1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
|
#include <torch/csrc/jit/passes/quantization/fusion_passes.h>
#include <torch/csrc/jit/passes/subgraph_rewrite.h>
namespace torch::jit {
namespace {
void fuseQuantizeAddReluImpl(std::shared_ptr<Graph>& graph) {
SubgraphRewriter fused_add_relu_rewriter;
std::string quantized_add_relu_pattern = R"(
graph(%a_quant, %b_quant, %scale, %zero_point):
%add_out = quantized::add(%a_quant, %b_quant, %scale, %zero_point)
%r = aten::relu(%add_out)
return (%r) )";
std::string fused_add_relu_pattern = R"(
graph(%a_quant, %b_quant, %scale, %zero_point):
%r = quantized::add_relu(%a_quant, %b_quant, %scale, %zero_point)
return (%r) )";
fused_add_relu_rewriter.RegisterRewritePattern(
quantized_add_relu_pattern, fused_add_relu_pattern);
std::string quantized_add_out_relu_pattern = R"(
graph(%a_quant, %b_quant, %out_quant):
%add_out = quantized::add_out(%a_quant, %b_quant, %out_quant)
%r = aten::relu(%add_out)
return (%r) )";
std::string fused_add_out_relu_pattern = R"(
graph(%a_quant, %b_quant, %out_quant):
%r = quantized::add_relu_out(%a_quant, %b_quant, %out_quant)
return (%r) )";
fused_add_relu_rewriter.RegisterRewritePattern(
quantized_add_out_relu_pattern, fused_add_out_relu_pattern);
std::string quantized_add_scalar_relu_pattern = R"(
graph(%a_quant, %b_scalar):
%add_out = quantized::add_scalar(%a_quant, %b_scalar)
%r = aten::relu(%add_out)
return (%r) )";
std::string fused_add_scalar_relu_pattern = R"(
graph(%a_quant, %b_scalar):
%r = quantized::add_scalar_relu(%a_quant, %b_scalar)
return (%r) )";
fused_add_relu_rewriter.RegisterRewritePattern(
quantized_add_scalar_relu_pattern, fused_add_scalar_relu_pattern);
std::string quantized_add_scalar_out_relu_pattern = R"(
graph(%a_quant, %b_scalar, %out_quant):
%add_out = quantized::add_scalar_out(%a_quant, %b_scalar, %out_quant)
%r = aten::relu(%add_out)
return (%r) )";
std::string fused_add_scalar_out_relu_pattern = R"(
graph(%a_quant, %b_scalar, %out_quant):
%r = quantized::add_scalar_relu_out(%a_quant, %b_scalar, %out_quant)
return (%r) )";
fused_add_relu_rewriter.RegisterRewritePattern(
quantized_add_scalar_out_relu_pattern, fused_add_scalar_out_relu_pattern);
fused_add_relu_rewriter.runOnGraph(graph);
}
} // namespace
void FuseQuantizedAddRelu(std::shared_ptr<Graph>& graph) {
fuseQuantizeAddReluImpl(graph);
}
} // namespace torch::jit
|
