Look through attributes and rates when determining by reference initialization #5023
Conversation
expipiplus1
force-pushed
the
uninit-global
branch
from
c0bb66c
to
614481a
Compare
source/slang/slang-ir-util.cpp
Outdated
| @@ -175,7 +175,9 @@ IRInst* maybeSpecializeWithGeneric(IRBuilder& builder, IRInst* genericToSpecaili | |||
| // Returns true if is not possible to produce side-effect from a value of `dataType`. | |||
| bool isValueType(IRInst* dataType) | |||
| { | |||
| dataType = getResolvedInstForDecorations(unwrapAttributedType(dataType)); | |||
| if(auto t = as<IRType>(dataType)) | |||
There was a problem hiding this comment.
I changed the type of unwrapAttributedType to be on IRType, so it's required now.
There was a problem hiding this comment.
That means we are checking the opcode twice, once for the dynamic cast and once for the actual unwrapping, which seems unnecessary.
There was a problem hiding this comment.
We should feel comfortable to accept IRInst where a type is expected, and rely on static cast to convert between IRInst and IRType because there is not a clean way to separate a type and an inst anyways. IRType has no more meaning other than clarifying what an IRInst is for in code, but for unwrapAttributeType, it is verify clear what the param is, so it accepting IRInst is not wrong and I don't see a need for the change.
There was a problem hiding this comment.
ok, I've reverted that change
Closes #5022