Hi @MaskRay, @craig.topper, @skan, reviewers, I've addressed your comments. Any more concerns?

rebase

update lit test

In D100085#2682080, @craig.topper wrote:

Thanks. I also found this https://github.com/gcc-mirror/gcc/commit/c422e5f81f42a0fc197f0715f4fcd81f1be90bff can you create a new patch to do the same for llvm/clang and rebase this patch on top of it.

Hi @craig.topper and @skan , THX for review! I tested that deleting sgx indeed leads to not generating "+sgx" in 'target-features', didn't know before:)

revert clang-format and revert deleting FeatureSGX def.

cancel the clang-format in constexpr ProcInfo Processors[] = {}

delete FeatureSGX in the backend since there are no IR intrinsics for SGX.

Hi @MaskRay , I tried to refactor the test file by assembling the common CHECKs. But I found it will lead to too many check-prefixes ine one RUN line. For example,

update lit test and clang-format

In D100085#2675965, @MaskRay wrote:

In D100085#2675919, @FreddyYe wrote:

Hi @MaskRay, I tried to refactor, but met some difficulties. Since these defines are dictionary ordered, a new #define may insert into a common CHECK. So it is difficult to let different RUN share common CHECKs.

Check prefixes of the same kind do not need to be contiguous.

// A:
// B:
// A:

Hi @MaskRay, I tried to refactor, but met some difficulties. Since these defines are dictionary ordered, a new #define may insert into a common CHECK. So it is difficult to let different RUN share common CHECKs.

Updating according to comments. Test refactoring not done.

THX for review!

No more comments?

THX for review!

delete extra spaces. THX for review!

Patch has solved this regression bug in Front End.

Patch has solved this regression bug in Front End.

refine clang-format

If Clobber is edx only, don't change "={eax}" to "=&{eax}".

Ping?

refine according to Lint

Fix Lint: Pre-merge checks

consider Clobber == "edx"

consider "=A" into the patch

I uploaded another patch in front-end D94466. It also can fix this regression bug. Both solutions are reasonable to me, since for now clang is weak to deal with MS inline asm.

THX for review!

simplify the MIR test

Add MIR test

This patch is to fix a regression bug (https://gcc.godbolt.org/z/GG44f6) on recent rewrite on RegAllocFast.cpp. The situation that implicit-def $eax and implicit-def early-clobber $eax both occur in MI is rare. But since old implement can handle it, so I add a some condition to check such MI.

Refine

Reorganize intrinsic orders to avoid using nested macros.

Use header file macros instead.

In D90822#2375423, @pengfei wrote:

Craig's method sounds good.
@FreddyYe , Why we check AVX512BW instead of AVX512F. I saw SDM says it depends on AVX512F.

I was referring to the old implement and test case. Seems like a error introduced before.

The fails are all unknown type errors on Windows, since those typedefs are declared in other header files.
The error message goes like:

ping?

Committed in https://github.com/llvm/llvm-test-suite/commit/33904171b17a6ecebb8822f1179fec559a7be9a1