Thanks! LGTM

Almost LGTM

LGTM

In rL366708#680974, @eugenis wrote:

We are still interested in gold compatibility, I think.

@pcc, @eugenis Maybe we should just switch bots to use LLD?

LGTM as-is with few questions and suggestions

It's does not change code, but looks like it can be improved in followup patches

Fix https://godbolt.org/z/-PinQP

Can you please add some trivial test under compiler-rt/test/sanitizer_common ? You can't check sanitization there, just check that does not crash and return reasonable results

In D65056#1603902, @ro wrote:

Ping? It's been a week now. Is there some policy not to xfail the test which as documented in the bug cannot pass on Solaris?

Is any test was broken?

Set of smaller patches may help you to bisect in case of regressions.

fix LTO on kernel

Rebase only

In D64675#1584016, @jfb wrote:

I would keep single stores because it's pretty silly codegen to copy from a global for that.

I would also verify that -O0 isn't affected too badly by this change.

Remove single store

compiler now crashes on this file from CTMark https://reviews.llvm.org/F9549420

undo unrelated line

0xFFFFFFFF

In D64597#1581654, @jfb wrote:

In D64597#1581605, @pcc wrote:

The problem with 0xaaaaaaaa on 32-bit is that it is likely to be a valid address.

When I discussed this with JF I proposed a pointer initialization of 0xffffffff which he agreed to. This value is very likely to trap when accessed (due to accesses likely wrapping to zero) and also has the benefit of being the same pattern as for floats.

Indeed.

Hi @jfb
Here I've added 32bit test and landed it without review as it's simple and does not change existing tests, I thought. Unfortunately later I realized that I've killed some alignment checks.
I don't think that they are important here. If you disagree, I can make something to return them.

In D64031#1581498, @eugenis wrote:

This seems correct. Any idea why it has not been done this way from the start?