Simplified test

In D144070#4131694, @sameerds wrote:

The commit description uses a number "60638" to refer to some issue, but there is no suitable context. If this is a github issue, just putting the whole URL might be better? And do note that this needs to be in the git commit, and not just the review description on this page.

Review comments

Reordered conditions as Jay mentioned

Addressed review comments

• Using isNeverUniform instead of isSourceOfDivergence.
• Removed the bit for always uniform.

In D143154#4099243, @arsenm wrote:

I’m not sure we really need a bit for just these 3 cases

Cleaned up MIR test

Addressed review comments, updated MIR test

Addressed Matt's comments

Addressed Matt's comments

Removed -O0 from run lines of all tests. legalize-uadde.mir, legalize-uaddo.mir, legalize-usube.mir, legalize-usubo.mir had to be updated

Removed -global-isel-abort=0

Addressed review comments, updated summary.

Since only v_add_co_u32_e32 can be converted to SDWA form there is no need to try convert v_addc_co_u32_e64 to it's VOP2 form. This will allow us to get rid of all the checks we were performing on v_addc_co_u32. The shrinking of this instr will be taken care by si-shrink-instructions.

Sorry! Accidently reopened the review with another patch I'm working on. Closing again.

• New changes abandon carry handling
• updated tests for updated rule
• clang format

Realized my mistake the carry output is properly updated (if condition, line 5104)

In D137407#3908174, @foad wrote:

In D137407#3907766, @yassingh wrote:

Where should I add the tests for this change?

Does AMDGPU use this code? If so you could put something in test/CodeGen/AMDGPU/GlobalISel/legalize-add.mir.

Where should I add the tests for this change?

Only shrink V_ADD_CO_U32_e32 instruction if V_ADDC_CO_U32_e32 is not shrinkable. Hence avoiding introducing redundant copy instruction as suggested by @arsenm and @foad

Pre-committed the tests and rebased the diff

Updated the predicate to check isAtomicRMW operations, added another one for CMPXCGH

removed unnecessary namespace identifier llvm:: as @arsenm suggested

In D134266#3802636, @arsenm wrote:

What's this going to be used for?

Removed extra markers

Moved string label template functions back to CFGPrinter.h

Moved common code to generate node label to template functions

Updating entire diff, since changes dropped the original patch, all syntax changes requested have been addressed.
Discussion about whether to move common cl-options for -dot-cfg and -dot-mcfg in single place is still open, but as @arsenm suggested it might not be worth the effort and I agree.

In D133709#3795640, @arsenm wrote:

Latest patch seems to have dropped the base patch

Syntax changes addressing reviewers comments

Changes addressing Matt's comments
Updated pass summary, code syntax and added tests