[Downstream Change][ARM]Disable memcpy LDM/STM inlining

[nasherm]

What is the change
Patch to disable memcpy LDM/STM inlining for Cortex-V7M

Why this change cannot be done upstream
An attempt was made to upstream this patch. Upstream reviewers wanted a better solution that could be more fine-grained for the issue at hand.

[pawosm-arm]

@paulwalker-arm FYI

[dcandler]

Upstream reviewers wanted a better solution that could be more fine-grained for the issue at hand.

Was there a reason this couldn't be done?

[pawosm-arm]

What is the link to the upstream PR this issue mentions?

[nasherm]

Upstream reviewers wanted a better solution that could be more fine-grained for the issue at hand.

Was there a reason this couldn't be done?

Upstream would prefer a different solution

[dcandler]

Upstream would prefer a different solution
And why can't we implement their preferred solution? We should be avoiding downstream changes if at all possible, and if it's not possible we should record the rationale.

Also, please don't close the issue, as it needs to track the change while it exists.

[nasherm]

And why can't we implement their preferred solution? We should be avoiding downstream changes if at all possible, and if it's not possible we should record the rationale.

The original patch existed in our downstream repo only and gave performance gain for a set of cores. When attempting to upstream, reviewers didn't see a benefit in accepting the change as they were unhappy with the codegen. The patch itself helps with getting our toolchain (LLVM ET at the time) performance closer to the downstream compiler. We made a judgement call to just add a downstream patch file rather than investigating and implementing a new solution that could give the same performance benefit due to time constraints.

Also, please don't close the issue, as it needs to track the change while it exists.

Oh sorry, my mistake