Skip to content

[sroa][profcheck] Propagate profile in unfoldGEPSelect #163318

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
mtrofin merged 1 commit into from
Oct 14, 2025
Merged

[sroa][profcheck] Propagate profile in unfoldGEPSelect #163318

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
18 changes: 15 additions & 3 deletions llvm/lib/Transforms/Scalar/SROA.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -118,9 +118,13 @@ STATISTIC(
STATISTIC(NumDeleted, "Number of instructions deleted");
STATISTIC(NumVectorized, "Number of vectorized aggregates");

namespace llvm {
/// Disable running mem2reg during SROA in order to test or debug SROA.
static cl::opt<bool> SROASkipMem2Reg("sroa-skip-mem2reg", cl::init(false),
cl::Hidden);
extern cl::opt<bool> ProfcheckDisableMetadataFixes;
} // namespace llvm

namespace {

class AllocaSliceRewriter;
Expand Down Expand Up @@ -1777,7 +1781,8 @@ static void speculateSelectInstLoads(SelectInst &SI, LoadInst &LI,
}

Value *V = IRB.CreateSelect(SI.getCondition(), TL, FL,
LI.getName() + ".sroa.speculated", &SI);
LI.getName() + ".sroa.speculated",
ProfcheckDisableMetadataFixes ? nullptr : &SI);

LLVM_DEBUG(dbgs() << " speculated to: " << *V << "\n");
LI.replaceAllUsesWith(V);
Expand Down Expand Up @@ -4360,10 +4365,13 @@ class AggLoadStoreRewriter : public InstVisitor<AggLoadStoreRewriter, bool> {
};

Value *Cond, *True, *False;
Instruction *MDFrom = nullptr;
if (auto *SI = dyn_cast<SelectInst>(Sel)) {
Cond = SI->getCondition();
True = SI->getTrueValue();
False = SI->getFalseValue();
if (!ProfcheckDisableMetadataFixes)
MDFrom = SI;
} else {
Cond = Sel->getOperand(0);
True = ConstantInt::get(Sel->getType(), 1);
Expand All @@ -4383,8 +4391,12 @@ class AggLoadStoreRewriter : public InstVisitor<AggLoadStoreRewriter, bool> {
IRB.CreateGEP(Ty, FalseOps[0], ArrayRef(FalseOps).drop_front(),
False->getName() + ".sroa.gep", NW);

Value *NSel =
IRB.CreateSelect(Cond, NTrue, NFalse, Sel->getName() + ".sroa.sel");
Value *NSel = MDFrom
? IRB.CreateSelect(Cond, NTrue, NFalse,
Sel->getName() + ".sroa.sel", MDFrom)
: IRB.CreateSelectWithUnknownProfile(
Cond, NTrue, NFalse, DEBUG_TYPE,
Sel->getName() + ".sroa.sel");
Copy link
Contributor

@nikic nikic Oct 14, 2025

Choose a reason for hiding this comment

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

We shouldn't do this. In profcheck mode MDFrom will be non-null, and otherwise we don't care.

Copy link
Member Author

@mtrofin mtrofin Oct 14, 2025

Choose a reason for hiding this comment

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

It can be null from the else at 4375.

Copy link
Contributor

@nikic nikic Oct 14, 2025

Choose a reason for hiding this comment

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

Oh, I see, this is for the zext case. Nevermind then.

Visited.erase(&GEPI);
GEPI.replaceAllUsesWith(NSel);
GEPI.eraseFromParent();
Expand Down
29 changes: 22 additions & 7 deletions llvm/test/Transforms/SROA/phi-gep.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,9 +1,12 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals
; RUN: opt -S -passes='sroa<preserve-cfg>' < %s | FileCheck %s --check-prefixes=CHECK,CHECK-PRESERVE-CFG
; RUN: opt -S -passes='sroa<modify-cfg>' < %s | FileCheck %s --check-prefixes=CHECK,CHECK-MODIFY-CFG

%pair = type { i32, i32 }

;.
; CHECK: @g = global %pair zeroinitializer, align 4
;.
define i32 @test_sroa_phi_gep(i1 %cond) {
; CHECK-LABEL: @test_sroa_phi_gep(
; CHECK-NEXT: entry:
Expand Down Expand Up @@ -334,18 +337,18 @@ exit:
unreachable
}

define void @test_sroa_gep_phi_select_same_block(i1 %c1, i1 %c2, ptr %ptr) {
define void @test_sroa_gep_phi_select_same_block(i1 %c1, i1 %c2, ptr %ptr) !prof !0 {
; CHECK-LABEL: @test_sroa_gep_phi_select_same_block(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [[PAIR:%.*]], align 8
; CHECK-NEXT: br label [[WHILE_BODY:%.*]]
; CHECK: while.body:
; CHECK-NEXT: [[PHI:%.*]] = phi ptr [ [[ALLOCA]], [[ENTRY:%.*]] ], [ [[SELECT:%.*]], [[WHILE_BODY]] ]
; CHECK-NEXT: [[SELECT]] = select i1 [[C1:%.*]], ptr [[PHI]], ptr [[PTR:%.*]]
; CHECK-NEXT: [[SELECT]] = select i1 [[C1:%.*]], ptr [[PHI]], ptr [[PTR:%.*]], !prof [[PROF1:![0-9]+]]
; CHECK-NEXT: [[PHI_SROA_GEP:%.*]] = getelementptr inbounds [[PAIR]], ptr [[PHI]], i64 1
; CHECK-NEXT: [[PTR_SROA_GEP:%.*]] = getelementptr inbounds [[PAIR]], ptr [[PTR]], i64 1
; CHECK-NEXT: [[SELECT_SROA_SEL:%.*]] = select i1 [[C1]], ptr [[PHI_SROA_GEP]], ptr [[PTR_SROA_GEP]]
; CHECK-NEXT: br i1 [[C2:%.*]], label [[EXIT:%.*]], label [[WHILE_BODY]]
; CHECK-NEXT: [[SELECT_SROA_SEL:%.*]] = select i1 [[C1]], ptr [[PHI_SROA_GEP]], ptr [[PTR_SROA_GEP]], !prof [[PROF1]]
; CHECK-NEXT: br i1 [[C2:%.*]], label [[EXIT:%.*]], label [[WHILE_BODY]], !prof [[PROF2:![0-9]+]]
; CHECK: exit:
; CHECK-NEXT: ret void
;
Expand All @@ -355,9 +358,9 @@ entry:

while.body:
%phi = phi ptr [ %alloca, %entry ], [ %select, %while.body ]
%select = select i1 %c1, ptr %phi, ptr %ptr
%select = select i1 %c1, ptr %phi, ptr %ptr, !prof !1
%gep = getelementptr inbounds %pair, ptr %select, i64 1
br i1 %c2, label %exit, label %while.body
br i1 %c2, label %exit, label %while.body, !prof !2

exit:
ret void
Expand Down Expand Up @@ -747,6 +750,18 @@ declare ptr @foo()
declare i32 @__gxx_personality_v0(...)

declare void @llvm.memcpy.p0.p0.i64(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i64, i1 immarg)

!0 = !{!"function_entry_count", i32 10}
!1 = !{!"branch_weights", i32 3, i32 5}
!2 = !{!"branch_weights", i32 7, i32 11}

;.
; CHECK: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
;.
; CHECK: [[META0:![0-9]+]] = !{!"function_entry_count", i32 10}
; CHECK: [[PROF1]] = !{!"branch_weights", i32 3, i32 5}
; CHECK: [[PROF2]] = !{!"branch_weights", i32 7, i32 11}
;.
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; CHECK-MODIFY-CFG: {{.*}}
; CHECK-PRESERVE-CFG: {{.*}}
19 changes: 16 additions & 3 deletions llvm/test/Transforms/SROA/select-gep.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals
; RUN: opt -S -passes='sroa<preserve-cfg>' < %s | FileCheck %s --check-prefixes=CHECK,CHECK-PRESERVE-CFG
; RUN: opt -S -passes='sroa<modify-cfg>' < %s | FileCheck %s --check-prefixes=CHECK,CHECK-MODIFY-CFG

Expand Down Expand Up @@ -203,10 +203,10 @@ define i32 @test_select_idx_mem2reg(i1 %c) {

; Test gep with a select-like zext index unfolding on an alloca that is
; splittable and promotable.
define i64 @test_select_like_zext_idx_mem2reg(i1 %c) {
define i64 @test_select_like_zext_idx_mem2reg(i1 %c) !prof !0 {
; CHECK-LABEL: @test_select_like_zext_idx_mem2reg(
; CHECK-NEXT: [[IDX:%.*]] = zext i1 [[C:%.*]] to i64
; CHECK-NEXT: [[RES:%.*]] = select i1 [[C]], i64 2, i64 1
; CHECK-NEXT: [[RES:%.*]] = select i1 [[C]], i64 2, i64 1, !prof [[PROF1:![0-9]+]]
; CHECK-NEXT: ret i64 [[RES]]
;
%alloca = alloca [2 x i64], align 8
Expand Down Expand Up @@ -352,3 +352,16 @@ define i32 @test_select_idx_not_constant3(i1 %c, ptr %p, i64 %arg) {
%res = load i32, ptr %gep, align 4
ret i32 %res
}

!0 = !{!"function_entry_count", i32 10}
;.
; CHECK-PRESERVE-CFG: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
;.
; CHECK-MODIFY-CFG: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }
;.
; CHECK-PRESERVE-CFG: [[META0:![0-9]+]] = !{!"function_entry_count", i32 10}
; CHECK-PRESERVE-CFG: [[PROF1]] = !{!"unknown", !"sroa"}
;.
; CHECK-MODIFY-CFG: [[META0:![0-9]+]] = !{!"function_entry_count", i32 10}
; CHECK-MODIFY-CFG: [[PROF1]] = !{!"unknown", !"sroa"}
;.