diff options
| author | Luke Hutton <luke.hutton@arm.com> | 2025-08-01 14:19:28 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-08-01 14:19:28 +0100 |
| commit | 03fd5f125b61ad6e80eb589b52db0bee22bdd540 (patch) | |
| tree | cedb0683fd7147313c67395115d8f5d3894a6f4c /llvm/docs/tutorial | |
| parent | 2ae996cbbe92f006d711eeeb8f29e0946fc1c1e8 (diff) | |
| download | llvm-03fd5f125b61ad6e80eb589b52db0bee22bdd540.zip llvm-03fd5f125b61ad6e80eb589b52db0bee22bdd540.tar.gz llvm-03fd5f125b61ad6e80eb589b52db0bee22bdd540.tar.bz2 | |
[mlir][tosa] Enable Constant Operand Check by Default in TOSA Validation Pass (#150598)
Previous behaviour was to only run the check for constant operands when
the "strict-op-spec-alignment" check was enabled. However, this type of
check seems more generally useful without this option enabled. For
example, tosa-to-linalg doesn't enable the "strict-op-spec-alignment"
option when running the tosa-validate pass, but it does expect operands
to be constant.
By enabling this type of checking by default, lowering that don't
support non constant operands and don't use the
"strict-op-spec-alignment" option will fail early, rather than later in
the lowering pipeline.
Should a use-case need to disable this type of checking,
`--tosa-validate="extension=dynamic"` can be used.
Diffstat (limited to 'llvm/docs/tutorial')
0 files changed, 0 insertions, 0 deletions