[llvm] 018a964 - [AArch64][GlobalISel] Fix a crash during selection of a G_ZEXT(s8 = G_LOAD)
Amara Emerson via llvm-commits
llvm-commits at lists.llvm.org
Fri May 28 16:35:31 PDT 2021
Author: Amara Emerson
Date: 2021-05-28T16:35:24-07:00
New Revision: 018a9641ff1a488f3ffbe1251666696fa5b29915
URL: https://github.com/llvm/llvm-project/commit/018a9641ff1a488f3ffbe1251666696fa5b29915
DIFF: https://github.com/llvm/llvm-project/commit/018a9641ff1a488f3ffbe1251666696fa5b29915.diff
LOG: [AArch64][GlobalISel] Fix a crash during selection of a G_ZEXT(s8 = G_LOAD)
We have special handling for a zext of a load <32b because the load does a zext
for free. In that case, we just select the G_ZEXT as if it were a copy but this
triggered the copy checking code to balk at the mismatched size.
This was being hidden because normally these get combined into G_ZEXTLOAD but
for atomics this doesn't happen. The test case here just uses a normal load
because the particular atomic isn't supported yet anyway.
Added:
llvm/test/CodeGen/AArch64/GlobalISel/select-zext-as-copy.mir
Modified:
llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
index 657e2e653c13..6c66f47629b0 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -958,6 +958,15 @@ static bool selectCopy(MachineInstr &I, const TargetInstrInfo &TII,
<< " operand\n");
return false;
}
+
+ // If this a GPR ZEXT that we want to just reduce down into a copy.
+ // The sizes will be mismatched with the source < 32b but that's ok.
+ if (I.getOpcode() == TargetOpcode::G_ZEXT) {
+ I.setDesc(TII.get(AArch64::COPY));
+ assert(SrcRegBank.getID() == AArch64::GPRRegBankID);
+ return selectCopy(I, TII, MRI, TRI, RBI);
+ }
+
I.setDesc(TII.get(AArch64::COPY));
return CheckCopy();
}
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/select-zext-as-copy.mir b/llvm/test/CodeGen/AArch64/GlobalISel/select-zext-as-copy.mir
new file mode 100644
index 000000000000..53d0147f57fe
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/select-zext-as-copy.mir
@@ -0,0 +1,35 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=aarch64-- -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=1 %s -o - | FileCheck %s
+---
+name: zext_of_load_copy
+alignment: 4
+exposesReturnsTwice: false
+legalized: true
+regBankSelected: true
+liveins:
+ - { reg: '$x0', virtual-reg: '' }
+ - { reg: '$x1', virtual-reg: '' }
+fixedStack: []
+stack: []
+callSites: []
+debugValueSubstitutions: []
+constants: []
+machineFunctionInfo: {}
+body: |
+ bb.1:
+ liveins: $x0, $x1
+
+ ; CHECK-LABEL: name: zext_of_load_copy
+ ; CHECK: [[DEF:%[0-9]+]]:gpr64common = IMPLICIT_DEF
+ ; CHECK: [[LDRBBui:%[0-9]+]]:gpr32 = LDRBBui [[DEF]], 0 :: (load 1)
+ ; CHECK: [[SUBREG_TO_REG:%[0-9]+]]:gpr64all = SUBREG_TO_REG 0, [[LDRBBui]], %subreg.sub_32
+ ; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY [[SUBREG_TO_REG]]
+ ; CHECK: [[ANDXri:%[0-9]+]]:gpr64sp = ANDXri [[COPY]], 4096
+ ; CHECK: $x0 = COPY [[ANDXri]]
+ %3:gpr(p0) = G_IMPLICIT_DEF
+ %2:gpr(s8) = G_LOAD %3(p0) :: (load 1)
+ %4:gpr(s64) = G_ZEXT %2(s8)
+ %5:gpr(s64) = G_CONSTANT i64 1
+ %6:gpr(s64) = G_AND %4, %5
+ $x0 = COPY %6
+...
More information about the llvm-commits
mailing list