cpu_quantize_pass.cc 39.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

15 16
#include "paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.h"

17
#include <sstream>
18 19
#include <utility>
#include <vector>
W
wanghuancoder 已提交
20

B
baoachun 已提交
21
#include "paddle/fluid/framework/ir/mkldnn/mkldnn_pass_util.h"
22
#include "paddle/fluid/platform/mkldnn_helper.h"
23 24 25 26 27 28
#include "paddle/fluid/string/pretty_log.h"

namespace paddle {
namespace framework {
namespace ir {

29
using EigenVectorArrayMap = Eigen::Map<Eigen::Array<double, Eigen::Dynamic, 1>>;
30 31
using EigenVectorArrayMapFloat =
    Eigen::Map<Eigen::Array<float, Eigen::Dynamic, 1>>;
32 33
using string::PrettyLogDetail;

34 35 36 37 38 39 40 41 42
namespace {

void UnlinkNodes(ir::Node* a, ir::Node* b) {
  a->outputs.erase(std::remove(a->outputs.begin(), a->outputs.end(), b),
                   a->outputs.end());
  b->inputs.erase(std::remove(b->inputs.begin(), b->inputs.end(), a),
                  b->inputs.end());
}

43
void MarkAndLogCannotQuantizeOp(Node* op, const char* details = nullptr) {
44 45 46
  std::stringstream msg_ss;
  msg_ss << "Cannot quantize operator " << op->Name()
         << " (type: " << op->Op()->Type() << ", id: " << op->id() << ").";
47
  if (details) msg_ss << " " << details;
48 49
  VLOG(2) << msg_ss.str().c_str();
  op->Op()->SetAttr("mkldnn_data_type", std::string("float32"));
50 51
}

52 53 54 55 56 57
void LogScaleIsMissingForVarName(const std::string& name) {
  VLOG(4) << "Quantization scale for the variable " << name << " is missing.";
}

void LogScaleIsMissingForVarNode(Node* node) {
  LogScaleIsMissingForVarName(node->Name());
58 59
}

60
void LogQuantizationDisabled(Node* op) {
61
  VLOG(2) << "Quantization skipped for operator " << op->Name()
62
          << " (type: " << op->Op()->Type() << ", id: " << op->id()
63
          << "). Attribute mkldnn_data_type != \"int8\".";
64 65
}

66 67
void LogQuantizedOpsCounter(const std::string& type,
                            const int counter,
68 69 70 71 72 73 74
                            const char* details = nullptr) {
  std::stringstream msg_ss;
  msg_ss << "---    quantized " << counter << " " << type << " ops";
  if (details) msg_ss << " " << details;
  PrettyLogDetail(msg_ss.str().c_str());
}

75 76 77 78
}  // namespace

enum { U8_MAX = 255, S8_MAX = 127 };

79 80 81 82 83
void CPUQuantizePass::QuantizeInput(Graph* g,
                                    Node* op,
                                    Node* input,
                                    std::string input_name,
                                    double scale_to_one,
84
                                    bool is_input_unsigned,
85 86
                                    std::string scale_attr_name,
                                    float shift,
87
                                    std::string shift_attr_name) const {
M
Michał Gallus 已提交
88 89 90
  auto inputs = op->Op()->InputNames();
  bool name_found =
      std::find(inputs.begin(), inputs.end(), input_name) != inputs.end();
91 92
  PADDLE_ENFORCE_EQ(name_found,
                    true,
93 94
                    platform::errors::InvalidArgument(
                        "Var(%s) isn't the input of the %s operator.",
95 96
                        input_name,
                        op->Op()->Type()));
97
  unsigned max = is_input_unsigned ? U8_MAX : S8_MAX;
98 99 100 101 102 103 104 105 106 107 108 109 110
  float scale = scale_to_one * max;

  // Create quantize output variable
  VarDesc quantize_out_desc(patterns::PDNodeName("quantize", "out"));
  auto* quantize_out_node = g->CreateVarNode(&quantize_out_desc);

  // create a quantize op node
  OpDesc q_desc;
  q_desc.SetType("quantize");
  q_desc.SetInput("Input", std::vector<std::string>({input->Name()}));
  q_desc.SetOutput("Output",
                   std::vector<std::string>({quantize_out_node->Name()}));
  q_desc.SetAttr("Scale", scale);
111 112
  q_desc.SetAttr("Shift", shift);
  q_desc.SetAttr("is_negative_input", !is_input_unsigned);
113

Z
Zuza 已提交
114 115 116
  // fix to fc format error
  if (op->Op()->Type() == "fc" &&
      op->Op()->GetAttrIfExists<int>("in_num_col_dims") == 2) {
117 118 119
    q_desc.SetAttr(
        "output_format",
        Has("data_layout") ? Get<std::string>("data_layout") : "NCHW");
Z
Zuza 已提交
120
  } else {
121 122 123
    q_desc.SetAttr(
        "output_format",
        Has("data_layout") ? Get<std::string>("data_layout") : "NHWC");
Z
Zuza 已提交
124
  }
125 126 127 128 129 130 131 132 133 134 135 136 137
  auto quantize_op = g->CreateOpNode(&q_desc);  // OpDesc will be copied.

  // update op's input
  op->Op()->SetInput(input_name,
                     std::vector<std::string>({quantize_out_node->Name()}));

  // link quantize op
  UnlinkNodes(input, op);
  IR_NODE_LINK_TO(input, quantize_op);
  IR_NODE_LINK_TO(quantize_op, quantize_out_node);
  IR_NODE_LINK_TO(quantize_out_node, op);

  if (!scale_attr_name.empty()) op->Op()->SetAttr(scale_attr_name, scale);
138
  if (!shift_attr_name.empty()) op->Op()->SetAttr(shift_attr_name, shift);
139 140
}

141 142 143
void CPUQuantizePass::QuantizeInputs(Graph* g,
                                     Node* op,
                                     std::string input_name,
144
                                     bool are_inputs_unsigned,
145 146
                                     std::string scale_attr_name,
                                     float shift,
147
                                     std::string shift_attr_name) const {
148
  auto inputs = op->inputs;
149
  auto output = op->outputs[0];
150 151
  PADDLE_ENFORCE_GE(inputs.size(),
                    1,
152 153
                    platform::errors::InvalidArgument(
                        "OP(%s)'s inputs(%d) must be equal or greater than 1.",
154 155 156 157
                        op->Name(),
                        inputs.size()));
  PADDLE_ENFORCE_EQ(op->outputs.size(),
                    1,
158
                    platform::errors::InvalidArgument(
159 160
                        "OP(%s)'s outputs(%d) must be equal to 1.",
                        op->Name(),
161
                        op->outputs.size()));
162 163 164 165 166 167 168 169

  // create a quantize op desc prototype
  OpDesc q_desc;
  q_desc.SetType("quantize");

  std::vector<Node*> quantize_out_nodes(inputs.size());
  std::vector<std::string> quantize_out_node_names(inputs.size());

170
  double scale_out = GetScaleValueForNode(output);
171
  unsigned max = are_inputs_unsigned ? U8_MAX : S8_MAX;
172
  float scale = scale_out * max;
173 174 175 176 177 178 179 180

  for (size_t i = 0; i < inputs.size(); i++) {
    // Create quantize output variable
    VarDesc quantize_out_desc(patterns::PDNodeName("quantize", "out"));
    quantize_out_nodes[i] = g->CreateVarNode(&quantize_out_desc);
    quantize_out_node_names[i] = quantize_out_nodes[i]->Name();

    q_desc.SetAttr("Scale", scale);
181
    q_desc.SetAttr("Shift", shift);
182 183 184
    q_desc.SetInput("Input", std::vector<std::string>({inputs[i]->Name()}));
    q_desc.SetOutput("Output",
                     std::vector<std::string>({quantize_out_node_names[i]}));
185
    q_desc.SetAttr("is_negative_input", !are_inputs_unsigned);
186 187 188 189 190 191 192 193 194 195 196 197 198
    auto quantize_op = g->CreateOpNode(&q_desc);  // OpDesc will be copied.

    // link quantize op
    UnlinkNodes(inputs[i], op);
    IR_NODE_LINK_TO(inputs[i], quantize_op);
    IR_NODE_LINK_TO(quantize_op, quantize_out_nodes[i]);
    IR_NODE_LINK_TO(quantize_out_nodes[i], op);
  }

  // update op's input
  op->Op()->SetInput(input_name, quantize_out_node_names);

  if (!scale_attr_name.empty()) op->Op()->SetAttr(scale_attr_name, scale);
199
  if (!shift_attr_name.empty()) op->Op()->SetAttr(shift_attr_name, shift);
200 201
}

202 203 204
void CPUQuantizePass::DequantizeOutput(Graph* g,
                                       Node* op,
                                       Node* output,
205
                                       std::string output_name,
206 207
                                       double scale_to_one,
                                       bool is_unsigned,
208
                                       std::string scale_attr_name) const {
M
Michał Gallus 已提交
209 210 211
  auto outputs = op->Op()->OutputNames();
  bool name_found =
      std::find(outputs.begin(), outputs.end(), output_name) != outputs.end();
212 213
  PADDLE_ENFORCE_EQ(name_found,
                    true,
M
Michał Gallus 已提交
214
                    platform::errors::InvalidArgument(
215
                        "Var(%s) isn't the output of the %s operator.",
216 217
                        output_name,
                        op->Op()->Type()));
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246
  unsigned max = is_unsigned ? U8_MAX : S8_MAX;
  float scale = scale_to_one * max;

  // Create dequantize input variable
  VarDesc dequantize_in_desc(patterns::PDNodeName("dequantize", "in"));
  auto* dequantize_in_node = g->CreateVarNode(&dequantize_in_desc);

  // create a dequantize op node for output.
  OpDesc deq_desc;
  deq_desc.SetType("dequantize");
  deq_desc.SetInput("Input",
                    std::vector<std::string>({dequantize_in_node->Name()}));
  deq_desc.SetOutput("Output", std::vector<std::string>({output->Name()}));
  deq_desc.SetAttr("Scale", scale);
  auto dequantize_op = g->CreateOpNode(&deq_desc);  // OpDesc will be copied.

  // update op's output
  op->Op()->SetOutput(output_name,
                      std::vector<std::string>({dequantize_in_node->Name()}));

  // link dequantize op
  UnlinkNodes(op, output);
  IR_NODE_LINK_TO(op, dequantize_in_node);
  IR_NODE_LINK_TO(dequantize_in_node, dequantize_op);
  IR_NODE_LINK_TO(dequantize_op, output);

  if (!scale_attr_name.empty()) op->Op()->SetAttr(scale_attr_name, scale);
}

247 248 249
bool CPUQuantizePass::AreScalesPresentForVarNames(
    std::vector<std::string> names) const {
  bool present = true;
B
baoachun 已提交
250 251 252 253 254 255 256 257 258 259 260 261 262 263
  if (var_quant_scales_->empty()) {
    auto& scales = Get<VarQuantScale>("quant_var_scales");
    for (auto name : names) {
      if (scales.find(name) == scales.end()) {
        present = false;
        LogScaleIsMissingForVarName(name);
      }
    }
  } else {
    for (auto name : names) {
      if (var_quant_scales_->find(name) == var_quant_scales_->end()) {
        present = false;
        LogScaleIsMissingForVarName(name);
      }
264 265 266 267 268
    }
  }
  return present;
}

269
bool CPUQuantizePass::AreScalesPresentForNodes(
270
    std::initializer_list<Node*> nodes) const {
271
  bool present = true;
B
baoachun 已提交
272 273 274 275 276 277 278 279 280 281 282 283 284 285
  if (var_quant_scales_->empty()) {
    auto& scales = Get<VarQuantScale>("quant_var_scales");
    for (auto node : nodes) {
      if (scales.count(node->Name()) == 0) {
        present = false;
        LogScaleIsMissingForVarNode(node);
      }
    }
  } else {
    for (auto node : nodes) {
      if (var_quant_scales_->count(node->Name()) == 0) {
        present = false;
        LogScaleIsMissingForVarNode(node);
      }
286 287 288 289 290
    }
  }
  return present;
}

291 292
std::pair<bool, LoDTensor> CPUQuantizePass::GetScaleDataByName(
    const std::string& name) const {
B
baoachun 已提交
293 294 295 296 297
  if (var_quant_scales_->empty()) {
    auto& scales = Get<VarQuantScale>("quant_var_scales");
    return scales.at(name);
  }
  return var_quant_scales_->at(name);
298 299
}

300 301
std::pair<bool, LoDTensor> CPUQuantizePass::GetScaleDataForNode(
    const Node* node) const {
302 303 304 305 306
  return GetScaleDataByName(node->Name());
}

LoDTensor CPUQuantizePass::GetScaleTensorByName(const std::string& name) const {
  return GetScaleDataByName(name).second;
307 308 309 310 311 312 313 314 315 316 317 318 319
}

LoDTensor CPUQuantizePass::GetScaleTensorForNode(const Node* node) const {
  return GetScaleDataForNode(node).second;
}

double CPUQuantizePass::GetScaleValueForNode(const Node* node,
                                             bool* is_unsigned) const {
  auto scale_data = GetScaleDataForNode(node);
  if (is_unsigned != nullptr) *is_unsigned = scale_data.first;
  return scale_data.second.data<double>()[0];
}

320 321
bool CPUQuantizePass::IsOpDequantized(const Node* node) const {
  return node->Op()->Type() == "dequantize" ||
322
         platform::HasOpINT8DataType(node->Op());
323 324 325
}

bool CPUQuantizePass::IsOpQuantized(const Node* node) const {
326 327 328 329 330 331
  // return true only if all of outputs are ops and their are either quantize or
  // have int8 data type
  return all_of(node->outputs.begin(), node->outputs.end(), [](Node* output) {
    return (output->IsOp() && (output->Op()->Type() == "quantize" ||
                               platform::HasOpINT8DataType(output->Op())));
  });
332 333
}

B
baoachun 已提交
334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
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));
  }
}

351 352 353 354 355 356 357 358 359 360 361 362 363 364
void CPUQuantizePass::QuantizeConv(Graph* graph,
                                   bool with_residual_data) const {
  GraphPatternDetector gpd;
  auto pattern = gpd.mutable_pattern();
  patterns::ConvResidual conv_pattern{pattern, name_scope_};
  conv_pattern(with_residual_data);

  int quantize_conv_count = 0;
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
    VLOG(4) << "Quantize conv2d op";
    GET_IR_NODE_FROM_SUBGRAPH(conv_op, conv_op, conv_pattern);

    // skip if should not be quantized
365
    if (!platform::HasOpINT8DataType(conv_op->Op())) {
366 367 368
      LogQuantizationDisabled(conv_op);
      return;
    }
369 370 371 372 373

    GET_IR_NODE_FROM_SUBGRAPH(conv_filter, conv_filter, conv_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(conv_input, conv_input, conv_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(conv_output, conv_output, conv_pattern);

374
    auto has_output_scale = AreScalesPresentForNodes({conv_output});
W
Wojciech Uss 已提交
375
    if (with_residual_data && !has_output_scale) {
376 377 378 379
      MarkAndLogCannotQuantizeOp(
          conv_op,
          "Conv op with ResidualData input cannot be quantized "
          "without output scale.");
W
Wojciech Uss 已提交
380 381 382
      return;
    }

383
    if (with_residual_data) {
384 385
      GET_IR_NODE_FROM_SUBGRAPH(
          conv_residual_data, conv_residual_data, conv_pattern);
386
      if (!AreScalesPresentForNodes(
387
              {conv_input, conv_filter, conv_residual_data})) {
388 389
        MarkAndLogCannotQuantizeOp(conv_op,
                                   "No scale available for the operator");
390
        return;
391
      }
392 393 394 395 396

      bool is_residual_unsigned{false};
      auto residual_scale =
          GetScaleValueForNode(conv_residual_data, &is_residual_unsigned);

397 398 399 400 401 402 403
      QuantizeInput(g,
                    conv_op,
                    conv_residual_data,
                    "ResidualData",
                    residual_scale,
                    is_residual_unsigned,
                    "Scale_in_eltwise");
404
    } else {
405
      if (!AreScalesPresentForNodes({conv_input, conv_filter})) {
406 407
        MarkAndLogCannotQuantizeOp(conv_op,
                                   "No scale available for the operator");
408
        return;
409
      }
410 411
    }

412 413
    bool is_input_unsigned{false};
    auto input_scale = GetScaleValueForNode(conv_input, &is_input_unsigned);
414 415 416 417 418 419 420
    QuantizeInput(g,
                  conv_op,
                  conv_input,
                  "Input",
                  input_scale,
                  is_input_unsigned,
                  "Scale_in");
421

422
    auto filter_scale_tensor = GetScaleTensorForNode(conv_filter);
423
    EigenVectorArrayMap eigen_tensor{filter_scale_tensor.data<double>(),
424
                                     filter_scale_tensor.numel()};
425 426 427 428 429 430 431
    eigen_tensor *= static_cast<double>(S8_MAX);
    std::vector<float> filter_scale{
        filter_scale_tensor.data<double>(),
        filter_scale_tensor.data<double>() + filter_scale_tensor.numel()};

    conv_op->Op()->SetAttr("Scale_weights", filter_scale);

432
    // if quantization scale is missing for output tensor, return fp32 data
W
Wojciech Uss 已提交
433
    if (has_output_scale) {
434 435 436
      bool is_output_unsigned{false};
      auto output_scale =
          GetScaleValueForNode(conv_output, &is_output_unsigned);
437 438 439 440 441 442 443
      DequantizeOutput(g,
                       conv_op,
                       conv_output,
                       "Output",
                       output_scale,
                       is_output_unsigned,
                       "Scale_out");
444 445 446
    } else {
      conv_op->Op()->SetAttr("force_fp32_output", true);
    }
447

448
    // change threshold in bounded ReLu
449 450
    if (conv_op->Op()->GetAttrIfExists<std::string>("fuse_activation") ==
        "relu6") {
451 452 453 454
      float scale_out =
          BOOST_GET_CONST(float, conv_op->Op()->GetAttr("Scale_out"));
      float threshold =
          BOOST_GET_CONST(float, conv_op->Op()->GetAttr("fuse_alpha"));
455
      conv_op->Op()->SetAttr("fuse_alpha", scale_out * threshold);
456 457
    }

458 459 460 461 462 463
    ++quantize_conv_count;
  };

  gpd(graph, handler);
  AddStatis(quantize_conv_count);

464
  LogQuantizedOpsCounter(
465 466
      "conv2d",
      quantize_conv_count,
467
      ((with_residual_data) ? "with residual connection" : ""));
468 469
}

M
Michał Gallus 已提交
470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486
void CPUQuantizePass::QuantizeFc(Graph* graph) const {
  GraphPatternDetector gpd;
  auto pattern = gpd.mutable_pattern();
  patterns::FCMKLDNN fc_pattern{pattern, name_scope_};
  auto* fc_input = gpd.mutable_pattern()
                       ->NewNode("fc_quantizer/input")
                       ->AsInput()
                       ->assert_is_op_input("fc", "Input");
  fc_pattern(fc_input, false);

  int quantize_fc_count = 0;
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
    VLOG(4) << "Quantize fc op";
    GET_IR_NODE_FROM_SUBGRAPH(fc, fc, fc_pattern);

    // skip if should not be quantized
487
    if (!platform::HasOpINT8DataType(fc->Op())) {
488 489 490
      LogQuantizationDisabled(fc);
      return;
    }
491
    if (!fc->Op()->GetAttrIfExists<bool>("use_mkldnn")) {
492
      MarkAndLogCannotQuantizeOp(fc, "use_mkldnn attribute set to false");
M
Michał Gallus 已提交
493
      return;
494
    }
M
Michał Gallus 已提交
495 496 497 498 499

    GET_IR_NODE_FROM_SUBGRAPH(weights, weights, fc_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(input, input, fc_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(output, output, fc_pattern);

500
    if (!AreScalesPresentForNodes({input, weights})) {
501
      MarkAndLogCannotQuantizeOp(fc, "No scale available for the operator");
502 503
      return;
    }
504

505 506
    bool is_input_unsigned{false};
    auto input_scale = GetScaleValueForNode(input, &is_input_unsigned);
507 508
    QuantizeInput(
        g, fc, input, "Input", input_scale, is_input_unsigned, "Scale_in");
M
Michał Gallus 已提交
509

510
    auto weight_scale_tensor = GetScaleTensorForNode(weights);
M
Michał Gallus 已提交
511
    EigenVectorArrayMap eigen_tensor{weight_scale_tensor.data<double>(),
512
                                     weight_scale_tensor.numel()};
M
Michał Gallus 已提交
513 514 515 516 517 518 519
    eigen_tensor *= static_cast<double>(S8_MAX);
    std::vector<float> filter_scale{
        weight_scale_tensor.data<double>(),
        weight_scale_tensor.data<double>() + weight_scale_tensor.numel()};

    fc->Op()->SetAttr("Scale_weights", filter_scale);

520
    // if quantization scale is missing for output tensor, return fp32 data
521
    if (AreScalesPresentForNodes({output})) {
522 523
      bool is_output_unsigned{false};
      auto output_scale = GetScaleValueForNode(output, &is_output_unsigned);
524 525
      DequantizeOutput(
          g, fc, output, "Out", output_scale, is_output_unsigned, "Scale_out");
526 527 528
    } else {
      fc->Op()->SetAttr("force_fp32_output", true);
    }
M
Michał Gallus 已提交
529 530 531 532 533 534

    ++quantize_fc_count;
  };

  gpd(graph, handler);
  AddStatis(quantize_fc_count);
535
  LogQuantizedOpsCounter("fc", quantize_fc_count);
M
Michał Gallus 已提交
536 537
}

538 539 540 541 542 543 544 545 546 547 548 549 550
void CPUQuantizePass::QuantizePool(Graph* graph) const {
  GraphPatternDetector gpd;
  auto pattern = gpd.mutable_pattern();
  patterns::Pool pool_pattern{pattern, name_scope_};
  pool_pattern();

  int quantize_pool_count = 0;
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
    VLOG(4) << "Quantize pool2d op";
    GET_IR_NODE_FROM_SUBGRAPH(pool_op, pool_op, pool_pattern);

    // skip if should not be quantized
551
    if (!platform::HasOpINT8DataType(pool_op->Op())) {
552 553 554
      LogQuantizationDisabled(pool_op);
      return;
    }
555 556 557 558

    GET_IR_NODE_FROM_SUBGRAPH(pool_input, pool_input, pool_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(pool_output, pool_output, pool_pattern);

559
    if (!AreScalesPresentForNodes({pool_input, pool_output})) {
560 561
      MarkAndLogCannotQuantizeOp(pool_op,
                                 "No scale available for the operator");
562 563
      return;
    }
564

565 566
    bool is_input_unsigned{false};
    auto input_scale = GetScaleValueForNode(pool_input, &is_input_unsigned);
567 568
    QuantizeInput(g, pool_op, pool_input, "X", input_scale, is_input_unsigned);

569 570
    bool is_output_unsigned{false};
    auto output_scale = GetScaleValueForNode(pool_output, &is_output_unsigned);
571 572
    DequantizeOutput(
        g, pool_op, pool_output, "Out", output_scale, is_output_unsigned);
573 574 575 576 577 578

    ++quantize_pool_count;
  };

  gpd(graph, handler);
  AddStatis(quantize_pool_count);
579
  LogQuantizedOpsCounter("pool2d", quantize_pool_count);
580 581
}

582 583 584 585 586 587 588 589 590 591 592 593 594
void CPUQuantizePass::QuantizeConcat(Graph* graph) const {
  GraphPatternDetector gpd;
  auto pattern = gpd.mutable_pattern();
  patterns::Concat concat_pattern{pattern, name_scope_};
  concat_pattern();

  int quantize_concat_count = 0;
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
    VLOG(4) << "Quantize concat op";
    GET_IR_NODE_FROM_SUBGRAPH(concat_op, concat_op, concat_pattern);

    // skip if should not be quantized
595
    if (!platform::HasOpINT8DataType(concat_op->Op())) {
596 597 598
      LogQuantizationDisabled(concat_op);
      return;
    }
599 600 601

    GET_IR_NODE_FROM_SUBGRAPH(concat_out, concat_out, concat_pattern);

602
    if (!AreScalesPresentForNodes({concat_out})) {
603 604
      MarkAndLogCannotQuantizeOp(concat_op,
                                 "No scale available for the operator");
605 606
      return;
    }
607

608 609
    // if all inputs were unsigned, then the output was set to unsigned
    // during the scale calculation step
610 611 612
    bool are_all_inputs_unsigned{false};
    auto output_scale =
        GetScaleValueForNode(concat_out, &are_all_inputs_unsigned);
613

614
    QuantizeInputs(g, concat_op, "X", are_all_inputs_unsigned);
615

616 617
    DequantizeOutput(
        g, concat_op, concat_out, "Out", output_scale, are_all_inputs_unsigned);
618 619 620 621 622 623

    ++quantize_concat_count;
  };

  gpd(graph, handler);
  AddStatis(quantize_concat_count);
624
  LogQuantizedOpsCounter("concat", quantize_concat_count);
625 626
}

627 628 629 630 631 632 633 634 635 636 637 638 639
void CPUQuantizePass::QuantizePriorBox(Graph* graph) const {
  GraphPatternDetector gpd;
  auto pattern = gpd.mutable_pattern();
  patterns::PriorBox prior_box_pattern{pattern, name_scope_};
  prior_box_pattern();

  int quantize_prior_box_count = 0;
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
    VLOG(4) << "Quantize prior_box op";
    GET_IR_NODE_FROM_SUBGRAPH(prior_box_op, prior_box_op, prior_box_pattern);

    // skip if should not be quantized
640
    if (!platform::HasOpINT8DataType(prior_box_op->Op())) {
641 642 643
      LogQuantizationDisabled(prior_box_op);
      return;
    }
644

645 646
    GET_IR_NODE_FROM_SUBGRAPH(
        prior_box_input, prior_box_input, prior_box_pattern);
647

648
    if (!AreScalesPresentForNodes({prior_box_input})) {
649 650
      MarkAndLogCannotQuantizeOp(prior_box_op,
                                 "No scale available for the operator");
651 652
      return;
    }
653

654 655 656
    bool is_input_unsigned{false};
    auto input_scale =
        GetScaleValueForNode(prior_box_input, &is_input_unsigned);
657 658 659 660 661
    QuantizeInput(g,
                  prior_box_op,
                  prior_box_input,
                  "Input",
                  input_scale,
662 663 664 665 666 667 668
                  is_input_unsigned);

    ++quantize_prior_box_count;
  };

  gpd(graph, handler);
  AddStatis(quantize_prior_box_count);
669
  LogQuantizedOpsCounter("prior_box", quantize_prior_box_count);
670 671
}

672 673 674
void CPUQuantizePass::QuantizeImmutable(Graph* graph,
                                        const std::string& immutable_type,
                                        const std::string& input_name) const {
675 676
  GraphPatternDetector gpd;
  auto pattern = gpd.mutable_pattern();
677 678
  patterns::Immutable immutable_pattern{pattern, name_scope_};
  immutable_pattern(immutable_type, input_name);
679

680
  int quantize_immutable_count = 0;
681 682
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
683 684
    VLOG(4) << "Quantize " + immutable_type + " op";
    GET_IR_NODE_FROM_SUBGRAPH(immutable_op, immutable_op, immutable_pattern);
685 686

    // skip if should not be quantized
687 688
    if (!platform::HasOpINT8DataType(immutable_op->Op())) {
      LogQuantizationDisabled(immutable_op);
689 690
      return;
    }
691 692 693
    GET_IR_NODE_FROM_SUBGRAPH(prev_op, prev_op, immutable_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(immutable_in, immutable_in, immutable_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(immutable_out, immutable_out, immutable_pattern);
694

695
    // skip if prev op and next op is not quantized
696 697
    if (!IsOpDequantized(prev_op) && !IsOpQuantized(immutable_out)) {
      MarkAndLogCannotQuantizeOp(immutable_op,
698
                                 "No other quantizable operators nearby");
699 700 701
      return;
    }

702 703
    if (!AreScalesPresentForNodes({immutable_out})) {
      MarkAndLogCannotQuantizeOp(immutable_op,
704
                                 "No scale available for the operator");
705
      return;
706
    }
707

708
    bool is_input_unsigned{false};
709 710 711 712 713 714 715 716
    auto input_scale = GetScaleValueForNode(immutable_out, &is_input_unsigned);

    QuantizeInput(g,
                  immutable_op,
                  immutable_in,
                  input_name,
                  input_scale,
                  is_input_unsigned);
717

718 719
    bool is_output_unsigned{false};
    auto output_scale =
720
        GetScaleValueForNode(immutable_out, &is_output_unsigned);
721
    DequantizeOutput(g,
722 723
                     immutable_op,
                     immutable_out,
724 725
                     "Out",
                     output_scale,
726 727
                     is_output_unsigned);

728
    ++quantize_immutable_count;
729 730 731
  };

  gpd(graph, handler);
732 733
  AddStatis(quantize_immutable_count);
  LogQuantizedOpsCounter(immutable_type, quantize_immutable_count);
Z
Zuza 已提交
734 735
}

736 737 738
void CPUQuantizePass::QuantizeMatmul(Graph* graph) const {
  GraphPatternDetector gpd;
  auto pattern = gpd.mutable_pattern();
739
  patterns::MatmulWithInputOps matmul_pattern{pattern, name_scope_};
740 741 742 743 744 745 746 747 748
  matmul_pattern();

  int quantize_matmul_count = 0;
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
    VLOG(4) << "Quantize matmul op";
    GET_IR_NODE_FROM_SUBGRAPH(matmul_op, matmul_op, matmul_pattern);

    // skip if should not be quantized
749
    if (!platform::HasOpINT8DataType(matmul_op->Op())) {
750
      LogQuantizationDisabled(matmul_op);
751 752 753 754 755 756 757
      return;
    }
    GET_IR_NODE_FROM_SUBGRAPH(prev_op_x, prev_op_x, matmul_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(prev_op_y, prev_op_y, matmul_pattern);

    // skip if prev ops are not quantized
    if (!IsOpDequantized(prev_op_x) || !IsOpDequantized(prev_op_y)) {
758 759
      MarkAndLogCannotQuantizeOp(matmul_op,
                                 "No other quantizable operators nearby");
760 761 762 763 764 765
      return;
    }
    GET_IR_NODE_FROM_SUBGRAPH(matmul_in_x, matmul_in_x, matmul_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(matmul_in_y, matmul_in_y, matmul_pattern);
    GET_IR_NODE_FROM_SUBGRAPH(matmul_out, matmul_out, matmul_pattern);

766
    if (!AreScalesPresentForNodes({matmul_in_x, matmul_in_y})) {
767 768
      MarkAndLogCannotQuantizeOp(matmul_op,
                                 "No scale available for the operator");
769
      return;
770
    }
771

772 773 774
    bool is_x_unsigned{false}, is_y_unsigned{false};
    auto input_x_scale = GetScaleValueForNode(matmul_in_x, &is_x_unsigned);
    auto input_y_scale = GetScaleValueForNode(matmul_in_y, &is_y_unsigned);
775 776
    PADDLE_ENFORCE_EQ(is_x_unsigned,
                      is_y_unsigned,
777 778 779 780
                      platform::errors::InvalidArgument(
                          "Matmul inputs should have the same "
                          "attribute of signed/unsigned, but they "
                          "are different: x(%d), y(%d).",
781 782 783 784 785 786 787 788
                          is_x_unsigned,
                          is_y_unsigned));
    QuantizeInput(g,
                  matmul_op,
                  matmul_in_x,
                  "X",
                  input_x_scale,
                  is_x_unsigned,
789
                  "Scale_x");
790 791 792 793 794 795
    QuantizeInput(g,
                  matmul_op,
                  matmul_in_y,
                  "Y",
                  input_y_scale,
                  is_y_unsigned,
796 797
                  "Scale_y");

798
    // if quantization scale is missing for output tensor, return fp32 data
799
    if (AreScalesPresentForNodes({matmul_out})) {
800 801
      bool is_output_unsigned{false};
      auto output_scale = GetScaleValueForNode(matmul_out, &is_output_unsigned);
802 803 804 805 806 807 808
      DequantizeOutput(g,
                       matmul_op,
                       matmul_out,
                       "Out",
                       output_scale,
                       is_output_unsigned,
                       "Scale_out");
809 810 811
    } else {
      matmul_op->Op()->SetAttr("force_fp32_output", true);
    }
812 813 814 815 816

    ++quantize_matmul_count;
  };
  gpd(graph, handler);
  AddStatis(quantize_matmul_count);
817
  LogQuantizedOpsCounter("matmul", quantize_matmul_count);
818 819
}

Z
Zuza 已提交
820
void CPUQuantizePass::QuantizeElementwise(
821
    Graph* graph, const std::string& elementwise_type) const {
822 823
  GraphPatternDetector gpd;
  auto pattern = gpd.mutable_pattern();
824
  patterns::ElementwiseOp elementwise_pattern{pattern, name_scope_};
825

826
  elementwise_pattern(elementwise_type);
827

Z
Zuza 已提交
828
  int quantize_elementwise_count = 0;
829 830
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
Z
Zuza 已提交
831
    VLOG(4) << "Quantize " + elementwise_type + " op";
832 833
    GET_IR_NODE_FROM_SUBGRAPH(
        elementwise_op, elementwise_op, elementwise_pattern);
834 835

    // skip if should not be quantized
Z
Zuza 已提交
836 837
    if (!platform::HasOpINT8DataType(elementwise_op->Op())) {
      LogQuantizationDisabled(elementwise_op);
838 839 840
      return;
    }

841 842 843 844 845 846 847 848 849 850 851 852
    auto x_name = elementwise_op->Op()->Input("X");
    auto y_name = elementwise_op->Op()->Input("Y");
    Node *elementwise_x, *elementwise_y;

    for (auto& input : elementwise_op->inputs) {
      if (input->Name() == x_name[0]) elementwise_x = input;
      if (input->Name() == y_name[0]) elementwise_y = input;
    }
    if (!elementwise_x || !elementwise_y) {
      return;
    }

853 854
    GET_IR_NODE_FROM_SUBGRAPH(
        elementwise_out, elementwise_out, elementwise_pattern);
855

856
    if (!AreScalesPresentForNodes(
Z
Zuza 已提交
857
            {elementwise_x, elementwise_y, elementwise_out})) {
858 859
      MarkAndLogCannotQuantizeOp(elementwise_op,
                                 "No scale available for the operator");
860 861 862 863
      return;
    }

    bool is_x_unsigned{false}, is_y_unsigned{false};
Z
Zuza 已提交
864 865
    auto input_x_scale = GetScaleValueForNode(elementwise_x, &is_x_unsigned);
    auto input_y_scale = GetScaleValueForNode(elementwise_y, &is_y_unsigned);
866 867 868

    // TODO(sfraczek): add support for different signness
    if (is_x_unsigned != is_y_unsigned) {
869 870
      MarkAndLogCannotQuantizeOp(
          elementwise_op, "Elementwise inputs must be of the same type.");
871 872 873
      return;
    }

874 875 876 877 878 879 880 881 882 883 884 885 886 887
    QuantizeInput(g,
                  elementwise_op,
                  elementwise_x,
                  "X",
                  input_x_scale,
                  is_x_unsigned,
                  "Scale_x");
    QuantizeInput(g,
                  elementwise_op,
                  elementwise_y,
                  "Y",
                  input_y_scale,
                  is_y_unsigned,
                  "Scale_y");
888

889 890
    bool is_output_unsigned{false};
    auto output_scale =
Z
Zuza 已提交
891
        GetScaleValueForNode(elementwise_out, &is_output_unsigned);
892

893 894 895 896 897 898 899
    DequantizeOutput(g,
                     elementwise_op,
                     elementwise_out,
                     "Out",
                     output_scale,
                     is_output_unsigned,
                     "Scale_out");
900

Z
Zuza 已提交
901
    ++quantize_elementwise_count;
902 903
  };
  gpd(graph, handler);
Z
Zuza 已提交
904
  AddStatis(quantize_elementwise_count);
905
  LogQuantizedOpsCounter(elementwise_type, quantize_elementwise_count);
906 907
}

908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929
void CPUQuantizePass::QuantizeFusionGru(Graph* graph) const {
  GraphPatternDetector gpd;
  patterns::FusionGru pattern{gpd.mutable_pattern(), name_scope_};
  pattern();

  int quantize_count = 0;
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
    VLOG(4) << "Quantize fusion_gru op";
    GET_IR_NODE_FROM_SUBGRAPH(op, op, pattern);

    // skip if should not be quantized
    if (!platform::HasOpINT8DataType(op->Op())) {
      LogQuantizationDisabled(op);
      return;
    }

    GET_IR_NODE_FROM_SUBGRAPH(x, x, pattern);
    GET_IR_NODE_FROM_SUBGRAPH(weight_h, weight_h, pattern);
    GET_IR_NODE_FROM_SUBGRAPH(weight_x, weight_x, pattern);
    GET_IR_NODE_FROM_SUBGRAPH(out, out, pattern);

930
    if (!AreScalesPresentForNodes({x, weight_x})) {
931
      MarkAndLogCannotQuantizeOp(op, "No scale available for the operator");
932 933 934 935 936 937 938 939 940
      return;
    }

    bool is_x_unsigned{false};
    auto input_x_scale = GetScaleValueForNode(x, &is_x_unsigned);

    double input_x_shift{128.};
    if (is_x_unsigned) input_x_shift = 0.;

941 942 943 944 945 946 947 948 949
    QuantizeInput(g,
                  op,
                  x,
                  "X",
                  input_x_scale,
                  is_x_unsigned,
                  "Scale_data",
                  input_x_shift,
                  "Shift_data");
950 951 952

    auto weight_scale_tensor = GetScaleTensorForNode(weight_x);
    EigenVectorArrayMap eigen_tensor{weight_scale_tensor.data<double>(),
953
                                     weight_scale_tensor.numel()};
954 955 956 957 958 959 960 961 962 963 964 965 966
    eigen_tensor *= static_cast<double>(S8_MAX);
    std::vector<float> scale_weights{
        weight_scale_tensor.data<double>(),
        weight_scale_tensor.data<double>() + weight_scale_tensor.numel()};

    op->Op()->SetAttr("Scale_weights", scale_weights);
    // return fp32 data
    op->Op()->SetAttr("force_fp32_output", true);

    ++quantize_count;
  };
  gpd(graph, handler);
  AddStatis(quantize_count);
967
  LogQuantizedOpsCounter("fusion_gru", quantize_count);
968 969
}

970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993
void CPUQuantizePass::QuantizeMultiGru(Graph* graph) const {
  GraphPatternDetector gpd;
  patterns::MultiGru pattern{gpd.mutable_pattern(), name_scope_};
  pattern();

  int quantize_count = 0;
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
    VLOG(4) << "Quantize multi_gru op";
    GET_IR_NODE_FROM_SUBGRAPH(gru, gru, pattern);

    // skip if should not be quantized
    if (!platform::HasOpINT8DataType(gru->Op())) {
      LogQuantizationDisabled(gru);
      return;
    }

    GET_IR_NODE_FROM_SUBGRAPH(x, x, pattern);
    GET_IR_NODE_FROM_SUBGRAPH(wx, wx, pattern);
    GET_IR_NODE_FROM_SUBGRAPH(h, h, pattern);

    auto wx_names = gru->Op()->Input("WeightX");
    if (!AreScalesPresentForNodes({x}) ||
        !AreScalesPresentForVarNames(wx_names)) {
994
      MarkAndLogCannotQuantizeOp(gru, "No scale available for the operator");
995 996 997 998 999 1000 1001 1002 1003
      return;
    }

    bool is_x_unsigned{false};
    auto input_x_scale = GetScaleValueForNode(x, &is_x_unsigned);

    double input_x_shift{128.};
    if (is_x_unsigned) input_x_shift = 0.;

1004 1005 1006 1007 1008 1009 1010 1011 1012
    QuantizeInput(g,
                  gru,
                  x,
                  "X",
                  input_x_scale,
                  is_x_unsigned,
                  "Scale_data",
                  input_x_shift,
                  "Shift_data");
1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023

    auto* scope = param_scope();
    int wx_size = wx_names.size();
    std::vector<std::string> w_scale_var_names;
    for (int i = 0; i < wx_size; ++i) {
      auto scale_tensor_src = GetScaleTensorByName(wx_names[i]);
      EigenVectorArrayMap eigen_tensor_src{scale_tensor_src.data<double>(),
                                           scale_tensor_src.numel()};

      VarDesc scale_var_desc(patterns::PDNodeName("multi_gru", "w_scale"));

1024
      scale_var_desc.SetShape(phi::vectorize(scale_tensor_src.dims()));
1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050
      scale_var_desc.SetDataType(proto::VarType::FP32);
      scale_var_desc.SetLoDLevel(scale_tensor_src.lod().size());
      scale_var_desc.SetPersistable(true);
      auto* w_scale_node = g->CreateVarNode(&scale_var_desc);

      auto* w_scale_tensor_dst =
          scope->Var(w_scale_node->Name())->GetMutable<LoDTensor>();
      w_scale_tensor_dst->Resize(scale_tensor_src.dims());
      auto* dst_data =
          w_scale_tensor_dst->mutable_data<float>(platform::CPUPlace());
      EigenVectorArrayMapFloat eigen_tensor_dst{dst_data,
                                                w_scale_tensor_dst->numel()};
      eigen_tensor_dst =
          eigen_tensor_src.cast<float>() * static_cast<float>(S8_MAX);
      w_scale_var_names.push_back(w_scale_node->Name());
      IR_NODE_LINK_TO(w_scale_node, gru);
    }

    gru->Op()->SetInput("Scale_weights", w_scale_var_names);
    // return fp32 data
    gru->Op()->SetAttr("force_fp32_output", true);

    ++quantize_count;
  };
  gpd(graph, handler);
  AddStatis(quantize_count);
1051
  LogQuantizedOpsCounter("multi_gru", quantize_count);
1052 1053
}

1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078
void CPUQuantizePass::QuantizeFusionLSTM(Graph* graph) const {
  GraphPatternDetector gpd;
  patterns::FusionLSTM pattern{gpd.mutable_pattern(), name_scope_};
  pattern();

  int quantize_count = 0;
  auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
                     Graph* g) {
    VLOG(4) << "Quantize fusion_lstm op";
    GET_IR_NODE_FROM_SUBGRAPH(op, op, pattern);

    // skip if should not be quantized
    if (!platform::HasOpINT8DataType(op->Op())) {
      LogQuantizationDisabled(op);
      return;
    }

    GET_IR_NODE_FROM_SUBGRAPH(x, x, pattern);
    GET_IR_NODE_FROM_SUBGRAPH(weight_h, weight_h, pattern);
    GET_IR_NODE_FROM_SUBGRAPH(weight_x, weight_x, pattern);
    GET_IR_NODE_FROM_SUBGRAPH(hidden, hidden, pattern);
    GET_IR_NODE_FROM_SUBGRAPH(cell, cell, pattern);

    // Starting from here there maybe issues
    if (!AreScalesPresentForNodes({x, weight_x})) {
1079
      MarkAndLogCannotQuantizeOp(op, "No scale available for the operator");
1080 1081 1082 1083 1084 1085 1086 1087 1088
      return;
    }

    bool is_x_unsigned{false};
    auto input_x_scale = GetScaleValueForNode(x, &is_x_unsigned);

    double input_x_shift{128.};
    if (is_x_unsigned) input_x_shift = 0.;

1089 1090 1091 1092 1093 1094 1095 1096 1097
    QuantizeInput(g,
                  op,
                  x,
                  "X",
                  input_x_scale,
                  is_x_unsigned,
                  "Scale_data",
                  input_x_shift,
                  "Shift_data");
1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114

    auto weight_scale_tensor = GetScaleTensorForNode(weight_x);
    EigenVectorArrayMap eigen_tensor{weight_scale_tensor.data<double>(),
                                     weight_scale_tensor.numel()};
    eigen_tensor *= static_cast<double>(S8_MAX);
    std::vector<float> scale_weights{
        weight_scale_tensor.data<double>(),
        weight_scale_tensor.data<double>() + weight_scale_tensor.numel()};

    op->Op()->SetAttr("Scale_weights", scale_weights);
    // return fp32 data
    op->Op()->SetAttr("force_fp32_output", true);

    ++quantize_count;
  };
  gpd(graph, handler);
  AddStatis(quantize_count);
1115
  LogQuantizedOpsCounter("fusion_lstm", quantize_count);
1116 1117
}

1118
void CPUQuantizePass::ApplyImpl(ir::Graph* graph) const {
1119
  VLOG(3) << "Quantizing the graph.";
1120 1121
  PADDLE_ENFORCE_NOT_NULL(
      graph, platform::errors::InvalidArgument("Graph cannot be nullptr."));
1122
  FusePassBase::Init(name_scope_, graph);
1123

1124 1125 1126
  PADDLE_ENFORCE_NOT_NULL(
      param_scope(),
      platform::errors::InvalidArgument("Scope cannot be nullptr."));
1127

B
baoachun 已提交
1128
  GetQuantInfo(graph);
1129 1130 1131
  QuantizeConv(graph, false /* with_residual_data */);
  QuantizeConv(graph, true /* with_residual_data */);
  QuantizePool(graph);
1132
  QuantizeConcat(graph);
1133
  QuantizePriorBox(graph);
M
Michał Gallus 已提交
1134
  QuantizeFc(graph);
1135
  QuantizeMatmul(graph);
1136 1137 1138
  QuantizeImmutable(graph, "reshape2", "X");
  QuantizeImmutable(graph, "transpose2", "X");
  QuantizeImmutable(graph, "slice", "Input");
1139
  QuantizeImmutable(graph, "shape", "Input");
1140 1141
  QuantizeImmutable(graph, "nearest_interp", "X");
  QuantizeImmutable(graph, "nearest_interp_v2", "X");
Z
Zuza 已提交
1142 1143
  QuantizeElementwise(graph, "elementwise_add");
  QuantizeElementwise(graph, "elementwise_mul");
1144
  QuantizeElementwise(graph, "elementwise_sub");
1145
  QuantizeFusionGru(graph);
1146
  QuantizeMultiGru(graph);
1147
  QuantizeFusionLSTM(graph);
1148 1149 1150 1151 1152 1153 1154 1155
}

}  // namespace ir
}  // namespace framework
}  // namespace paddle

REGISTER_PASS(cpu_quantize_pass, paddle::framework::ir::CPUQuantizePass)
    .RequirePassAttr("quant_var_scales");