 ->

To solve this issue, we have enabled ClipQuantFusion exclusively on the
CPU execution provider.
---
 onnxruntime/core/optimizer/graph_transformer_utils.cc         | 3 +--
 .../core/optimizer/qdq_transformer/clip_quantizelinear.cc     | 4 +++-
 onnxruntime/test/optimizer/qdq_transformer_test.cc            | 2 +-
 3 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/onnxruntime/core/optimizer/graph_transformer_utils.cc b/onnxruntime/core/optimizer/graph_transformer_utils.cc
index 9341ade0a2f1d..66ac676fa2f6d 100644
--- a/onnxruntime/core/optimizer/graph_transformer_utils.cc
+++ b/onnxruntime/core/optimizer/graph_transformer_utils.cc
@@ -132,14 +132,13 @@ InlinedVector<std::unique_ptr<RewriteRule>> GenerateRewriteRules(
       rules.push_back(std::make_unique<ConvBNFusion>());
       rules.push_back(std::make_unique<PadFusion>());
       rules.push_back(std::make_unique<MatmulBNFusion>());
-      rules.push_back(std::make_unique<ClipQuantFusion>());
       rules.push_back(std::make_unique<ReluQuantFusion>());
       rules.push_back(std::make_unique<LabelEncoderFusion>());
       break;
 
     case TransformerLevel::Level2:
+      rules.push_back(std::make_unique<ClipQuantFusion>());
       rules.push_back(std::make_unique<GemmTransposeFusion>());
-      // No level2 rules available today
       break;
 
     case TransformerLevel::Level3:
diff --git a/onnxruntime/core/optimizer/qdq_transformer/clip_quantizelinear.cc b/onnxruntime/core/optimizer/qdq_transformer/clip_quantizelinear.cc
index 50653b368857d..72ca1cb74f1fd 100644
--- a/onnxruntime/core/optimizer/qdq_transformer/clip_quantizelinear.cc
+++ b/onnxruntime/core/optimizer/qdq_transformer/clip_quantizelinear.cc
@@ -83,13 +83,15 @@ static bool GetQConstantLowerUpper(const Graph& graph, const Node& node, float&
 
 bool ClipQuantFusion::SatisfyCondition(const Graph& graph, const Node& node, const logging::Logger& /*logger*/) const {
   if (!graph_utils::IsSupportedOptypeVersionAndDomain(node, "Clip", {1, 6, 11, 12, 13}) ||
+      !graph_utils::IsSupportedProvider(node, {kCpuExecutionProvider}) ||
       !optimizer_utils::CheckOutputEdges(graph, node, 1)) {
     return false;
   }
 
   // if Clip is followed by QuantizeLinear, it can be fused into QuantizeLinear potentially
   const auto& next_node = *node.OutputNodesBegin();
-  if (!QDQ::MatchQNode(next_node)) {
+  if (!graph_utils::IsSupportedProvider(next_node, {kCpuExecutionProvider}) ||
+      !QDQ::MatchQNode(next_node)) {
     return false;
   }
 
diff --git a/onnxruntime/test/optimizer/qdq_transformer_test.cc b/onnxruntime/test/optimizer/qdq_transformer_test.cc
index ae263a7ca7d35..8c138b22bd52b 100644
--- a/onnxruntime/test/optimizer/qdq_transformer_test.cc
+++ b/onnxruntime/test/optimizer/qdq_transformer_test.cc
@@ -2565,7 +2565,7 @@ TEST(QDQTransformerTests, Clip) {
 
     TransformerTester(build_test_case, check_clip_graph,
                       TransformerLevel::Default,
-                      TransformerLevel::Level1,
+                      TransformerLevel::Level2,
                       opset_version,
                       epsilon,
                       epsilon);