[PATCH] D127308: [Statepoints] FixupStatepoint: Clear isKill flag if COPY is not deleted.
Denis Antrushin via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Wed Jun 8 07:50:40 PDT 2022
dantrushin created this revision.
dantrushin added a reviewer: skatkov.
Herald added subscribers: jsji, pengfei, hiraditya.
Herald added a project: All.
dantrushin requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.
When spilling CSRs, FixupStatepoint pass does simple copy propagation,
trying to find COPY instruction which defines register being spilled
and spill COPY source instead. I.e., if we have CSR $x and found
$x = COPY $y
we will spill $y instead.
But we may be unable to delete COPY instruction for some reason.
Then, spill will be inserted after it, adding another use of $y.
If COPY instruction was last use of $y (killed it), after insertion of
the spill it is not, so `isKill` flag must be cleared. We failed to do
so and this patch fixes this issue.
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D127308
Files:
llvm/lib/CodeGen/FixupStatepointCallerSaved.cpp
llvm/test/CodeGen/X86/statepoint-fixup-copy-prop.mir
Index: llvm/test/CodeGen/X86/statepoint-fixup-copy-prop.mir
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/X86/statepoint-fixup-copy-prop.mir
@@ -0,0 +1,34 @@
+# RUN: llc -o - %s -run-pass fixup-statepoint-caller-saved -verify-machineinstrs | FileCheck %s
+
+--- |
+ target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+ target triple = "x86_64-pc-linux-gnu"
+
+ declare void @foo(i64)
+
+ define i8 addrspace(1)* @test_cp(i64 %a, i64 %b, i64 %c, i8 addrspace(1)* %p) gc "statepoint-example" {
+ ret i8 addrspace(1)* undef
+ }
+...
+---
+name: test_cp
+alignment: 16
+tracksRegLiveness: true
+liveins:
+ - { reg: '$rdi', virtual-reg: '' }
+frameInfo:
+body: |
+ bb.0:
+ liveins: $rax, $rdi, $rbp
+
+ ; CHECK: renamable $rdi = COPY renamable $rax
+ ; CHECK: MOV64mr %stack.1, 1, $noreg, 0, $noreg, killed $rax :: (store (s64) into %stack.1)
+ ; CHECK: renamable $rax = MOV64rm renamable $rdi, 1, $noreg, 16, $noreg
+ ; CHECK: MOV64mr %stack.0, 1, $noreg, 0, $noreg, killed $rax :: (store (s64) into %stack.0)
+ ; CHECK: STATEPOINT 0, 0, 0, @foo, 2, 0, 2, 0, 2, 1, 1, 8, %stack.0, 0, 2, 1, 1, 8, %stack.1, 0, 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp :: (load (s64) from %stack.0), (load store (s64) on %stack.1)
+ renamable $rdi = COPY killed renamable $rax
+ renamable $rax = MOV64rm renamable $rdi, 1, $noreg, 16, $noreg
+ renamable $rdi = STATEPOINT 0, 0, 0, @foo, 2, 0, 2, 0, 2, 1, renamable $rax, 2, 1, killed renamable $rdi(tied-def 0), 2, 0, 2, 1, 0, 0, csr_64, implicit-def $rsp, implicit-def $ssp
+ $rax = COPY killed renamable $rdi
+ RET 0, killed $rax
+...
Index: llvm/lib/CodeGen/FixupStatepointCallerSaved.cpp
===================================================================
--- llvm/lib/CodeGen/FixupStatepointCallerSaved.cpp
+++ llvm/lib/CodeGen/FixupStatepointCallerSaved.cpp
@@ -153,12 +153,17 @@
RI = ++MachineBasicBlock::iterator(Def);
IsKill = DestSrc->Source->isKill();
- // There are no uses of original register between COPY and STATEPOINT.
- // There can't be any after STATEPOINT, so we can eliminate Def.
if (!Use) {
+ // There are no uses of original register between COPY and STATEPOINT.
+ // There can't be any after STATEPOINT, so we can eliminate Def.
LLVM_DEBUG(dbgs() << "spillRegisters: removing dead copy " << *Def);
Def->eraseFromParent();
+ } else if (IsKill) {
+ // COPY will remain in place, spill will be inserted *after* it, so it is
+ // not a kill of source anymore.
+ const_cast<MachineOperand *>(DestSrc->Source)->setIsKill(false);
}
+
return SrcReg;
}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D127308.435159.patch
Type: text/x-patch
Size: 2739 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220608/7b70d67b/attachment.bin>
More information about the llvm-commits
mailing list