[PATCH] D25366: AMDGPU/SI: Emit fixups for long branches
Tom Stellard via llvm-commits
llvm-commits at lists.llvm.org
Fri Oct 7 06:31:37 PDT 2016
tstellarAMD created this revision.
tstellarAMD added a reviewer: arsenm.
tstellarAMD added a subscriber: llvm-commits.
Herald added subscribers: tony-tye, yaxunl, nhaehnle, wdng, kzhuravl.
https://reviews.llvm.org/D25366
Files:
lib/Target/AMDGPU/MCTargetDesc/AMDGPUAsmBackend.cpp
Index: lib/Target/AMDGPU/MCTargetDesc/AMDGPUAsmBackend.cpp
===================================================================
--- lib/Target/AMDGPU/MCTargetDesc/AMDGPUAsmBackend.cpp
+++ lib/Target/AMDGPU/MCTargetDesc/AMDGPUAsmBackend.cpp
@@ -107,6 +107,24 @@
const MCFixup &Fixup, const MCFragment *DF,
const MCValue &Target, uint64_t &Value,
bool &IsResolved) {
+ MCValue Res;
+
+ // When we have complex expressions like: BB0_1 + (BB0_2 - 4), which are
+ // used for long branches, this function will be called with
+ // IsResolved = false and Value set to some pre-computed value. In
+ // the example above, the value would be:
+ // (BB0_1 + (BB0_2 - 4)) - CurrentOffsetFromStartOfFunction.
+ // This is not what we want. We just want the expression computation
+ // only. The reason the MC layer subtracts the current offset from the
+ // expression is because the fixup is of kind FK_PCRel_4.
+ // For these scenarios, evaluateAsValue gives us the computation that we
+ // want.
+ if (!IsResolved && Fixup.getValue()->evaluateAsValue(Res, Layout) &&
+ Res.isAbsolute()) {
+ Value = Res.getConstant();
+ IsResolved = true;
+
+ }
if (IsResolved)
Value = adjustFixupValue(Fixup, Value, &Asm.getContext());
}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D25366.73924.patch
Type: text/x-patch
Size: 1373 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20161007/80991e5e/attachment.bin>
More information about the llvm-commits
mailing list