Thanks for the suggestion

• Added a test to check that hlfir is not output without the flag
• clang-formatted CLOptions.inc

Added a test

Thanks for the review. Changes are as follows:

• Move createHLFIRToFIRPassPipeline before #if !defined(FLANG_EXCCLUDE_CODEGEN
• Only canonicalize HLFIR (creating hlfir.matmul_transpose) if we are optimizing for speed
• Unconditionally run HLFIR passes
• Add tests

In D145502#4201384, @jeanPerier wrote:

Thanks, LGTM. Please beware there is a bot failure, but I think it is unrelated to your patch (you could rebase if this was fixed to check that).

Changes:

• Spell out auto
• Allow unknown operations - this prevents the pass from having to add every MLIR dialect used in Flang (and from breaking when new ones are added).

Use braced initialization

Thanks for your comment.

• Undo previous change
• Add braces around if blocks

Thanks for review. I've updated to initialize using braces.

LGTM, thanks for addressing my comments and answering my questions.

Changed the signature of MatmulTransposeOPConversion::matchAndRewrite to match
the signature in required by the new base class mlir::OpRewritePattern (due to
changes in D145503).

Thanks for review, that's very helpful to know.

In D145503#4189416, @clementval wrote:

Can you add some tests?

In D142420#4080619, @mnadeem wrote:

We have tried full LTO on Aarch64 (without this patch) and have seen a few 3-8% improvements in SPEC and a 100+% improvement in leslie3d.
There were a couple of additional failures in SPEC that we have yet to inspect but AFAICT no issue with respect to bitcode/assembly changes.

Update with rebase on upstream (a header moved)

Looks good to me

Please could you add tests for the flag forwarding logic in flang/tests/Driver/frontend-forwarding.f90. For example see https://reviews.llvm.org/rGd0d4b635786d510cd919cadbeb7e5e19983242cf

Thanks for implementing this.

Thanks for your changes. LGTM but wait for somebody else to look over it

Fix broken test

Thanks for review

Thanks for review.

LGTM

Thanks for reviewing this. Changes are:

• Share argument dereferencing with hlfir.sum
• Share result type calculation with hlfir.sum

Looks good to me

In D143930#4125273, @jeanPerier wrote:

Makes sense. Would adding let dependentDialects = ["fir::FIROpsDialect"]; in hlfir_Dialect declaration in [1] also work ?
If so, that is probably more in line with the tablegen dialect definition approach.

[1]: https://github.com/llvm/llvm-project/blob/3ad26060e4bceb2cf9f4959d659cbb29d88344cf/flang/include/flang/Optimizer/HLFIR/HLFIROpBase.td#L39.

Updated to make the change in tablegen

LGTM, thanks for catching this

@jeanPerier there's a problem building this with -DBUILD_SHARED_LIBS=ON. Putting the new helper functions in Builder/HLFIRTools.cpp requires a dependency from libHLFIRDialect.so to libFIRBuilder.so. But libFIRBuilder.so also depends upon libHLFIRDialect.so, making it circular.