Skip to content

[llvm][CodeGen] Fixed max cycle calculation with zero-cost instructions for window scheduler #99454

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 2 commits into from
Jul 24, 2024
Merged

[llvm][CodeGen] Fixed max cycle calculation with zero-cost instructions for window scheduler #99454

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
bcaa0eb
[llvm][CodeGen] Fixed max cycle calculation with zero-cost instructio…
kaiyan96 Jul 18, 2024
54a0491
[llvm][CodeGen] Modifications made based on review comments
kaiyan96 Jul 19, 2024
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
16 changes: 10 additions & 6 deletions llvm/lib/CodeGen/WindowScheduler.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -437,12 +437,16 @@ int WindowScheduler::calculateMaxCycle(ScheduleDAGInstrs &DAG,
int PredCycle = getOriCycle(PredMI);
ExpectCycle = std::max(ExpectCycle, PredCycle + (int)Pred.getLatency());
}
// ResourceManager can be used to detect resource conflicts between the
// current MI and the previously inserted MIs.
while (!RM.canReserveResources(*SU, CurCycle) || CurCycle < ExpectCycle) {
++CurCycle;
if (CurCycle == (int)WindowIILimit)
return CurCycle;
// Zero cost instructions do not need to check resource.
if (!TII->isZeroCost(MI.getOpcode())) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why does this hook exist? How is it different from isMetaInstruction?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Also copy should need resource checks?

Copy link
Contributor Author

@kaiyan96 kaiyan96 Jul 19, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why does this hook exist? How is it different from isMetaInstruction?

  • Metainstructions do not participate in scheduling, but zero-cost instructions do. COPY is a typical example of a zero-cost instruction

Also copy should need resource checks?

  • COPY do not occupy hardware resources so they do not require resource checks, they still need to be handled during instruction scheduling to ensure proper execution.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

COPY do not occupy hardware resources

This is a very target specific assumption but ok

// ResourceManager can be used to detect resource conflicts between the
// current MI and the previously inserted MIs.
while (!RM.canReserveResources(*SU, CurCycle) || CurCycle < ExpectCycle) {
++CurCycle;
if (CurCycle == (int)WindowIILimit)
return CurCycle;
}
RM.reserveResources(*SU, CurCycle);
}
RM.reserveResources(*SU, CurCycle);
OriToCycle[getOriMI(&MI)] = CurCycle;
Expand Down
45 changes: 45 additions & 0 deletions llvm/test/CodeGen/Hexagon/swp-ws-zero-cost.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,45 @@
# REQUIRES: asserts
# RUN: llc --march=hexagon %s -run-pass=pipeliner -debug-only=pipeliner \
# RUN: -window-sched=force -filetype=null -verify-machineinstrs 2>&1 \
# RUN: | FileCheck %s

# CHECK-NOT: Can't find a valid II. Keep searching...
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A check-not check for a specific message is fragile and practically useless. This needs positive checks

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Updated

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

CHECK-NEXT and -implicit-check-not are more reliable than putting a CHECK-NOT in one place

# CHECK: Start analyzing II
# CHECK: Start scheduling Phis
# CHECK: Current window Offset is {{[0-9]+}} and II is {{[0-9]+}}

---
name: relu
tracksRegLiveness: true
body: |
bb.0:
successors: %bb.2(0x30000000), %bb.1(0x50000000)
liveins: $r0, $r1, $r2
%0:intregs = COPY $r2
%1:intregs = COPY $r1
%2:intregs = COPY $r0
%3:predregs = C2_cmpeqi %2, 0
J2_jumpt killed %3, %bb.2, implicit-def dead $pc
J2_jump %bb.1, implicit-def dead $pc
bb.1:
successors: %bb.3(0x80000000)
%4:hvxvr = V6_vd0
%5:intregs = A2_addi %2, 31
%6:intregs = S2_lsr_i_r %5, 5
%7:intregs = COPY %6
J2_loop0r %bb.3, %7, implicit-def $lc0, implicit-def $sa0, implicit-def $usr
J2_jump %bb.3, implicit-def dead $pc
bb.2:
PS_jmpret $r31, implicit-def dead $pc
bb.3 (machine-block-address-taken):
successors: %bb.3(0x7c000000), %bb.2(0x04000000)
%8:intregs = PHI %1, %bb.1, %9, %bb.3
%10:intregs = PHI %0, %bb.1, %14, %bb.3
%11:hvxvr, %9:intregs = V6_vL32b_pi %8, 128
%12:intregs = COPY %10
%13:hvxvr = V6_vmaxw killed %11, %4
%14:intregs = V6_vS32b_pi %12, 128, killed %13
ENDLOOP0 %bb.3, implicit-def $pc, implicit-def $lc0, implicit $sa0, implicit $lc0
J2_jump %bb.2, implicit-def dead $pc
...

Loading