[Parser] Parse call_ref #6103
[Parser] Parse call_ref #6103
Conversation
|
Current dependencies on/for this PR:
This stack of pull requests is managed by Graphite. |
| ;; CHECK-NEXT: (local $0 (ref null $void)) | ||
| ;; CHECK-NEXT: (local $1 (ref null $ret2)) | ||
| ;; CHECK-NEXT: (local $2 (ref null $many)) | ||
| ;; CHECK-NEXT: (local $scratch (i32 i32)) |
There was a problem hiding this comment.
Why were these scratch values created?
There was a problem hiding this comment.
Because the call to many returns a tuple, which is then accessed by multiple different drop instructions. Since we have an expression tree, each value can only be consumed by one expression (its parent) normally. Since tuples need to be accessible by multiple expressions, they need to be stashed in locals first. The scratch locals are the locals IRBuilder introduces to hold these tuples.
| auto data = dataidx(ctx); | ||
| CHECK_ERR(data); | ||
| return ctx.makeArrayNewElem(pos, *type, *data); | ||
| return ctx.in.err("unimplemented instruction"); |
There was a problem hiding this comment.
Why is this now unimplemented?
There was a problem hiding this comment.
I noticed that it incorrectly used dataidx above, when it should have used elemidx, which doesn't exist yet. I removed the incorrect implementation with the standard "unimplemented" error.
tlively
force-pushed
the
parser-array-new-fixed
branch
from
80587bb to
c557024
Compare
tlively
force-pushed
the
parser-call-ref
branch
from
0076179 to
50ffc2b
Compare
tlively
force-pushed
the
parser-array-new-fixed
branch
from
c557024 to
d8f004f
Compare
Also mark array.new_elem as unimplemented as a drive-by; it previously had an incorrect implementation.
tlively
force-pushed
the
parser-call-ref
branch
from
50ffc2b to
08bc7a9
Compare
Also mark array.new_elem as unimplemented as a drive-by; it previously had an incorrect implementation.
Also mark array.new_elem as unimplemented as a drive-by; it previously had an
incorrect implementation.