refine the implementation

This commit is contained in:
Haicheng Wu
2021-09-08 13:14:08 +00:00
parent 4e8af93da1
commit 59e2aa505a
45 changed files with 1593 additions and 706 deletions

View File

@@ -105,6 +105,18 @@ public:
return status;
}
static int const kAlignmentC = ImplicitGemmKernel::Epilogue::OutputTileIterator::kElementsPerAccess;
if (kConvolutionalOperator == conv::Operator::kFprop) {
if (args.problem_size.K % kAlignmentC)
return Status::kErrorMisalignedOperand;
} else if (kConvolutionalOperator == conv::Operator::kDgrad) {
if (args.problem_size.C % kAlignmentC)
return Status::kErrorMisalignedOperand;
} else if (kConvolutionalOperator == conv::Operator::kWgrad) {
if (args.problem_size.C % kAlignmentC)
return Status::kErrorMisalignedOperand;
}
// check for unsupported problem sizes for strided dgrad implementation
if (kConvolutionalOperator == conv::Operator::kDgrad &&
kStrideSupport == conv::StrideSupport::kStrided) {