Skip to content

[VPlan] Fold NOT into predicate of wide compares. #129430

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 4 commits into from
Mar 7, 2025
Merged

[VPlan] Fold NOT into predicate of wide compares. #129430

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
16041bb
[VPlan] Fold NOT into predicate of wide compares.
fhahn Feb 26, 2025
89c97c7
Merge remote-tracking branch 'origin/main' into vplan-fold-not-into-c…
fhahn Mar 4, 2025
f1fe3a1
!fixup move code to other negation handling
fhahn Mar 4, 2025
2702792
Merge branch 'main' into vplan-fold-not-into-cmp-pred
fhahn Mar 6, 2025
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
9 changes: 9 additions & 0 deletions llvm/lib/Transforms/Vectorize/VPlan.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -459,6 +459,9 @@ class VPRecipeBase : public ilist_node_with_parent<VPRecipeBase, VPBasicBlock>,
/// Returns the debug location of the recipe.
DebugLoc getDebugLoc() const { return DL; }

/// Set the recipe's debug location to \p NewDL.
void setDebugLoc(DebugLoc NewDL) { DL = NewDL; }

protected:
/// Compute the cost of this recipe either using a recipe's specialized
/// implementation or using the legacy cost model and the underlying
Expand Down Expand Up @@ -793,6 +796,12 @@ class VPRecipeWithIRFlags : public VPSingleDefRecipe {
return CmpPredicate;
}

void setPredicate(CmpInst::Predicate Pred) {
Copy link
Collaborator

@ayalz ayalz Mar 2, 2025

Choose a reason for hiding this comment

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

Suggested change
void setPredicate(CmpInst::Predicate Pred) {
/// ...
void resetPredicate(CmpInst::Predicate Pred) {

Recipe expected to have a predicate, which this method replaces?

Copy link
Contributor

@lukel97 lukel97 Mar 3, 2025

Choose a reason for hiding this comment

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

I presume this was meant to mirror the LLVM IR method in CmpInst::setPredicate. Is it a thing where we try to keep the methods on VPRecipe in sync with their Instruction counterparts?

Copy link
Contributor Author

@fhahn fhahn Mar 6, 2025

Choose a reason for hiding this comment

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

Yep this was to mirror the corresponding API at CmpInst.

assert(OpType == OperationType::Cmp &&
"recipe doesn't have a compare predicate");
CmpPredicate = Pred;
}

GEPNoWrapFlags getGEPNoWrapFlags() const { return GEPFlags; }

/// Returns true if the recipe has fast-math flags.
Expand Down
15 changes: 15 additions & 0 deletions llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -978,6 +978,21 @@ static void simplifyRecipe(VPRecipeBase &R, VPTypeAnalysis &TypeInfo) {
TypeInfo.inferScalarType(R.getOperand(1)) ==
TypeInfo.inferScalarType(R.getVPSingleValue()))
return R.getVPSingleValue()->replaceAllUsesWith(R.getOperand(1));

if (match(&R, m_Not(m_VPValue(A))) && isa<VPWidenRecipe>(A) &&
Copy link
Collaborator

@ayalz ayalz Mar 3, 2025

Choose a reason for hiding this comment

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

Independently: could also fold Not-Not here, if that ever occurs(?)

Copy link
Contributor

@david-arm david-arm Mar 3, 2025

Choose a reason for hiding this comment

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

Isn't that already done on line 970 without being tied to widen recipes?

Copy link
Contributor Author

@fhahn fhahn Mar 4, 2025

Choose a reason for hiding this comment

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

I moved the code to where NOT(NOT()) is handled to clarify this, thanks

A->getNumUsers() == 1) {
Copy link
Collaborator

@ayalz ayalz Mar 3, 2025

Choose a reason for hiding this comment

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

The transformation modifies A in-place, hence checking that it has a single user (interested in this change). However, the intention is to collect R - worth also check if it has but a single user?
A comment could be added explaining this.

Copy link
Contributor Author

@fhahn fhahn Mar 4, 2025

Choose a reason for hiding this comment

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

Added, thanks

auto *WideCmp = cast<VPWidenRecipe>(A);
if (WideCmp->getOpcode() == Instruction::ICmp ||
WideCmp->getOpcode() == Instruction::FCmp) {
WideCmp->setPredicate(
CmpInst::getInversePredicate(WideCmp->getPredicate()));
R.getVPSingleValue()->replaceAllUsesWith(WideCmp);
// If WideCmp doesn't have a debug location, use the one from the
// negation, to preserve the location.
if (!WideCmp->getDebugLoc())
Copy link
Collaborator

@ayalz ayalz Mar 2, 2025

Choose a reason for hiding this comment

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

nit: worth also checking

Suggested change
if (!WideCmp->getDebugLoc())
if (!WideCmp->getDebugLoc() && R.getDebugLoc())

?

Copy link
Contributor Author

@fhahn fhahn Mar 4, 2025

Choose a reason for hiding this comment

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

Done, thanks!

WideCmp->setDebugLoc(R.getDebugLoc());
}
}
}

void VPlanTransforms::simplifyRecipes(VPlan &Plan, Type &CanonicalIVTy) {
Expand Down
12 changes: 4 additions & 8 deletions llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1037,8 +1037,7 @@ define void @test_widen_exp_v2(ptr noalias %p2, ptr noalias %p, i64 %n) #5 {
; TFALWAYS-NEXT: [[TMP6:%.*]] = tail call double @llvm.exp.f64(double [[TMP4]]) #[[ATTR7]]
; TFALWAYS-NEXT: [[TMP7:%.*]] = insertelement <2 x double> poison, double [[TMP5]], i32 0
; TFALWAYS-NEXT: [[TMP8:%.*]] = insertelement <2 x double> [[TMP7]], double [[TMP6]], i32 1
; TFALWAYS-NEXT: [[TMP9:%.*]] = fcmp ogt <2 x double> [[TMP8]], zeroinitializer
; TFALWAYS-NEXT: [[TMP10:%.*]] = xor <2 x i1> [[TMP9]], splat (i1 true)
; TFALWAYS-NEXT: [[TMP10:%.*]] = fcmp ule <2 x double> [[TMP8]], zeroinitializer
; TFALWAYS-NEXT: [[TMP11:%.*]] = select <2 x i1> [[ACTIVE_LANE_MASK]], <2 x i1> [[TMP10]], <2 x i1> zeroinitializer
; TFALWAYS-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP11]], <2 x double> splat (double 1.000000e+00), <2 x double> zeroinitializer
; TFALWAYS-NEXT: [[TMP12:%.*]] = extractelement <2 x i1> [[ACTIVE_LANE_MASK]], i32 0
Expand Down Expand Up @@ -1082,8 +1081,7 @@ define void @test_widen_exp_v2(ptr noalias %p2, ptr noalias %p, i64 %n) #5 {
; TFFALLBACK-NEXT: [[TMP6:%.*]] = tail call double @llvm.exp.f64(double [[TMP4]]) #[[ATTR7]]
; TFFALLBACK-NEXT: [[TMP7:%.*]] = insertelement <2 x double> poison, double [[TMP5]], i32 0
; TFFALLBACK-NEXT: [[TMP8:%.*]] = insertelement <2 x double> [[TMP7]], double [[TMP6]], i32 1
; TFFALLBACK-NEXT: [[TMP9:%.*]] = fcmp ogt <2 x double> [[TMP8]], zeroinitializer
; TFFALLBACK-NEXT: [[TMP10:%.*]] = xor <2 x i1> [[TMP9]], splat (i1 true)
; TFFALLBACK-NEXT: [[TMP10:%.*]] = fcmp ule <2 x double> [[TMP8]], zeroinitializer
; TFFALLBACK-NEXT: [[TMP11:%.*]] = select <2 x i1> [[ACTIVE_LANE_MASK]], <2 x i1> [[TMP10]], <2 x i1> zeroinitializer
; TFFALLBACK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP11]], <2 x double> splat (double 1.000000e+00), <2 x double> zeroinitializer
; TFFALLBACK-NEXT: [[TMP12:%.*]] = extractelement <2 x i1> [[ACTIVE_LANE_MASK]], i32 0
Expand Down Expand Up @@ -1133,10 +1131,8 @@ define void @test_widen_exp_v2(ptr noalias %p2, ptr noalias %p, i64 %n) #5 {
; TFA_INTERLEAVE-NEXT: [[TMP10:%.*]] = tail call double @llvm.exp.f64(double [[TMP4]]) #[[ATTR7]]
; TFA_INTERLEAVE-NEXT: [[TMP11:%.*]] = insertelement <2 x double> poison, double [[TMP9]], i32 0
; TFA_INTERLEAVE-NEXT: [[TMP12:%.*]] = insertelement <2 x double> [[TMP11]], double [[TMP10]], i32 1
; TFA_INTERLEAVE-NEXT: [[TMP13:%.*]] = fcmp ogt <2 x double> [[TMP8]], zeroinitializer
; TFA_INTERLEAVE-NEXT: [[TMP14:%.*]] = fcmp ogt <2 x double> [[TMP12]], zeroinitializer
; TFA_INTERLEAVE-NEXT: [[TMP15:%.*]] = xor <2 x i1> [[TMP13]], splat (i1 true)
; TFA_INTERLEAVE-NEXT: [[TMP16:%.*]] = xor <2 x i1> [[TMP14]], splat (i1 true)
; TFA_INTERLEAVE-NEXT: [[TMP15:%.*]] = fcmp ule <2 x double> [[TMP8]], zeroinitializer
; TFA_INTERLEAVE-NEXT: [[TMP16:%.*]] = fcmp ule <2 x double> [[TMP12]], zeroinitializer
; TFA_INTERLEAVE-NEXT: [[TMP17:%.*]] = select <2 x i1> [[ACTIVE_LANE_MASK]], <2 x i1> [[TMP15]], <2 x i1> zeroinitializer
; TFA_INTERLEAVE-NEXT: [[TMP18:%.*]] = select <2 x i1> [[ACTIVE_LANE_MASK2]], <2 x i1> [[TMP16]], <2 x i1> zeroinitializer
; TFA_INTERLEAVE-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP17]], <2 x double> splat (double 1.000000e+00), <2 x double> zeroinitializer
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/Transforms/LoopVectorize/AArch64/masked-op-cost.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,8 +5,8 @@
target triple = "aarch64-unknown-linux-gnu"

; CHECK-COST: Checking a loop in 'fixed_width'
; CHECK-COST: Cost of 10 for VF 2: WIDEN store vp<%6>, ir<2>, vp<%5>
; CHECK-COST: Cost of 20 for VF 4: WIDEN store vp<%6>, ir<2>, vp<%5>
; CHECK-COST: Cost of 10 for VF 2: WIDEN store vp<{{.+}}>, ir<2>, ir<{{.+}}>
; CHECK-COST: Cost of 20 for VF 4: WIDEN store vp<{{.+}}>, ir<2>, ir<{{.+}}>
; CHECK-COST: Selecting VF: 1.

; We should decide this loop is not worth vectorising using fixed width vectors
Expand Down
15 changes: 5 additions & 10 deletions llvm/test/Transforms/LoopVectorize/AArch64/simple_early_exit.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,9 +43,8 @@ define i64 @same_exit_block_pre_inc_use1() #1 {
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[TMP10]]
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[TMP13]], i32 0
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 16 x i8>, ptr [[TMP14]], align 1
; CHECK-NEXT: [[TMP15:%.*]] = icmp eq <vscale x 16 x i8> [[WIDE_LOAD]], [[WIDE_LOAD2]]
; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <vscale x 16 x i8> [[WIDE_LOAD]], [[WIDE_LOAD2]]
; CHECK-NEXT: [[INDEX_NEXT3]] = add nuw i64 [[INDEX1]], [[TMP5]]
; CHECK-NEXT: [[TMP16:%.*]] = xor <vscale x 16 x i1> [[TMP15]], splat (i1 true)
; CHECK-NEXT: [[TMP17:%.*]] = call i1 @llvm.vector.reduce.or.nxv16i1(<vscale x 16 x i1> [[TMP16]])
; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT3]], [[N_VEC]]
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 16 x i64> [[VEC_IND]], [[DOTSPLAT]]
Expand Down Expand Up @@ -126,9 +125,8 @@ define i64 @same_exit_block_pre_inc_use4() {
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i64, ptr [[P1]], i64 [[TMP0]]
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i64, ptr [[TMP1]], i32 0
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP2]], align 1
; CHECK-NEXT: [[TMP3:%.*]] = icmp ult <2 x i64> [[VEC_IND]], [[WIDE_LOAD]]
; CHECK-NEXT: [[TMP4:%.*]] = icmp uge <2 x i64> [[VEC_IND]], [[WIDE_LOAD]]
; CHECK-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX1]], 2
; CHECK-NEXT: [[TMP4:%.*]] = xor <2 x i1> [[TMP3]], splat (i1 true)
; CHECK-NEXT: [[TMP5:%.*]] = call i1 @llvm.vector.reduce.or.v2i1(<2 x i1> [[TMP4]])
; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT2]], 64
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
Expand Down Expand Up @@ -204,9 +202,8 @@ define i64 @loop_contains_safe_call() #1 {
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[TMP1]], i32 0
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP2]], align 1
; CHECK-NEXT: [[TMP3:%.*]] = call fast <4 x float> @llvm.sqrt.v4f32(<4 x float> [[WIDE_LOAD]])
; CHECK-NEXT: [[TMP4:%.*]] = fcmp fast ult <4 x float> [[TMP3]], splat (float 3.000000e+00)
; CHECK-NEXT: [[TMP5:%.*]] = fcmp fast oge <4 x float> [[TMP3]], splat (float 3.000000e+00)
; CHECK-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX1]], 4
; CHECK-NEXT: [[TMP5:%.*]] = xor <4 x i1> [[TMP4]], splat (i1 true)
; CHECK-NEXT: [[TMP6:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP5]])
; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT2]], 64
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
Expand Down Expand Up @@ -299,9 +296,8 @@ define i64 @loop_contains_safe_div() #1 {
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 0
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP2]], align 1
; CHECK-NEXT: [[TMP13:%.*]] = udiv <vscale x 4 x i32> [[WIDE_LOAD]], splat (i32 20000)
; CHECK-NEXT: [[TMP14:%.*]] = icmp eq <vscale x 4 x i32> [[TMP13]], splat (i32 1)
; CHECK-NEXT: [[TMP15:%.*]] = icmp ne <vscale x 4 x i32> [[TMP13]], splat (i32 1)
; CHECK-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX2]], [[TMP5]]
; CHECK-NEXT: [[TMP15:%.*]] = xor <vscale x 4 x i1> [[TMP14]], splat (i1 true)
; CHECK-NEXT: [[TMP6:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[TMP15]])
; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT2]], [[INDEX1]]
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[DOTSPLAT]]
Expand Down Expand Up @@ -381,12 +377,11 @@ define i64 @loop_contains_load_after_early_exit(ptr dereferenceable(1024) align(
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[P1]], i64 [[TMP0]]
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 0
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP2]], align 1
; CHECK-NEXT: [[TMP3:%.*]] = icmp eq <4 x i32> [[WIDE_LOAD]], splat (i32 1)
; CHECK-NEXT: [[TMP6:%.*]] = icmp ne <4 x i32> [[WIDE_LOAD]], splat (i32 1)
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i64, ptr [[P2]], i64 [[TMP0]]
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i64, ptr [[TMP4]], i32 0
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <4 x i64>, ptr [[TMP5]], align 8
; CHECK-NEXT: [[INDEX_NEXT3]] = add nuw i64 [[INDEX1]], 4
; CHECK-NEXT: [[TMP6:%.*]] = xor <4 x i1> [[TMP3]], splat (i1 true)
; CHECK-NEXT: [[TMP7:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP6]])
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT3]], 64
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
Expand Down
35 changes: 14 additions & 21 deletions llvm/test/Transforms/LoopVectorize/AArch64/sve-select-cmp.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,9 +10,8 @@ define i32 @select_const_i32_from_icmp(ptr nocapture readonly %v, i64 %n) #0 {
; CHECK-VF4IC1: vector.body:
; CHECK-VF4IC1: [[VEC_PHI:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL:%.*]], %vector.body ]
; CHECK-VF4IC1: [[VEC_LOAD:%.*]] = load <vscale x 4 x i32>
; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = icmp eq <vscale x 4 x i32> [[VEC_LOAD]], splat (i32 3)
; CHECK-VF4IC1-NEXT: [[NOT:%*]] = xor <vscale x 4 x i1> [[VEC_ICMP]], splat (i1 true)
; CHECK-VF4IC1-NEXT: [[VEC_SEL]] = or <vscale x 4 x i1> [[VEC_PHI]], [[NOT]]
; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = icmp ne <vscale x 4 x i32> [[VEC_LOAD]], splat (i32 3)
; CHECK-VF4IC1-NEXT: [[VEC_SEL]] = or <vscale x 4 x i1> [[VEC_PHI]], [[VEC_ICMP]]
; CHECK-VF4IC1: middle.block:
; CHECK-VF4IC1-NEXT: [[OR_RDX:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[VEC_SEL]])
; CHECK-VF4IC1-NEXT: [[FR:%.*]] = freeze i1 [[OR_RDX]]
Expand All @@ -25,18 +24,14 @@ define i32 @select_const_i32_from_icmp(ptr nocapture readonly %v, i64 %n) #0 {
; CHECK-VF4IC4: [[VEC_PHI2:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL2:%.*]], %vector.body ]
; CHECK-VF4IC4: [[VEC_PHI3:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL3:%.*]], %vector.body ]
; CHECK-VF4IC4: [[VEC_PHI4:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL4:%.*]], %vector.body ]
; CHECK-VF4IC4: [[VEC_ICMP1:%.*]] = icmp eq <vscale x 4 x i32> {{.*}}, splat (i32 3)
; CHECK-VF4IC4-NEXT: [[VEC_ICMP2:%.*]] = icmp eq <vscale x 4 x i32> {{.*}}, splat (i32 3)
; CHECK-VF4IC4-NEXT: [[VEC_ICMP3:%.*]] = icmp eq <vscale x 4 x i32> {{.*}}, splat (i32 3)
; CHECK-VF4IC4-NEXT: [[VEC_ICMP4:%.*]] = icmp eq <vscale x 4 x i32> {{.*}}, splat (i32 3)
; CHECK-VF4IC4-NEXT: [[NOT1:%.*]] = xor <vscale x 4 x i1> [[VEC_ICMP1]], splat (i1 true)
; CHECK-VF4IC4-NEXT: [[NOT2:%.*]] = xor <vscale x 4 x i1> [[VEC_ICMP2]], splat (i1 true)
; CHECK-VF4IC4-NEXT: [[NOT3:%.*]] = xor <vscale x 4 x i1> [[VEC_ICMP3]], splat (i1 true)
; CHECK-VF4IC4-NEXT: [[NOT4:%.*]] = xor <vscale x 4 x i1> [[VEC_ICMP4]], splat (i1 true)
; CHECK-VF4IC4-NEXT: [[VEC_SEL1:%.*]] = or <vscale x 4 x i1> [[VEC_PHI1]], [[NOT1]]
; CHECK-VF4IC4-NEXT: [[VEC_SEL2:%.*]] = or <vscale x 4 x i1> [[VEC_PHI2]], [[NOT2]]
; CHECK-VF4IC4-NEXT: [[VEC_SEL3:%.*]] = or <vscale x 4 x i1> [[VEC_PHI3]], [[NOT3]]
; CHECK-VF4IC4-NEXT: [[VEC_SEL4:%.*]] = or <vscale x 4 x i1> [[VEC_PHI4]], [[NOT4]]
; CHECK-VF4IC4: [[VEC_ICMP1:%.*]] = icmp ne <vscale x 4 x i32> {{.*}}, splat (i32 3)
; CHECK-VF4IC4-NEXT: [[VEC_ICMP2:%.*]] = icmp ne <vscale x 4 x i32> {{.*}}, splat (i32 3)
; CHECK-VF4IC4-NEXT: [[VEC_ICMP3:%.*]] = icmp ne <vscale x 4 x i32> {{.*}}, splat (i32 3)
; CHECK-VF4IC4-NEXT: [[VEC_ICMP4:%.*]] = icmp ne <vscale x 4 x i32> {{.*}}, splat (i32 3)
; CHECK-VF4IC4-NEXT: [[VEC_SEL1:%.*]] = or <vscale x 4 x i1> [[VEC_PHI1]], [[VEC_ICMP1]]
; CHECK-VF4IC4-NEXT: [[VEC_SEL2:%.*]] = or <vscale x 4 x i1> [[VEC_PHI2]], [[VEC_ICMP2]]
; CHECK-VF4IC4-NEXT: [[VEC_SEL3:%.*]] = or <vscale x 4 x i1> [[VEC_PHI3]], [[VEC_ICMP3]]
; CHECK-VF4IC4-NEXT: [[VEC_SEL4:%.*]] = or <vscale x 4 x i1> [[VEC_PHI4]], [[VEC_ICMP4]]
; CHECK-VF4IC4: middle.block:
; CHECK-VF4IC4-NEXT: [[OR1:%.*]] = or <vscale x 4 x i1> [[VEC_SEL2]], [[VEC_SEL1]]
; CHECK-VF4IC4-NEXT: [[OR2:%.*]] = or <vscale x 4 x i1> [[VEC_SEL3]], [[OR1]]
Expand Down Expand Up @@ -71,9 +66,8 @@ define i32 @select_i32_from_icmp(ptr nocapture readonly %v, i32 %a, i32 %b, i64
; CHECK-VF4IC1: vector.body:
; CHECK-VF4IC1: [[VEC_PHI:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL:%.*]], %vector.body ]
; CHECK-VF4IC1: [[VEC_LOAD:%.*]] = load <vscale x 4 x i32>
; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = icmp eq <vscale x 4 x i32> [[VEC_LOAD]], splat (i32 3)
; CHECK-VF4IC1-NEXT: [[NOT:%*]] = xor <vscale x 4 x i1> [[VEC_ICMP]], splat (i1 true)
; CHECK-VF4IC1-NEXT: [[VEC_SEL]] = or <vscale x 4 x i1> [[VEC_PHI]], [[NOT]]
; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = icmp ne <vscale x 4 x i32> [[VEC_LOAD]], splat (i32 3)
; CHECK-VF4IC1-NEXT: [[VEC_SEL]] = or <vscale x 4 x i1> [[VEC_PHI]], [[VEC_ICMP]]
; CHECK-VF4IC1: middle.block:
; CHECK-VF4IC1-NEXT: [[OR_RDX:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[VEC_SEL]])
; CHECK-VF4IC1-NEXT: [[FR:%.*]] = freeze i1 [[OR_RDX]]
Expand Down Expand Up @@ -105,9 +99,8 @@ define i32 @select_const_i32_from_fcmp(ptr nocapture readonly %v, i64 %n) #0 {
; CHECK-VF4IC1: vector.body:
; CHECK-VF4IC1: [[VEC_PHI:%.*]] = phi <vscale x 4 x i1> [ zeroinitializer, %vector.ph ], [ [[VEC_SEL:%.*]], %vector.body ]
; CHECK-VF4IC1: [[VEC_LOAD:%.*]] = load <vscale x 4 x float>
; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = fcmp fast ueq <vscale x 4 x float> [[VEC_LOAD]], splat (float 3.000000e+00)
; CHECK-VF4IC1-NEXT: [[NOT:%*]] = xor <vscale x 4 x i1> [[VEC_ICMP]], splat (i1 true)
; CHECK-VF4IC1-NEXT: [[VEC_SEL]] = or <vscale x 4 x i1> [[VEC_PHI]], [[NOT]]
; CHECK-VF4IC1-NEXT: [[VEC_ICMP:%.*]] = fcmp fast one <vscale x 4 x float> [[VEC_LOAD]], splat (float 3.000000e+00)
; CHECK-VF4IC1-NEXT: [[VEC_SEL]] = or <vscale x 4 x i1> [[VEC_PHI]], [[VEC_ICMP]]
; CHECK-VF4IC1: middle.block:
; CHECK-VF4IC1-NEXT: [[OR_RDX:%.*]] = call i1 @llvm.vector.reduce.or.nxv4i1(<vscale x 4 x i1> [[VEC_SEL]])
; CHECK-VF4IC1-NEXT: [[FR:%.*]] = freeze i1 [[OR_RDX]]
Expand Down
3 changes: 1 addition & 2 deletions llvm/test/Transforms/LoopVectorize/AArch64/sve-tail-folding.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -460,8 +460,7 @@ define void @cond_uniform_load(ptr noalias %dst, ptr noalias readonly %src, ptr
; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[COND:%.*]], i64 [[TMP10]]
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, ptr [[TMP11]], i32 0
; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0(ptr [[TMP12]], i32 4, <vscale x 4 x i1> [[ACTIVE_LANE_MASK]], <vscale x 4 x i32> poison)
; CHECK-NEXT: [[TMP13:%.*]] = icmp eq <vscale x 4 x i32> [[WIDE_MASKED_LOAD]], zeroinitializer
; CHECK-NEXT: [[TMP14:%.*]] = xor <vscale x 4 x i1> [[TMP13]], splat (i1 true)
; CHECK-NEXT: [[TMP14:%.*]] = icmp ne <vscale x 4 x i32> [[WIDE_MASKED_LOAD]], zeroinitializer
; CHECK-NEXT: [[TMP15:%.*]] = select <vscale x 4 x i1> [[ACTIVE_LANE_MASK]], <vscale x 4 x i1> [[TMP14]], <vscale x 4 x i1> zeroinitializer
; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> [[BROADCAST_SPLAT]], i32 4, <vscale x 4 x i1> [[TMP15]], <vscale x 4 x i32> poison)
; CHECK-NEXT: [[PREDPHI:%.*]] = select <vscale x 4 x i1> [[TMP15]], <vscale x 4 x i32> [[WIDE_MASKED_GATHER]], <vscale x 4 x i32> zeroinitializer
Expand Down
5 changes: 2 additions & 3 deletions llvm/test/Transforms/LoopVectorize/AArch64/tail-fold-uniform-memops.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,9 +62,8 @@ define void @cond_uniform_load(ptr nocapture %dst, ptr nocapture readonly %src,
; CHECK-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <4 x i1> [ [[INIT_ACTIVE_LANE_MASK]], %vector.ph ], [ [[NEXT_ACTIVE_LANE_MASK:%.*]], %vector.body ]
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[IDX]], 0
; CHECK: [[COND_LOAD:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr {{%.*}}, i32 4, <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i32> poison)
; CHECK-NEXT: [[TMP4:%.*]] = icmp eq <4 x i32> [[COND_LOAD]], zeroinitializer
; CHECK-NEXT: [[TMP5:%.*]] = xor <4 x i1> [[TMP4]], splat (i1 true)
; CHECK-NEXT: [[MASK:%.*]] = select <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i1> [[TMP5]], <4 x i1> zeroinitializer
; CHECK-NEXT: [[TMP4:%.*]] = icmp ne <4 x i32> [[COND_LOAD]], zeroinitializer
; CHECK-NEXT: [[MASK:%.*]] = select <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i1> [[TMP4]], <4 x i1> zeroinitializer
; CHECK-NEXT: call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> [[SRC_SPLAT]], i32 4, <4 x i1> [[MASK]], <4 x i32> poison)
entry:
br label %for.body
Expand Down
Loading