[PowerPC] fix float ABI selection on ppcle #154773
Conversation
llvmbot
added
clang
|
@llvm/pr-subscribers-clang-codegen Author: None (DanilaZhebryakov) Changessoft float ABI selection was not taking effect on little-endian powerPC with embedded vectors (e.g. e500v2) leading to errors. Full diff: https://github.com/llvm/llvm-project/pull/154773.diff 1 Files Affected:
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 677d8bc82cb0a..8b85b93b5d568 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -188,7 +188,8 @@ createTargetCodeGenInfo(CodeGenModule &CGM) {
return createPPC32TargetCodeGenInfo(CGM, IsSoftFloat);
}
case llvm::Triple::ppcle: {
- bool IsSoftFloat = CodeGenOpts.FloatABI == "soft";
+ bool IsSoftFloat =
+ CodeGenOpts.FloatABI == "soft" || Target.hasFeature("spe");
return createPPC32TargetCodeGenInfo(CGM, IsSoftFloat);
}
case llvm::Triple::ppc64:
|
|
@llvm/pr-subscribers-clang Author: None (DanilaZhebryakov) Changessoft float ABI selection was not taking effect on little-endian powerPC with embedded vectors (e.g. e500v2) leading to errors. Full diff: https://github.com/llvm/llvm-project/pull/154773.diff 1 Files Affected:
diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 677d8bc82cb0a..8b85b93b5d568 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -188,7 +188,8 @@ createTargetCodeGenInfo(CodeGenModule &CGM) {
return createPPC32TargetCodeGenInfo(CGM, IsSoftFloat);
}
case llvm::Triple::ppcle: {
- bool IsSoftFloat = CodeGenOpts.FloatABI == "soft";
+ bool IsSoftFloat =
+ CodeGenOpts.FloatABI == "soft" || Target.hasFeature("spe");
return createPPC32TargetCodeGenInfo(CGM, IsSoftFloat);
}
case llvm::Triple::ppc64:
|
There was a problem hiding this comment.
Please add a regression test.
I'm not sure the Target.hasFeature("spe") is really what we want to be doing here. It looks like spe is actually a mutually incompatible ABI, not just a CPU feature, and we should model it as such. But it matches big-endian ppc, so I won't block this patch on fixing that.
DanilaZhebryakov
force-pushed
the
ppcle-float-abi-fix
branch
from
96fde1d to
8e05a06
Compare
|
Added test. |
DanilaZhebryakov
force-pushed
the
ppcle-float-abi-fix
branch
from
3995f52 to
542f85b
Compare
|
Done. It appears that existing checks in clang/test/CodeGen/PowerPC/ppc-sfvarargs.c are sufficient to check that correct ABI is used. |
DanilaZhebryakov
force-pushed
the
ppcle-float-abi-fix
branch
from
33ac551 to
8ad5395
Compare
|
also oops, some weird changes are shown due to accidental merge rather than rebase. |
|
We have automation that's supposed to trigger, but since it isn't:
|
Done. If everything is ok, please merge it. |
soft float ABI selection was not taking effect on little-endian powerPC with embedded vectors (e.g. e500v2) leading to errors.
(embedded vectors use "extended" GPRs to store floating-point values, and this caused issues with variadic arguments assuming dedicated floating-point registers with hard-float ABI)