Skip to content

Optimize dot product of Matmul_v2 #37037

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Nov 15, 2021
Merged

Optimize dot product of Matmul_v2 #37037

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 18 additions & 20 deletions paddle/pten/kernels/functions/math/matmul_func.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,9 +17,7 @@ limitations under the License. */
#include "paddle/fluid/operators/math/blas.h"
#include "paddle/fluid/operators/math/complex_functors.h"

#include "paddle/fluid/operators/eigen/eigen_function.h"
#include "paddle/pten/core/dense_tensor.h"
#include "paddle/pten/kernels/functions/eigen/common.h"

namespace pten {
namespace math {
Expand Down Expand Up @@ -105,34 +103,34 @@ void MatMulFunction(const DeviceContext& dev_ctx,
const T* x_data = X.data<T>();
const T* y_data = Y.data<T>();

auto blas = paddle::operators::math::GetBlas<DeviceContext, T>(dev_ctx);

if (x_ndim == 1 && y_ndim == 1) {
const int M = X.numel();
const int N = Y.numel();
PADDLE_ENFORCE_EQ(
X.numel(),
Y.numel(),
M,
N,
paddle::platform::errors::InvalidArgument(
"X's numbers must be equal to Y's numbers,"
"when X/Y's dims =1. But received X has [%d] elements,"
"received Y has [%d] elements",
X.numel(),
Y.numel()));
M,
N));
VLOG(3) << "MatMul's case 1";
Out->Resize({1});
Out->mutable_data<T>();
auto out_eigen = EigenScalar<T>::From(*Out);
auto x_eigen = EigenVector<T>::Flatten(X);
auto y_eigen = EigenVector<T>::Flatten(Y);

auto& dev = *dev_ctx.eigen_device();
if (flag) {
out_eigen.device(dev) = (x_eigen * y_eigen).sum() + out_eigen;
} else {
out_eigen.device(dev) = (x_eigen * y_eigen).sum();
}
blas.GEMM(CblasNoTrans,
CblasTrans,
1,
1,
M,
static_cast<T>(1),
y_data,
x_data,
static_cast<T>(flag),
Out->mutable_data<T>());
return;
}

auto blas = paddle::operators::math::GetBlas<DeviceContext, T>(dev_ctx);

if (x_ndim == 1) {
const int N = X.numel();
if (trans_y) {
Expand Down