Just to check that I understand this correctly.
Is the intention behind the function prefix to make it configurable in the end, so that one can also check for Legalized selection DAG and others?

Yes, looks great, thanks

Sorry for taking so long to review this.

This looks like it was fixed by D114652, but that got reverted?

The change of PrivateLabelPrefix was pushed in D114273.

I left a nit inline. Apart from that, LGTM.

It would be nice if we can put the common code with AArch64 into some generic place instead of duplicating it.
Other than that, looks good.

LGTM, thanks

If these registers keep changing, we probably want to switch between architectures at some point. But it’s easier to look at (slightly off) names than at hex numbers, so LGTM.

We have slight problems when it comes to differences between hardware versions.
COMPUTE_USER_ACCUM_1 on gfx9 has the same number as COMPUTE_SHADER_CHKSUM on gfx10.
Also, some have slightly different names. E.g. IA_MULTI_VGT_PARAM for gfx9 is called IA_MULTI_VGT_PARAM_PIPED for gfx10.

Also stop passing them for amdgpu_gfx, since the DAG path seems to skip these. I'm unclear on what amdgpu_gfx's expectations are.

Looks good to me, I left some nits inline.
Someone who is more familiar with GlobalISel should review this.

ping :)

ping

Add test with loop

Rebase

Add pre-commited tests.

Fix lld test failure

Ah, the dynamically sized alloca provoked a message that it is unsupported.
An alloca in a branch works fine.

Fix code and add more tests.

Rebased

Looks good to me.
I tested it with the amdvlk vulkan driver (needs a pal-specific patch) and a short Vulkan CTS test ran through fine (except for pal-related failures).

Also set MEM_ORDERED and WGP_MODE for supported PGMRSrc1 registers.

Fix comments

I don’t see a way to add a test case. It fails an assertion on Windows when compiling with msvc.

Disallow calls with C calling convention from shaders

Update from internal review comments.

Looks good, I have one comment.

Fix return value handling, also need to copy COPYs following the call.

So someone has a preference :)

friendly ping for review

Remove debug dumps

Thanks for the heads-up, I reverted it for now.

Fix formatting

Add fixme that wait on function entry/return should depend on calling convention.

Improve is-widened check in CustomWidenLowerNode to determine if a value was widened or not.

Having this one use a different multiclass is weird looking. Why can't it directly use the same multiclass as the other cases?

I would expect this to look something like

class MTBUF_LoadIntrinsicPta<SDPatternOperator node, ValueType memvt, ValueType vt = memvt>
and then only override the vt in the weird v3 cases

Move is gs-done check to own function.

Improve comment as suggested

Use DebugLoc from call for waitcnt and return early.

Wow, good catch. Looks good to me.

Looks good, thanks!

Right, changed unsigned to uint8_t for offsets in ImageDimIntrinsicInfo.

LGTM

Address review comments: Move patterns to SIInstrInfo.td and use MemoryVT.

Preserve fast-math flags and add test that ensures a16 combining is not done on gfx8.

Fix review comments

Thanks for the notification @davezarzycki, an auto-bisecting bot is cool!

I’m also trying to get it working properly (currently for SDag). I think I got the legalization/widening part working but I’m still trying to figure out how to select the right instruction patterns.

Rebased and fix triple for Thumb2 tests as suggested.

Here you go.

Rebased and added some docs.

Rebased (no conflicts this time).

Rebased and removed a few includes as suggested.
Make the TargetTransformInfo a private member of InstCombiner because it should not be used in general inst combines.
Move CreateOverflowTuple out of InstCombiner and make CreateNonTerminatorUnreachable static.

Rebased and call target-specific combining only for target-specific intrinsics as suggested.
Add Function::isTargetIntrinsic() for this purpose.