Skip to content

Commit f37a84c

Browse files
ElvisWang123ElvisWang123
committed
[RISCV][TTI] Implement getAddressComputationCost() in RISCV TTI.
This patch impelemt the `getAddressComputationCost()` in RISCV TTI which make the gather/scatter with address calculation more expansive that stride cost. Note that the only user of `getAddressComputationCost()` with vector type is in `VPWidenMemoryRecipe::computeCost()`. So this patch make some LV tests changes. I've checked the tests changes in LV and seems those changes can be divided into two groups. * gather/scatter with uniform vector ptr, seems can be optimized to masked.load. * can optimize to stirde load/store.
1 parent effc9f8 commit f37a84c

File tree

5 files changed

+55
-114
lines changed

5 files changed

+55
-114
lines changed

‎llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp

Lines changed: 13 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1566,6 +1566,19 @@ RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
15661566
return BaseT::getIntrinsicInstrCost(ICA, CostKind);
15671567
}
15681568

1569+
InstructionCost RISCVTTIImpl::getAddressComputationCost(Type *Ty,
1570+
ScalarEvolution *SE,
1571+
const SCEV *Ptr) const {
1572+
// Address computations with vector type are usually for indexed load/store
1573+
// which is likely more expensive.
1574+
if (ST->hasVInstructions() && Ty->isVectorTy())
1575+
return getArithmeticInstrCost(
1576+
Instruction::Add, Ty, TTI::TCK_RecipThroughput,
1577+
{TTI::OK_AnyValue, TTI::OP_None}, {TTI::OK_AnyValue, TTI::OP_None}, {});
1578+
1579+
return BaseT::getAddressComputationCost(Ty, SE, Ptr);
1580+
}
1581+
15691582
InstructionCost RISCVTTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
15701583
Type *Src,
15711584
TTI::CastContextHint CCH,

‎llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -177,6 +177,10 @@ class RISCVTTIImpl final : public BasicTTIImplBase<RISCVTTIImpl> {
177177
getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
178178
TTI::TargetCostKind CostKind) const override;
179179

180+
InstructionCost
181+
getAddressComputationCost(Type *Ty, ScalarEvolution *SE = nullptr,
182+
const SCEV *Ptr = nullptr) const override;
183+
180184
InstructionCost getInterleavedMemoryOpCost(
181185
unsigned Opcode, Type *VecTy, unsigned Factor, ArrayRef<unsigned> Indices,
182186
Align Alignment, unsigned AddressSpace, TTI::TargetCostKind CostKind,

‎llvm/test/Transforms/LoopVectorize/RISCV/dead-ops-cost.ll

Lines changed: 13 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -93,8 +93,7 @@ define i8 @dead_live_out_due_to_scalar_epilogue_required(ptr %src, ptr %dst) {
9393
; CHECK-SAME: ptr [[SRC:%.*]], ptr [[DST:%.*]]) #[[ATTR0]] {
9494
; CHECK-NEXT: [[ENTRY:.*]]:
9595
; CHECK-NEXT: [[TMP0:%.*]] = call i32 @llvm.vscale.i32()
96-
; CHECK-NEXT: [[TMP1:%.*]] = mul nuw i32 [[TMP0]], 4
97-
; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.umax.i32(i32 6, i32 [[TMP1]])
96+
; CHECK-NEXT: [[TMP2:%.*]] = mul nuw i32 [[TMP0]], 8
9897
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i32 252, [[TMP2]]
9998
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]]
10099
; CHECK: [[VECTOR_MEMCHECK]]:
@@ -106,29 +105,29 @@ define i8 @dead_live_out_due_to_scalar_epilogue_required(ptr %src, ptr %dst) {
106105
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
107106
; CHECK: [[VECTOR_PH]]:
108107
; CHECK-NEXT: [[TMP3:%.*]] = call i32 @llvm.vscale.i32()
109-
; CHECK-NEXT: [[TMP4:%.*]] = mul nuw i32 [[TMP3]], 4
108+
; CHECK-NEXT: [[TMP4:%.*]] = mul nuw i32 [[TMP3]], 8
110109
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 252, [[TMP4]]
111110
; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i32 [[N_MOD_VF]], 0
112111
; CHECK-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 [[N_MOD_VF]]
113112
; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 252, [[TMP6]]
114113
; CHECK-NEXT: [[TMP7:%.*]] = call i32 @llvm.vscale.i32()
115-
; CHECK-NEXT: [[TMP8:%.*]] = mul nuw i32 [[TMP7]], 4
114+
; CHECK-NEXT: [[TMP8:%.*]] = mul nuw i32 [[TMP7]], 8
116115
; CHECK-NEXT: [[IND_END:%.*]] = mul i32 [[N_VEC]], 4
117-
; CHECK-NEXT: [[TMP9:%.*]] = call <vscale x 4 x i32> @llvm.stepvector.nxv4i32()
118-
; CHECK-NEXT: [[TMP11:%.*]] = mul <vscale x 4 x i32> [[TMP9]], splat (i32 4)
119-
; CHECK-NEXT: [[INDUCTION:%.*]] = add <vscale x 4 x i32> zeroinitializer, [[TMP11]]
116+
; CHECK-NEXT: [[TMP10:%.*]] = call <vscale x 8 x i32> @llvm.stepvector.nxv8i32()
117+
; CHECK-NEXT: [[TMP11:%.*]] = mul <vscale x 8 x i32> [[TMP10]], splat (i32 4)
118+
; CHECK-NEXT: [[INDUCTION:%.*]] = add <vscale x 8 x i32> zeroinitializer, [[TMP11]]
120119
; CHECK-NEXT: [[TMP14:%.*]] = mul i32 4, [[TMP8]]
121-
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 4 x i32> poison, i32 [[TMP14]], i64 0
122-
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <vscale x 4 x i32> [[DOTSPLATINSERT]], <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer
120+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[TMP14]], i64 0
121+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
123122
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
124123
; CHECK: [[VECTOR_BODY]]:
125124
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
126-
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i32> [ [[INDUCTION]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
127-
; CHECK-NEXT: [[TMP15:%.*]] = sext <vscale x 4 x i32> [[VEC_IND]] to <vscale x 4 x i64>
128-
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[DST]], <vscale x 4 x i64> [[TMP15]]
129-
; CHECK-NEXT: call void @llvm.masked.scatter.nxv4i8.nxv4p0(<vscale x 4 x i8> zeroinitializer, <vscale x 4 x ptr> [[TMP16]], i32 1, <vscale x 4 x i1> splat (i1 true)), !alias.scope [[META5:![0-9]+]], !noalias [[META8:![0-9]+]]
125+
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 8 x i32> [ [[INDUCTION]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
126+
; CHECK-NEXT: [[TMP12:%.*]] = sext <vscale x 8 x i32> [[VEC_IND]] to <vscale x 8 x i64>
127+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[DST]], <vscale x 8 x i64> [[TMP12]]
128+
; CHECK-NEXT: call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> zeroinitializer, <vscale x 8 x ptr> [[TMP13]], i32 1, <vscale x 8 x i1> splat (i1 true)), !alias.scope [[META5:![0-9]+]], !noalias [[META8:![0-9]+]]
130129
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], [[TMP8]]
131-
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i32> [[VEC_IND]], [[DOTSPLAT]]
130+
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 8 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
132131
; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
133132
; CHECK-NEXT: br i1 [[TMP17]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
134133
; CHECK: [[MIDDLE_BLOCK]]:

‎llvm/test/Transforms/LoopVectorize/RISCV/pr88802.ll

Lines changed: 13 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -34,22 +34,22 @@ define void @test(ptr %p, i64 %a, i8 %b) {
3434
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 2 x i32> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[FOR_COND]] ]
3535
; CHECK-NEXT: [[AVL:%.*]] = phi i32 [ 9, [[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], [[FOR_COND]] ]
3636
; CHECK-NEXT: [[TMP11:%.*]] = call i32 @llvm.experimental.get.vector.length.i32(i32 [[AVL]], i32 2, i1 true)
37-
; CHECK-NEXT: [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <vscale x 2 x i32> poison, i32 [[TMP11]], i64 0
38-
; CHECK-NEXT: [[BROADCAST_SPLAT6:%.*]] = shufflevector <vscale x 2 x i32> [[BROADCAST_SPLATINSERT5]], <vscale x 2 x i32> poison, <vscale x 2 x i32> zeroinitializer
39-
; CHECK-NEXT: [[TMP20:%.*]] = mul i32 1, [[TMP11]]
40-
; CHECK-NEXT: [[BROADCAST_SPLATINSERT7:%.*]] = insertelement <vscale x 2 x i32> poison, i32 [[TMP20]], i64 0
37+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT7:%.*]] = insertelement <vscale x 2 x i32> poison, i32 [[TMP11]], i64 0
4138
; CHECK-NEXT: [[BROADCAST_SPLAT8:%.*]] = shufflevector <vscale x 2 x i32> [[BROADCAST_SPLATINSERT7]], <vscale x 2 x i32> poison, <vscale x 2 x i32> zeroinitializer
42-
; CHECK-NEXT: [[TMP19:%.*]] = call <vscale x 2 x i32> @llvm.stepvector.nxv2i32()
43-
; CHECK-NEXT: [[TMP13:%.*]] = icmp ult <vscale x 2 x i32> [[TMP19]], [[BROADCAST_SPLAT6]]
44-
; CHECK-NEXT: [[TMP14:%.*]] = icmp sge <vscale x 2 x i32> [[VEC_IND]], splat (i32 2)
45-
; CHECK-NEXT: [[TMP15:%.*]] = select <vscale x 2 x i1> [[TMP13]], <vscale x 2 x i1> [[TMP14]], <vscale x 2 x i1> zeroinitializer
46-
; CHECK-NEXT: [[PREDPHI:%.*]] = select <vscale x 2 x i1> [[TMP15]], <vscale x 2 x i32> [[TMP7]], <vscale x 2 x i32> [[TMP8]]
47-
; CHECK-NEXT: [[TMP16:%.*]] = shl <vscale x 2 x i32> [[PREDPHI]], splat (i32 8)
48-
; CHECK-NEXT: [[TMP17:%.*]] = trunc <vscale x 2 x i32> [[TMP16]] to <vscale x 2 x i8>
49-
; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i8.nxv2p0(<vscale x 2 x i8> [[TMP17]], <vscale x 2 x ptr> align 1 [[BROADCAST_SPLAT4]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP11]])
39+
; CHECK-NEXT: [[TMP20:%.*]] = mul i32 1, [[TMP11]]
40+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <vscale x 2 x i32> poison, i32 [[TMP20]], i64 0
41+
; CHECK-NEXT: [[BROADCAST_SPLAT6:%.*]] = shufflevector <vscale x 2 x i32> [[BROADCAST_SPLATINSERT5]], <vscale x 2 x i32> poison, <vscale x 2 x i32> zeroinitializer
42+
; CHECK-NEXT: [[TMP13:%.*]] = call <vscale x 2 x i32> @llvm.stepvector.nxv2i32()
43+
; CHECK-NEXT: [[TMP14:%.*]] = icmp ult <vscale x 2 x i32> [[TMP13]], [[BROADCAST_SPLAT8]]
44+
; CHECK-NEXT: [[TMP15:%.*]] = icmp sge <vscale x 2 x i32> [[VEC_IND]], splat (i32 2)
45+
; CHECK-NEXT: [[TMP16:%.*]] = select <vscale x 2 x i1> [[TMP14]], <vscale x 2 x i1> [[TMP15]], <vscale x 2 x i1> zeroinitializer
46+
; CHECK-NEXT: [[PREDPHI:%.*]] = select <vscale x 2 x i1> [[TMP16]], <vscale x 2 x i32> [[TMP7]], <vscale x 2 x i32> [[TMP8]]
47+
; CHECK-NEXT: [[TMP17:%.*]] = shl <vscale x 2 x i32> [[PREDPHI]], splat (i32 8)
48+
; CHECK-NEXT: [[TMP19:%.*]] = trunc <vscale x 2 x i32> [[TMP17]] to <vscale x 2 x i8>
49+
; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i8.nxv2p0(<vscale x 2 x i8> [[TMP19]], <vscale x 2 x ptr> align 1 [[BROADCAST_SPLAT4]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP11]])
5050
; CHECK-NEXT: [[INDEX_EVL_NEXT]] = add nuw i32 [[TMP11]], [[EVL_BASED_IV]]
5151
; CHECK-NEXT: [[AVL_NEXT]] = sub nuw i32 [[AVL]], [[TMP11]]
52-
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT8]]
52+
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT6]]
5353
; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i32 [[INDEX_EVL_NEXT]], 9
5454
; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_COND]], !llvm.loop [[LOOP0:![0-9]+]]
5555
; CHECK: middle.block:

‎llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-gather-scatter.ll

Lines changed: 12 additions & 87 deletions
Original file line numberDiff line numberDiff line change
@@ -10,110 +10,35 @@
1010
define void @gather_scatter(ptr noalias %in, ptr noalias %out, ptr noalias %index, i64 %n) {
1111
; IF-EVL-LABEL: @gather_scatter(
1212
; IF-EVL-NEXT: entry:
13-
; IF-EVL-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[ENTRY:%.*]]
14-
; IF-EVL: vector.ph:
15-
; IF-EVL-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
16-
; IF-EVL-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP4]], 2
17-
; IF-EVL-NEXT: [[TMP6:%.*]] = sub i64 [[TMP5]], 1
18-
; IF-EVL-NEXT: [[N_RND_UP:%.*]] = add i64 [[N:%.*]], [[TMP6]]
19-
; IF-EVL-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N_RND_UP]], [[TMP5]]
20-
; IF-EVL-NEXT: [[N_VEC:%.*]] = sub i64 [[N_RND_UP]], [[N_MOD_VF]]
21-
; IF-EVL-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64()
22-
; IF-EVL-NEXT: [[TMP8:%.*]] = mul nuw i64 [[TMP7]], 2
23-
; IF-EVL-NEXT: [[TMP9:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
24-
; IF-EVL-NEXT: [[TMP10:%.*]] = mul <vscale x 2 x i64> [[TMP9]], splat (i64 1)
25-
; IF-EVL-NEXT: [[INDUCTION:%.*]] = add <vscale x 2 x i64> zeroinitializer, [[TMP10]]
26-
; IF-EVL-NEXT: br label [[FOR_BODY:%.*]]
27-
; IF-EVL: vector.body:
28-
; IF-EVL-NEXT: [[EVL_BASED_IV:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[INDEX_EVL_NEXT:%.*]], [[FOR_BODY]] ]
29-
; IF-EVL-NEXT: [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[INDUCTION]], [[ENTRY]] ], [ [[VEC_IND_NEXT:%.*]], [[FOR_BODY]] ]
30-
; IF-EVL-NEXT: [[AVL:%.*]] = phi i64 [ [[N]], [[ENTRY]] ], [ [[AVL_NEXT:%.*]], [[FOR_BODY]] ]
31-
; IF-EVL-NEXT: [[TMP11:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 2, i1 true)
32-
; IF-EVL-NEXT: [[TMP12:%.*]] = zext i32 [[TMP11]] to i64
33-
; IF-EVL-NEXT: [[TMP13:%.*]] = mul i64 1, [[TMP12]]
34-
; IF-EVL-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP13]], i64 0
35-
; IF-EVL-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
36-
; IF-EVL-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[INDEX:%.*]], <vscale x 2 x i64> [[VEC_IND]]
37-
; IF-EVL-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 2 x i64> @llvm.vp.gather.nxv2i64.nxv2p0(<vscale x 2 x ptr> align 8 [[TMP14]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP11]])
38-
; IF-EVL-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, ptr [[IN:%.*]], <vscale x 2 x i64> [[WIDE_MASKED_GATHER]]
39-
; IF-EVL-NEXT: [[WIDE_MASKED_GATHER2:%.*]] = call <vscale x 2 x float> @llvm.vp.gather.nxv2f32.nxv2p0(<vscale x 2 x ptr> align 4 [[TMP15]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP11]])
40-
; IF-EVL-NEXT: [[TMP16:%.*]] = getelementptr inbounds float, ptr [[OUT:%.*]], <vscale x 2 x i64> [[WIDE_MASKED_GATHER]]
41-
; IF-EVL-NEXT: call void @llvm.vp.scatter.nxv2f32.nxv2p0(<vscale x 2 x float> [[WIDE_MASKED_GATHER2]], <vscale x 2 x ptr> align 4 [[TMP16]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP11]])
42-
; IF-EVL-NEXT: [[TMP17:%.*]] = zext i32 [[TMP11]] to i64
43-
; IF-EVL-NEXT: [[INDEX_EVL_NEXT]] = add i64 [[TMP17]], [[EVL_BASED_IV]]
44-
; IF-EVL-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP17]]
45-
; IF-EVL-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
46-
; IF-EVL-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_EVL_NEXT]], [[N]]
47-
; IF-EVL-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
48-
; IF-EVL: middle.block:
49-
; IF-EVL-NEXT: br label [[FOR_END:%.*]]
50-
; IF-EVL: scalar.ph:
51-
; IF-EVL-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 0, [[ENTRY1:%.*]] ]
5213
; IF-EVL-NEXT: br label [[FOR_BODY1:%.*]]
5314
; IF-EVL: for.body:
54-
; IF-EVL-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ 0, [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT1:%.*]], [[FOR_BODY1]] ]
55-
; IF-EVL-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, ptr [[INDEX]], i64 [[INDVARS_IV1]]
15+
; IF-EVL-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ 0, [[SCALAR_PH:%.*]] ], [ [[INDVARS_IV_NEXT1:%.*]], [[FOR_BODY1]] ]
16+
; IF-EVL-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, ptr [[INDEX:%.*]], i64 [[INDVARS_IV1]]
5617
; IF-EVL-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX3]], align 8
57-
; IF-EVL-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, ptr [[IN]], i64 [[TMP0]]
18+
; IF-EVL-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, ptr [[IN:%.*]], i64 [[TMP0]]
5819
; IF-EVL-NEXT: [[TMP1:%.*]] = load float, ptr [[ARRAYIDX5]], align 4
59-
; IF-EVL-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[OUT]], i64 [[TMP0]]
20+
; IF-EVL-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[OUT:%.*]], i64 [[TMP0]]
6021
; IF-EVL-NEXT: store float [[TMP1]], ptr [[ARRAYIDX7]], align 4
6122
; IF-EVL-NEXT: [[INDVARS_IV_NEXT1]] = add nuw nsw i64 [[INDVARS_IV1]], 1
62-
; IF-EVL-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT1]], [[N]]
63-
; IF-EVL-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY1]], !llvm.loop [[LOOP4:![0-9]+]]
23+
; IF-EVL-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT1]], [[N:%.*]]
24+
; IF-EVL-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END:%.*]], label [[FOR_BODY1]]
6425
; IF-EVL: for.end:
6526
; IF-EVL-NEXT: ret void
6627
;
6728
; NO-VP-LABEL: @gather_scatter(
6829
; NO-VP-NEXT: entry:
69-
; NO-VP-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64()
70-
; NO-VP-NEXT: [[TMP14:%.*]] = mul nuw i64 [[TMP13]], 2
71-
; NO-VP-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP14]]
72-
; NO-VP-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[ENTRY:%.*]]
73-
; NO-VP: vector.ph:
74-
; NO-VP-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
75-
; NO-VP-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 2
76-
; NO-VP-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
77-
; NO-VP-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
78-
; NO-VP-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
79-
; NO-VP-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP4]], 2
80-
; NO-VP-NEXT: [[TMP6:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
81-
; NO-VP-NEXT: [[TMP7:%.*]] = mul <vscale x 2 x i64> [[TMP6]], splat (i64 1)
82-
; NO-VP-NEXT: [[INDUCTION:%.*]] = add <vscale x 2 x i64> zeroinitializer, [[TMP7]]
83-
; NO-VP-NEXT: [[TMP8:%.*]] = mul i64 1, [[TMP5]]
84-
; NO-VP-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP8]], i64 0
85-
; NO-VP-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
86-
; NO-VP-NEXT: br label [[FOR_BODY:%.*]]
87-
; NO-VP: vector.body:
88-
; NO-VP-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
89-
; NO-VP-NEXT: [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[INDUCTION]], [[ENTRY]] ], [ [[VEC_IND_NEXT:%.*]], [[FOR_BODY]] ]
90-
; NO-VP-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[INDEX:%.*]], <vscale x 2 x i64> [[VEC_IND]]
91-
; NO-VP-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 2 x i64> @llvm.masked.gather.nxv2i64.nxv2p0(<vscale x 2 x ptr> [[TMP9]], i32 8, <vscale x 2 x i1> splat (i1 true), <vscale x 2 x i64> poison)
92-
; NO-VP-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[IN:%.*]], <vscale x 2 x i64> [[WIDE_MASKED_GATHER]]
93-
; NO-VP-NEXT: [[WIDE_MASKED_GATHER2:%.*]] = call <vscale x 2 x float> @llvm.masked.gather.nxv2f32.nxv2p0(<vscale x 2 x ptr> [[TMP10]], i32 4, <vscale x 2 x i1> splat (i1 true), <vscale x 2 x float> poison)
94-
; NO-VP-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, ptr [[OUT:%.*]], <vscale x 2 x i64> [[WIDE_MASKED_GATHER]]
95-
; NO-VP-NEXT: call void @llvm.masked.scatter.nxv2f32.nxv2p0(<vscale x 2 x float> [[WIDE_MASKED_GATHER2]], <vscale x 2 x ptr> [[TMP11]], i32 4, <vscale x 2 x i1> splat (i1 true))
96-
; NO-VP-NEXT: [[INDVARS_IV_NEXT]] = add nuw i64 [[INDVARS_IV]], [[TMP5]]
97-
; NO-VP-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
98-
; NO-VP-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[N_VEC]]
99-
; NO-VP-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
100-
; NO-VP: middle.block:
101-
; NO-VP-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
102-
; NO-VP-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
103-
; NO-VP: scalar.ph:
104-
; NO-VP-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY1:%.*]] ]
10530
; NO-VP-NEXT: br label [[FOR_BODY1:%.*]]
10631
; NO-VP: for.body:
107-
; NO-VP-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT1:%.*]], [[FOR_BODY1]] ]
108-
; NO-VP-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, ptr [[INDEX]], i64 [[INDVARS_IV1]]
32+
; NO-VP-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT1:%.*]], [[FOR_BODY1]] ]
33+
; NO-VP-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, ptr [[INDEX:%.*]], i64 [[INDVARS_IV1]]
10934
; NO-VP-NEXT: [[TMP0:%.*]] = load i64, ptr [[ARRAYIDX3]], align 8
110-
; NO-VP-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, ptr [[IN]], i64 [[TMP0]]
35+
; NO-VP-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, ptr [[IN:%.*]], i64 [[TMP0]]
11136
; NO-VP-NEXT: [[TMP1:%.*]] = load float, ptr [[ARRAYIDX5]], align 4
112-
; NO-VP-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[OUT]], i64 [[TMP0]]
37+
; NO-VP-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[OUT:%.*]], i64 [[TMP0]]
11338
; NO-VP-NEXT: store float [[TMP1]], ptr [[ARRAYIDX7]], align 4
11439
; NO-VP-NEXT: [[INDVARS_IV_NEXT1]] = add nuw nsw i64 [[INDVARS_IV1]], 1
115-
; NO-VP-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT1]], [[N]]
116-
; NO-VP-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY1]], !llvm.loop [[LOOP3:![0-9]+]]
40+
; NO-VP-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT1]], [[N:%.*]]
41+
; NO-VP-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END:%.*]], label [[FOR_BODY1]]
11742
; NO-VP: for.end:
11843
; NO-VP-NEXT: ret void
11944
;

0 commit comments

Comments
 (0)