Skip to content

GitLab

Commit 8665c111 authored by wenjh's avatar wenjh
Browse files

Fix hipblaslt_goupedgemm 


Signed-off-by: wenjh's avatarwenjh <wenjh@sugon.com>
parent 14aaa48a
Show whitespace changes
Inline Side-by-side
Showing with 2 additions and 3 deletions
transformer_engine/common/gemm/rocm_gemm.cu
View file @ 8665c111
......@@ -1603,6 +1603,8 @@ void hipblaslt_goupedgemm(std::vector<const Tensor*>& inputA, std::vector<const
return;
}
// Make sure to initialize everytime the algo changes
NVTE_CHECK_HIPBLASLT(groupedgemm.initialize(heuristicResult[0].algo, workspace));
// Get the default values from the grouepdgemm object
groupedgemm.getDefaultValueForDeviceUserArguments(userArgs);
// Copy them to device memory
......@@ -1612,9 +1614,6 @@ void hipblaslt_goupedgemm(std::vector<const Tensor*>& inputA, std::vector<const
userArgs,
m.size() * sizeof(hipblaslt_ext::UserArguments),
hipMemcpyHostToDevice, stream));
// Make sure to initialize everytime the algo changes
NVTE_CHECK_HIPBLASLT(groupedgemm.initialize(heuristicResult[0].algo, workspace));
NVTE_CHECK_HIPBLASLT(groupedgemm.run(d_userArgs, stream));
// NVTE_CHECK_HIPBLASLT(groupedgemm.initialize(heuristicResult[0].algo, workspace, false, stream));
// NVTE_CHECK_HIPBLASLT(groupedgemm.run(stream));
......
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