forked from jiuyuan/InfiniTensor
Fix: ignore transpose in CudaGraph since no kernel
This commit is contained in:
parent
0865f8d823
commit
a732b6f176
|
@ -49,7 +49,7 @@ template <typename R, typename... Args> class Functor<R(Args...)> {
|
|||
virtual R visit_(const Tensor &c, Args... args) FUNCTOR_DEFAULT;
|
||||
virtual R visit_(const Func &c, Args... args) FUNCTOR_DEFAULT;
|
||||
virtual R visitDefault(const Expr &c, [[maybe_unused]] Args... args) {
|
||||
dbg(*c);
|
||||
dbg(*c, c->getType());
|
||||
nnet_assert(0, "Reach unimplemented visit function.");
|
||||
return R();
|
||||
};
|
||||
|
|
|
@ -141,7 +141,9 @@ double CudaRuntimeObj::timeWithCudaGraph(Graph graph) {
|
|||
kernel->compute(op, perfData, this);
|
||||
else
|
||||
kernel->compute(op, this);
|
||||
if (!ctcMap.at(op->getGuid()) && op->getOpType() != OpType::Reshape)
|
||||
// FIXME: transpose
|
||||
if (!ctcMap.at(op->getGuid()) && op->getOpType() != OpType::Transpose &&
|
||||
op->getOpType() != OpType::Reshape)
|
||||
kernels.emplace_back(op, kernel, perfData);
|
||||
}
|
||||
for (auto &[op, kernel, perfData] : kernels) {
|
||||
|
|
Loading…
Reference in New Issue