[CIR] [Lowering] [X86_64] VaArg for long double in x86 #1087
Conversation
ChuanqiXu9
changed the title
ChuanqiXu9
force-pushed
the
VAArgForLongDoubleInX86
branch
from
2b69de9 to
7d28be4
Compare
Fixed |
Wow, this is amazing. Is that only for x86_64 or did you try other arches? |
|
I'll take a look after the next PR update (conflicts + stacked approach, or after the other one lands) |
I only tested on X86_64 |
|
Close to create stacked PR |
|
@bcardosolopes after I rebased, I noticed the return value of |
That might be related to #1089? That reverted a previously landed commit, and there's some discussion there. |
|
It looks like the case. Thanks. |
Thanks. BTW, can we add a rule like the upstream that we can only create branches starts with |
This patch implements transformations for VAArg in X86_64 ABI **in shape**. `In shape` means it can't work properly due to the dependent X86_64 ABI is not robust. e.g., when we want to use VAArg with `long double`, we need #1087. This patch literally implement https://github.com/llvm/llvm-project/blob/d233fedfb0de882353c348cd1ac57dab619efa6d/clang/lib/CodeGen/Targets/X86.cpp#L3015-L3240 in CIR. There some differences due to the traditional pipeline are converting AST to LLVM and we're transforming CIR to CIR. And also to get the ABI Info, I moved `X86_64ABIInfo` to the header.
This is the following of #1086. Since I can't create branches to this repo so that I can't start real stacked PRs. After #1086 got merged, the PR will show the real change. This PR is for preview only.
After #1086, when we want to use
LongDoubleforVAArg, we will be in trouble due to details in X86_64's ABI and this patch tries to address this.The practical impact the patch is, after this patch, with #1088 and a small following up fix, we can build and run all C's benchmark in SpecCPU 2017. I think it is a milestone.