Skip to content

Reapply "[CodeGen] Fix register pressure computation in MachinePipeli… #87312

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Apr 3, 2024
Merged

Reapply "[CodeGen] Fix register pressure computation in MachinePipeli… #87312

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion llvm/lib/CodeGen/MachinePipeliner.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1268,7 +1268,7 @@ class HighRegisterPressureDetector {
// Calculate the upper limit of each pressure set
void computePressureSetLimit(const RegisterClassInfo &RCI) {
for (unsigned PSet = 0; PSet < PSetNum; PSet++)
PressureSetLimit[PSet] = RCI.getRegPressureSetLimit(PSet);
PressureSetLimit[PSet] = TRI->getRegPressureSetLimit(MF, PSet);

// We assume fixed registers, such as stack pointer, are already in use.
// Therefore subtracting the weight of the fixed registers from the limit of
Expand Down
160 changes: 160 additions & 0 deletions llvm/test/CodeGen/AArch64/sms-regpress.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,160 @@
# RUN: llc --verify-machineinstrs -mtriple=aarch64 -o - %s -run-pass pipeliner -aarch64-enable-pipeliner -pipeliner-max-mii=40 -pipeliner-register-pressure -pipeliner-ii-search-range=30 -debug-only=pipeliner 2>&1 | FileCheck %s

# REQUIRES: asserts

# Check that if the register pressure is too high, the schedule is rejected, II is incremented, and scheduling continues.
# The specific value of II is not important.

# CHECK: {{^ *}}Try to schedule with {{[0-9]+$}}
# CHECK: {{^ *}}Rejected the schedule because of too high register pressure{{$}}
# CHECK: {{^ *}}Try to schedule with {{[0-9]+$}}
# CHECK: {{^ *}}Schedule Found? 1 (II={{[0-9]+}}){{$}}

--- |
target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"

define dso_local double @kernel(ptr nocapture noundef readonly %a, ptr nocapture noundef readonly %b, i32 noundef %n) local_unnamed_addr {
entry:
%0 = load double, ptr %a, align 8
%arrayidx1 = getelementptr inbounds i8, ptr %a, i64 8
%1 = load double, ptr %arrayidx1, align 8
%cmp133 = icmp sgt i32 %n, 0
br i1 %cmp133, label %for.body.preheader, label %for.cond.cleanup

for.body.preheader: ; preds = %entry
%wide.trip.count = zext nneg i32 %n to i64
br label %for.body

for.cond.cleanup: ; preds = %for.body, %entry
%res.0.lcssa = phi double [ 0.000000e+00, %entry ], [ %add54, %for.body ]
ret double %res.0.lcssa

for.body: ; preds = %for.body.preheader, %for.body
%lsr.iv137 = phi i64 [ %wide.trip.count, %for.body.preheader ], [ %lsr.iv.next, %for.body ]
%lsr.iv = phi ptr [ %b, %for.body.preheader ], [ %scevgep, %for.body ]
%res.0135 = phi double [ 0.000000e+00, %for.body.preheader ], [ %add54, %for.body ]
%2 = load double, ptr %lsr.iv, align 8
%3 = tail call double @llvm.fmuladd.f64(double %0, double %2, double %0)
%4 = tail call double @llvm.fmuladd.f64(double %3, double %2, double %3)
%5 = tail call double @llvm.fmuladd.f64(double %4, double %2, double %4)
%6 = tail call double @llvm.fmuladd.f64(double %5, double %2, double %5)
%7 = tail call double @llvm.fmuladd.f64(double %6, double %2, double %6)
%8 = tail call double @llvm.fmuladd.f64(double %7, double %2, double %7)
%9 = tail call double @llvm.fmuladd.f64(double %8, double %2, double %8)
%10 = tail call double @llvm.fmuladd.f64(double %9, double %2, double %9)
%11 = tail call double @llvm.fmuladd.f64(double %10, double %2, double %10)
%12 = tail call double @llvm.fmuladd.f64(double %11, double %2, double %11)
%13 = tail call double @llvm.fmuladd.f64(double %12, double %2, double %12)
%14 = tail call double @llvm.fmuladd.f64(double %13, double %2, double %13)
%15 = tail call double @llvm.fmuladd.f64(double %14, double %2, double %14)
%16 = tail call double @llvm.fmuladd.f64(double %15, double %2, double %15)
%17 = tail call double @llvm.fmuladd.f64(double %16, double %2, double %16)
%18 = tail call double @llvm.fmuladd.f64(double %17, double %2, double %17)
%add = fadd double %17, %18
%19 = tail call double @llvm.fmuladd.f64(double %18, double %2, double %add)
%add35 = fadd double %10, %19
%20 = tail call double @llvm.fmuladd.f64(double %3, double %2, double %add35)
%add38 = fadd double %11, %20
%21 = tail call double @llvm.fmuladd.f64(double %4, double %2, double %add38)
%add41 = fadd double %12, %21
%22 = tail call double @llvm.fmuladd.f64(double %5, double %2, double %add41)
%add44 = fadd double %14, %15
%add45 = fadd double %13, %add44
%add46 = fadd double %add45, %22
%23 = tail call double @llvm.fmuladd.f64(double %6, double %2, double %add46)
%mul = fmul double %2, %7
%mul51 = fmul double %1, %mul
%24 = tail call double @llvm.fmuladd.f64(double %mul51, double %9, double %23)
%25 = tail call double @llvm.fmuladd.f64(double %8, double %1, double %24)
%add54 = fadd double %res.0135, %25
%scevgep = getelementptr i8, ptr %lsr.iv, i64 8
%lsr.iv.next = add nsw i64 %lsr.iv137, -1
%exitcond.not = icmp eq i64 %lsr.iv.next, 0
br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
}

declare double @llvm.fmuladd.f64(double, double, double)

...
---
name: kernel
tracksRegLiveness: true
liveins:
- { reg: '$x0', virtual-reg: '%10' }
- { reg: '$x1', virtual-reg: '%11' }
- { reg: '$w2', virtual-reg: '%12' }
body: |
bb.0.entry:
successors: %bb.1, %bb.4
liveins: $x0, $x1, $w2

%12:gpr32common = COPY $w2
%11:gpr64 = COPY $x1
%10:gpr64common = COPY $x0
dead $wzr = SUBSWri %12, 1, 0, implicit-def $nzcv
Bcc 10, %bb.1, implicit $nzcv

bb.4:
%13:fpr64 = FMOVD0
B %bb.2

bb.1.for.body.preheader:
%0:fpr64 = LDRDui %10, 0 :: (load (s64) from %ir.a)
%1:fpr64 = LDRDui %10, 1 :: (load (s64) from %ir.arrayidx1)
%16:gpr32 = ORRWrs $wzr, %12, 0
%2:gpr64all = SUBREG_TO_REG 0, killed %16, %subreg.sub_32
%15:fpr64 = FMOVD0
B %bb.3

bb.2.for.cond.cleanup:
%3:fpr64 = PHI %13, %bb.4, %7, %bb.3
$d0 = COPY %3
RET_ReallyLR implicit $d0

bb.3.for.body:
successors: %bb.2, %bb.3

%4:gpr64sp = PHI %2, %bb.1, %9, %bb.3
%5:gpr64sp = PHI %11, %bb.1, %8, %bb.3
%6:fpr64 = PHI %15, %bb.1, %7, %bb.3
early-clobber %17:gpr64sp, %18:fpr64 = LDRDpost %5, 8 :: (load (s64) from %ir.lsr.iv)
%19:fpr64 = nofpexcept FMADDDrrr %0, %18, %0, implicit $fpcr
%20:fpr64 = nofpexcept FMADDDrrr %19, %18, %19, implicit $fpcr
%21:fpr64 = nofpexcept FMADDDrrr %20, %18, %20, implicit $fpcr
%22:fpr64 = nofpexcept FMADDDrrr %21, %18, %21, implicit $fpcr
%23:fpr64 = nofpexcept FMADDDrrr %22, %18, %22, implicit $fpcr
%24:fpr64 = nofpexcept FMADDDrrr %23, %18, %23, implicit $fpcr
%25:fpr64 = nofpexcept FMADDDrrr %24, %18, %24, implicit $fpcr
%26:fpr64 = nofpexcept FMADDDrrr %25, %18, %25, implicit $fpcr
%27:fpr64 = nofpexcept FMADDDrrr %26, %18, %26, implicit $fpcr
%28:fpr64 = nofpexcept FMADDDrrr %27, %18, %27, implicit $fpcr
%29:fpr64 = nofpexcept FMADDDrrr %28, %18, %28, implicit $fpcr
%30:fpr64 = nofpexcept FMADDDrrr %29, %18, %29, implicit $fpcr
%31:fpr64 = nofpexcept FMADDDrrr %30, %18, %30, implicit $fpcr
%32:fpr64 = nofpexcept FMADDDrrr %31, %18, %31, implicit $fpcr
%33:fpr64 = nofpexcept FMADDDrrr %32, %18, %32, implicit $fpcr
%34:fpr64 = nofpexcept FMADDDrrr %33, %18, %33, implicit $fpcr
%35:fpr64 = nofpexcept FADDDrr %33, %34, implicit $fpcr
%36:fpr64 = nofpexcept FMADDDrrr %34, %18, killed %35, implicit $fpcr
%37:fpr64 = nofpexcept FADDDrr %26, killed %36, implicit $fpcr
%38:fpr64 = nofpexcept FMADDDrrr %19, %18, killed %37, implicit $fpcr
%39:fpr64 = nofpexcept FADDDrr %27, killed %38, implicit $fpcr
%40:fpr64 = nofpexcept FMADDDrrr %20, %18, killed %39, implicit $fpcr
%41:fpr64 = nofpexcept FADDDrr %28, killed %40, implicit $fpcr
%42:fpr64 = nofpexcept FMADDDrrr %21, %18, killed %41, implicit $fpcr
%43:fpr64 = nofpexcept FADDDrr %30, %31, implicit $fpcr
%44:fpr64 = nofpexcept FADDDrr %29, killed %43, implicit $fpcr
%45:fpr64 = nofpexcept FADDDrr killed %44, killed %42, implicit $fpcr
%46:fpr64 = nofpexcept FMADDDrrr %22, %18, killed %45, implicit $fpcr
%47:fpr64 = nofpexcept FMULDrr %18, %23, implicit $fpcr
%48:fpr64 = nofpexcept FMULDrr %1, killed %47, implicit $fpcr
%49:fpr64 = nofpexcept FMADDDrrr killed %48, %25, killed %46, implicit $fpcr
%50:fpr64 = nofpexcept FMADDDrrr %24, %1, killed %49, implicit $fpcr
%7:fpr64 = nofpexcept FADDDrr %6, killed %50, implicit $fpcr
%8:gpr64all = COPY %17
%51:gpr64 = nsw SUBSXri %4, 1, 0, implicit-def $nzcv
%9:gpr64all = COPY %51
Bcc 0, %bb.2, implicit $nzcv
B %bb.3

...
Loading