Adding a comments to the early exits in expandVSXLoadForLE and expandVSXStoreForLE.

Updating patch to delay SVX load and store combine until after legalization.

LGTM as well aside from Zarko's comment.

Rebasing with main.

I did not remove that on purpose, so adding it back makes sense to me.

I've updated the patch to remove the path without the /root/usr suffix and exclusively add the path with the root/usr suffix.

Addressing review comments. Removing path without /root/usr prefix and changing unit test to match.

Hi, this patch broke one of the PowerPC bots (https://lab.llvm.org/buildbot/#/builders/121). unix03-sigpipe-exit.test is failing. Here is the first build with this patch: https://lab.llvm.org/buildbot/#/builders/121/builds/20219

Thank you @nemanjai! I've updated the patch based on your suggestion and tested it for both the existing testcase and the RHEL buildbot failure.

Addressing review comments.

Other than my comment, LGTM

Fixing failing tests added in this patch.

Small change for consistency.

Small change for consistency. llvm::Value -> Value

Updating patch with a fix for emitting builtin arguments in an unspecified order.

Re-opening the revision so that I can update it with a fix.

In D121637#3437371, @thakis wrote:

Looks like this breaks tests on windows: http://45.33.8.238/win/55893/step_7.txt

Please take a look and revert for now if it takes a while to fix.

Fixing a set of builtins added by the rebase with main.

Rebasing with main.

Fixing some testcases that broke due to re-ordering IR in my last update.

Adressing review comments. Added a testcase which fails prior to this patch due to the arguments being emmited multiple times. Refactored some calls to EmitScalarExpr() to be more consistent.

Moving testcases that require code generation from clang/test/Sema/ppc-pair-mma-types.c to clang/test/CodeGen/PowerPC/ppc-mma-types.c and clang/test/CodeGen/PowerPC/builtins-ppc-pair-mma-types.c. This is because some of these tests use the line Ops.push_back(EmitArrayToPointerDecay(E->getArg(i)).getPointer()); which I initially removed from EmitPPCbuiltinExpr. I have added this back in the switch statement where it is needed.

LGTM

Fixing a failing test case.

Minor formatting update.

LGTM. @Chia-hungDuan I've verified that this fixes the issue. Thanks again for implementing this!

Hi @Chia-hungDuan, while doing PowerPC Linux release testing for LLVM 14.0.0-rc1 I encountered some differing object files between Phase2 and Phase3 of the bootstrap build:

# Comparing Phase 2 and Phase 3 files
file ArithmeticOps.cpp.o differs between phase 2 and phase 3
file TestPatterns.cpp.o differs between phase 2 and phase 3

I used git bisect to find that this patch was the first instance of the differing files. We are hoping to fix this for the LLVM 14.0.0 release. I've created this issue in the llvm project. It would be great if you could take a look at it. Thanks!

lgtm

Adressing review comments. Added register alias sp for the register r1.
Added a front end testcase to test the location information and all of the
aliases for r1. Merged the back end test cases so that we are testing both
r1 and x1 in 32bit and 64bit.

Addressing review comments. Added a comment about why we are not using getReservedRegs(). Added implementation for PPC64.

LGTM

@sammccall could you take a look at the other changes when you have time? Thanks!

@Paul-C-Anagnostopoulos could you take a look at the new changes when you have time? Thanks!

@schweitz could you take a look at the new changes when you have time? Thanks!