Mikhail Dvoretckii comments

Results 10 comments of


                                            Mikhail Dvoretckii

[GPU] Transpose strides for padded GEMM

Changed the existing oneDNN GEMM transposed padding test to include a case for padding on all dimensions.

[GPU] Transpose strides for padded GEMM

build_jenkins

[GPU] Transpose strides for padded GEMM

@yeonbok Please rerun the jenkins pipeline. I've confirmed tests to be passing locally with the latest change.

[GPU] Recognize parameters as valid inputs for compressed weights

build_jenkins

[GPU] Recognize parameters as valid inputs for compressed weights

build_jenkins

[GPU] Recognize parameters as valid inputs for compressed weights

@CuriousPanCake please review.

[GPU] Recognize parameters as valid inputs for compressed weights

@CuriousPanCake please review.

[GPU] Recognize parameters as valid inputs for compressed weights

Optimized kernels are called for the matrix multiplication itself, but there is a perf cost from transpose nodes for the parameter weights that cannot be optimized away.

[GPU] Recognize parameters as valid inputs for compressed weights

> Consider add new tests Tests added.

[GPU] Recognize parameters as valid inputs for compressed weights

@Lyamin-Roman please review.