In src/tir/transforms/lower_tvm_builtin.cc:
if (device_type_.defined()) {
if (const auto* dev_type = device_type_.as<IntImmNode>()) {
if (dev_type->value == kDLCPU) {
int32_t constant_size = op->constant_allocation_size();
if (constant_size > 0 && constant_size * nbytes < runtime::kMaxStackAlloca) {
return stmt;
}
}
}
}If the above optimization code is here, it will raise LLVM function signature errors when there are multiple targets. This piece of code was removed in #6126 to fix this issue. But it was reverted back in #8274, as removal of this code will cause some VM test failed.
To quick fix, the multiple target unittests are disabled in the current main branch. But we need further investigation on how to resolve this nicely.
cc @echuraev @elvin-n
In
src/tir/transforms/lower_tvm_builtin.cc:If the above optimization code is here, it will raise LLVM function signature errors when there are multiple targets. This piece of code was removed in #6126 to fix this issue. But it was reverted back in #8274, as removal of this code will cause some VM test failed.
To quick fix, the multiple target unittests are disabled in the current main branch. But we need further investigation on how to resolve this nicely.
cc @echuraev @elvin-n