Fix MATMUL: respect --realloc-lhs-arrays flag #9089
Conversation
|
... still have to wait how this plays with #9039 , let's merge it after that one. |
| real, allocatable :: A(:,:), B(:,:), C(:,:) | ||
| real :: expected | ||
|
|
||
| allocate(A(3, 4), B(4, 2), C(3, 2)) |
There was a problem hiding this comment.
Let's also add a test where C is not allocated, and ensure it works with --realloc-lhs-arrays, and that it gives a runtime error without i.
baebadb to
72e8757
Compare
|
Added
The integration test runs with the flag. The error case is verified manually but not added as a reference test due to test framework complications with runtime errors. |
|
Update: Both tests now added as requested:
Both tests pass locally. |
HarshitaKalani
left a comment
There was a problem hiding this comment.
Looks good to me. Thanks!
Remove unconditional allocation from instantiate_MatMul. Previously, MATMUL would allocate the result array inside the generated function regardless of the --realloc-lhs-arrays flag setting. This violated LFortran's stricter-by-default behavior where allocatable LHS arrays must be pre-allocated unless --realloc-lhs-arrays is specified. Now MATMUL respects the flag like regular array assignment: - Without flag: error if LHS is not allocated - With flag: auto-(re)allocate LHS as needed Add matmul_03 integration test covering pre-allocated result case. Fixes: lfortran#9086
Addresses review comment: test that matmul works when C is not pre-allocated, using --realloc-lhs-arrays flag. Without the flag, runtime error is raised as expected: Runtime Error: Array 'result' is indexed but not allocated.
Verifies that matmul to unallocated allocatable result gives: Runtime Error: Array 'result' is indexed but not allocated.
c1b3cb8 to
01e1adb
Compare
certik
left a comment
There was a problem hiding this comment.
Looks great, let's see if tests pass.
| @@ -0,0 +1,2 @@ | |||
| At 10:5 of file tests/errors/matmul_unallocated_01.f90 | |||
| Runtime Error: Array 'result' is indexed but not allocated. | |||
There was a problem hiding this comment.
I had to think a while about what this error means: I think we transform matmul to loops, then we index A, and that's where this error comes from. That's fine for now.
Fixes #9086.
Follow-up to #9050: remove unconditional allocation from
instantiate_MatMul. The allocation/reallocation of allocatable LHS should be handled by the assignment pass based on--realloc-lhs-arrays, not inside the intrinsic.Behavior now matches regular array assignment and other intrinsics.