Skip to content

[ValueTracking] Compute known bits from recursive select/phi #113707

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.

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 8 commits into from
Nov 2, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
24 changes: 17 additions & 7 deletions llvm/lib/Analysis/ValueTracking.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -1566,6 +1566,22 @@ static void computeKnownBitsFromOperator(const Operator *I,
// Skip direct self references.
if (IncValue == P) continue;

// Recurse, but cap the recursion to one level, because we don't
// want to waste time spinning around in loops.
// TODO: See if we can base recursion limiter on number of incoming phi
// edges so we don't overly clamp analysis.
unsigned IncDepth = MaxAnalysisRecursionDepth - 1;

// If the Use is a select of this phi, use the knownbit of the other
// operand to break the recursion.
if (auto *SI = dyn_cast<SelectInst>(IncValue)) {
if (SI->getTrueValue() == P || SI->getFalseValue() == P) {
IncValue = SI->getTrueValue() == P ? SI->getFalseValue()
: SI->getTrueValue();
IncDepth = Depth + 1;
Copy link
Contributor

Choose a reason for hiding this comment

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

Forwarding the question from #114689 (review) because the issue was introduced here: Why do we use the full depth for the select case?

As far as I can tell, a) there may be multiple incoming values in this form, resulting in a fan-out problem and b) we are not actually guaranteed that this breaks recursion (because the other select operand may also be based on the phi indirectly) so the fan-out may be recursive as well.

Copy link
Contributor

Choose a reason for hiding this comment

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

Answered in #114689 (comment), lets keep the conversion in one place.

}
}

// Change the context instruction to the "edge" that flows into the
// phi. This is important because that is where the value is actually
// "evaluated" even though it is used later somewhere else. (see also
Expand All @@ -1574,13 +1590,7 @@ static void computeKnownBitsFromOperator(const Operator *I,
RecQ.CxtI = P->getIncomingBlock(u)->getTerminator();

Known2 = KnownBits(BitWidth);

// Recurse, but cap the recursion to one level, because we don't
// want to waste time spinning around in loops.
// TODO: See if we can base recursion limiter on number of incoming phi
// edges so we don't overly clamp analysis.
computeKnownBits(IncValue, DemandedElts, Known2,
MaxAnalysisRecursionDepth - 1, RecQ);
computeKnownBits(IncValue, DemandedElts, Known2, IncDepth, RecQ);

// See if we can further use a conditional branch into the phi
// to help us determine the range of the value.
Expand Down
115 changes: 115 additions & 0 deletions llvm/test/Transforms/InstCombine/known-phi-recurse.ll
Original file line number Diff line number Diff line change
Expand Up @@ -142,3 +142,118 @@ end:
ret i32 %res
}

define i32 @knownbits_phi_select_test1(ptr %p1, ptr %p2, i8 %x) {
; CHECK-LABEL: @knownbits_phi_select_test1(
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[INDVAR1:%.*]] = phi i8 [ [[LOAD2:%.*]], [[BB2:%.*]] ], [ [[X:%.*]], [[ENTRY:%.*]] ]
; CHECK-NEXT: [[INDVAR3:%.*]] = phi ptr [ [[INDVAR3_NEXT:%.*]], [[BB2]] ], [ [[P1:%.*]], [[ENTRY]] ]
; CHECK-NEXT: [[INDVAR4:%.*]] = phi i32 [ [[INDVAR4_NEXT:%.*]], [[BB2]] ], [ 0, [[ENTRY]] ]
; CHECK-NEXT: [[INDVAR5:%.*]] = phi i32 [ [[INDVAR5_NEXT:%.*]], [[BB2]] ], [ 0, [[ENTRY]] ]
; CHECK-NEXT: switch i8 [[INDVAR1]], label [[DEFAULT:%.*]] [
; CHECK-NEXT: i8 0, label [[EXIT:%.*]]
; CHECK-NEXT: i8 59, label [[BB1:%.*]]
; CHECK-NEXT: i8 35, label [[BB1]]
; CHECK-NEXT: ]
; CHECK: default:
; CHECK-NEXT: [[EXT:%.*]] = sext i8 [[INDVAR1]] to i64
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds i16, ptr [[P2:%.*]], i64 [[EXT]]
; CHECK-NEXT: [[LOAD1:%.*]] = load i16, ptr [[GEP1]], align 2
; CHECK-NEXT: [[MASK:%.*]] = and i16 [[LOAD1]], 8192
; CHECK-NEXT: [[CMP1:%.*]] = icmp eq i16 [[MASK]], 0
; CHECK-NEXT: br i1 [[CMP1]], label [[BB2]], label [[BB1]]
; CHECK: bb1:
; CHECK-NEXT: [[CMP2:%.*]] = icmp ne i32 [[INDVAR4]], 0
; CHECK-NEXT: [[CMP3:%.*]] = icmp ne i32 [[INDVAR5]], 0
; CHECK-NEXT: [[OR_COND:%.*]] = select i1 [[CMP2]], i1 true, i1 [[CMP3]]
; CHECK-NEXT: br i1 [[OR_COND]], label [[BB2]], label [[EXIT]]
; CHECK: bb2:
; CHECK-NEXT: [[CMP4:%.*]] = icmp eq i8 [[INDVAR1]], 39
; CHECK-NEXT: [[EXT2:%.*]] = zext i1 [[CMP4]] to i32
; CHECK-NEXT: [[INDVAR4_NEXT]] = xor i32 [[INDVAR4]], [[EXT2]]
; CHECK-NEXT: [[CMP6:%.*]] = icmp eq i8 [[INDVAR1]], 34
; CHECK-NEXT: [[EXT3:%.*]] = zext i1 [[CMP6]] to i32
; CHECK-NEXT: [[INDVAR5_NEXT]] = xor i32 [[INDVAR5]], [[EXT3]]
; CHECK-NEXT: [[INDVAR3_NEXT]] = getelementptr inbounds i8, ptr [[INDVAR3]], i64 1
; CHECK-NEXT: [[LOAD2]] = load i8, ptr [[INDVAR3_NEXT]], align 1
; CHECK-NEXT: br label [[LOOP]]
; CHECK: exit:
; CHECK-NEXT: ret i32 [[INDVAR5]]
;
entry:
br label %loop

loop:
%indvar1 = phi i8 [ %load2, %bb2 ], [ %x, %entry ]
%indvar2 = phi i64 [ %indvar2.next, %bb2 ], [ 0, %entry ]
%indvar3 = phi ptr [ %indvar3.next, %bb2 ], [ %p1, %entry ]
%indvar4 = phi i32 [ %indvar4.next, %bb2 ], [ 0, %entry ]
%indvar5 = phi i32 [ %indvar5.next, %bb2 ], [ 0, %entry ]
switch i8 %indvar1, label %default [
i8 0, label %exit
i8 59, label %bb1
i8 35, label %bb1
]

default:
%ext = sext i8 %indvar1 to i64
%gep1 = getelementptr inbounds i16, ptr %p2, i64 %ext
%load1 = load i16, ptr %gep1, align 2
%mask = and i16 %load1, 8192
%cmp1 = icmp eq i16 %mask, 0
br i1 %cmp1, label %bb2, label %bb1

bb1:
%cmp2 = icmp ne i32 %indvar4, 0
%cmp3 = icmp ne i32 %indvar5, 0
%or.cond = select i1 %cmp2, i1 true, i1 %cmp3
br i1 %or.cond, label %bb2, label %exit

bb2:
%cmp4 = icmp eq i8 %indvar1, 39
%cmp5 = icmp eq i32 %indvar4, 0
%ext2 = zext i1 %cmp5 to i32
%indvar4.next = select i1 %cmp4, i32 %ext2, i32 %indvar4
%cmp6 = icmp eq i8 %indvar1, 34
%cmp7 = icmp eq i32 %indvar5, 0
%ext3 = zext i1 %cmp7 to i32
%indvar5.next = select i1 %cmp6, i32 %ext3, i32 %indvar5
%indvar3.next = getelementptr inbounds i8, ptr %indvar3, i64 1
%indvar2.next = add i64 %indvar2, 1
%load2 = load i8, ptr %indvar3.next, align 1
br label %loop

exit:
ret i32 %indvar5
}

define i8 @knownbits_phi_select_test2() {
; CHECK-LABEL: @knownbits_phi_select_test2(
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[INDVAR:%.*]] = phi i8 [ 0, [[ENTRY:%.*]] ], [ [[CONTAIN:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[COND0:%.*]] = call i1 @cond()
; CHECK-NEXT: [[CONTAIN]] = select i1 [[COND0]], i8 1, i8 [[INDVAR]]
; CHECK-NEXT: [[COND1:%.*]] = call i1 @cond()
; CHECK-NEXT: br i1 [[COND1]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
; CHECK-NEXT: ret i8 [[CONTAIN]]
;
entry:
br label %loop

loop:
%indvar = phi i8 [ 0, %entry ], [ %contain, %loop ]
%cond0 = call i1 @cond()
%contain = select i1 %cond0, i8 1, i8 %indvar
%cond1 = call i1 @cond()
br i1 %cond1, label %exit, label %loop

exit:
%bool = and i8 %contain, 1
ret i8 %bool
}

declare i1 @cond()
Loading