Skip to content

N
ngraph
Commit 8eb31b78 authored by fenglei.tian's avatar fenglei.tian
Browse files
Options

clang format

parent d172456c
Show whitespace changes
Inline Side-by-side
Showing with 69 additions and 58 deletions
src/ngraph/runtime/gpu/gpu_emitter.cpp
View file @ 8eb31b78
...@@ -104,22 +104,22 @@ namespace ngraph ...@@ -104,22 +104,22 @@ namespace ngraph
{ {
namespace gpu namespace gpu
{ {
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Abs) void GPU_Emitter::EMITTER_DECL(ngraph::op::Abs)
{ {
writer << "{ // " << node->get_name() << "\n"; writer << "{ // " << node->get_name() << "\n";
writer.indent++; writer.indent++;
writer << "int count = " << out[0].get_size() << ";\n"; writer << "int count = " << out[0].get_size() << ";\n";
writer << "if(count == 0) return;\n"; writer << "if(count == 0) return;\n";
writer << "ngraph::runtime::gpu::emit_abs((void*) " << args[0].get_name() << ", (void*) " writer << "ngraph::runtime::gpu::emit_abs((void*) " << args[0].get_name()
<< out[0].get_name() << ", count);\n"; << ", (void*) " << out[0].get_name() << ", count);\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Add) void GPU_Emitter::EMITTER_DECL(ngraph::op::Add)
{ {
writer << "{ // " << node->get_name() << "\n"; writer << "{ // " << node->get_name() << "\n";
writer.indent++; writer.indent++;
writer << "int count = " << out[0].get_size() << ";\n"; writer << "int count = " << out[0].get_size() << ";\n";
...@@ -154,11 +154,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc, ...@@ -154,11 +154,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc,
<< "descriptor," << out[0].get_name() << ");\n"; << "descriptor," << out[0].get_name() << ");\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Dot) void GPU_Emitter::EMITTER_DECL(ngraph::op::Dot)
{ {
const ngraph::op::Dot* dot = static_cast<const ngraph::op::Dot*>(node); const ngraph::op::Dot* dot = static_cast<const ngraph::op::Dot*>(node);
const Shape& arg0_shape = args[0].get_shape(); const Shape& arg0_shape = args[0].get_shape();
const Shape& arg1_shape = args[1].get_shape(); const Shape& arg1_shape = args[1].get_shape();
...@@ -176,7 +176,8 @@ template <> ...@@ -176,7 +176,8 @@ template <>
<< "1," << out[0].get_name() << ", 1);\n"; << "1," << out[0].get_name() << ", 1);\n";
writer << "cublasSscal(" writer << "cublasSscal("
<< "cublas_handle," << "cublas_handle,"
<< "count ," << first.get_name() << "," << out[0].get_name() << ", 1);\n"; << "count ," << first.get_name() << "," << out[0].get_name()
<< ", 1);\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
return; return;
...@@ -198,8 +199,8 @@ template <> ...@@ -198,8 +199,8 @@ template <>
{ {
writer << "{ // " << node->get_name() << "\n"; writer << "{ // " << node->get_name() << "\n";
writer.indent++; writer.indent++;
writer << "runtime::gpu::cuda_memset(" << out[0].get_name() << ", 0, " << out[0].get_size() writer << "runtime::gpu::cuda_memset(" << out[0].get_name() << ", 0, "
<< " * sizeof(float));\n"; << out[0].get_size() << " * sizeof(float));\n";
writer << "return;\n"; writer << "return;\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
...@@ -228,7 +229,8 @@ template <> ...@@ -228,7 +229,8 @@ template <>
<< "cublas_handle," << "cublas_handle,"
<< "CUBLAS_OP_T," << arg0_shape[0] << "," << arg0_shape[1] << "," << "CUBLAS_OP_T," << arg0_shape[0] << "," << arg0_shape[1] << ","
<< "&alpha," // Alpha << "&alpha," // Alpha
<< args[0].get_name() << "," << arg0_shape[1] << "," << args[1].get_name() << "," << args[0].get_name() << "," << arg0_shape[1] << ","
<< args[1].get_name() << ","
<< "1," << "1,"
<< "&beta," // beta << "&beta," // beta
<< out[0].get_name() << "," << out[0].get_name() << ","
...@@ -274,13 +276,14 @@ template <> ...@@ -274,13 +276,14 @@ template <>
} }
else else
{ {
throw std::runtime_error(node->get_name() + " with more then 2D is not implemented."); throw std::runtime_error(node->get_name() +
" with more then 2D is not implemented.");
}
} }
}
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Maximum) void GPU_Emitter::EMITTER_DECL(ngraph::op::Maximum)
{ {
writer << "{ // " << node->get_name() << "\n"; writer << "{ // " << node->get_name() << "\n";
writer.indent++; writer.indent++;
writer << "int count = " << out[0].get_size() << ";\n"; writer << "int count = " << out[0].get_size() << ";\n";
...@@ -315,11 +318,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc, ...@@ -315,11 +318,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc,
<< "descriptor," << out[0].get_name() << ");\n"; << "descriptor," << out[0].get_name() << ");\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Minimum) void GPU_Emitter::EMITTER_DECL(ngraph::op::Minimum)
{ {
writer << "{ // " << node->get_name() << "\n"; writer << "{ // " << node->get_name() << "\n";
writer.indent++; writer.indent++;
writer << "int count = " << out[0].get_size() << ";\n"; writer << "int count = " << out[0].get_size() << ";\n";
...@@ -354,11 +357,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc, ...@@ -354,11 +357,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc,
<< "descriptor," << out[0].get_name() << ");\n"; << "descriptor," << out[0].get_name() << ");\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Negative) void GPU_Emitter::EMITTER_DECL(ngraph::op::Negative)
{ {
writer << "{ // " << node->get_name() << "\n"; writer << "{ // " << node->get_name() << "\n";
writer.indent++; writer.indent++;
writer << "int count = " << out[0].get_size() << ";\n"; writer << "int count = " << out[0].get_size() << ";\n";
...@@ -393,11 +396,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc, ...@@ -393,11 +396,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc,
<< "descriptor," << out[0].get_name() << ");\n"; << "descriptor," << out[0].get_name() << ");\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Broadcast) void GPU_Emitter::EMITTER_DECL(ngraph::op::Broadcast)
{ {
auto broadcast = static_cast<const ngraph::op::Broadcast*>(node); auto broadcast = static_cast<const ngraph::op::Broadcast*>(node);
auto arg_shape = args[0].get_shape(); auto arg_shape = args[0].get_shape();
auto result_shape = out[0].get_shape(); auto result_shape = out[0].get_shape();
...@@ -408,8 +411,9 @@ template <> ...@@ -408,8 +411,9 @@ template <>
{ {
writer << "{ // " << node->get_name() << " \n"; writer << "{ // " << node->get_name() << " \n";
writer.indent++; writer.indent++;
writer << "runtime::gpu::cuda_memcpyDtD(" << out[0].get_name() << ", " << args[0].get_name() writer << "runtime::gpu::cuda_memcpyDtD(" << out[0].get_name() << ", "
<< ", " << out[0].get_size() << " * " << out[0].get_element_type().size() << ");\n"; << args[0].get_name() << ", " << out[0].get_size() << " * "
<< out[0].get_element_type().size() << ");\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
return; return;
...@@ -447,9 +451,9 @@ template <> ...@@ -447,9 +451,9 @@ template <>
writer << "{ // " << node->get_name() << " \n"; writer << "{ // " << node->get_name() << " \n";
writer.indent++; writer.indent++;
writer << "runtime::gpu::emit_broadcast(" << args[0].get_name() << ", " << out[0].get_name() writer << "runtime::gpu::emit_broadcast(" << args[0].get_name() << ", "
<< ", " << repeat_size << ", " << repeat_times << ", " << out[0].get_size() << out[0].get_name() << ", " << repeat_size << ", " << repeat_times
<< ");\n"; << ", " << out[0].get_size() << ");\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
...@@ -457,16 +461,16 @@ template <> ...@@ -457,16 +461,16 @@ template <>
{ {
throw std::runtime_error(node->get_name() + " is not implemented."); throw std::runtime_error(node->get_name() + " is not implemented.");
} }
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Constant) void GPU_Emitter::EMITTER_DECL(ngraph::op::Constant)
{ {
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Reshape) void GPU_Emitter::EMITTER_DECL(ngraph::op::Reshape)
{ {
auto reshape = static_cast<const op::Reshape*>(node); auto reshape = static_cast<const op::Reshape*>(node);
writer << "{ // " << node->get_name() << "\n"; writer << "{ // " << node->get_name() << "\n";
writer.indent++; writer.indent++;
...@@ -491,8 +495,9 @@ template <> ...@@ -491,8 +495,9 @@ template <>
{ {
writer << "{ // " << node->get_name() << " 1\n"; writer << "{ // " << node->get_name() << " 1\n";
writer.indent++; writer.indent++;
writer << "runtime::gpu::cuda_memcpyDtD(" << out[0].get_name() << ", " << args[0].get_name() writer << "runtime::gpu::cuda_memcpyDtD(" << out[0].get_name() << ", "
<< ", " << out[0].get_size() << " * " << out[0].get_element_type().size() << ");\n"; << args[0].get_name() << ", " << out[0].get_size() << " * "
<< out[0].get_element_type().size() << ");\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
...@@ -512,8 +517,8 @@ template <> ...@@ -512,8 +517,8 @@ template <>
<< "&alpha," // Alpha << "&alpha," // Alpha
<< args[0].get_name() << "," << arg_shape[1] << "," << args[0].get_name() << "," << arg_shape[1] << ","
<< "&beta," // beta << "&beta," // beta
<< args[0].get_name() << "," << arg_shape[1] << "," << out[0].get_name() << "," << args[0].get_name() << "," << arg_shape[1] << "," << out[0].get_name()
<< result_shape[1] << ");\n"; << "," << result_shape[1] << ");\n";
writer << "cublasSetPointerMode(cublas_handle, CUBLAS_POINTER_MODE_DEVICE);\n"; writer << "cublasSetPointerMode(cublas_handle, CUBLAS_POINTER_MODE_DEVICE);\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
...@@ -522,20 +527,21 @@ template <> ...@@ -522,20 +527,21 @@ template <>
else else
{ {
throw runtime_error( throw runtime_error(
"Axis permutation in reshape is not implemented yet for tensors with rank>2"); "Axis permutation in reshape is not implemented yet for tensors with "
"rank>2");
} }
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::FunctionCall) void GPU_Emitter::EMITTER_DECL(ngraph::op::FunctionCall)
{ {
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Multiply) void GPU_Emitter::EMITTER_DECL(ngraph::op::Multiply)
{ {
writer << "{ // " << node->get_name() << "\n"; writer << "{ // " << node->get_name() << "\n";
writer.indent++; writer.indent++;
writer << "int count = " << out[0].get_size() << ";\n"; writer << "int count = " << out[0].get_size() << ";\n";
...@@ -570,11 +576,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc, ...@@ -570,11 +576,11 @@ cudnnSetOpTensorDescriptor(opTensorDesc,
<< "descriptor," << out[0].get_name() << ");\n"; << "descriptor," << out[0].get_name() << ");\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
template <> template <>
void GPU_Emitter::EMITTER_DECL(ngraph::op::Sqrt) void GPU_Emitter::EMITTER_DECL(ngraph::op::Sqrt)
{ {
writer << "{ // " << node->get_name() << "\n"; writer << "{ // " << node->get_name() << "\n";
writer.indent++; writer.indent++;
writer << "int count = " << out[0].get_size() << ";\n"; writer << "int count = " << out[0].get_size() << ";\n";
...@@ -609,7 +615,7 @@ cudnnSetOpTensorDescriptor(opTensorDesc, ...@@ -609,7 +615,7 @@ cudnnSetOpTensorDescriptor(opTensorDesc,
<< "descriptor," << out[0].get_name() << ");\n"; << "descriptor," << out[0].get_name() << ");\n";
writer.indent--; writer.indent--;
writer << "}\n"; writer << "}\n";
} }
} }
} }
} }
src/ngraph/runtime/gpu/gpu_external_function.cpp
View file @ 8eb31b78
...@@ -169,7 +169,8 @@ static const runtime::gpu::OpMap dispatcher{ ...@@ -169,7 +169,8 @@ static const runtime::gpu::OpMap dispatcher{
{TI(ngraph::op::Concat), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Concat>}, {TI(ngraph::op::Concat), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Concat>},
{TI(ngraph::op::Divide), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Divide>}, {TI(ngraph::op::Divide), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Divide>},
{TI(ngraph::op::Equal), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Equal>}, {TI(ngraph::op::Equal), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Equal>},
{TI(ngraph::op::GetOutputElement), &runtime::gpu::GPU_Emitter::emit<ngraph::op::GetOutputElement>}, {TI(ngraph::op::GetOutputElement),
&runtime::gpu::GPU_Emitter::emit<ngraph::op::GetOutputElement>},
{TI(ngraph::op::Greater), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Greater>}, {TI(ngraph::op::Greater), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Greater>},
{TI(ngraph::op::GreaterEq), &runtime::gpu::GPU_Emitter::emit<ngraph::op::GreaterEq>}, {TI(ngraph::op::GreaterEq), &runtime::gpu::GPU_Emitter::emit<ngraph::op::GreaterEq>},
{TI(ngraph::op::Less), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Less>}, {TI(ngraph::op::Less), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Less>},
...@@ -216,13 +217,17 @@ static const runtime::gpu::OpMap dispatcher{ ...@@ -216,13 +217,17 @@ static const runtime::gpu::OpMap dispatcher{
{TI(ngraph::op::Reverse), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Reverse>}, {TI(ngraph::op::Reverse), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Reverse>},
{TI(ngraph::op::Result), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Result>}, {TI(ngraph::op::Result), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Result>},
{TI(ngraph::op::ReduceWindow), &runtime::gpu::GPU_Emitter::emit<ngraph::op::ReduceWindow>}, {TI(ngraph::op::ReduceWindow), &runtime::gpu::GPU_Emitter::emit<ngraph::op::ReduceWindow>},
{TI(ngraph::op::SelectAndScatter), &runtime::gpu::GPU_Emitter::emit<ngraph::op::SelectAndScatter>}, {TI(ngraph::op::SelectAndScatter),
&runtime::gpu::GPU_Emitter::emit<ngraph::op::SelectAndScatter>},
{TI(ngraph::op::AvgPool), &runtime::gpu::GPU_Emitter::emit<ngraph::op::AvgPool>}, {TI(ngraph::op::AvgPool), &runtime::gpu::GPU_Emitter::emit<ngraph::op::AvgPool>},
{TI(ngraph::op::AvgPoolBackprop), &runtime::gpu::GPU_Emitter::emit<ngraph::op::AvgPoolBackprop>}, {TI(ngraph::op::AvgPoolBackprop),
&runtime::gpu::GPU_Emitter::emit<ngraph::op::AvgPoolBackprop>},
{TI(ngraph::op::Pad), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Pad>}, {TI(ngraph::op::Pad), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Pad>},
{TI(ngraph::op::BatchNorm), &runtime::gpu::GPU_Emitter::emit<ngraph::op::BatchNorm>}, {TI(ngraph::op::BatchNorm), &runtime::gpu::GPU_Emitter::emit<ngraph::op::BatchNorm>},
{TI(ngraph::op::BatchNormBackprop), &runtime::gpu::GPU_Emitter::emit<ngraph::op::BatchNormBackprop>}, {TI(ngraph::op::BatchNormBackprop),
{TI(ngraph::op::MaxPoolBackprop), &runtime::gpu::GPU_Emitter::emit<ngraph::op::MaxPoolBackprop>}, &runtime::gpu::GPU_Emitter::emit<ngraph::op::BatchNormBackprop>},
{TI(ngraph::op::MaxPoolBackprop),
&runtime::gpu::GPU_Emitter::emit<ngraph::op::MaxPoolBackprop>},
{TI(ngraph::op::Product), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Product>}, {TI(ngraph::op::Product), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Product>},
{TI(ngraph::op::Max), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Max>}, {TI(ngraph::op::Max), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Max>},
{TI(ngraph::op::Min), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Min>}, {TI(ngraph::op::Min), &runtime::gpu::GPU_Emitter::emit<ngraph::op::Min>},
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment