Possibly. But more fixes are needed as preconditions to enable this, as you mentioned here (fix oneUse check). Without them, we cannot push this futher.

removed line

Any advices about next steps for this patch?

There is no compiler warning to catch this UB?

I wonder hot it looks now. Can you rebase it?

Also please update
https://clang.llvm.org/cxx_status.html

It would be good if anybody familiar with llvm patternmatch could fix m_OneUse to ignore llvm.assume.

Since llvm.assume is not a really true use, oneUse check fails? Then we should really fix it to not count it.

Update old tests

I am not familiar with clang-tidy’s codebase and I personally prefer good compiler warnings than dependency on another tool (clang-tidy). I mean the cases when diagnostic check is easy to do in the compiler.

Some coding guidelines may require switch to have always default label. Even if devs know that default is not reachable, they can add default: abort(); or assert to increase safety (and warning will be silenced).

GCC warns even for for that case, so this patch implements GCC's behaviour for now.

Removed redundant line in DiagnosticGroups

Rebased

Ok, reclaiming patch.

@spatel maybe missing pattern in instcombine?

No review ?

Maybe you can add more tests with these intrinsics?

Do not use builder.

• Added and precommited more tests

Ping @rsmith

Do we have any checker to recommend llvm functions over std ?

Maybe Attributor can infer tail call marks too?

Couldn’t passes like FunctionAttrs / Attributor infer “tail call mark” ?
Not sure how many times these passes go, indeed.

We should avoid this transformation for glibc environment..

Please rebase/update

Plesse try to add more reviewers. This seems to be very promising.

Test?

There is a regression with -O3 -march=native on Haswell. It is slower than standard -O3. GCC 9 with -O3 -march=native produces faster code than standard -O3.

int cmpcomp(char *s, char *p) {

return memcmp(s,p,10) == bcmp(s,p,10) ;

}

Did you check the 462.libquantum 's regression?

Please check also https://bugs.llvm.org/show_bug.cgi?id=40030

Ping @aaron.ballman @rsmith

offtopic a bit..

Cannot we canonicalize to have a smaller index as lhs ?

Ping

Ping

There are no tests which would catch this issue ?

Addressed notes.