[MachineOutliner] Preserve regmasks in calls to outlined functions #120940
+119
−0
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
When emitting calls to an outlined function, the register masks from the outlined sequence are lost. The AArch64CollectLOH pass, which I plan to move to PreEmitPass2 (positioned after MachineOutliner), relies on accurate register masks. This patch ensures that regmasks are correctly preserved in the outlined calls, maintaining the required accuracy for subsequent passes.
|
@llvm/pr-subscribers-backend-aarch64 Author: Zhaoxuan Jiang (nocchijiang) ChangesWhen emitting calls to an outlined function, the register masks from the outlined sequence are lost. The AArch64CollectLOH pass, which I plan to move to PreEmitPass2 (positioned after MachineOutliner), relies on accurate register masks. This patch ensures that regmasks are correctly preserved in the outlined calls, maintaining the required accuracy for subsequent passes. Full diff: https://github.com/llvm/llvm-project/pull/120940.diff 2 Files Affected:
diff --git a/llvm/lib/CodeGen/MachineOutliner.cpp b/llvm/lib/CodeGen/MachineOutliner.cpp
index 4c5489434c69bb..c01d9b04a89d93 100644
--- a/llvm/lib/CodeGen/MachineOutliner.cpp
+++ b/llvm/lib/CodeGen/MachineOutliner.cpp
@@ -1117,6 +1117,7 @@ bool MachineOutliner::outline(
// instruction. It also updates call site information for moved
// code.
SmallSet<Register, 2> UseRegs, DefRegs;
+ SmallPtrSet<const uint32_t *, 2> RegMasks;
// Copy over the defs in the outlined range.
// First inst in outlined range <-- Anything that's defined in this
// ... .. range has to be added as an
@@ -1130,6 +1131,12 @@ bool MachineOutliner::outline(
MachineInstr *MI = &*Iter;
SmallSet<Register, 2> InstrUseRegs;
for (MachineOperand &MOP : MI->operands()) {
+ // Collect all regmasks. Merge them in the end.
+ if (MOP.isRegMask()) {
+ RegMasks.insert(MOP.getRegMask());
+ continue;
+ }
+
// Skip over anything that isn't a register.
if (!MOP.isReg())
continue;
@@ -1153,6 +1160,24 @@ bool MachineOutliner::outline(
MI->getMF()->eraseCallSiteInfo(MI);
}
+ if (!RegMasks.empty()) {
+ if (RegMasks.size() == 1) {
+ CallInst->addOperand(
+ MachineOperand::CreateRegMask(*RegMasks.begin()));
+ } else {
+ uint32_t *RegMask = MF->allocateRegMask();
+ unsigned NumRegs =
+ MF->getSubtarget().getRegisterInfo()->getNumRegs();
+ unsigned Size = MachineOperand::getRegMaskSize(NumRegs);
+ memset(RegMask, UINT32_MAX, Size * sizeof(RegMask[0]));
+ for (const uint32_t *Mask : RegMasks) {
+ for (unsigned I = 0; I < NumRegs; ++I)
+ RegMask[I] &= Mask[I];
+ }
+ CallInst->addOperand(MachineOperand::CreateRegMask(RegMask));
+ }
+ }
+
for (const Register &I : DefRegs)
// If it's a def, add it to the call instruction.
CallInst->addOperand(
diff --git a/llvm/test/CodeGen/AArch64/machine-outliner-regmask.mir b/llvm/test/CodeGen/AArch64/machine-outliner-regmask.mir
new file mode 100644
index 00000000000000..047a73f81dd2ae
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/machine-outliner-regmask.mir
@@ -0,0 +1,94 @@
+# RUN: llc -mtriple=aarch64-apple-ios -run-pass=prologepilog -run-pass=machine-outliner %s -o - | FileCheck %s
+--- |
+ declare swiftcc void @bar()
+ declare void @baz(i32, i32, i32) #0
+
+ define void @test_same_regmask() #0 {
+ ret void
+ }
+ define void @test_different_regmasks() #0 {
+ ret void
+ }
+ define void @foo(i32, i32, i32, i32, i32, i32, i32, i32) #0 {
+ ret void
+ }
+
+...
+---
+name: foo
+tracksRegLiveness: true
+body: |
+ bb.0:
+ RET undef $lr
+
+
+...
+---
+name: test_same_regmask
+tracksRegLiveness: true
+body: |
+ ; CHECK-LABEL: name: test_same_regmask
+ ; CHECK: bb.1:
+ ; CHECK-NEXT: BL @OUTLINED_FUNCTION_1, csr_aarch64_aapcs
+
+ bb.0:
+ $sp = frame-setup SUBXri $sp, 16, 0
+
+ bb.1:
+ $w0 = MOVZWi 1, 0
+ $w1 = MOVZWi 2, 0
+ $w2 = MOVZWi 3, 0
+ BL @baz, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit killed $w0, implicit killed $w1, implicit killed $w2, implicit-def $sp
+ $w0 = MOVZWi 1, 0
+ $w1 = MOVZWi 2, 0
+ $w2 = MOVZWi 3, 0
+ BL @baz, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit killed $w0, implicit killed $w1, implicit killed $w2, implicit-def $sp
+ $sp = ADDXri $sp, 16, 0
+ RET undef $lr
+
+
+...
+---
+name: test_different_regmasks
+tracksRegLiveness: true
+body: |
+ ; CHECK-LABEL: name: test_different_regmasks
+ ; CHECK: bb.1:
+ ; CHECK-NEXT: BL @OUTLINED_FUNCTION_0, CustomRegMask($fp,$lr,$wzr,$wzr_hi,$xzr,$b8,$b9,$b10,$b11,$b12,$b13,$b14,$b15,$d8,$d9,$d10,$d11,$d12,$d13,$d14,$d15,$h8,$h9,$h10,$h11,$h12,$h13,$h14,$h15,$s8,$s9,$s10,$s11,$s12,$s13,$s14,$s15,$w19,$w20,$w22,$w23,$w24,$w25,$w26,$w27,$w28,$w29,$w30,$x19,$x20,$x22,$x23,$x24,$x25,$x26,$x27,$x28,$b8_hi,$b9_hi,$b10_hi,$b11_hi,$b12_hi,$b13_hi,$b14_hi,$b15_hi,$h8_hi,$h9_hi,$h10_hi,$h11_hi,$h12_hi,$h13_hi,$h14_hi,$h15_hi,$s8_hi,$s9_hi,$s10_hi,$s11_hi,$s12_hi,$s13_hi,$s14_hi,$s15_hi,$w19_hi,$w20_hi,$w22_hi,$w23_hi,$w24_hi,$w25_hi,$w26_hi,$w27_hi,$w28_hi,$w29_hi,$w30_hi,$d8_d9,$d9_d10,$d10_d11,$d11_d12,$d12_d13,$d13_d14,$d14_d15,$d8_d9_d10_d11,$d9_d10_d11_d12,$d10_d11_d12_d13,$d11_d12_d13_d14,$d12_d13_d14_d15,$d8_d9_d10,$d9_d10_d11,$d10_d11_d12,$d11_d12_d13,$d12_d13_d14,$d13_d14_d15,$x22_x23_x24_x25_x26_x27_x28_fp,$w22_w23,$w24_w25,$w26_w27,$w28_w29,$x28_fp,$x22_x23,$x24_x25,$x26_x27)
+
+ bb.0:
+ $sp = frame-setup SUBXri $sp, 16, 0
+
+ bb.1:
+ $w0 = MOVZWi 1, 0
+ $w1 = MOVZWi 2, 0
+ $w2 = MOVZWi 3, 0
+ $w3 = MOVZWi 4, 0
+ $w4 = MOVZWi 5, 0
+ $w5 = MOVZWi 6, 0
+ $w6 = MOVZWi 7, 0
+ $w7 = MOVZWi 8, 0
+ BL @foo, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit killed $w0, implicit killed $w1, implicit killed $w2, implicit killed $w3, implicit killed $w4, implicit killed $w5, implicit killed $w6, implicit killed $w7, implicit-def $sp
+ BL @bar, csr_aarch64_aapcs_swifterror, implicit-def dead $lr, implicit $sp, implicit-def $sp
+ $w0 = MOVZWi 1, 0
+ $w1 = MOVZWi 2, 0
+ $w2 = MOVZWi 3, 0
+ $w3 = MOVZWi 4, 0
+ $w4 = MOVZWi 5, 0
+ $w5 = MOVZWi 6, 0
+ $w6 = MOVZWi 7, 0
+ $w7 = MOVZWi 8, 0
+ BL @foo, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit killed $w0, implicit killed $w1, implicit killed $w2, implicit killed $w3, implicit killed $w4, implicit killed $w5, implicit killed $w6, implicit killed $w7, implicit-def $sp
+ BL @bar, csr_aarch64_aapcs_swifterror, implicit-def dead $lr, implicit $sp, implicit-def $sp
+ $w0 = MOVZWi 1, 0
+ $w1 = MOVZWi 2, 0
+ $w2 = MOVZWi 3, 0
+ $w3 = MOVZWi 4, 0
+ $w4 = MOVZWi 5, 0
+ $w5 = MOVZWi 6, 0
+ $w6 = MOVZWi 7, 0
+ $w7 = MOVZWi 8, 0
+ BL @foo, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit killed $w0, implicit killed $w1, implicit killed $w2, implicit killed $w3, implicit killed $w4, implicit killed $w5, implicit killed $w6, implicit killed $w7, implicit-def $sp
+ BL @bar, csr_aarch64_aapcs_swifterror, implicit-def dead $lr, implicit $sp, implicit-def $sp
+ $sp = ADDXri $sp, 16, 0
+ RET undef $lr
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When emitting calls to an outlined function, the register masks from the outlined sequence are lost. The AArch64CollectLOH pass, which I plan to move to PreEmitPass2 (positioned after MachineOutliner), relies on accurate register masks. This patch ensures that regmasks are correctly preserved in the outlined calls, maintaining the required accuracy for subsequent passes.