Skip to content

Commit 47d831f

Browse files
arsenmarsenm
authored
TTI: Check legalization cost of min/max ISD nodes (#100514)
Instead of counting the cost of the assumed expansion. The AMDGPU costs for the i64 case look too high to me. Preserve default expansion logic
1 parent ea1dee7 commit 47d831f

File tree

9 files changed

+226
-268
lines changed

9 files changed

+226
-268
lines changed

‎llvm/include/llvm/CodeGen/BasicTTIImpl.h‎

Lines changed: 25 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -2135,21 +2135,17 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
21352135
return Cost;
21362136
}
21372137
case Intrinsic::smax:
2138+
ISD = ISD::SMAX;
2139+
break;
21382140
case Intrinsic::smin:
2141+
ISD = ISD::SMIN;
2142+
break;
21392143
case Intrinsic::umax:
2140-
case Intrinsic::umin: {
2141-
// minmax(X,Y) = select(icmp(X,Y),X,Y)
2142-
Type *CondTy = RetTy->getWithNewBitWidth(1);
2143-
bool IsUnsigned = IID == Intrinsic::umax || IID == Intrinsic::umin;
2144-
CmpInst::Predicate Pred =
2145-
IsUnsigned ? CmpInst::ICMP_UGT : CmpInst::ICMP_SGT;
2146-
InstructionCost Cost = 0;
2147-
Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, RetTy, CondTy,
2148-
Pred, CostKind);
2149-
Cost += thisT()->getCmpSelInstrCost(BinaryOperator::Select, RetTy, CondTy,
2150-
Pred, CostKind);
2151-
return Cost;
2152-
}
2144+
ISD = ISD::UMAX;
2145+
break;
2146+
case Intrinsic::umin:
2147+
ISD = ISD::UMIN;
2148+
break;
21532149
case Intrinsic::sadd_sat:
21542150
ISD = ISD::SADDSAT;
21552151
break;
@@ -2345,6 +2341,22 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
23452341
return thisT()->getIntrinsicInstrCost(FMulAttrs, CostKind) +
23462342
thisT()->getIntrinsicInstrCost(FAddAttrs, CostKind);
23472343
}
2344+
case Intrinsic::smin:
2345+
case Intrinsic::smax:
2346+
case Intrinsic::umin:
2347+
case Intrinsic::umax: {
2348+
// minmax(X,Y) = select(icmp(X,Y),X,Y)
2349+
Type *CondTy = RetTy->getWithNewBitWidth(1);
2350+
bool IsUnsigned = IID == Intrinsic::umax || IID == Intrinsic::umin;
2351+
CmpInst::Predicate Pred =
2352+
IsUnsigned ? CmpInst::ICMP_UGT : CmpInst::ICMP_SGT;
2353+
InstructionCost Cost = 0;
2354+
Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, RetTy, CondTy,
2355+
Pred, CostKind);
2356+
Cost += thisT()->getCmpSelInstrCost(BinaryOperator::Select, RetTy, CondTy,
2357+
Pred, CostKind);
2358+
return Cost;
2359+
}
23482360
case Intrinsic::sadd_sat:
23492361
case Intrinsic::ssub_sat: {
23502362
// Assume a default expansion.

‎llvm/test/Analysis/CostModel/AMDGPU/arith-sminmax.ll‎

Lines changed: 58 additions & 106 deletions
Large diffs are not rendered by default.

‎llvm/test/Analysis/CostModel/AMDGPU/arith-uminmax.ll‎

Lines changed: 58 additions & 106 deletions
Large diffs are not rendered by default.

‎llvm/test/Analysis/CostModel/ARM/cmps.ll‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -227,7 +227,7 @@ define void @minmax() {
227227
; CHECK-V8R-RECIP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %c3 = icmp slt i32 undef, undef
228228
; CHECK-V8R-RECIP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %s3 = select i1 %c3, i32 undef, i32 undef
229229
; CHECK-V8R-RECIP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %c4 = icmp slt <4 x i32> undef, undef
230-
; CHECK-V8R-RECIP-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %s4 = select <4 x i1> %c4, <4 x i32> undef, <4 x i32> undef
230+
; CHECK-V8R-RECIP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %s4 = select <4 x i1> %c4, <4 x i32> undef, <4 x i32> undef
231231
; CHECK-V8R-RECIP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %c5 = icmp slt ptr undef, undef
232232
; CHECK-V8R-RECIP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %s5 = select i1 %c5, ptr undef, ptr undef
233233
; CHECK-V8R-RECIP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %c6 = icmp slt <4 x ptr> undef, undef
@@ -287,7 +287,7 @@ define void @minmax() {
287287
; CHECK-V8R-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %c3 = icmp slt i32 undef, undef
288288
; CHECK-V8R-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %s3 = select i1 %c3, i32 undef, i32 undef
289289
; CHECK-V8R-SIZE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %c4 = icmp slt <4 x i32> undef, undef
290-
; CHECK-V8R-SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %s4 = select <4 x i1> %c4, <4 x i32> undef, <4 x i32> undef
290+
; CHECK-V8R-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %s4 = select <4 x i1> %c4, <4 x i32> undef, <4 x i32> undef
291291
; CHECK-V8R-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %c5 = icmp slt ptr undef, undef
292292
; CHECK-V8R-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %s5 = select i1 %c5, ptr undef, ptr undef
293293
; CHECK-V8R-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %c6 = icmp slt <4 x ptr> undef, undef

‎llvm/test/Analysis/CostModel/ARM/reduce-smax.ll‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -48,7 +48,7 @@ define i32 @reduce_i32(i32 %arg) {
4848
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
4949
;
5050
; NEON-LABEL: 'reduce_i32'
51-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i32 @llvm.vector.reduce.smax.v2i32(<2 x i32> undef)
51+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i32 @llvm.vector.reduce.smax.v2i32(<2 x i32> undef)
5252
; NEON-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> undef)
5353
; NEON-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8 = call i32 @llvm.vector.reduce.smax.v8i32(<8 x i32> undef)
5454
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16 = call i32 @llvm.vector.reduce.smax.v16i32(<16 x i32> undef)
@@ -82,8 +82,8 @@ define i32 @reduce_i16(i32 %arg) {
8282
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
8383
;
8484
; NEON-LABEL: 'reduce_i16'
85-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i16 @llvm.vector.reduce.smax.v2i16(<2 x i16> undef)
86-
; NEON-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V4 = call i16 @llvm.vector.reduce.smax.v4i16(<4 x i16> undef)
85+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i16 @llvm.vector.reduce.smax.v2i16(<2 x i16> undef)
86+
; NEON-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4 = call i16 @llvm.vector.reduce.smax.v4i16(<4 x i16> undef)
8787
; NEON-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> undef)
8888
; NEON-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16 = call i16 @llvm.vector.reduce.smax.v16i16(<16 x i16> undef)
8989
; NEON-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32 = call i16 @llvm.vector.reduce.smax.v32i16(<32 x i16> undef)
@@ -120,9 +120,9 @@ define i32 @reduce_i8(i32 %arg) {
120120
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
121121
;
122122
; NEON-LABEL: 'reduce_i8'
123-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i8 @llvm.vector.reduce.smax.v2i8(<2 x i8> undef)
124-
; NEON-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V4 = call i8 @llvm.vector.reduce.smax.v4i8(<4 x i8> undef)
125-
; NEON-NEXT: Cost Model: Found an estimated cost of 153 for instruction: %V8 = call i8 @llvm.vector.reduce.smax.v8i8(<8 x i8> undef)
123+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i8 @llvm.vector.reduce.smax.v2i8(<2 x i8> undef)
124+
; NEON-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4 = call i8 @llvm.vector.reduce.smax.v4i8(<4 x i8> undef)
125+
; NEON-NEXT: Cost Model: Found an estimated cost of 150 for instruction: %V8 = call i8 @llvm.vector.reduce.smax.v8i8(<8 x i8> undef)
126126
; NEON-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16 = call i8 @llvm.vector.reduce.smax.v16i8(<16 x i8> undef)
127127
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32 = call i8 @llvm.vector.reduce.smax.v32i8(<32 x i8> undef)
128128
; NEON-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64 = call i8 @llvm.vector.reduce.smax.v64i8(<64 x i8> undef)

‎llvm/test/Analysis/CostModel/ARM/reduce-smin.ll‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -48,7 +48,7 @@ define i32 @reduce_i32(i32 %arg) {
4848
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
4949
;
5050
; NEON-LABEL: 'reduce_i32'
51-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i32 @llvm.vector.reduce.smin.v2i32(<2 x i32> undef)
51+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i32 @llvm.vector.reduce.smin.v2i32(<2 x i32> undef)
5252
; NEON-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call i32 @llvm.vector.reduce.smin.v4i32(<4 x i32> undef)
5353
; NEON-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8 = call i32 @llvm.vector.reduce.smin.v8i32(<8 x i32> undef)
5454
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16 = call i32 @llvm.vector.reduce.smin.v16i32(<16 x i32> undef)
@@ -82,8 +82,8 @@ define i32 @reduce_i16(i32 %arg) {
8282
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
8383
;
8484
; NEON-LABEL: 'reduce_i16'
85-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i16 @llvm.vector.reduce.smin.v2i16(<2 x i16> undef)
86-
; NEON-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V4 = call i16 @llvm.vector.reduce.smin.v4i16(<4 x i16> undef)
85+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i16 @llvm.vector.reduce.smin.v2i16(<2 x i16> undef)
86+
; NEON-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4 = call i16 @llvm.vector.reduce.smin.v4i16(<4 x i16> undef)
8787
; NEON-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call i16 @llvm.vector.reduce.smin.v8i16(<8 x i16> undef)
8888
; NEON-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16 = call i16 @llvm.vector.reduce.smin.v16i16(<16 x i16> undef)
8989
; NEON-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32 = call i16 @llvm.vector.reduce.smin.v32i16(<32 x i16> undef)
@@ -120,9 +120,9 @@ define i32 @reduce_i8(i32 %arg) {
120120
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
121121
;
122122
; NEON-LABEL: 'reduce_i8'
123-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i8 @llvm.vector.reduce.smin.v2i8(<2 x i8> undef)
124-
; NEON-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V4 = call i8 @llvm.vector.reduce.smin.v4i8(<4 x i8> undef)
125-
; NEON-NEXT: Cost Model: Found an estimated cost of 153 for instruction: %V8 = call i8 @llvm.vector.reduce.smin.v8i8(<8 x i8> undef)
123+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i8 @llvm.vector.reduce.smin.v2i8(<2 x i8> undef)
124+
; NEON-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4 = call i8 @llvm.vector.reduce.smin.v4i8(<4 x i8> undef)
125+
; NEON-NEXT: Cost Model: Found an estimated cost of 150 for instruction: %V8 = call i8 @llvm.vector.reduce.smin.v8i8(<8 x i8> undef)
126126
; NEON-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16 = call i8 @llvm.vector.reduce.smin.v16i8(<16 x i8> undef)
127127
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32 = call i8 @llvm.vector.reduce.smin.v32i8(<32 x i8> undef)
128128
; NEON-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64 = call i8 @llvm.vector.reduce.smin.v64i8(<64 x i8> undef)

‎llvm/test/Analysis/CostModel/ARM/reduce-umax.ll‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -48,7 +48,7 @@ define i32 @reduce_i32(i32 %arg) {
4848
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
4949
;
5050
; NEON-LABEL: 'reduce_i32'
51-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i32 @llvm.vector.reduce.umax.v2i32(<2 x i32> undef)
51+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i32 @llvm.vector.reduce.umax.v2i32(<2 x i32> undef)
5252
; NEON-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call i32 @llvm.vector.reduce.umax.v4i32(<4 x i32> undef)
5353
; NEON-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8 = call i32 @llvm.vector.reduce.umax.v8i32(<8 x i32> undef)
5454
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16 = call i32 @llvm.vector.reduce.umax.v16i32(<16 x i32> undef)
@@ -82,8 +82,8 @@ define i32 @reduce_i16(i32 %arg) {
8282
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
8383
;
8484
; NEON-LABEL: 'reduce_i16'
85-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i16 @llvm.vector.reduce.umax.v2i16(<2 x i16> undef)
86-
; NEON-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V4 = call i16 @llvm.vector.reduce.umax.v4i16(<4 x i16> undef)
85+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i16 @llvm.vector.reduce.umax.v2i16(<2 x i16> undef)
86+
; NEON-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4 = call i16 @llvm.vector.reduce.umax.v4i16(<4 x i16> undef)
8787
; NEON-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call i16 @llvm.vector.reduce.umax.v8i16(<8 x i16> undef)
8888
; NEON-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16 = call i16 @llvm.vector.reduce.umax.v16i16(<16 x i16> undef)
8989
; NEON-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32 = call i16 @llvm.vector.reduce.umax.v32i16(<32 x i16> undef)
@@ -120,9 +120,9 @@ define i32 @reduce_i8(i32 %arg) {
120120
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
121121
;
122122
; NEON-LABEL: 'reduce_i8'
123-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i8 @llvm.vector.reduce.umax.v2i8(<2 x i8> undef)
124-
; NEON-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V4 = call i8 @llvm.vector.reduce.umax.v4i8(<4 x i8> undef)
125-
; NEON-NEXT: Cost Model: Found an estimated cost of 153 for instruction: %V8 = call i8 @llvm.vector.reduce.umax.v8i8(<8 x i8> undef)
123+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i8 @llvm.vector.reduce.umax.v2i8(<2 x i8> undef)
124+
; NEON-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4 = call i8 @llvm.vector.reduce.umax.v4i8(<4 x i8> undef)
125+
; NEON-NEXT: Cost Model: Found an estimated cost of 150 for instruction: %V8 = call i8 @llvm.vector.reduce.umax.v8i8(<8 x i8> undef)
126126
; NEON-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16 = call i8 @llvm.vector.reduce.umax.v16i8(<16 x i8> undef)
127127
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32 = call i8 @llvm.vector.reduce.umax.v32i8(<32 x i8> undef)
128128
; NEON-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64 = call i8 @llvm.vector.reduce.umax.v64i8(<64 x i8> undef)

‎llvm/test/Analysis/CostModel/ARM/reduce-umin.ll‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -48,7 +48,7 @@ define i32 @reduce_i32(i32 %arg) {
4848
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
4949
;
5050
; NEON-LABEL: 'reduce_i32'
51-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i32 @llvm.vector.reduce.umin.v2i32(<2 x i32> undef)
51+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i32 @llvm.vector.reduce.umin.v2i32(<2 x i32> undef)
5252
; NEON-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call i32 @llvm.vector.reduce.umin.v4i32(<4 x i32> undef)
5353
; NEON-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8 = call i32 @llvm.vector.reduce.umin.v8i32(<8 x i32> undef)
5454
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16 = call i32 @llvm.vector.reduce.umin.v16i32(<16 x i32> undef)
@@ -82,8 +82,8 @@ define i32 @reduce_i16(i32 %arg) {
8282
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
8383
;
8484
; NEON-LABEL: 'reduce_i16'
85-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i16 @llvm.vector.reduce.umin.v2i16(<2 x i16> undef)
86-
; NEON-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V4 = call i16 @llvm.vector.reduce.umin.v4i16(<4 x i16> undef)
85+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i16 @llvm.vector.reduce.umin.v2i16(<2 x i16> undef)
86+
; NEON-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4 = call i16 @llvm.vector.reduce.umin.v4i16(<4 x i16> undef)
8787
; NEON-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call i16 @llvm.vector.reduce.umin.v8i16(<8 x i16> undef)
8888
; NEON-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16 = call i16 @llvm.vector.reduce.umin.v16i16(<16 x i16> undef)
8989
; NEON-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32 = call i16 @llvm.vector.reduce.umin.v32i16(<32 x i16> undef)
@@ -120,9 +120,9 @@ define i32 @reduce_i8(i32 %arg) {
120120
; V8M-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
121121
;
122122
; NEON-LABEL: 'reduce_i8'
123-
; NEON-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2 = call i8 @llvm.vector.reduce.umin.v2i8(<2 x i8> undef)
124-
; NEON-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V4 = call i8 @llvm.vector.reduce.umin.v4i8(<4 x i8> undef)
125-
; NEON-NEXT: Cost Model: Found an estimated cost of 153 for instruction: %V8 = call i8 @llvm.vector.reduce.umin.v8i8(<8 x i8> undef)
123+
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2 = call i8 @llvm.vector.reduce.umin.v2i8(<2 x i8> undef)
124+
; NEON-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4 = call i8 @llvm.vector.reduce.umin.v4i8(<4 x i8> undef)
125+
; NEON-NEXT: Cost Model: Found an estimated cost of 150 for instruction: %V8 = call i8 @llvm.vector.reduce.umin.v8i8(<8 x i8> undef)
126126
; NEON-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16 = call i8 @llvm.vector.reduce.umin.v16i8(<16 x i8> undef)
127127
; NEON-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32 = call i8 @llvm.vector.reduce.umin.v32i8(<32 x i8> undef)
128128
; NEON-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64 = call i8 @llvm.vector.reduce.umin.v64i8(<64 x i8> undef)

0 commit comments

Comments
 (0)