Skip to content

Commit ca2e43c

Browse files
Mel-ChenMel-Chen
committed
[LoopVectorize] Vectorize fixed-order recurrence with vscale x 1.
1 parent c738308 commit ca2e43c

File tree

3 files changed

+95
-17
lines changed

3 files changed

+95
-17
lines changed

‎llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

Lines changed: 12 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -6139,11 +6139,6 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I,
61396139

61406140
// First-order recurrences are replaced by vector shuffles inside the loop.
61416141
if (VF.isVector() && Legal->isFixedOrderRecurrence(Phi)) {
6142-
// For <vscale x 1 x i64>, if vscale = 1 we are unable to extract the
6143-
// penultimate value of the recurrence.
6144-
// TODO: Consider vscale_range info.
6145-
if (VF.isScalable() && VF.getKnownMinValue() == 1)
6146-
return InstructionCost::getInvalid();
61476142
SmallVector<int> Mask(VF.getKnownMinValue());
61486143
std::iota(Mask.begin(), Mask.end(), VF.getKnownMinValue() - 1);
61496144
return TTI.getShuffleCost(TargetTransformInfo::SK_Splice,
@@ -8564,13 +8559,17 @@ addUsersInExitBlocks(VPlan &Plan,
85648559
/// users in the original exit block using the VPIRInstruction wrapping to the
85658560
/// LCSSA phi.
85668561
static void addExitUsersForFirstOrderRecurrences(
8567-
VPlan &Plan, SetVector<VPIRInstruction *> &ExitUsersToFix) {
8562+
VPlan &Plan, SetVector<VPIRInstruction *> &ExitUsersToFix, VFRange &Range) {
85688563
VPRegionBlock *VectorRegion = Plan.getVectorLoopRegion();
85698564
auto *ScalarPHVPBB = Plan.getScalarPreheader();
85708565
auto *MiddleVPBB = Plan.getMiddleBlock();
85718566
VPBuilder ScalarPHBuilder(ScalarPHVPBB);
85728567
VPBuilder MiddleBuilder(MiddleVPBB, MiddleVPBB->getFirstNonPhi());
85738568

8569+
auto IsScalableOne = [](ElementCount VF) -> bool {
8570+
return VF.isScalable() && VF.getKnownMinValue() == 1;
8571+
};
8572+
85748573
for (auto &HeaderPhi : VectorRegion->getEntryBasicBlock()->phis()) {
85758574
auto *FOR = dyn_cast<VPFirstOrderRecurrencePHIRecipe>(&HeaderPhi);
85768575
if (!FOR)
@@ -8652,6 +8651,12 @@ static void addExitUsersForFirstOrderRecurrences(
86528651
for (VPIRInstruction *ExitIRI : ExitUsersToFix) {
86538652
if (ExitIRI->getOperand(0) != FOR)
86548653
continue;
8654+
// For VF vscale x 1, if vscale = 1, we are unable to extract the
8655+
// penultimate value of the recurrence. Instead, we can extract the last
8656+
// element directly from VPInstruction::FirstOrderRecurrenceSplice.
8657+
if (LoopVectorizationPlanner::getDecisionAndClampRange(IsScalableOne,
8658+
Range))
8659+
return;
86558660
VPValue *PenultimateElement = MiddleBuilder.createNaryOp(
86568661
VPInstruction::ExtractPenultimateElement, {FOR->getBackedgeValue()},
86578662
{}, "vector.recur.extract.for.phi");
@@ -8867,7 +8872,7 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(
88678872
addScalarResumePhis(RecipeBuilder, *Plan, IVEndValues);
88688873
SetVector<VPIRInstruction *> ExitUsersToFix =
88698874
collectUsersInLatchExitBlock(*Plan);
8870-
addExitUsersForFirstOrderRecurrences(*Plan, ExitUsersToFix);
8875+
addExitUsersForFirstOrderRecurrences(*Plan, ExitUsersToFix, Range);
88718876
addUsersInExitBlocks(*Plan, ExitUsersToFix);
88728877

88738878
// ---------------------------------------------------------------------------

‎llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

Lines changed: 0 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -3830,9 +3830,6 @@ VPFirstOrderRecurrencePHIRecipe::computeCost(ElementCount VF,
38303830
if (VF.isScalar())
38313831
return Ctx.TTI.getCFInstrCost(Instruction::PHI, Ctx.CostKind);
38323832

3833-
if (VF.isScalable() && VF.getKnownMinValue() == 1)
3834-
return InstructionCost::getInvalid();
3835-
38363833
return 0;
38373834
}
38383835

‎llvm/test/Transforms/LoopVectorize/first-order-recurrence-scalable-vf1.ll

Lines changed: 83 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -8,17 +8,51 @@ define i64 @pr97452_scalable_vf1_for_live_out(ptr %src) {
88
; CHECK-LABEL: define i64 @pr97452_scalable_vf1_for_live_out(
99
; CHECK-SAME: ptr [[SRC:%.*]]) {
1010
; CHECK-NEXT: [[ENTRY:.*]]:
11+
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
12+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 23, [[TMP0]]
13+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
14+
; CHECK: [[VECTOR_PH]]:
15+
; CHECK-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
16+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 23, [[TMP1]]
17+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 23, [[N_MOD_VF]]
18+
; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
19+
; CHECK-NEXT: [[TMP3:%.*]] = call i32 @llvm.vscale.i32()
20+
; CHECK-NEXT: [[TMP4:%.*]] = sub i32 [[TMP3]], 1
21+
; CHECK-NEXT: [[VECTOR_RECUR_INIT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 0, i32 [[TMP4]]
22+
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
23+
; CHECK: [[VECTOR_BODY]]:
24+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
25+
; CHECK-NEXT: [[VECTOR_RECUR:%.*]] = phi <vscale x 1 x i64> [ [[VECTOR_RECUR_INIT]], %[[VECTOR_PH]] ], [ [[WIDE_LOAD:%.*]], %[[VECTOR_BODY]] ]
26+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[SRC]], i64 [[INDEX]]
27+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[TMP5]], i32 0
28+
; CHECK-NEXT: [[WIDE_LOAD]] = load <vscale x 1 x i64>, ptr [[TMP6]], align 8
29+
; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 1 x i64> @llvm.vector.splice.nxv1i64(<vscale x 1 x i64> [[VECTOR_RECUR]], <vscale x 1 x i64> [[WIDE_LOAD]], i32 -1)
30+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP2]]
31+
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
32+
; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
33+
; CHECK: [[MIDDLE_BLOCK]]:
34+
; CHECK-NEXT: [[TMP9:%.*]] = call i32 @llvm.vscale.i32()
35+
; CHECK-NEXT: [[TMP10:%.*]] = sub i32 [[TMP9]], 1
36+
; CHECK-NEXT: [[TMP11:%.*]] = extractelement <vscale x 1 x i64> [[TMP7]], i32 [[TMP10]]
37+
; CHECK-NEXT: [[TMP12:%.*]] = call i32 @llvm.vscale.i32()
38+
; CHECK-NEXT: [[TMP13:%.*]] = sub i32 [[TMP12]], 1
39+
; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <vscale x 1 x i64> [[WIDE_LOAD]], i32 [[TMP13]]
40+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 23, [[N_VEC]]
41+
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
42+
; CHECK: [[SCALAR_PH]]:
43+
; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i64 [ [[VECTOR_RECUR_EXTRACT]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
44+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
1145
; CHECK-NEXT: br label %[[LOOP:.*]]
1246
; CHECK: [[LOOP]]:
13-
; CHECK-NEXT: [[FOR:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[L:%.*]], %[[LOOP]] ]
14-
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
47+
; CHECK-NEXT: [[FOR:%.*]] = phi i64 [ [[SCALAR_RECUR_INIT]], %[[SCALAR_PH]] ], [ [[L:%.*]], %[[LOOP]] ]
48+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
1549
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
1650
; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i64, ptr [[SRC]], i64 [[IV]]
1751
; CHECK-NEXT: [[L]] = load i64, ptr [[GEP]], align 8
1852
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV]], 22
19-
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]]
53+
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
2054
; CHECK: [[EXIT]]:
21-
; CHECK-NEXT: [[RES:%.*]] = phi i64 [ [[FOR]], %[[LOOP]] ]
55+
; CHECK-NEXT: [[RES:%.*]] = phi i64 [ [[FOR]], %[[LOOP]] ], [ [[TMP11]], %[[MIDDLE_BLOCK]] ]
2256
; CHECK-NEXT: ret i64 [[RES]]
2357
;
2458
entry:
@@ -43,17 +77,51 @@ define void @pr97452_scalable_vf1_for_no_live_out(ptr %src, ptr noalias %dst) {
4377
; CHECK-LABEL: define void @pr97452_scalable_vf1_for_no_live_out(
4478
; CHECK-SAME: ptr [[SRC:%.*]], ptr noalias [[DST:%.*]]) {
4579
; CHECK-NEXT: [[ENTRY:.*]]:
80+
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
81+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 23, [[TMP0]]
82+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
83+
; CHECK: [[VECTOR_PH]]:
84+
; CHECK-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
85+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 23, [[TMP1]]
86+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 23, [[N_MOD_VF]]
87+
; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
88+
; CHECK-NEXT: [[TMP3:%.*]] = call i32 @llvm.vscale.i32()
89+
; CHECK-NEXT: [[TMP4:%.*]] = sub i32 [[TMP3]], 1
90+
; CHECK-NEXT: [[VECTOR_RECUR_INIT:%.*]] = insertelement <vscale x 1 x i64> poison, i64 0, i32 [[TMP4]]
91+
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
92+
; CHECK: [[VECTOR_BODY]]:
93+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
94+
; CHECK-NEXT: [[VECTOR_RECUR:%.*]] = phi <vscale x 1 x i64> [ [[VECTOR_RECUR_INIT]], %[[VECTOR_PH]] ], [ [[WIDE_LOAD:%.*]], %[[VECTOR_BODY]] ]
95+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i64, ptr [[SRC]], i64 [[INDEX]]
96+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i64, ptr [[TMP5]], i32 0
97+
; CHECK-NEXT: [[WIDE_LOAD]] = load <vscale x 1 x i64>, ptr [[TMP6]], align 8
98+
; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 1 x i64> @llvm.vector.splice.nxv1i64(<vscale x 1 x i64> [[VECTOR_RECUR]], <vscale x 1 x i64> [[WIDE_LOAD]], i32 -1)
99+
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i64, ptr [[DST]], i64 [[INDEX]]
100+
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i64, ptr [[TMP8]], i32 0
101+
; CHECK-NEXT: store <vscale x 1 x i64> [[TMP7]], ptr [[TMP9]], align 8
102+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP2]]
103+
; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
104+
; CHECK-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
105+
; CHECK: [[MIDDLE_BLOCK]]:
106+
; CHECK-NEXT: [[TMP11:%.*]] = call i32 @llvm.vscale.i32()
107+
; CHECK-NEXT: [[TMP12:%.*]] = sub i32 [[TMP11]], 1
108+
; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <vscale x 1 x i64> [[WIDE_LOAD]], i32 [[TMP12]]
109+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 23, [[N_VEC]]
110+
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
111+
; CHECK: [[SCALAR_PH]]:
112+
; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i64 [ [[VECTOR_RECUR_EXTRACT]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
113+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
46114
; CHECK-NEXT: br label %[[LOOP:.*]]
47115
; CHECK: [[LOOP]]:
48-
; CHECK-NEXT: [[FOR:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[L:%.*]], %[[LOOP]] ]
49-
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
116+
; CHECK-NEXT: [[FOR:%.*]] = phi i64 [ [[SCALAR_RECUR_INIT]], %[[SCALAR_PH]] ], [ [[L:%.*]], %[[LOOP]] ]
117+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
50118
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
51119
; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i64, ptr [[SRC]], i64 [[IV]]
52120
; CHECK-NEXT: [[L]] = load i64, ptr [[GEP]], align 8
53121
; CHECK-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i64, ptr [[DST]], i64 [[IV]]
54122
; CHECK-NEXT: store i64 [[FOR]], ptr [[GEP_DST]], align 8
55123
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV]], 22
56-
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]]
124+
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
57125
; CHECK: [[EXIT]]:
58126
; CHECK-NEXT: ret void
59127
;
@@ -74,3 +142,11 @@ loop:
74142
exit:
75143
ret void
76144
}
145+
;.
146+
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
147+
; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
148+
; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
149+
; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
150+
; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
151+
; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]}
152+
;.

0 commit comments

Comments
 (0)