dyno: Add initial support for _ and de-tupled loop index variables
#25418
Conversation
dlongnecke-cray
force-pushed
the
dyno-detuple-throwaway-vars
branch
from
7939772
to
cab3eab
Compare
|
Adjusting this branch to call |
frontend/lib/resolution/Resolver.cpp
Outdated
| // If the element is an index variable, then use the RHS's intent. | ||
| if (auto var = namedDecl->toVarLikeDecl()) { | ||
| if (var->storageKind() == uast::Qualifier::INDEX) { | ||
| auto& rr = byPostorder.byAst(var); | ||
| auto kind = rhsEltType.kind(); | ||
|
|
||
| // If the kind is a value, set it to 'const'. | ||
| if (!rhsEltType.isRef() && !rhsEltType.isParam() && | ||
| !rhsEltType.isType()) { | ||
| auto combiner = KindProperties::fromKind(kind); | ||
| combiner.setConst(true); | ||
| kind = combiner.toKind(); | ||
| } | ||
| rr.setType({ kind, rr.type().type() }); | ||
| } | ||
| } |
frontend/lib/resolution/Resolver.cpp
Outdated
| auto combiner = KindProperties::fromKind(kind); | ||
| combiner.setConst(true); | ||
| kind = combiner.toKind(); |
There was a problem hiding this comment.
This too could be a static helper on KindProperties, probbaly called addConstness.
frontend/lib/resolution/Resolver.cpp
Outdated
| } else if (auto namedDecl = actual->toNamedDecl()) { | ||
| resolveNamedDecl(namedDecl, rhsEltType.type()); | ||
|
|
||
| // If the element is an index variable, then use the RHS's intent. |
There was a problem hiding this comment.
Not clear to me why this is needed, could you explain?
There was a problem hiding this comment.
If I don't do this then the component will have the INDEX intent, which isn't concrete.
There was a problem hiding this comment.
But why do you make it const?
There was a problem hiding this comment.
Ah, I see your comment below.
| assert(qtTup.kind() == QualifiedType::CONST_VAR); | ||
| assert(qtTup.type()->isTupleType()); |
There was a problem hiding this comment.
You should probably be checking the component kinds too, since your code adjusts them one-by-one.
| assert(m["i"].kind() == QualifiedType::CONST_VAR); | ||
| assert(m["i"].type()->isIntType()); | ||
| assert(m["j"].kind() == QualifiedType::CONST_VAR); | ||
| assert(m["j"].type()->isIntType()); |
There was a problem hiding this comment.
Could you explain why these are const?
There was a problem hiding this comment.
They're const because they're const in production, if the RHS is not a reference or a product of a zip() expression it's usually const.
| assert(m["b"].kind() == QualifiedType::CONST_REF); | ||
| assert(m["b"].type()->isIntType()); | ||
| } | ||
|
|
||
| int main() { |
There was a problem hiding this comment.
Could you add one more test that ensures that = is not resolved for records when they are assigned to _? You added that logic but I don't see it tested here.
There was a problem hiding this comment.
Good reminder, thanks!!!
Signed-off-by: David Longnecker <[email protected]>
Signed-off-by: David Longnecker <[email protected]>
Signed-off-by: David Longnecker <[email protected]>
Signed-off-by: David Longnecker <[email protected]>
Signed-off-by: David Longnecker <[email protected]>
Signed-off-by: David Longnecker <[email protected]>
dlongnecke-cray
force-pushed
the
dyno-detuple-throwaway-vars
branch
from
896c6b6
to
c1cbad0
Compare
This PR adds support for
_throwaway variables, de-tupled loop index variables e.g.,for (i, _) in foo() do;, andreftuples.