Skip to content

[llvm][CodeGen] Fixed a bug in stall cycle calculation for window scheduler #99451

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 25, 2024
Merged

[llvm][CodeGen] Fixed a bug in stall cycle calculation for window scheduler #99451

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
504d488
[llvm][CodeGen] Fixed a bug in stall cycle calculation for window sch…
kaiyan96 Jul 18, 2024
85594e8
[llvm][CodeGen] Modifications made based on review comments for bugfi…
kaiyan96 Jul 22, 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
8 changes: 5 additions & 3 deletions llvm/lib/CodeGen/WindowScheduler.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -485,15 +485,17 @@ int WindowScheduler::calculateMaxCycle(ScheduleDAGInstrs &DAG,
// ========================================
int WindowScheduler::calculateStallCycle(unsigned Offset, int MaxCycle) {
int MaxStallCycle = 0;
int CurrentII = MaxCycle + 1;
Copy link
Contributor

Choose a reason for hiding this comment

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

Please provide a schematic diagram for the calculation of stall cycles.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

To make the changes in this PR much clear, here is an example of StallCycle calculation.

Without adding 1 to MaxCycle, StallCycle would be 4 instead of 3, which is not correct.
image

Copy link
Contributor

Choose a reason for hiding this comment

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

LGTM

auto Range = getScheduleRange(Offset, SchedInstrNum);
for (auto &MI : Range) {
auto *SU = TripleDAG->getSUnit(&MI);
int DefCycle = getOriCycle(&MI);
for (auto &Succ : SU->Succs) {
if (Succ.isWeak() || Succ.getSUnit() == &TripleDAG->ExitSU)
continue;
// If the expected cycle does not exceed MaxCycle, no check is needed.
if (DefCycle + (int)Succ.getLatency() <= MaxCycle)
// If the expected cycle does not exceed loop initiation interval, no
Copy link
Contributor

Choose a reason for hiding this comment

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

-> "does not exceed CurrentII"

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.

// check is needed.
if (DefCycle + (int)Succ.getLatency() <= CurrentII)
continue;
// If the cycle of the scheduled MI A is less than that of the scheduled
// MI B, the scheduling will fail because the lifetime of the
Expand All @@ -503,7 +505,7 @@ int WindowScheduler::calculateStallCycle(unsigned Offset, int MaxCycle) {
if (DefCycle < UseCycle)
return WindowIILimit;
// Get the stall cycle introduced by the register between two trips.
int StallCycle = DefCycle + (int)Succ.getLatency() - MaxCycle - UseCycle;
int StallCycle = DefCycle + (int)Succ.getLatency() - CurrentII - UseCycle;
MaxStallCycle = std::max(MaxStallCycle, StallCycle);
}
}
Expand Down
59 changes: 59 additions & 0 deletions llvm/test/CodeGen/Hexagon/swp-ws-stall-cycle.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,59 @@
# REQUIRES: asserts
# RUN: llc --march=hexagon %s -run-pass=pipeliner -debug-only=pipeliner \
# RUN: -window-sched=force -filetype=null -verify-machineinstrs \
# RUN: -window-region-limit=1 -window-search-ratio=100 -window-diff-limit=0 \
# RUN: 2>&1 | FileCheck %s

# CHECK-LABEL: Start analyzing II
# CHECK: MaxStallCycle is 0
# CHECK-LABEL: Start analyzing II
# CHECK: MaxStallCycle is 0
# CHECK-LABEL: Start analyzing II
# CHECK: MaxStallCycle is 0

---
name: test_window_stall_cycle
tracksRegLiveness: true
body: |
bb.0:
successors: %bb.3(0x40000000), %bb.1(0x40000000)
liveins: $r0, $r1

%0:intregs = COPY $r1
%1:intregs = COPY $r0
%2:intregs = nsw A2_add %0, %1
%3:intregs = S2_lsr_i_r_acc %2, %2, 31
%4:intregs = S2_asr_i_r killed %3, 1
%5:predregs = C2_cmpgt %1, %4
%6:intregs = A2_tfrsi 0
J2_jumpt killed %5, %bb.3, implicit-def dead $pc
J2_jump %bb.1, implicit-def dead $pc

bb.1:
successors: %bb.2(0x80000000)

%7:intregs = A2_addi %4, 2
%8:intregs = A2_tfrsi 0
%9:intregs = A2_sub %4, %1
%10:intregs = A2_addi %9, 1
%11:intregs = COPY %10
J2_loop0r %bb.2, %11, implicit-def $lc0, implicit-def $sa0, implicit-def $usr

bb.2 (machine-block-address-taken):
successors: %bb.3(0x04000000), %bb.2(0x7c000000)

%12:intregs = PHI %7, %bb.1, %13, %bb.2
%14:intregs = PHI %8, %bb.1, %15, %bb.2
%16:intregs = PHI %8, %bb.1, %17, %bb.2
%18:intregs, %13:intregs = L2_loadri_pi %12, -4
%17:intregs = nsw A2_add killed %18, %16
%15:intregs = A2_max %17, %14
ENDLOOP0 %bb.2, implicit-def $pc, implicit-def $lc0, implicit $sa0, implicit $lc0
J2_jump %bb.3, implicit-def dead $pc

bb.3:
%19:intregs = PHI %6, %bb.0, %15, %bb.2
$r0 = COPY %19
PS_jmpret $r31, implicit-def dead $pc, implicit $r0

...
Loading