Remove changes to demangle test

ping :)

Thanks!! The problem here is when we hit a non clobbering memPhi, we should just continue. Instead we are trying to cast it to a memUseOrDef which causes the failure.

Update : fill mask with poison mask elem in shufflevector

ping :)

I don't think doing phi-of-ops on the following is valid :

Perhaps we could add this case as a check to OpIsSafeForPHIOfOps. While walking the operands we could check if they are used in an assume until we hit a dominator. Not sure this makes complete sense, but we already do something similar to prevent InstSimplify from code walking through non-dominating phis.

Add assert, as per Alina's request.

This patch isn't really useful, the value will be moved to the correct congruence class regardless. Sorry for wasting your time.

Thanks for the review!

In D156055#4541541, @kmitropoulou wrote:

LGTM :)

Forgot to add the diff link to the commit message, closing manually.

Remove old checks.

Minor simplifications and comment rewriting.

We don't need to check that the memPhi is in the same block to deem it unsafe.
A clobbering memPhi that doesn't strictly dominate the original access implies that there are different versions of memory reaching the load and that using the loaded value between iterations isn't safe.

Adding Guozhi Wei as a reviewer.

Hello @aqjune thanks for the comments !! Some users of SimplifyDemandedVectorElts might need some more tweaking to prevent regressions for the cases you noted. I'll work on this as soon as I can!

ping :)

ping :)

rebase & ping

Handle vector-of-ptrs.

Uploaded the fixed tests by mistake.
Address @nikic 's comments.

Ping

Address comments.
Cannot use poison for undefined part of result in SSE4A intrinsics.
Regressions:

• x86-sse4a-inseltpoison.ll
• x86-sse4a.ll

In D142388#4425246, @aaron.ballman wrote:

In D142388#4415349, @rsmith wrote:

I think __builtin_any_value works pretty well, and emphasizes that this can really return any value. I'd also be OK with __builtin_convenient_value, to emphasize that the compiler will pick something that's convenient for it, but I prefer __builtin_any_value.

I can get behind __builtin_any_value as the name.

@ManuelJBrito: would you be willing to make these modifications?

In D149548#4413762, @uweigand wrote:

In D149548#4413639, @nlopes wrote:

If a vector is fully initialized with insertvector (i.e., one operation per index), then the value of the base vector is irrelevant. It can be poison.
Poison in vectors is element-wise. <poison, 42> doesn't propagate to <poison, poison>.

OK, that should be fine for SystemZ then. Thanks!

Completely drop alignment.

In D149548#4405173, @qiucf wrote:

Why this changes IR output of following case?

// RUN: clang vecpromote.c -S -o - -O0 -target s390x-linux-gnu -fzvector -emit-llvm
#include <vecintrin.h>

vector int si;
int g;
int i;

void foo() {
  si = vec_promote(g, i);
}

// store <4 x i32> undef, ptr %__vec.i, align 16, !noalias !6  ; this undef becomes poison after this patch
// %2 = load i32, ptr %__scalar.addr.i, align 4, !noalias !6
// %3 = load i32, ptr %__index.addr.i, align 4, !noalias !6
// %4 = load <4 x i32>, ptr %__vec.i, align 16, !noalias !6
// %vecins.i = insertelement <4 x i32> %4, i32 %2, i32 %and.i

Adding some PowerPC folks to review the change in codegen for vec_promote.
AFAICT its OK to have the undefined values be poison.

Remove redundant check

Hello Nemanja is the change in codegen for vec_promote OK?

This change is visible in clang in particular in codegen for the vec_promote builtin.
It now produces poison instead of undef for the undefined elements.
I think this is OK according to IBM's documentation[https://www.ibm.com/docs/en/epfz/5.3?topic=book-semantics]:

Address comments:

• simplify wordings
• remove undef to poison recommendation

Address comments

Split patch into print update and internal representation.

In D149210#4298124, @nikic wrote:

Could you please split the change to printing + the test updates from all the other changes? The code changes get lost in the large diff.

Thanks for the review! I'll simplify the masks.

Any further comments @RKSimon ?

Update to remove multiple uses of freeze poison.
I am unsure about the code style used in the 128 to 512 casts. Any comments are appreciated.

Implementing the 128 to 512 casts by filling the rest of the vector with the same definition of a nondeterministic_value is not correct because :

Reverted. I'll reassess what can be done here.