[llvm] 048fc2b - [LiveIntervals] Ignore artificial regs when adding kill flags (#116963)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Dec 4 12:25:35 PST 2024
Author: Sander de Smalen
Date: 2024-12-04T20:25:31Z
New Revision: 048fc2bc102cff806613592829ff275c0f2b826f
URL: https://github.com/llvm/llvm-project/commit/048fc2bc102cff806613592829ff275c0f2b826f
DIFF: https://github.com/llvm/llvm-project/commit/048fc2bc102cff806613592829ff275c0f2b826f.diff
LOG: [LiveIntervals] Ignore artificial regs when adding kill flags (#116963)
If parts of a physical register for a given liverange, as assigned by
the register allocator, can be used to store other values not
represented by this liverange, then `LiveIntervals::addKillFlags`
normally avoids adding a kill flag on the use of this register
when the value's liverange ends.
However, if all the other regunits are artificial, then we can
still safely add the kill flag, since those parts of the register
can never be accessed independently.
Added:
Modified:
llvm/include/llvm/MC/MCRegisterInfo.h
llvm/lib/CodeGen/LiveIntervals.cpp
llvm/lib/MC/MCRegisterInfo.cpp
llvm/test/CodeGen/AArch64/arm64-addrmode.ll
llvm/test/CodeGen/AArch64/nested-iv-regalloc.mir
llvm/test/CodeGen/AArch64/preserve_nonecc_varargs_darwin.ll
Removed:
################################################################################
diff --git a/llvm/include/llvm/MC/MCRegisterInfo.h b/llvm/include/llvm/MC/MCRegisterInfo.h
index 73f29d0f521edf..164ef1ef44bbba 100644
--- a/llvm/include/llvm/MC/MCRegisterInfo.h
+++ b/llvm/include/llvm/MC/MCRegisterInfo.h
@@ -404,6 +404,11 @@ class MCRegisterInfo {
/// be modelled, such as the top 16-bits of a 32-bit GPR.
bool isArtificial(MCRegister RegNo) const { return get(RegNo).IsArtificial; }
+ /// Returns true when the given register unit is considered artificial.
+ /// Register units are considered artificial when at least one of the
+ /// root registers is artificial.
+ bool isArtificialRegUnit(MCRegUnit Unit) const;
+
/// Return the number of registers this target has (useful for
/// sizing arrays holding per register information)
unsigned getNumRegs() const {
diff --git a/llvm/lib/CodeGen/LiveIntervals.cpp b/llvm/lib/CodeGen/LiveIntervals.cpp
index a0b6bf445fa8af..f9ee6e4563f8d6 100644
--- a/llvm/lib/CodeGen/LiveIntervals.cpp
+++ b/llvm/lib/CodeGen/LiveIntervals.cpp
@@ -730,7 +730,12 @@ void LiveIntervals::addKillFlags(const VirtRegMap *VRM) {
// Find the regunit intervals for the assigned register. They may overlap
// the virtual register live range, cancelling any kills.
RU.clear();
- for (MCRegUnit Unit : TRI->regunits(PhysReg)) {
+ LaneBitmask ArtificialLanes;
+ for (MCRegUnitMaskIterator UI(PhysReg, TRI); UI.isValid(); ++UI) {
+ auto [Unit, Bitmask] = *UI;
+ // Record lane mask for all artificial RegUnits for this physreg.
+ if (TRI->isArtificialRegUnit(Unit))
+ ArtificialLanes |= Bitmask;
const LiveRange &RURange = getRegUnit(Unit);
if (RURange.empty())
continue;
@@ -782,7 +787,11 @@ void LiveIntervals::addKillFlags(const VirtRegMap *VRM) {
LaneBitmask DefinedLanesMask;
if (LI.hasSubRanges()) {
// Compute a mask of lanes that are defined.
- DefinedLanesMask = LaneBitmask::getNone();
+ // Artificial regunits are not independently allocatable so the
+ // register allocator cannot have used them to represent any other
+ // values. That's why we mark them as 'defined' here, as this
+ // otherwise prevents kill flags from being added.
+ DefinedLanesMask = ArtificialLanes;
for (const LiveInterval::SubRange &SR : LI.subranges())
for (const LiveRange::Segment &Segment : SR.segments) {
if (Segment.start >= RI->end)
diff --git a/llvm/lib/MC/MCRegisterInfo.cpp b/llvm/lib/MC/MCRegisterInfo.cpp
index 178b1d21e5200a..4a9bacdbc8fc47 100644
--- a/llvm/lib/MC/MCRegisterInfo.cpp
+++ b/llvm/lib/MC/MCRegisterInfo.cpp
@@ -220,3 +220,10 @@ bool MCRegisterInfo::regsOverlap(MCRegister RegA, MCRegister RegB) const {
} while (*IA < *IB ? ++IA != EA : ++IB != EB);
return false;
}
+
+bool MCRegisterInfo::isArtificialRegUnit(unsigned Unit) const {
+ for (MCRegUnitRootIterator Root(Unit, this); Root.isValid(); ++Root)
+ if (isArtificial(*Root))
+ return true;
+ return false;
+}
diff --git a/llvm/test/CodeGen/AArch64/arm64-addrmode.ll b/llvm/test/CodeGen/AArch64/arm64-addrmode.ll
index bfef61abd8c129..f8695b62619c09 100644
--- a/llvm/test/CodeGen/AArch64/arm64-addrmode.ll
+++ b/llvm/test/CodeGen/AArch64/arm64-addrmode.ll
@@ -1,5 +1,5 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=arm64-eabi < %s | FileCheck %s
+; RUN: llc -aarch64-enable-subreg-liveness-tracking -mtriple=arm64-eabi < %s | FileCheck %s
; rdar://10232252
@object = external hidden global i64, section "__DATA, __objc_ivar", align 8
diff --git a/llvm/test/CodeGen/AArch64/nested-iv-regalloc.mir b/llvm/test/CodeGen/AArch64/nested-iv-regalloc.mir
index 3bd8f83d27c2da..ff29c78b5a0ce5 100644
--- a/llvm/test/CodeGen/AArch64/nested-iv-regalloc.mir
+++ b/llvm/test/CodeGen/AArch64/nested-iv-regalloc.mir
@@ -1,5 +1,5 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple aarch64 --run-pass=greedy,virtregrewriter -verify-machineinstrs %s -o - | FileCheck %s
+# RUN: llc -mtriple aarch64 -aarch64-enable-subreg-liveness-tracking --run-pass=greedy,virtregrewriter -verify-machineinstrs %s -o - | FileCheck %s
# We should ideally not spill around any of the SUBSWri in the loop exit blocks (if.end and if.end27).
@@ -219,7 +219,7 @@ body: |
; CHECK-NEXT: liveins: $w10, $w11, $x2, $x8
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: STRXui renamable $x8, %stack.1, 0 :: (store (s64) into %stack.1)
- ; CHECK-NEXT: renamable $w9 = MOVi32imm 36, implicit-def $x9
+ ; CHECK-NEXT: renamable $w9 = MOVi32imm 36
; CHECK-NEXT: renamable $x8 = MADDXrrr killed renamable $x8, killed renamable $x9, $xzr
; CHECK-NEXT: renamable $x9 = MOVaddr target-flags(aarch64-page) @g, target-flags(aarch64-pageoff, aarch64-nc) @g
; CHECK-NEXT: renamable $w8 = LDRWroX killed renamable $x9, killed renamable $x8, 0, 0 :: (load (s32) from %ir.arrayidx9)
@@ -244,7 +244,7 @@ body: |
; CHECK-NEXT: successors: %bb.5(0x50000000), %bb.8(0x30000000)
; CHECK-NEXT: liveins: $w10, $w11, $x2, $x12
; CHECK-NEXT: {{ $}}
- ; CHECK-NEXT: renamable $w8 = MOVi32imm 36, implicit-def $x8
+ ; CHECK-NEXT: renamable $w8 = MOVi32imm 36
; CHECK-NEXT: renamable $x8 = MADDXrrr renamable $x12, killed renamable $x8, $xzr
; CHECK-NEXT: renamable $x9 = MOVaddr target-flags(aarch64-page) @g, target-flags(aarch64-pageoff, aarch64-nc) @g
; CHECK-NEXT: renamable $w8 = LDRWroX killed renamable $x9, killed renamable $x8, 0, 0 :: (load (s32) from %ir.arrayidx14)
diff --git a/llvm/test/CodeGen/AArch64/preserve_nonecc_varargs_darwin.ll b/llvm/test/CodeGen/AArch64/preserve_nonecc_varargs_darwin.ll
index e227f14542cc11..2a77d4dd33fe53 100644
--- a/llvm/test/CodeGen/AArch64/preserve_nonecc_varargs_darwin.ll
+++ b/llvm/test/CodeGen/AArch64/preserve_nonecc_varargs_darwin.ll
@@ -1,5 +1,5 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc -mtriple=aarch64-apple-darwin < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-apple-darwin -aarch64-enable-subreg-liveness-tracking < %s | FileCheck %s
define preserve_nonecc i32 @callee(i32 %a1, i32 %a2, i32 %a3, i32 %a4, i32 %a5, ...) nounwind noinline ssp {
; CHECK-LABEL: callee:
@@ -27,12 +27,11 @@ define i32 @caller() nounwind ssp {
; CHECK-NEXT: sub sp, sp, #208
; CHECK-NEXT: mov w8, #10 ; =0xa
; CHECK-NEXT: mov w9, #9 ; =0x9
-; CHECK-NEXT: mov w0, #1 ; =0x1
+; CHECK-NEXT: mov w10, #8 ; =0x8
; CHECK-NEXT: stp x9, x8, [sp, #24]
-; CHECK-NEXT: mov w8, #8 ; =0x8
-; CHECK-NEXT: mov w9, #6 ; =0x6
-; CHECK-NEXT: str x8, [sp, #16]
; CHECK-NEXT: mov w8, #7 ; =0x7
+; CHECK-NEXT: mov w9, #6 ; =0x6
+; CHECK-NEXT: mov w0, #1 ; =0x1
; CHECK-NEXT: mov w1, #2 ; =0x2
; CHECK-NEXT: mov w2, #3 ; =0x3
; CHECK-NEXT: mov w3, #4 ; =0x4
@@ -47,7 +46,8 @@ define i32 @caller() nounwind ssp {
; CHECK-NEXT: stp x22, x21, [sp, #160] ; 16-byte Folded Spill
; CHECK-NEXT: stp x20, x19, [sp, #176] ; 16-byte Folded Spill
; CHECK-NEXT: stp x29, x30, [sp, #192] ; 16-byte Folded Spill
-; CHECK-NEXT: stp x9, x8, [sp]
+; CHECK-NEXT: stp x8, x10, [sp, #8]
+; CHECK-NEXT: str x9, [sp]
; CHECK-NEXT: bl _callee
; CHECK-NEXT: ldp x29, x30, [sp, #192] ; 16-byte Folded Reload
; CHECK-NEXT: ldp x20, x19, [sp, #176] ; 16-byte Folded Reload
More information about the llvm-commits
mailing list