Skip to content

JIT: Run new block layout only in backend #107634

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 7 commits into from
Oct 21, 2024
Merged

JIT: Run new block layout only in backend #107634

Show file tree
Hide file tree
Changes from 6 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
a4c3369
Run new block layout once
amanasifkhalid Sep 10, 2024
e86c950
Add back fgUpdateFlowGraph call
amanasifkhalid Sep 10, 2024
8cddc67
Remove compaction change
amanasifkhalid Sep 10, 2024
9e17e9a
Merge from main
amanasifkhalid Sep 20, 2024
217c173
Undo whitespace change
amanasifkhalid Sep 20, 2024
54bce96
Merge from main
amanasifkhalid Oct 15, 2024
395df26
Merge from main
amanasifkhalid Oct 20, 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
14 changes: 10 additions & 4 deletions src/coreclr/jit/compiler.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5235,8 +5235,7 @@ void Compiler::compCompile(void** methodCodePtr, uint32_t* methodCodeSize, JitFl
m_pLowering->FinalizeOutgoingArgSpace();

// We can not add any new tracked variables after this point.
lvaTrackedFixed = true;
const unsigned numBlocksBeforeLSRA = fgBBcount;
lvaTrackedFixed = true;

// Now that lowering is completed we can proceed to perform register allocation
//
Expand All @@ -5250,8 +5249,10 @@ void Compiler::compCompile(void** methodCodePtr, uint32_t* methodCodeSize, JitFl

if (opts.OptimizationEnabled())
{
// LSRA may introduce new blocks. If it does, rerun layout.
if ((fgBBcount != numBlocksBeforeLSRA) && JitConfig.JitDoReversePostOrderLayout())
// We won't introduce new blocks from here on out,
// so run the new block layout.
//
if (JitConfig.JitDoReversePostOrderLayout())
{
auto lateLayoutPhase = [this] {
fgDoReversePostOrderLayout();
Expand All @@ -5272,7 +5273,12 @@ void Compiler::compCompile(void** methodCodePtr, uint32_t* methodCodeSize, JitFl
DoPhase(this, PHASE_DETERMINE_FIRST_COLD_BLOCK, &Compiler::fgDetermineFirstColdBlock);

// Now that the flowgraph is finalized, run post-layout optimizations.
//
DoPhase(this, PHASE_OPTIMIZE_POST_LAYOUT, &Compiler::optOptimizePostLayout);

// Determine start of cold region if we are hot/cold splitting
//
DoPhase(this, PHASE_DETERMINE_FIRST_COLD_BLOCK, &Compiler::fgDetermineFirstColdBlock);
}

#if FEATURE_LOOP_ALIGN
Expand Down
18 changes: 4 additions & 14 deletions src/coreclr/jit/fgopt.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3341,22 +3341,12 @@ bool Compiler::fgReorderBlocks(bool useProfile)

if (useProfile)
{
// Don't run the new layout until we get to the backend,
// since LSRA can introduce new blocks, and lowering can churn the flowgraph.
//
if (JitConfig.JitDoReversePostOrderLayout())
{
fgDoReversePostOrderLayout();
fgMoveColdBlocks();

if (compHndBBtabCount != 0)
{
fgFindEHRegionEnds();
}

// Renumber blocks to facilitate LSRA's order of block visitation
// TODO: Consider removing this, and using traversal order in lSRA
//
fgRenumberBlocks();

return true;
return (newRarelyRun || movedBlocks || optimizedSwitches);
}

// We will be reordering blocks, so ensure the false target of a BBJ_COND block is its next block
Expand Down
Loading