Add unsigned int8 scale propagation (#46378)

* Add unsigned int8 propagation

* Add or modify unit tests

* Correct concat scale checking

* Apply review suggestions

* Corrections

paddle/fluid/framework/ir/mkldnn/compute_propagate_scales_mkldnn_pass.cc

@@ -19,7 +19,6 @@
 #include <algorithm>
 
 #include "paddle/fluid/framework/ir/graph_helper.h"
-#include "paddle/fluid/framework/ir/mkldnn/mkldnn_pass_util.h"
 #include "paddle/fluid/framework/op_version_registry.h"
 
 namespace paddle {
@@ -394,8 +393,13 @@ std::unordered_set<std::string> ComputePropagateScalesMkldnnPass::UpdateScales(
       auto out_iter = var_quant_scales->find(op_node->Op()->Output("Out")[0]);
       if (out_iter != var_quant_scales->end()) {
         std::vector<std::string> input_names = op_node->Op()->Input("X");
-        for (auto input_name : input_names)
-          (*var_quant_scales)[input_name] = out_iter->second;
+        for (auto input_name : input_names) {
+          auto concat_in_iter = var_quant_scales->find(input_name);
+          if (concat_in_iter == var_quant_scales->end())
+            (*var_quant_scales)[input_name] = out_iter->second;
+          else
+            (*var_quant_scales)[input_name].second = out_iter->second.second;
+        }
       }
     } else if (op_name == "scale") {
       const std::string output_name = op_node->Op()->Output("Out")[0];
@@ -409,6 +413,40 @@ std::unordered_set<std::string> ComputePropagateScalesMkldnnPass::UpdateScales(
   }
   return waiting_for_scale;
 }
+void ComputePropagateScalesMkldnnPass::UpdateReluOutputScales(
+    ir::Graph* graph, StringPairMap* var_quant_scales) const {
+  for (auto* op_node :
+       ir::TopologyVarientSort(*graph, static_cast<ir::SortKind>(0))) {
+    if (!op_node->IsOp()) continue;
+    auto op = op_node->Op();
+    bool is_unsigned = false;
+    std::string output_name = "Out";
+    std::string act_name;
+    if (op->Type() == "relu") {
+      is_unsigned = true;
+    } else {
+      if (op->Type() == "conv2d") {
+        act_name = "fuse_activation";
+        output_name = "Output";
+      } else if (op->Type() == "fc") {
+        act_name = "activation_type";
+      }
+      if (!act_name.empty()) {
+        auto act = op->GetAttrIfExists<std::string>(act_name);
+        if (act == "relu" || act == "relu6") {
+          is_unsigned = true;
+        }
+      }
+    }
+    if (is_unsigned) {
+      std::string output_var_name = op->Output(output_name)[0];
+      auto out_iter = var_quant_scales->find(output_var_name);
+      if (out_iter != var_quant_scales->end()) {
+        (*var_quant_scales)[output_var_name].first = true;
+      }
+    }
+  }
+}
 
 void ComputePropagateScalesMkldnnPass::PropagateScales(
     ir::Graph* graph,
@@ -427,21 +465,6 @@ void ComputePropagateScalesMkldnnPass::PropagateScales(
   }
 }
 
-void ComputePropagateScalesMkldnnPass::ConvertStringPairMap(
-    const StringPairMap& var_quant_scales,
-    std::unordered_map<std::string, std::vector<float>>* info_map) const {
-  for (auto iter = var_quant_scales.begin(); iter != var_quant_scales.end();
-       iter++) {
-    auto* data = iter->second.second.data<float>();
-    std::vector<float> data_v;
-    for (int i = 0; i < iter->second.second.numel(); i++) {
-      data_v.push_back(data[i]);
-    }
-
-    info_map->insert(std::make_pair(iter->first, data_v));
-  }
-}
-
 void ComputePropagateScalesMkldnnPass::ApplyImpl(ir::Graph* graph) const {
   VLOG(3) << "Convert paddle model to mkldnn quantized model.";
   const std::string pattern_name = "compute_propagate_scales_mkldnn_pass";
@@ -461,13 +484,13 @@ void ComputePropagateScalesMkldnnPass::ApplyImpl(ir::Graph* graph) const {
   auto* scope = param_scope();
   GetQuantInfo(graph, &var_quant_scales);
   ComputeWeightScales(graph, scope, &var_quant_scales);
+  UpdateReluOutputScales(graph, &var_quant_scales);
   PropagateScales(graph, &var_quant_scales, scale_immutable_ops);
 
   // save var_quant_scales in the first op's attr
   // for cpu_quantize_pass
-  std::unordered_map<std::string, std::vector<float>> info_map;
-  ConvertStringPairMap(var_quant_scales, &info_map);
-  SaveInfoInTheFirstOp(graph, "has_quant_info", "var_quant_scales", info_map);
+  SaveInfoInTheFirstOp(
+      graph, "has_quant_info", "var_quant_scales", var_quant_scales);
 }
 
 }  // namespace ir

paddle/fluid/framework/ir/mkldnn/compute_propagate_scales_mkldnn_pass.h

@@ -17,14 +17,12 @@
 #include <string>
 
 #include "paddle/fluid/framework/ir/fuse_pass_base.h"
+#include "paddle/fluid/framework/ir/mkldnn/mkldnn_pass_util.h"
 
 namespace paddle {
 namespace framework {
 namespace ir {
 
-using StringPairMap =
-    std::unordered_map<std::string, std::pair<bool, phi::DenseTensor>>;
-
 class ComputePropagateScalesMkldnnPass : public FusePassBase {
  public:
   ComputePropagateScalesMkldnnPass() = default;
@@ -78,6 +76,9 @@ class ComputePropagateScalesMkldnnPass : public FusePassBase {
                            Scope* scope,
                            StringPairMap* var_quant_scales) const;
 
+  void UpdateReluOutputScales(ir::Graph* graph,
+                              StringPairMap* var_quant_scales) const;
+
   void UpdateScaleOpInScale(Node* op_node,
                             const std::string& input_name,
                             const std::string& output_name,
@@ -92,10 +93,6 @@ class ComputePropagateScalesMkldnnPass : public FusePassBase {
       ir::Graph* graph,
       StringPairMap* var_quant_scales,
       const std::unordered_set<std::string>& scale_immutable_ops) const;
-
-  void ConvertStringPairMap(
-      const StringPairMap& var_quant_scales,
-      std::unordered_map<std::string, std::vector<float>>* info_map) const;
 };
 }  // namespace ir
 }  // namespace framework

paddle/fluid/framework/ir/mkldnn/compute_propagate_scales_mkldnn_pass_tester.cc

@@ -13,6 +13,7 @@
 // limitations under the License.
 
 #include <gtest/gtest.h>
+#include <unordered_map>
 
 #include "paddle/fluid/framework/ir/mkldnn/compute_propagate_scales_mkldnn_pass.h"
 #include "paddle/fluid/framework/naive_executor.h"
@@ -91,11 +92,16 @@ class ComputePropagateScalesMkldnnPassTest : public testing::Test {
         graph, scope, wx_name, wh_name, var_quant_scales);
   }
 
+  void UpdateReluOutputScales(ir::Graph* graph,
+                              StringPairMap* var_quant_scales) const {
+    pass->UpdateReluOutputScales(graph, var_quant_scales);
+  }
+
   void InitTensorHolder(Scope* scope,
                         const paddle::platform::Place& place,
                         const std::string& var_name) {
     auto x = scope->Var(var_name);
-    auto tensor = x->GetMutable<LoDTensor>();
+    auto tensor = x->GetMutable<phi::DenseTensor>();
     auto tensor_size = 1;
     if (var_name == "filter") {
       tensor_size = positive_and_negative_values.size();
@@ -124,7 +130,6 @@ class ComputePropagateScalesMkldnnPassTest : public testing::Test {
   }
 
   void ComputeRnnWeightScalesTest(const std::string& type,
-                                  const std::initializer_list<std::string>& ops,
                                   const framework::ProgramDesc& prog,
                                   std::vector<double> scales) {
     ir::Graph* graph(new ir::Graph(prog));
@@ -140,7 +145,7 @@ class ComputePropagateScalesMkldnnPassTest : public testing::Test {
     StringPairMap var_quant_scales;
 
     auto* wx_var = scope.FindVar(wx_var_names);
-    auto* wx_tensor = wx_var->GetMutable<LoDTensor>();
+    auto* wx_tensor = wx_var->GetMutable<phi::DenseTensor>();
     wx_tensor->Resize(phi::make_dim(wx.size(), wx[0].size()));
     for (size_t i = 0; i < wx.size(); i++)
       std::copy(begin(wx[i]),
@@ -149,7 +154,7 @@ class ComputePropagateScalesMkldnnPassTest : public testing::Test {
                     i * wx[0].size());
 
     auto* wh_var = scope.FindVar(wh_var_names);
-    auto* wh_tensor = wh_var->GetMutable<LoDTensor>();
+    auto* wh_tensor = wh_var->GetMutable<phi::DenseTensor>();
     wh_tensor->Resize(phi::make_dim(wh.size(), wh[0].size()));
     for (size_t i = 0; i < wh.size(); i++)
       std::copy(begin(wh[i]),
@@ -174,6 +179,24 @@ class ComputePropagateScalesMkldnnPassTest : public testing::Test {
     }
   }
 
+  void UpdateReluOutputScaleTest(
+      const framework::ProgramDesc& prog,
+      StringPairMap* var_quant_scales,
+      const std::initializer_list<std::string>& variable_names) {
+    ir::Graph* graph(new ir::Graph(prog));
+    Scope scope;
+
+    PrepareGraph(graph, prog, &scope, conv_variable_names);
+
+    UpdateReluOutputScales(graph, var_quant_scales);
+
+    for (auto& var_name : variable_names) {
+      auto iter = var_quant_scales->find(var_name);
+      ASSERT_NE(iter, var_quant_scales->end());
+      ASSERT_EQ((*var_quant_scales)[var_name].first, true);
+    }
+  }
+
  private:
   std::unique_ptr<ComputePropagateScalesMkldnnPass> pass;
 };
@@ -182,11 +205,15 @@ void SetOp(ProgramDesc* prog,
            const std::string& type,
            const std::string& name,
            const std::vector<std::string>& inputs,
-           const std::vector<std::string>& outputs) {
+           const std::vector<std::string>& outputs,
+           const std::unordered_map<std::string, std::string>& attrs = {}) {
   auto* op = prog->MutableBlock(0)->AppendOp();
   op->SetType(type);
   op->SetAttr("use_mkldnn", true);
   op->SetAttr("name", name);
+  if (!attrs.empty())
+    for (auto& attr : attrs) op->SetAttr(attr.first, attr.second);
+
   if (type == "conv2d") {
     op->SetInput("Input", {inputs[0]});
     if (inputs.size() > 1) op->SetInput("Filter", {inputs[1]});
@@ -211,6 +238,23 @@ ProgramDesc BuildConv2dProgramDesc() {
   return prog;
 }
 
+ProgramDesc BuildConv2dReluProgramDesc() {
+  ProgramDesc prog;
+  for (auto& v : conv_variable_names) {
+    prog.MutableBlock(0)->Var(v);
+  }
+  std::unordered_map<std::string, std::string> attrs = {
+      {"fuse_activation", "relu"}};
+  SetOp(&prog,
+        "conv2d",
+        "Conv2d",
+        {"conv_in", "filter", "bias"},
+        {"conv_out"},
+        attrs);
+
+  return prog;
+}
+
 ProgramDesc BuildFusionGruProgramDesc() {
   ProgramDesc prog;
   for (auto& v : rnn_variable_names) {
@@ -262,7 +306,7 @@ TEST_F(ComputePropagateScalesMkldnnPassTest, compute_var_scales) {
   StringPairMap var_quant_scales;
 
   auto* var = scope.FindVar(weight_var_name);
-  auto* weight_tensor = var->GetMutable<LoDTensor>();
+  auto* weight_tensor = var->GetMutable<phi::DenseTensor>();
   weight_tensor->Resize(phi::make_dim(1, values.size()));
   std::copy(begin(values),
             end(values),
@@ -283,15 +327,24 @@ TEST_F(ComputePropagateScalesMkldnnPassTest, compute_var_scales) {
 }
 
 TEST_F(ComputePropagateScalesMkldnnPassTest, compute_gru_weight_scales) {
-  ComputeRnnWeightScalesTest("gru",
-                             {"fusion_gru", "multi_gru"},
-                             BuildFusionGruProgramDesc(),
-                             gru_scales);
+  ComputeRnnWeightScalesTest("gru", BuildFusionGruProgramDesc(), gru_scales);
 }
 
 TEST_F(ComputePropagateScalesMkldnnPassTest, compute_lstm_weight_scales) {
-  ComputeRnnWeightScalesTest(
-      "lstm", {"fusion_lstm"}, BuildFusionLstmProgramDesc(), lstm_scales);
+  ComputeRnnWeightScalesTest("lstm", BuildFusionLstmProgramDesc(), lstm_scales);
+}
+
+TEST_F(ComputePropagateScalesMkldnnPassTest, update_relu_output_scales) {
+  StringPairMap var_quant_scales;
+  for (auto& var_name : conv_variable_names) {
+    phi::DenseTensor tensor;
+    auto* data = tensor.mutable_data<float>({1}, platform::CPUPlace());
+    data[0] = 10;
+    auto pair = std::make_pair(false, tensor);
+    var_quant_scales.insert(std::make_pair(var_name, pair));
+  }
+  UpdateReluOutputScaleTest(
+      BuildConv2dReluProgramDesc(), &var_quant_scales, {"conv_out"});
 }
 
 }  // namespace ir

paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.cc

@@ -229,6 +229,7 @@ void CPUQuantizePass::DequantizeOutput(Graph* g,
                     std::vector<std::string>({dequantize_in_node->Name()}));
   deq_desc.SetOutput("Output", std::vector<std::string>({output->Name()}));
   deq_desc.SetAttr("Scale", scale);
+  deq_desc.SetAttr("is_negative_input", !is_unsigned);
   auto dequantize_op = g->CreateOpNode(&deq_desc);  // OpDesc will be copied.
 
   // update op's output
@@ -332,20 +333,8 @@ bool CPUQuantizePass::IsOpQuantized(const Node* node) const {
 }
 
 void CPUQuantizePass::GetQuantInfo(Graph* graph) const {
-  std::unordered_map<std::string, std::vector<float>> info_map{};
-  GetInfoFromTheFirstOp(graph, "has_quant_info", "var_quant_scales", &info_map);
-
-  for (auto iter = info_map.begin(); iter != info_map.end(); iter++) {
-    LoDTensor tensor;
-    const int size = static_cast<int>(iter->second.size());
-    auto* data = tensor.mutable_data<double>({size}, platform::CPUPlace());
-    for (int i = 0; i < size; i++) {
-      data[i] = static_cast<double>(iter->second[i]);
-    }
-
-    auto pair = std::make_pair(false, tensor);
-    var_quant_scales_->insert(std::make_pair(iter->first, pair));
-  }
+  GetInfoFromTheFirstOp(
+      graph, "has_quant_info", "var_quant_scales", var_quant_scales_);
 }
 
 void CPUQuantizePass::QuantizeConv(Graph* graph,
@@ -593,6 +582,20 @@ void CPUQuantizePass::QuantizeConcat(Graph* graph) const {
       return;
     }
 
+    bool are_all_inputs_unsigned{true};
+    // if all inputs were unsigned, then the output was set to unsigned
+    // during the scale calculation step
+    auto inputs = concat_op->inputs;
+    for (size_t i = 0; i < inputs.size(); i++) {
+      if (AreScalesPresentForVarNames({inputs[i]->Name()})) {
+        auto scale_data = GetScaleDataByName(inputs[i]->Name());
+        if (scale_data.first == false) {
+          are_all_inputs_unsigned = false;
+          break;
+        }
+      }
+    }
+
     GET_IR_NODE_FROM_SUBGRAPH(concat_out, concat_out, concat_pattern);
 
     if (!AreScalesPresentForNodes({concat_out})) {
@@ -601,17 +604,12 @@ void CPUQuantizePass::QuantizeConcat(Graph* graph) const {
       return;
     }
 
-    // if all inputs were unsigned, then the output was set to unsigned
-    // during the scale calculation step
-    bool are_all_inputs_unsigned{false};
-    auto output_scale =
-        GetScaleValueForNode(concat_out, &are_all_inputs_unsigned);
+    auto output_scale = GetScaleValueForNode(concat_out);
 
     QuantizeInputs(g, concat_op, "X", are_all_inputs_unsigned);
 
     DequantizeOutput(
         g, concat_op, concat_out, "Out", output_scale, are_all_inputs_unsigned);
-
     ++quantize_concat_count;
   };
 

paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass.cc

@@ -105,51 +105,24 @@ void CPUQuantizeSquashPass::FindNodesToKeep(
   AddStatis(found_count);
 }
 
-bool CPUQuantizeSquashPass::IsDequantizeInputUint8(
-    const Node* dequant_in) const {
-  PADDLE_ENFORCE_EQ(
-      dequant_in->inputs.size(),
-      1,
-      platform::errors::InvalidArgument(
-          "Dequantize (id: %f) should have only one input.", dequant_in->id()));
-  if (dequant_in->inputs[0]->IsOp()) {
-    auto prev_op = dequant_in->inputs[0]->Op();
-    std::string act_name;
-    if (prev_op->Type() == "relu") {
-      return true;
-    } else {
-      if (prev_op->Type() == "conv2d") {
-        act_name = "fuse_activation";
-      } else if (prev_op->Type() == "fc") {
-        act_name = "activation_type";
-      }
-      if (!act_name.empty()) {
-        auto act = prev_op->GetAttrIfExists<std::string>(act_name);
-        if (act == "relu" || act == "relu6") {
-          return true;
-        }
-      }
-    }
-  }
-  return false;
-}
-
 bool CPUQuantizeSquashPass::IsDequantizeQuantizeIncompatible(
-    Node* quant_op, Node* dequant_in, Node* next_op) const {
-  bool is_concat_signed =
+    Node* quant_op, Node* dequant_op, Node* next_op) const {
+  bool is_next_op_signed =
       quant_op->Op()->GetAttrIfExists<bool>("is_negative_input");
-  bool is_input_unsigned = IsDequantizeInputUint8(dequant_in);
+  bool is_input_signed =
+      dequant_op->Op()->GetAttrIfExists<bool>("is_negative_input");
+
   /* TODO(sfraczek): remove elementwise from this condition when BinaryMKLDNN
    kernel will support two different input data types */
   bool is_next_op_concat_or_elementwise =
       next_op->Op()->Type() == "concat" ||
       next_op->Op()->Type().find("elementwise") == 0;
-  if (is_next_op_concat_or_elementwise && is_concat_signed &&
-      is_input_unsigned) {
+  if (is_next_op_concat_or_elementwise &&
+      (is_next_op_signed ^ is_input_signed)) {
     VLOG(4) << "Do not squash dequant-quant, because "
             << "next_op is: " << next_op->Op()->Type()
-            << ", is_concat_signed: " << is_concat_signed
-            << ", is_input_unsigned: " << is_input_unsigned << ".";
+            << ", is_next_op_signed: " << is_next_op_signed
+            << ", is_input_signed: " << is_input_signed << ".";
     return true;
   }
   return false;
@@ -174,7 +147,7 @@ void CPUQuantizeSquashPass::DequantQuantSquash(
     GET_IR_NODE_FROM_SUBGRAPH(quant_out, quant_out, squash_pattern);
     GET_IR_NODE_FROM_SUBGRAPH(next_op, next_op, squash_pattern);
 
-    if (IsDequantizeQuantizeIncompatible(quant_op, dequant_in, next_op)) {
+    if (IsDequantizeQuantizeIncompatible(quant_op, dequant_op, next_op)) {
       return;
     }
 

paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass.h

@@ -43,11 +43,6 @@ class CPUQuantizeSquashPass : public FusePassBase {
       Graph* graph,
       std::unordered_map<const Node*, int>* nodes_keep_counter) const;
 
-  /*
-   * Check if input to dequantize is uint8
-   */
-  bool IsDequantizeInputUint8(const Node* dequant_in) const;
-
   /*
    * Don't squash unsigned dequantize with signed quantize.
    * This is important for concat and elementwise ops.
@@ -55,7 +50,7 @@ class CPUQuantizeSquashPass : public FusePassBase {
    * elementwise assumes first input type.
    */
   bool IsDequantizeQuantizeIncompatible(Node* quant_op,
-                                        Node* dequant_in,
+                                        Node* dequant_op,
                                         Node* next_op) const;
 
   /*

paddle/fluid/framework/ir/mkldnn/cpu_quantize_squash_pass_tester.cc

@@ -68,15 +68,11 @@ void SetOp(ProgramDesc* prog,
     op->SetAttr("padding_algorithm", std::string("EXPLICIT"));
     op->SetAttr("data_format", std::string("NCHW"));
     op->SetAttr("force_fp32_output", false);
-  } else if (type == "quantize") {
+  } else if (type == "quantize" || type == "dequantize") {
     op->SetInput("Input", {inputs[0]});
     op->SetOutput("Output", {outputs[0]});
     op->SetAttr("Scale", scale[0]);
     op->SetAttr("is_negative_input", is_negative_input);
-  } else if (type == "dequantize") {
-    op->SetInput("Input", {inputs[0]});
-    op->SetOutput("Output", {outputs[0]});
-    op->SetAttr("Scale", scale[0]);
   } else if (type == "requantize") {
     op->SetInput("Input", {inputs[0]});
     op->SetOutput("Output", {outputs[0]});
@@ -303,31 +299,22 @@ ProgramDesc BuildConvMultiRequantProgramDesc(bool use_mkldnn,
   return prog;
 }
 
-/* a->relu->b->Dequant->c(u8)->Quant->d-\
- * e->relu->f->Dequant->g(u8)->Quant->h--Concat1->x
- * i->relu->j->Dequant->k(u8)->Quant->l-/
+/* a->relu->b->Dequant(u8)->c->Quant(u8)->d-\
+ * e->relu->f->Dequant(u8)->g->Quant(u8)->h--Concat1->i
  */
-ProgramDesc BuildU8U8U8ConcatProgramDesc(float scale_out, float scale) {
+ProgramDesc BuildU8U8ConcatProgramDesc(float scale_out, float scale) {
   ProgramDesc prog;
   for (auto& v : variable_names) {
     prog.MutableBlock(0)->Var(v);
   }
   SetOp(&prog, "relu", "Relu1", {"a"}, {"b"}, true, {scale, scale_out});
   SetOp(&prog, "relu", "Relu2", {"e"}, {"f"}, true, {scale, scale_out});
-  SetOp(&prog, "relu", "Relu3", {"i"}, {"j"}, true, {scale, scale_out});
-
-  SetOp(
-      &prog, "dequantize", "Dequant1", {"b"}, {"c"}, true, {scale, scale_out});
-  SetOp(
-      &prog, "dequantize", "Dequant2", {"f"}, {"g"}, true, {scale, scale_out});
-  SetOp(
-      &prog, "dequantize", "Dequant3", {"j"}, {"k"}, true, {scale, scale_out});
 
   SetOp(&prog,
-        "quantize",
-        "Quant1",
+        "dequantize",
+        "Dequant1",
+        {"b"},
         {"c"},
-        {"d"},
         true,
         {scale, scale_out},
         0.0f,
@@ -336,10 +323,23 @@ ProgramDesc BuildU8U8U8ConcatProgramDesc(float scale_out, float scale) {
         1,
         false);  // is_negative_input = false
   SetOp(&prog,
-        "quantize",
-        "Quant2",
+        "dequantize",
+        "Dequant2",
+        {"f"},
         {"g"},
-        {"h"},
+        true,
+        {scale, scale_out},
+        0.0f,
+        "float32",
+        false,
+        1,
+        false);  // is_negative_input = false
+
+  SetOp(&prog,
+        "quantize",
+        "Quant1",
+        {"c"},
+        {"d"},
         true,
         {scale, scale_out},
         0.0f,
@@ -349,9 +349,9 @@ ProgramDesc BuildU8U8U8ConcatProgramDesc(float scale_out, float scale) {
         false);  // is_negative_input = false
   SetOp(&prog,
         "quantize",
-        "Quant3",
-        {"k"},
-        {"l"},
+        "Quant2",
+        {"g"},
+        {"h"},
         true,
         {scale, scale_out},
         0.0f,
@@ -360,27 +360,47 @@ ProgramDesc BuildU8U8U8ConcatProgramDesc(float scale_out, float scale) {
         1,
         false);  // is_negative_input = false
 
-  SetOp(&prog, "concat", "Concat1", {"d", "h", "l"}, {"x"}, true);
+  SetOp(&prog, "concat", "Concat1", {"d", "h"}, {"i"}, true);
   return prog;
 }
 
-/* a->relu->b->Dequant->c(u8)->Quant->d-\
- * e->relu->f->Dequant->g(u8)->Quant->h--Concat1->x
- * i->pool2d->j->Dequant->k(s8)->Quant->l-/
+/* a->relu->b->Dequant(u8)->c->Quant(s8)->d-\
+ * e->relu->f->Dequant(u8)->g->Quant(s8)->h--Concat1->x
+ * i->pool2d->j->Dequant(s8)->k->Quant(s8)->l-/
  */
 ProgramDesc BuildU8U8S8ConcatProgramDesc(float scale_out, float scale) {
   ProgramDesc prog;
   for (auto& v : variable_names) {
     prog.MutableBlock(0)->Var(v);
   }
-  SetOp(&prog, "relu", "Pool2d1", {"a"}, {"b"}, true, {scale, scale_out});
-  SetOp(&prog, "relu", "Relu1", {"e"}, {"f"}, true, {scale, scale_out});
+  SetOp(&prog, "relu", "Relu1", {"a"}, {"b"}, true, {scale, scale_out});
+  SetOp(&prog, "relu", "Relu2", {"e"}, {"f"}, true, {scale, scale_out});
   SetOp(&prog, "pool2d", "Pool2d2", {"i"}, {"j"}, true, {scale, scale_out});
 
-  SetOp(
-      &prog, "dequantize", "Dequant1", {"b"}, {"c"}, true, {scale, scale_out});
-  SetOp(
-      &prog, "dequantize", "Dequant2", {"f"}, {"g"}, true, {scale, scale_out});
+  SetOp(&prog,
+        "dequantize",
+        "Dequant1",
+        {"b"},
+        {"c"},
+        true,
+        {scale, scale_out},
+        0.0f,
+        "float32",
+        false,
+        1,
+        false);  // is_negative_input = false
+  SetOp(&prog,
+        "dequantize",
+        "Dequant2",
+        {"f"},
+        {"g"},
+        true,
+        {scale, scale_out},
+        0.0f,
+        "float32",
+        false,
+        1,
+        false);  // is_negative_input = false
   SetOp(
       &prog, "dequantize", "Dequant3", {"j"}, {"k"}, true, {scale, scale_out});
 
@@ -392,9 +412,9 @@ ProgramDesc BuildU8U8S8ConcatProgramDesc(float scale_out, float scale) {
   return prog;
 }
 
-/* a->pool2d->b->Dequant->c(s8)->Quant->d-\
- * e->relu->f->Dequant->g(u8)->Quant->h--Concat1->x
- * i->pool2d->j->Dequant->k(s8)->Quant->l-/
+/* a->pool2d->b->Dequant(s8)->c->Quant(s8)->d-\
+ * e->relu->f->Dequant(u8)->g->Quant(s8)->h--Concat1->x
+ * i->pool2d->j->Dequant(s8)->k->Quant(s8)->l-/
  */
 ProgramDesc BuildS8U8S8ConcatProgramDesc(float scale_out, float scale) {
   ProgramDesc prog;
@@ -407,8 +427,18 @@ ProgramDesc BuildS8U8S8ConcatProgramDesc(float scale_out, float scale) {
 
   SetOp(
       &prog, "dequantize", "Dequant1", {"b"}, {"c"}, true, {scale, scale_out});
-  SetOp(
-      &prog, "dequantize", "Dequant2", {"f"}, {"g"}, true, {scale, scale_out});
+  SetOp(&prog,
+        "dequantize",
+        "Dequant2",
+        {"f"},
+        {"g"},
+        true,
+        {scale, scale_out},
+        0.0f,
+        "float32",
+        false,
+        1,
+        false);  // is_negative_input = false
   SetOp(
       &prog, "dequantize", "Dequant3", {"j"}, {"k"}, true, {scale, scale_out});
 
@@ -1141,13 +1171,12 @@ TEST(CpuQuantizeSquashPass, squash_all_s8_input_to_concat1) {
 }
 
 TEST(CpuQuantizeSquashPass, squash_all_u8_input_to_concat2) {
-  // removed 3 x 4 (dequantize_op, dequantize_out, quantize, quantize_out)
-  auto remove_nodes = 12;
+  // removed 2 x 4 (dequantize_op, dequantize_out, quantize, quantize_out)
+  auto remove_nodes = 8;
   std::unordered_map<std::string, int> expected_operators = {
-      {"concat", 1}, {"quantize", 0}, {"dequantize", 0}, {"relu", 3}};
-  CheckNodesTest(BuildU8U8U8ConcatProgramDesc(1.2f, 1.2f),
-                 expected_operators,
-                 remove_nodes);
+      {"concat", 1}, {"quantize", 0}, {"dequantize", 0}, {"relu", 2}};
+  CheckNodesTest(
+      BuildU8U8ConcatProgramDesc(1.2f, 1.2f), expected_operators, remove_nodes);
 }
 
 }  // namespace ir

paddle/fluid/framework/ir/mkldnn/mkldnn_pass_util.h

@@ -22,6 +22,9 @@ namespace paddle {
 namespace framework {
 namespace ir {
 
+using StringPairMap =
+    std::unordered_map<std::string, std::pair<bool, phi::DenseTensor>>;
+
 static void SaveInfoInTheFirstOp(
     ir::Graph* graph,
     const std::string& flag,
@@ -44,6 +47,31 @@ static void SaveInfoInTheFirstOp(
   }
 }
 
+static void SaveInfoInTheFirstOp(ir::Graph* graph,
+                                 const std::string& flag,
+                                 const std::string& key_suffix,
+                                 const StringPairMap& info_map) {
+  VLOG(3) << "save variables in the first op's attr";
+
+  const std::string suffix = "_" + key_suffix + "_" + flag;
+  for (auto* op_node :
+       ir::TopologyVarientSort(*graph, static_cast<ir::SortKind>(0))) {
+    if (!op_node->IsOp() || op_node->Op()->Type() == "feed" ||
+        op_node->Op()->Type() == "fetch")
+      continue;
+
+    op_node->Op()->SetAttr(flag, true);
+    for (auto iter = info_map.begin(); iter != info_map.end(); ++iter) {
+      auto* data = iter->second.second.data<float>();
+      std::vector<float> data_v(data, data + iter->second.second.numel());
+      op_node->Op()->SetAttr(iter->first + suffix + "_unsigned",
+                             iter->second.first);
+      op_node->Op()->SetAttr(iter->first + suffix, data_v);
+    }
+    break;
+  }
+}
+
 static void GetInfoFromTheFirstOp(
     ir::Graph* graph,
     const std::string& flag,
@@ -77,6 +105,54 @@ static void GetInfoFromTheFirstOp(
   }
 }
 
+static void GetInfoFromTheFirstOp(ir::Graph* graph,
+                                  const std::string& flag,
+                                  const std::string& key_suffix,
+                                  StringPairMap* info_map) {
+  VLOG(3) << "get variables from the first op's attr";
+  const std::string unsigned_flag = "_unsigned";
+  const std::string suffix = "_" + key_suffix + "_" + flag;
+  const std::string suffix_is_unsigned = suffix + unsigned_flag;
+  for (auto* op_node :
+       ir::TopologyVarientSort(*graph, static_cast<ir::SortKind>(0))) {
+    if (!op_node->IsOp() || op_node->Op()->Type() == "feed" ||
+        op_node->Op()->Type() == "fetch")
+      continue;
+
+    auto* op_desc = op_node->Op();
+    if (op_desc->GetAttrIfExists<bool>(flag)) {
+      op_desc->RemoveAttr(flag);
+      std::vector<std::string> attr_names = op_desc->AttrNames();
+      for (auto fake_name : attr_names) {
+        auto is_unsigned = false;
+        size_t pos = fake_name.find(suffix_is_unsigned);
+
+        if (pos != std::string::npos) {
+          std::string unsigned_var_name = fake_name;
+          is_unsigned =
+              PADDLE_GET_CONST(bool, op_desc->GetAttr(unsigned_var_name));
+
+          std::string var_name = fake_name.substr(0, pos);
+          size_t unsigned_pos = fake_name.find(unsigned_flag);
+          std::string vector_name =
+              fake_name.erase(unsigned_pos, unsigned_flag.length());
+          auto scales_vector = PADDLE_GET_CONST(std::vector<float>,
+                                                op_desc->GetAttr(vector_name));
+          phi::DenseTensor tensor;
+          const int size = static_cast<int>(scales_vector.size());
+          auto data = tensor.mutable_data<double>({size}, platform::CPUPlace());
+          std::copy(scales_vector.begin(), scales_vector.end(), data);
+          auto pair = std::make_pair(is_unsigned, tensor);
+          info_map->insert(std::make_pair(var_name, pair));
+          op_desc->RemoveAttr(unsigned_var_name);
+          op_desc->RemoveAttr(vector_name);
+        }
+      }
+      break;
+    }
+  }
+}
+
 }  // namespace ir
 }  // namespace framework
 }  // namespace paddle