In D145808#4197028, @kiranchandramohan wrote:

@jdoerfert Would using functions from compiler-rt cause issues with target offloading?
Context: We are thinking of using functions from compiler-rt for implementing the complex division operation. The existing implementations in Flang (an inline one) and the one on MLIR both have precision issues whereas the compiler-rt ones work fine.

ping

LGTM

LGTM

Add checks for negative and positive zero
Add tests for f16 and bf16

Fold constant divide at compile time before MLIR

Use suggested check for constant argument instead

Remove unnecessary casts to f32 for other real types

Clarify in unit test that in the f64 case we convert to f32 first

I think we just want to get rid of this. Is there an additional check that needs doing somewhere in Semantic analysis or anything? I couldn't find a case that shouldn't be allowed that is when I tried.

Remove check entirely

LGTM

I think the other suggestion from the warning should be applied, as I don't think it _is_ intended that "quote" is matched by either of these. It would actually be odd if it did as you'd have two matches.

The new failure doesn't look related to this patch, unless I'm mistaken?

Thanks for the suggested fix, I was wondering why this wasn't failing for me locally. Hopefully it will work this time!

Add dependencies to CMake test file

Thanks for fixing this!

Rebase on top of runtime

This appears to be breaking the buildbots (and causing failures for me locally), for example here: https://lab.llvm.org/buildbot/#/builders/160/builds/15142. Could we revert until it can be fixed?

Spell out auto variables

Rebase and spelling correction

Rebase and spelling correction

Fix error string comparison check

Thanks for the review!

Use Destroy over Deallocate
Don't swap the descriptors
Add handling for MOVE_ALLOC(x,x)

Note: this doesn't actually make LTO builds work correctly. There appears to be a bug where LTO builds sometimes cause segmentation faults at flang runtime. This patch just allows LTO builds to build correctly.

@jdoerfert I've had to change this quite a bit to cover some corner cases that came up, are you still happy with it?

Fix failing MLIR lit tests

Make compatible with C and add unit test.

Fix failing unit test

LGTM, if this supports getting more flang build bots up then I do think it has value
Thanks!

This is intended to fix https://github.com/llvm/llvm-project/issues/58233 .
Does this look like the right sort of approach to push the fork handling into the runtime?

On MSVC flang uses a completely custom int128 implementation to get around it lacking int128 support, see: flang/include/flang/Common/uint128.h
I think with clang-cl on Windows we're better off using the builtins and using compiler-rt than using the custom implementation, so fixing this looks sensible.

Add clarification about what is done for unsupported operations.

In D135882#3883183, @vzakhari wrote:

Thanks! LGTM

Fix includes
Add size suggestion to SmallVector

clang-format

Add flag to control new behaviour.
Take results of operations as mlir::complex and convert to fir::complex.

Glad to see this flag being added to flang-new! As a note to self I've written some tests that should be updated once this lands that currently don't pass through the flang-new driver.

Good to see HLFIR getting going! Do you have a plan to support this in the flang-new driver as well as bbc?

I've uploaded D135882 now which shows this approach in action

Remove unnecessary convert.