[X86] Create all-one vector(v8i32) for TESTC(X,~X) == TESTC(X,-1) if …

[yubingex007-a11y]

…X is v8f32

[AlexeySachkov]

Is there any reason this is not pushed to the upstream llvm/llvm-project repo?

[yubingex007-a11y]

Is there any reason this is not pushed to the upstream llvm/llvm-project repo?
some customer would like to fix a bug ASAP since it block their work

[yubingex007-a11y]

i encounter a issue that i can't cherry-pick  https://reviews.llvm.org/D148608 to intel/llvm since the testcase doesn't match. i guess there is some commits in llorg change the testcase and these commits are not in intel/llvm . 

REF:

llvm/llvm/test/CodeGen/X86/combine-testpd.ll

Lines 1 to 6 in 48abe38

	; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
	; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s
	;
	; testz(~X,Y) -> testc(X,Y)
	;

https://github.com/llvm/llvm-project/blob/9356097206a17eb781180d5f895b52f8b0333f8e/llvm/test/CodeGen/X86/combine-testpd.ll#L1-L7

[yubingex007-a11y]

i saw there are 3 commits modified llvm/llvm/test/CodeGen/X86/combine-testpd.ll in llorg but these commits are not in intel/llvm.
so it is brutal to cherry-pick all of them. Thus, i remove the testcase changes.
when pulldown coordinator see the commit land on xmain-web it lead to conflict and he will add back the testcases.
@AlexeySachkov

[yubingex007-a11y]

@AlexeySachkov ping?

[yubingex007-a11y]

ping? @AlexeySachkov

[AlexeySachkov]

I'm ok with cherry-picking this earlier to unblock a customer.

@againull, @FreddyLeaf: FYI. This is a partial cherry-pick of D148608, because we had to adjust tests to avoid cherry-picking more commits. If you encounter conflicts with this patch, they should be resolved by picking community version of the code

[againull]

USM/usm_pooling.cpp reported here: #9148