Skip to content

P
Paddle

机器未来 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
未验证 提交 d7be46b3 编写于 作者: Z zhangyikun02 提交者： GitHub
浏览文件
操作

add implement of resnet_basic_block op for XPU2, test=kunlun (#44143)

上级 337bb47b
展开全部 隐藏空白更改
内联 并排
Showing with 997 addition and 18 deletion
paddle/fluid/operators/fused/resnet_basic_block_op.cc
浏览文件 @ d7be46b3
...@@ -258,24 +258,25 @@ class ResNetBasicBlockOp : public framework::OperatorWithKernel { ...@@ -258,24 +258,25 @@ class ResNetBasicBlockOp : public framework::OperatorWithKernel {
class ResNetBasicBlockOpMaker : public framework::OpProtoAndCheckerMaker { class ResNetBasicBlockOpMaker : public framework::OpProtoAndCheckerMaker {
public: public:
void Make() { void Make() {
// has_shortcut = True: X else: X // has_shortcut = True: else:
// / / // X X
// | | | | // / /
// CONV1 | CONV1 | // | | | |
// | | | | // CONV1 | CONV1 |
// BN1 | BN1 | // | | | |
// | | | | // BN1 | BN1 |
// RELU1 | RELU1 | // | | | |
// | | | | // RELU1 | RELU1 |
// CONV2 CONV3 CONV2 | // | | | |
// | | | | // CONV2 CONV3 CONV2 |
// BN2 BN3 BN2 | // | | | |
// \ / \ / // BN2 BN3 BN2 |
// ADD ADD // \ / \ /
// | | // ADD ADD
// RELU RELU // | |
// | | // RELU RELU
// Y Y // | |
// Y Y
AddInput("X", "Input tensor of conv 1"); AddInput("X", "Input tensor of conv 1");
AddInput("Filter1", "Filter tensor of conv 1"); AddInput("Filter1", "Filter tensor of conv 1");
AddInput("Scale1", "Scale tensor of bn 1"); AddInput("Scale1", "Scale tensor of bn 1");
......
paddle/fluid/platform/device/xpu/xpu2_op_list.h
浏览文件 @ d7be46b3
...@@ -505,6 +505,14 @@ XPUOpMap& get_kl2_ops() { ...@@ -505,6 +505,14 @@ XPUOpMap& get_kl2_ops() {
XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})}, XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
{"sequence_conv_grad", {"sequence_conv_grad",
XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})}, XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace())})},
// Fused op
{"resnet_basic_block_grad",
XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace()),
pOpKernelType(vartype::FP16, XPUPlace())})},
{"resnet_basic_block",
XPUKernelSet({pOpKernelType(vartype::FP32, XPUPlace()),
pOpKernelType(vartype::FP16, XPUPlace())})},
}; };
return s_xpu2_kernels; return s_xpu2_kernels;
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或