xref: /llvm-project/llvm/test/Analysis/CostModel/ARM/arith-overflow.ll (revision 0bc640774880121e3e59d11011313d3ee8c52b01) 
10ef62fc2SSam Parker; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
24178e334SSimon Pilgrim; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -mtriple=thumbv8m.main < %s | FileCheck %s --check-prefix=V8M-RECIP
34178e334SSimon Pilgrim; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -mtriple=armv8a-linux-gnueabihf < %s | FileCheck %s --check-prefix=NEON-RECIP
44178e334SSimon Pilgrim; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -mtriple=armv8.1m.main -mattr=+mve < %s | FileCheck %s --check-prefix=MVE-RECIP
54178e334SSimon Pilgrim; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mtriple=thumbv8m.main < %s | FileCheck %s --check-prefix=V8M-SIZE
64178e334SSimon Pilgrim; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mtriple=armv8a-linux-gnueabihf < %s | FileCheck %s --check-prefix=NEON-SIZE
74178e334SSimon Pilgrim; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mtriple=armv8.1m.main -mattr=+mve < %s | FileCheck %s --check-prefix=MVE-SIZE
80ef62fc2SSam Parker
9309f1e4aSDavid Greentarget datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
10309f1e4aSDavid Green
110ef62fc2SSam Parkerdeclare {i64, i1}              @llvm.sadd.with.overflow.i64(i64, i64)
120ef62fc2SSam Parkerdeclare {<2 x i64>, <2 x i1>}  @llvm.sadd.with.overflow.v2i64(<2 x i64>, <2 x i64>)
130ef62fc2SSam Parkerdeclare {<4 x i64>, <4 x i1>}  @llvm.sadd.with.overflow.v4i64(<4 x i64>, <4 x i64>)
140ef62fc2SSam Parkerdeclare {<8 x i64>, <8 x i1>}  @llvm.sadd.with.overflow.v8i64(<8 x i64>, <8 x i64>)
150ef62fc2SSam Parker
160ef62fc2SSam Parkerdeclare {i32, i1}               @llvm.sadd.with.overflow.i32(i32, i32)
170ef62fc2SSam Parkerdeclare {<4 x i32>, <4 x i1>}   @llvm.sadd.with.overflow.v4i32(<4 x i32>, <4 x i32>)
180ef62fc2SSam Parkerdeclare {<8 x i32>, <8 x i1>}   @llvm.sadd.with.overflow.v8i32(<8 x i32>, <8 x i32>)
190ef62fc2SSam Parkerdeclare {<16 x i32>, <16 x i1>} @llvm.sadd.with.overflow.v16i32(<16 x i32>, <16 x i32>)
200ef62fc2SSam Parker
210ef62fc2SSam Parkerdeclare {i16, i1}               @llvm.sadd.with.overflow.i16(i16, i16)
220ef62fc2SSam Parkerdeclare {<8 x i16>,  <8 x i1>}  @llvm.sadd.with.overflow.v8i16(<8 x i16>, <8 x i16>)
230ef62fc2SSam Parkerdeclare {<16 x i16>, <16 x i1>} @llvm.sadd.with.overflow.v16i16(<16 x i16>, <16 x i16>)
240ef62fc2SSam Parkerdeclare {<32 x i16>, <32 x i1>} @llvm.sadd.with.overflow.v32i16(<32 x i16>, <32 x i16>)
250ef62fc2SSam Parker
260ef62fc2SSam Parkerdeclare {i8, i1}                @llvm.sadd.with.overflow.i8(i8, i8)
270ef62fc2SSam Parkerdeclare {<16 x i8>, <16 x i1>}  @llvm.sadd.with.overflow.v16i8(<16 x i8>, <16 x i8>)
280ef62fc2SSam Parkerdeclare {<32 x i8>, <32 x i1>}  @llvm.sadd.with.overflow.v32i8(<32 x i8>, <32 x i8>)
290ef62fc2SSam Parkerdeclare {<64 x i8>, <64 x i1>}  @llvm.sadd.with.overflow.v64i8(<64 x i8>, <64 x i8>)
300ef62fc2SSam Parker
310ef62fc2SSam Parkerdefine i32 @sadd(i32 %arg) {
3264cfb8a8SSam Parker; V8M-RECIP-LABEL: 'sadd'
33*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 undef, i64 undef)
34*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.sadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
35*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.sadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
36*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.sadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
37*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 undef, i32 undef)
38*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.sadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
39*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.sadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
40*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.sadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
41*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.sadd.with.overflow.i16(i16 undef, i16 undef)
42*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.sadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
43*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.sadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
44*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.sadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
45*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 undef, i8 undef)
46*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.sadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
47*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.sadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
48*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.sadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
492596da31SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
500ef62fc2SSam Parker;
5164cfb8a8SSam Parker; NEON-RECIP-LABEL: 'sadd'
52*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 undef, i64 undef)
5376534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.sadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
5476534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.sadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
5576534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.sadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
56*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 undef, i32 undef)
5776534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.sadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
5876534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.sadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
5976534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.sadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
60*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.sadd.with.overflow.i16(i16 undef, i16 undef)
6176534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.sadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
6276534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.sadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
6376534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.sadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
64*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 undef, i8 undef)
6576534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.sadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
6676534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.sadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
6776534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.sadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
68f2675ab4SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
690ef62fc2SSam Parker;
7064cfb8a8SSam Parker; MVE-RECIP-LABEL: 'sadd'
71*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 undef, i64 undef)
72255ad734SDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.sadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
7376534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 218 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.sadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
7476534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 434 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.sadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
75*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 undef, i32 undef)
7676534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.sadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
7776534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.sadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
7876534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 154 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.sadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
79*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.sadd.with.overflow.i16(i16 undef, i16 undef)
8076534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.sadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
8176534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 142 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.sadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
8276534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 284 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.sadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
83*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 undef, i8 undef)
8476534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.sadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
8576534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 272 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.sadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
8676534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 544 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.sadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
87f2675ab4SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
8864cfb8a8SSam Parker;
8964cfb8a8SSam Parker; V8M-SIZE-LABEL: 'sadd'
90*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 undef, i64 undef)
9176534829SCraig Topper; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.sadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
92*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.sadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
93*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.sadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
94*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 undef, i32 undef)
95*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.sadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
96*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.sadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
97*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.sadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
98*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.sadd.with.overflow.i16(i16 undef, i16 undef)
99*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.sadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
100*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.sadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
101*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.sadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
102*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 undef, i8 undef)
103*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.sadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
104*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.sadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
105*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.sadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
10664cfb8a8SSam Parker; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
10764cfb8a8SSam Parker;
10864cfb8a8SSam Parker; NEON-SIZE-LABEL: 'sadd'
10976534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 undef, i64 undef)
11076534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.sadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
11176534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.sadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
11276534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.sadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
113*0bc64077SMatt Arsenault; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 undef, i32 undef)
11476534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.sadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
11576534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.sadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
11676534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.sadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
117*0bc64077SMatt Arsenault; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.sadd.with.overflow.i16(i16 undef, i16 undef)
11876534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.sadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
11976534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.sadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
12076534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.sadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
121*0bc64077SMatt Arsenault; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 undef, i8 undef)
12276534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.sadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
12376534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.sadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
12476534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.sadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
12564cfb8a8SSam Parker; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
12664cfb8a8SSam Parker;
12764cfb8a8SSam Parker; MVE-SIZE-LABEL: 'sadd'
128*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 undef, i64 undef)
129255ad734SDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.sadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
13076534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 75 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.sadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
13176534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 147 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.sadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
132*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 undef, i32 undef)
13376534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.sadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
13476534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.sadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
13576534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 141 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.sadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
136*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.sadd.with.overflow.i16(i16 undef, i16 undef)
13776534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.sadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
13876534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.sadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
13976534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 270 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.sadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
140*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.sadd.with.overflow.i8(i8 undef, i8 undef)
14176534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.sadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
14276534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 264 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.sadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
14376534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 528 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.sadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
14464cfb8a8SSam Parker; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
1450ef62fc2SSam Parker;
1460ef62fc2SSam Parker  %I64 = call {i64, i1} @llvm.sadd.with.overflow.i64(i64 undef, i64 undef)
1470ef62fc2SSam Parker  %V2I64 = call {<2 x i64>, <2 x i1>} @llvm.sadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
1480ef62fc2SSam Parker  %V4I64 = call {<4 x i64>, <4 x i1>} @llvm.sadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
1490ef62fc2SSam Parker  %V8I64 = call {<8 x i64>, <8 x i1>} @llvm.sadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
1500ef62fc2SSam Parker
1510ef62fc2SSam Parker  %I32 = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 undef, i32 undef)
1520ef62fc2SSam Parker  %V4I32  = call {<4 x i32>, <4 x i1>}  @llvm.sadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
1530ef62fc2SSam Parker  %V8I32  = call {<8 x i32>, <8 x i1>}  @llvm.sadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
1540ef62fc2SSam Parker  %V16I32 = call {<16 x i32>, <16 x i1>} @llvm.sadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
1550ef62fc2SSam Parker
1560ef62fc2SSam Parker  %I16 = call {i16, i1} @llvm.sadd.with.overflow.i16(i16 undef, i16 undef)
1570ef62fc2SSam Parker  %V8I16  = call {<8 x i16>, <8 x i1>}  @llvm.sadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
1580ef62fc2SSam Parker  %V16I16 = call {<16 x i16>, <16 x i1>} @llvm.sadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
1590ef62fc2SSam Parker  %V32I16 = call {<32 x i16>, <32 x i1>} @llvm.sadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
1600ef62fc2SSam Parker
1610ef62fc2SSam Parker  %I8 = call {i8, i1} @llvm.sadd.with.overflow.i8(i8 undef, i8 undef)
1620ef62fc2SSam Parker  %V16I8 = call {<16 x i8>, <16 x i1>} @llvm.sadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
1630ef62fc2SSam Parker  %V32I8 = call {<32 x i8>, <32 x i1>} @llvm.sadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
1640ef62fc2SSam Parker  %V64I8 = call {<64 x i8>, <64 x i1>} @llvm.sadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
1650ef62fc2SSam Parker
1660ef62fc2SSam Parker  ret i32 undef
1670ef62fc2SSam Parker}
1680ef62fc2SSam Parker
1690ef62fc2SSam Parkerdeclare {i64, i1}              @llvm.uadd.with.overflow.i64(i64, i64)
1700ef62fc2SSam Parkerdeclare {<2 x i64>, <2 x i1>}  @llvm.uadd.with.overflow.v2i64(<2 x i64>, <2 x i64>)
1710ef62fc2SSam Parkerdeclare {<4 x i64>, <4 x i1>}  @llvm.uadd.with.overflow.v4i64(<4 x i64>, <4 x i64>)
1720ef62fc2SSam Parkerdeclare {<8 x i64>, <8 x i1>}  @llvm.uadd.with.overflow.v8i64(<8 x i64>, <8 x i64>)
1730ef62fc2SSam Parker
1740ef62fc2SSam Parkerdeclare {i32, i1}               @llvm.uadd.with.overflow.i32(i32, i32)
1750ef62fc2SSam Parkerdeclare {<4 x i32>, <4 x i1>}   @llvm.uadd.with.overflow.v4i32(<4 x i32>, <4 x i32>)
1760ef62fc2SSam Parkerdeclare {<8 x i32>, <8 x i1>}   @llvm.uadd.with.overflow.v8i32(<8 x i32>, <8 x i32>)
1770ef62fc2SSam Parkerdeclare {<16 x i32>, <16 x i1>} @llvm.uadd.with.overflow.v16i32(<16 x i32>, <16 x i32>)
1780ef62fc2SSam Parker
1790ef62fc2SSam Parkerdeclare {i16, i1}               @llvm.uadd.with.overflow.i16(i16, i16)
1800ef62fc2SSam Parkerdeclare {<8 x i16>,  <8 x i1>}  @llvm.uadd.with.overflow.v8i16(<8 x i16>, <8 x i16>)
1810ef62fc2SSam Parkerdeclare {<16 x i16>, <16 x i1>} @llvm.uadd.with.overflow.v16i16(<16 x i16>, <16 x i16>)
1820ef62fc2SSam Parkerdeclare {<32 x i16>, <32 x i1>} @llvm.uadd.with.overflow.v32i16(<32 x i16>, <32 x i16>)
1830ef62fc2SSam Parker
1840ef62fc2SSam Parkerdeclare {i8, i1}                @llvm.uadd.with.overflow.i8(i8, i8)
1850ef62fc2SSam Parkerdeclare {<16 x i8>, <16 x i1>}  @llvm.uadd.with.overflow.v16i8(<16 x i8>, <16 x i8>)
1860ef62fc2SSam Parkerdeclare {<32 x i8>, <32 x i1>}  @llvm.uadd.with.overflow.v32i8(<32 x i8>, <32 x i8>)
1870ef62fc2SSam Parkerdeclare {<64 x i8>, <64 x i1>}  @llvm.uadd.with.overflow.v64i8(<64 x i8>, <64 x i8>)
1880ef62fc2SSam Parker
1890ef62fc2SSam Parkerdefine i32 @uadd(i32 %arg) {
19064cfb8a8SSam Parker; V8M-RECIP-LABEL: 'uadd'
191*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 undef, i64 undef)
192*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.uadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
193*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.uadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
194*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.uadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
19564cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 undef, i32 undef)
196*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.uadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
197*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.uadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
198*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.uadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
19964cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.uadd.with.overflow.i16(i16 undef, i16 undef)
200*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.uadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
201*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.uadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
202*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.uadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
20364cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 undef, i8 undef)
204*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.uadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
205*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.uadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
206*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.uadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
2072596da31SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
2080ef62fc2SSam Parker;
20964cfb8a8SSam Parker; NEON-RECIP-LABEL: 'uadd'
210*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 undef, i64 undef)
21164cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.uadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
21264cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.uadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
21364cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.uadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
21464cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 undef, i32 undef)
21564cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.uadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
21664cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.uadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
21764cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.uadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
21864cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.uadd.with.overflow.i16(i16 undef, i16 undef)
21964cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.uadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
22064cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.uadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
22164cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.uadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
22264cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 undef, i8 undef)
22364cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.uadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
22464cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.uadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
22564cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.uadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
226f2675ab4SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
2270ef62fc2SSam Parker;
22864cfb8a8SSam Parker; MVE-RECIP-LABEL: 'uadd'
229*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 undef, i64 undef)
230a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.uadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
231a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.uadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
23264cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 288 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.uadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
23364cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 undef, i32 undef)
23464cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.uadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
235a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.uadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
236a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.uadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
23764cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.uadd.with.overflow.i16(i16 undef, i16 undef)
23864cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.uadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
239a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.uadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
240a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.uadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
24164cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 undef, i8 undef)
24264cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.uadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
243a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 136 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.uadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
244a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 272 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.uadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
245f2675ab4SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
24664cfb8a8SSam Parker;
24764cfb8a8SSam Parker; V8M-SIZE-LABEL: 'uadd'
248*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 undef, i64 undef)
249*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.uadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
250*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.uadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
251*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.uadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
252e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 undef, i32 undef)
253*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.uadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
254*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.uadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
255*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.uadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
256e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.uadd.with.overflow.i16(i16 undef, i16 undef)
257*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.uadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
258*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.uadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
259*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.uadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
260e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 undef, i8 undef)
261*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.uadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
262*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.uadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
263*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.uadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
26464cfb8a8SSam Parker; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
26564cfb8a8SSam Parker;
26664cfb8a8SSam Parker; NEON-SIZE-LABEL: 'uadd'
267*0bc64077SMatt Arsenault; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 undef, i64 undef)
268e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.uadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
269e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.uadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
270e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.uadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
271e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 undef, i32 undef)
272e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.uadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
273e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.uadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
274e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.uadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
275e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.uadd.with.overflow.i16(i16 undef, i16 undef)
276e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.uadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
277e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.uadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
278e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.uadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
279e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 undef, i8 undef)
280e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.uadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
281e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.uadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
282e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.uadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
28364cfb8a8SSam Parker; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
28464cfb8a8SSam Parker;
28564cfb8a8SSam Parker; MVE-SIZE-LABEL: 'uadd'
286*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 undef, i64 undef)
287a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.uadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
288a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 73 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.uadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
289e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 145 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.uadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
290e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 undef, i32 undef)
291e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.uadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
292a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.uadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
293a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.uadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
294e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.uadd.with.overflow.i16(i16 undef, i16 undef)
295e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.uadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
296a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.uadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
297a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 136 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.uadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
298e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.uadd.with.overflow.i8(i8 undef, i8 undef)
299e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.uadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
300a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.uadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
301a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 264 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.uadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
30264cfb8a8SSam Parker; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
3030ef62fc2SSam Parker;
3040ef62fc2SSam Parker  %I64 = call {i64, i1} @llvm.uadd.with.overflow.i64(i64 undef, i64 undef)
3050ef62fc2SSam Parker  %V2I64 = call {<2 x i64>, <2 x i1>} @llvm.uadd.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
3060ef62fc2SSam Parker  %V4I64 = call {<4 x i64>, <4 x i1>} @llvm.uadd.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
3070ef62fc2SSam Parker  %V8I64 = call {<8 x i64>, <8 x i1>} @llvm.uadd.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
3080ef62fc2SSam Parker
3090ef62fc2SSam Parker  %I32 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 undef, i32 undef)
3100ef62fc2SSam Parker  %V4I32  = call {<4 x i32>, <4 x i1>}  @llvm.uadd.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
3110ef62fc2SSam Parker  %V8I32  = call {<8 x i32>, <8 x i1>}  @llvm.uadd.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
3120ef62fc2SSam Parker  %V16I32 = call {<16 x i32>, <16 x i1>} @llvm.uadd.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
3130ef62fc2SSam Parker
3140ef62fc2SSam Parker  %I16 = call {i16, i1} @llvm.uadd.with.overflow.i16(i16 undef, i16 undef)
3150ef62fc2SSam Parker  %V8I16  = call {<8 x i16>, <8 x i1>}  @llvm.uadd.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
3160ef62fc2SSam Parker  %V16I16 = call {<16 x i16>, <16 x i1>} @llvm.uadd.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
3170ef62fc2SSam Parker  %V32I16 = call {<32 x i16>, <32 x i1>} @llvm.uadd.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
3180ef62fc2SSam Parker
3190ef62fc2SSam Parker  %I8 = call {i8, i1} @llvm.uadd.with.overflow.i8(i8 undef, i8 undef)
3200ef62fc2SSam Parker  %V16I8 = call {<16 x i8>, <16 x i1>} @llvm.uadd.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
3210ef62fc2SSam Parker  %V32I8 = call {<32 x i8>, <32 x i1>} @llvm.uadd.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
3220ef62fc2SSam Parker  %V64I8 = call {<64 x i8>, <64 x i1>} @llvm.uadd.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
3230ef62fc2SSam Parker
3240ef62fc2SSam Parker  ret i32 undef
3250ef62fc2SSam Parker}
3260ef62fc2SSam Parker
3270ef62fc2SSam Parkerdeclare {i64, i1}              @llvm.ssub.with.overflow.i64(i64, i64)
3280ef62fc2SSam Parkerdeclare {<2 x i64>, <2 x i1>}  @llvm.ssub.with.overflow.v2i64(<2 x i64>, <2 x i64>)
3290ef62fc2SSam Parkerdeclare {<4 x i64>, <4 x i1>}  @llvm.ssub.with.overflow.v4i64(<4 x i64>, <4 x i64>)
3300ef62fc2SSam Parkerdeclare {<8 x i64>, <8 x i1>}  @llvm.ssub.with.overflow.v8i64(<8 x i64>, <8 x i64>)
3310ef62fc2SSam Parker
3320ef62fc2SSam Parkerdeclare {i32, i1}               @llvm.ssub.with.overflow.i32(i32, i32)
3330ef62fc2SSam Parkerdeclare {<4 x i32>, <4 x i1>}   @llvm.ssub.with.overflow.v4i32(<4 x i32>, <4 x i32>)
3340ef62fc2SSam Parkerdeclare {<8 x i32>, <8 x i1>}   @llvm.ssub.with.overflow.v8i32(<8 x i32>, <8 x i32>)
3350ef62fc2SSam Parkerdeclare {<16 x i32>, <16 x i1>} @llvm.ssub.with.overflow.v16i32(<16 x i32>, <16 x i32>)
3360ef62fc2SSam Parker
3370ef62fc2SSam Parkerdeclare {i16, i1}               @llvm.ssub.with.overflow.i16(i16, i16)
3380ef62fc2SSam Parkerdeclare {<8 x i16>,  <8 x i1>}  @llvm.ssub.with.overflow.v8i16(<8 x i16>, <8 x i16>)
3390ef62fc2SSam Parkerdeclare {<16 x i16>, <16 x i1>} @llvm.ssub.with.overflow.v16i16(<16 x i16>, <16 x i16>)
3400ef62fc2SSam Parkerdeclare {<32 x i16>, <32 x i1>} @llvm.ssub.with.overflow.v32i16(<32 x i16>, <32 x i16>)
3410ef62fc2SSam Parker
3420ef62fc2SSam Parkerdeclare {i8, i1}                @llvm.ssub.with.overflow.i8(i8, i8)
3430ef62fc2SSam Parkerdeclare {<16 x i8>, <16 x i1>}  @llvm.ssub.with.overflow.v16i8(<16 x i8>, <16 x i8>)
3440ef62fc2SSam Parkerdeclare {<32 x i8>, <32 x i1>}  @llvm.ssub.with.overflow.v32i8(<32 x i8>, <32 x i8>)
3450ef62fc2SSam Parkerdeclare {<64 x i8>, <64 x i1>}  @llvm.ssub.with.overflow.v64i8(<64 x i8>, <64 x i8>)
3460ef62fc2SSam Parker
3470ef62fc2SSam Parkerdefine i32 @ssub(i32 %arg) {
34864cfb8a8SSam Parker; V8M-RECIP-LABEL: 'ssub'
349*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 undef, i64 undef)
350*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.ssub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
351*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.ssub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
352*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.ssub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
353*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 undef, i32 undef)
354*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.ssub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
355*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.ssub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
356*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.ssub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
357*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.ssub.with.overflow.i16(i16 undef, i16 undef)
358*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.ssub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
359*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.ssub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
360*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.ssub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
361*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 undef, i8 undef)
362*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.ssub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
363*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.ssub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
364*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.ssub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
3652596da31SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
3660ef62fc2SSam Parker;
36764cfb8a8SSam Parker; NEON-RECIP-LABEL: 'ssub'
368*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 undef, i64 undef)
36976534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.ssub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
37076534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.ssub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
37176534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.ssub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
372*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 undef, i32 undef)
37376534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.ssub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
37476534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.ssub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
37576534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.ssub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
376*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.ssub.with.overflow.i16(i16 undef, i16 undef)
37776534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.ssub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
37876534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.ssub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
37976534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.ssub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
380*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 undef, i8 undef)
38176534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.ssub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
38276534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.ssub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
38376534829SCraig Topper; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.ssub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
384f2675ab4SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
3850ef62fc2SSam Parker;
38664cfb8a8SSam Parker; MVE-RECIP-LABEL: 'ssub'
387*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 undef, i64 undef)
388255ad734SDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.ssub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
38976534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 218 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.ssub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
39076534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 434 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.ssub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
391*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 undef, i32 undef)
39276534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.ssub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
39376534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.ssub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
39476534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 154 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.ssub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
395*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.ssub.with.overflow.i16(i16 undef, i16 undef)
39676534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.ssub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
39776534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 142 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.ssub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
39876534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 284 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.ssub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
399*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 undef, i8 undef)
40076534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.ssub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
40176534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 272 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.ssub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
40276534829SCraig Topper; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 544 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.ssub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
403f2675ab4SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
40464cfb8a8SSam Parker;
40564cfb8a8SSam Parker; V8M-SIZE-LABEL: 'ssub'
406*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 undef, i64 undef)
40776534829SCraig Topper; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.ssub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
408*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.ssub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
409*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.ssub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
410*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 undef, i32 undef)
411*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.ssub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
412*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.ssub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
413*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.ssub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
414*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.ssub.with.overflow.i16(i16 undef, i16 undef)
415*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.ssub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
416*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.ssub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
417*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.ssub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
418*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 undef, i8 undef)
419*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.ssub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
420*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.ssub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
421*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.ssub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
42264cfb8a8SSam Parker; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
42364cfb8a8SSam Parker;
42464cfb8a8SSam Parker; NEON-SIZE-LABEL: 'ssub'
42576534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 undef, i64 undef)
42676534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.ssub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
42776534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.ssub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
42876534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.ssub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
429*0bc64077SMatt Arsenault; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 undef, i32 undef)
43076534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.ssub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
43176534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.ssub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
43276534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.ssub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
433*0bc64077SMatt Arsenault; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.ssub.with.overflow.i16(i16 undef, i16 undef)
43476534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.ssub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
43576534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.ssub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
43676534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.ssub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
437*0bc64077SMatt Arsenault; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 undef, i8 undef)
43876534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.ssub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
43976534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.ssub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
44076534829SCraig Topper; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.ssub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
44164cfb8a8SSam Parker; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
44264cfb8a8SSam Parker;
44364cfb8a8SSam Parker; MVE-SIZE-LABEL: 'ssub'
444*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 undef, i64 undef)
445255ad734SDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.ssub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
44676534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 75 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.ssub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
44776534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 147 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.ssub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
448*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 undef, i32 undef)
44976534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.ssub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
45076534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.ssub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
45176534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 141 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.ssub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
452*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.ssub.with.overflow.i16(i16 undef, i16 undef)
45376534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.ssub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
45476534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.ssub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
45576534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 270 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.ssub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
456*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.ssub.with.overflow.i8(i8 undef, i8 undef)
45776534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.ssub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
45876534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 264 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.ssub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
45976534829SCraig Topper; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 528 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.ssub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
46064cfb8a8SSam Parker; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
4610ef62fc2SSam Parker;
4620ef62fc2SSam Parker  %I64 = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 undef, i64 undef)
4630ef62fc2SSam Parker  %V2I64 = call {<2 x i64>, <2 x i1>} @llvm.ssub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
4640ef62fc2SSam Parker  %V4I64 = call {<4 x i64>, <4 x i1>} @llvm.ssub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
4650ef62fc2SSam Parker  %V8I64 = call {<8 x i64>, <8 x i1>} @llvm.ssub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
4660ef62fc2SSam Parker
4670ef62fc2SSam Parker  %I32 = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 undef, i32 undef)
4680ef62fc2SSam Parker  %V4I32  = call {<4 x i32>, <4 x i1>}  @llvm.ssub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
4690ef62fc2SSam Parker  %V8I32  = call {<8 x i32>, <8 x i1>}  @llvm.ssub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
4700ef62fc2SSam Parker  %V16I32 = call {<16 x i32>, <16 x i1>} @llvm.ssub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
4710ef62fc2SSam Parker
4720ef62fc2SSam Parker  %I16 = call {i16, i1} @llvm.ssub.with.overflow.i16(i16 undef, i16 undef)
4730ef62fc2SSam Parker  %V8I16  = call {<8 x i16>, <8 x i1>}  @llvm.ssub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
4740ef62fc2SSam Parker  %V16I16 = call {<16 x i16>, <16 x i1>} @llvm.ssub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
4750ef62fc2SSam Parker  %V32I16 = call {<32 x i16>, <32 x i1>} @llvm.ssub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
4760ef62fc2SSam Parker
4770ef62fc2SSam Parker  %I8 = call {i8, i1} @llvm.ssub.with.overflow.i8(i8 undef, i8 undef)
4780ef62fc2SSam Parker  %V16I8 = call {<16 x i8>, <16 x i1>} @llvm.ssub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
4790ef62fc2SSam Parker  %V32I8 = call {<32 x i8>, <32 x i1>} @llvm.ssub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
4800ef62fc2SSam Parker  %V64I8 = call {<64 x i8>, <64 x i1>} @llvm.ssub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
4810ef62fc2SSam Parker
4820ef62fc2SSam Parker  ret i32 undef
4830ef62fc2SSam Parker}
4840ef62fc2SSam Parker
4850ef62fc2SSam Parkerdeclare {i64, i1}              @llvm.usub.with.overflow.i64(i64, i64)
4860ef62fc2SSam Parkerdeclare {<2 x i64>, <2 x i1>}  @llvm.usub.with.overflow.v2i64(<2 x i64>, <2 x i64>)
4870ef62fc2SSam Parkerdeclare {<4 x i64>, <4 x i1>}  @llvm.usub.with.overflow.v4i64(<4 x i64>, <4 x i64>)
4880ef62fc2SSam Parkerdeclare {<8 x i64>, <8 x i1>}  @llvm.usub.with.overflow.v8i64(<8 x i64>, <8 x i64>)
4890ef62fc2SSam Parker
4900ef62fc2SSam Parkerdeclare {i32, i1}               @llvm.usub.with.overflow.i32(i32, i32)
4910ef62fc2SSam Parkerdeclare {<4 x i32>, <4 x i1>}   @llvm.usub.with.overflow.v4i32(<4 x i32>, <4 x i32>)
4920ef62fc2SSam Parkerdeclare {<8 x i32>, <8 x i1>}   @llvm.usub.with.overflow.v8i32(<8 x i32>, <8 x i32>)
4930ef62fc2SSam Parkerdeclare {<16 x i32>, <16 x i1>} @llvm.usub.with.overflow.v16i32(<16 x i32>, <16 x i32>)
4940ef62fc2SSam Parker
4950ef62fc2SSam Parkerdeclare {i16, i1}               @llvm.usub.with.overflow.i16(i16, i16)
4960ef62fc2SSam Parkerdeclare {<8 x i16>,  <8 x i1>}  @llvm.usub.with.overflow.v8i16(<8 x i16>, <8 x i16>)
4970ef62fc2SSam Parkerdeclare {<16 x i16>, <16 x i1>} @llvm.usub.with.overflow.v16i16(<16 x i16>, <16 x i16>)
4980ef62fc2SSam Parkerdeclare {<32 x i16>, <32 x i1>} @llvm.usub.with.overflow.v32i16(<32 x i16>, <32 x i16>)
4990ef62fc2SSam Parker
5000ef62fc2SSam Parkerdeclare {i8, i1}                @llvm.usub.with.overflow.i8(i8, i8)
5010ef62fc2SSam Parkerdeclare {<16 x i8>, <16 x i1>}  @llvm.usub.with.overflow.v16i8(<16 x i8>, <16 x i8>)
5020ef62fc2SSam Parkerdeclare {<32 x i8>, <32 x i1>}  @llvm.usub.with.overflow.v32i8(<32 x i8>, <32 x i8>)
5030ef62fc2SSam Parkerdeclare {<64 x i8>, <64 x i1>}  @llvm.usub.with.overflow.v64i8(<64 x i8>, <64 x i8>)
5040ef62fc2SSam Parker
5050ef62fc2SSam Parkerdefine i32 @usub(i32 %arg) {
50664cfb8a8SSam Parker; V8M-RECIP-LABEL: 'usub'
507*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 undef, i64 undef)
508*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.usub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
509*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.usub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
510*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.usub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
51164cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 undef, i32 undef)
512*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.usub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
513*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.usub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
514*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.usub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
51564cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.usub.with.overflow.i16(i16 undef, i16 undef)
516*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.usub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
517*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.usub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
518*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.usub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
51964cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 undef, i8 undef)
520*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.usub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
521*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.usub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
522*0bc64077SMatt Arsenault; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.usub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
5232596da31SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
5240ef62fc2SSam Parker;
52564cfb8a8SSam Parker; NEON-RECIP-LABEL: 'usub'
526*0bc64077SMatt Arsenault; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 undef, i64 undef)
52764cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.usub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
52864cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.usub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
52964cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.usub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
53064cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 undef, i32 undef)
53164cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.usub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
53264cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.usub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
53364cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.usub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
53464cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.usub.with.overflow.i16(i16 undef, i16 undef)
53564cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.usub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
53664cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.usub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
53764cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.usub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
53864cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 undef, i8 undef)
53964cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.usub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
54064cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.usub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
54164cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.usub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
542f2675ab4SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
5430ef62fc2SSam Parker;
54464cfb8a8SSam Parker; MVE-RECIP-LABEL: 'usub'
545*0bc64077SMatt Arsenault; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 undef, i64 undef)
546a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.usub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
547a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.usub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
54864cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 288 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.usub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
54964cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 undef, i32 undef)
55064cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.usub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
551a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.usub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
552a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.usub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
55364cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.usub.with.overflow.i16(i16 undef, i16 undef)
55464cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.usub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
555a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.usub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
556a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.usub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
55764cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 undef, i8 undef)
55864cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.usub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
559a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 136 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.usub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
560a2e0312cSDavid Green; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 272 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.usub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
561f2675ab4SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
56264cfb8a8SSam Parker;
56364cfb8a8SSam Parker; V8M-SIZE-LABEL: 'usub'
564*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 undef, i64 undef)
565*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.usub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
566*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.usub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
567*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.usub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
568e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 undef, i32 undef)
569*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.usub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
570*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.usub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
571*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.usub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
572e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.usub.with.overflow.i16(i16 undef, i16 undef)
573*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.usub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
574*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.usub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
575*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.usub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
576e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 undef, i8 undef)
577*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.usub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
578*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.usub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
579*0bc64077SMatt Arsenault; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.usub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
58064cfb8a8SSam Parker; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
58164cfb8a8SSam Parker;
58264cfb8a8SSam Parker; NEON-SIZE-LABEL: 'usub'
583*0bc64077SMatt Arsenault; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 undef, i64 undef)
584e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.usub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
585e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.usub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
586e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.usub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
587e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 undef, i32 undef)
588e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.usub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
589e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.usub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
590e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.usub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
591e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.usub.with.overflow.i16(i16 undef, i16 undef)
592e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.usub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
593e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.usub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
594e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.usub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
595e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 undef, i8 undef)
596e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.usub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
597e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.usub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
598e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.usub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
59964cfb8a8SSam Parker; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
60064cfb8a8SSam Parker;
60164cfb8a8SSam Parker; MVE-SIZE-LABEL: 'usub'
602*0bc64077SMatt Arsenault; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 undef, i64 undef)
603a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.usub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
604a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 73 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.usub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
605e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 145 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.usub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
606e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I32 = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 undef, i32 undef)
607e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.usub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
608a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.usub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
609a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.usub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
610e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I16 = call { i16, i1 } @llvm.usub.with.overflow.i16(i16 undef, i16 undef)
611e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.usub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
612a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.usub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
613a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 136 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.usub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
614e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %I8 = call { i8, i1 } @llvm.usub.with.overflow.i8(i8 undef, i8 undef)
615e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.usub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
616a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.usub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
617a2e0312cSDavid Green; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 264 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.usub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
61864cfb8a8SSam Parker; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
6190ef62fc2SSam Parker;
6200ef62fc2SSam Parker  %I64 = call {i64, i1} @llvm.usub.with.overflow.i64(i64 undef, i64 undef)
6210ef62fc2SSam Parker  %V2I64 = call {<2 x i64>, <2 x i1>} @llvm.usub.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
6220ef62fc2SSam Parker  %V4I64 = call {<4 x i64>, <4 x i1>} @llvm.usub.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
6230ef62fc2SSam Parker  %V8I64 = call {<8 x i64>, <8 x i1>} @llvm.usub.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
6240ef62fc2SSam Parker
6250ef62fc2SSam Parker  %I32 = call {i32, i1} @llvm.usub.with.overflow.i32(i32 undef, i32 undef)
6260ef62fc2SSam Parker  %V4I32  = call {<4 x i32>, <4 x i1>}  @llvm.usub.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
6270ef62fc2SSam Parker  %V8I32  = call {<8 x i32>, <8 x i1>}  @llvm.usub.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
6280ef62fc2SSam Parker  %V16I32 = call {<16 x i32>, <16 x i1>} @llvm.usub.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
6290ef62fc2SSam Parker
6300ef62fc2SSam Parker  %I16 = call {i16, i1} @llvm.usub.with.overflow.i16(i16 undef, i16 undef)
6310ef62fc2SSam Parker  %V8I16  = call {<8 x i16>, <8 x i1>}  @llvm.usub.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
6320ef62fc2SSam Parker  %V16I16 = call {<16 x i16>, <16 x i1>} @llvm.usub.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
6330ef62fc2SSam Parker  %V32I16 = call {<32 x i16>, <32 x i1>} @llvm.usub.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
6340ef62fc2SSam Parker
6350ef62fc2SSam Parker  %I8 = call {i8, i1} @llvm.usub.with.overflow.i8(i8 undef, i8 undef)
6360ef62fc2SSam Parker  %V16I8 = call {<16 x i8>, <16 x i1>} @llvm.usub.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
6370ef62fc2SSam Parker  %V32I8 = call {<32 x i8>, <32 x i1>} @llvm.usub.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
6380ef62fc2SSam Parker  %V64I8 = call {<64 x i8>, <64 x i1>} @llvm.usub.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
6390ef62fc2SSam Parker
6400ef62fc2SSam Parker  ret i32 undef
6410ef62fc2SSam Parker}
6420ef62fc2SSam Parker
6430ef62fc2SSam Parkerdeclare {i64, i1}              @llvm.smul.with.overflow.i64(i64, i64)
6440ef62fc2SSam Parkerdeclare {<2 x i64>, <2 x i1>}  @llvm.smul.with.overflow.v2i64(<2 x i64>, <2 x i64>)
6450ef62fc2SSam Parkerdeclare {<4 x i64>, <4 x i1>}  @llvm.smul.with.overflow.v4i64(<4 x i64>, <4 x i64>)
6460ef62fc2SSam Parkerdeclare {<8 x i64>, <8 x i1>}  @llvm.smul.with.overflow.v8i64(<8 x i64>, <8 x i64>)
6470ef62fc2SSam Parker
6480ef62fc2SSam Parkerdeclare {i32, i1}               @llvm.smul.with.overflow.i32(i32, i32)
6490ef62fc2SSam Parkerdeclare {<4 x i32>, <4 x i1>}   @llvm.smul.with.overflow.v4i32(<4 x i32>, <4 x i32>)
6500ef62fc2SSam Parkerdeclare {<8 x i32>, <8 x i1>}   @llvm.smul.with.overflow.v8i32(<8 x i32>, <8 x i32>)
6510ef62fc2SSam Parkerdeclare {<16 x i32>, <16 x i1>} @llvm.smul.with.overflow.v16i32(<16 x i32>, <16 x i32>)
6520ef62fc2SSam Parker
6530ef62fc2SSam Parkerdeclare {i16, i1}               @llvm.smul.with.overflow.i16(i16, i16)
6540ef62fc2SSam Parkerdeclare {<8 x i16>,  <8 x i1>}  @llvm.smul.with.overflow.v8i16(<8 x i16>, <8 x i16>)
6550ef62fc2SSam Parkerdeclare {<16 x i16>, <16 x i1>} @llvm.smul.with.overflow.v16i16(<16 x i16>, <16 x i16>)
6560ef62fc2SSam Parkerdeclare {<32 x i16>, <32 x i1>} @llvm.smul.with.overflow.v32i16(<32 x i16>, <32 x i16>)
6570ef62fc2SSam Parker
6580ef62fc2SSam Parkerdeclare {i8, i1}                @llvm.smul.with.overflow.i8(i8, i8)
6590ef62fc2SSam Parkerdeclare {<16 x i8>, <16 x i1>}  @llvm.smul.with.overflow.v16i8(<16 x i8>, <16 x i8>)
6600ef62fc2SSam Parkerdeclare {<32 x i8>, <32 x i1>}  @llvm.smul.with.overflow.v32i8(<32 x i8>, <32 x i8>)
6610ef62fc2SSam Parkerdeclare {<64 x i8>, <64 x i1>}  @llvm.smul.with.overflow.v64i8(<64 x i8>, <64 x i8>)
6620ef62fc2SSam Parker
6630ef62fc2SSam Parkerdefine i32 @smul(i32 %arg) {
66464cfb8a8SSam Parker; V8M-RECIP-LABEL: 'smul'
6657bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %I64 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 undef, i64 undef)
6667bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.smul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
6677bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 164 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.smul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
6687bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 328 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.smul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
6697bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %I32 = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 undef, i32 undef)
670309f1e4aSDavid Green; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
671309f1e4aSDavid Green; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 136 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.smul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
672309f1e4aSDavid Green; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 272 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
67364cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call { i16, i1 } @llvm.smul.with.overflow.i16(i16 undef, i16 undef)
67464cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.smul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
67564cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
67664cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.smul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
67764cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I8 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 undef, i8 undef)
67864cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.smul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
67964cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.smul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
68064cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.smul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
6812596da31SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
6820ef62fc2SSam Parker;
68364cfb8a8SSam Parker; NEON-RECIP-LABEL: 'smul'
6847bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %I64 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 undef, i64 undef)
6857bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.smul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
6867bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 162 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.smul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
6877bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.smul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
6887bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %I32 = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 undef, i32 undef)
6897bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
6907bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.smul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
6917bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
69264cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call { i16, i1 } @llvm.smul.with.overflow.i16(i16 undef, i16 undef)
6937bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.smul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
6947bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
6957bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.smul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
69664cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I8 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 undef, i8 undef)
6977bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.smul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
6987bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.smul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
6997bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.smul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
700f2675ab4SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
7010ef62fc2SSam Parker;
70264cfb8a8SSam Parker; MVE-RECIP-LABEL: 'smul'
7037bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %I64 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 undef, i64 undef)
7047bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 508 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.smul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
7057bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1016 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.smul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
7067bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2032 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.smul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
7077bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %I32 = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 undef, i32 undef)
7087bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 284 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
7097bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 872 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.smul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
7107bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2832 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
71164cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call { i16, i1 } @llvm.smul.with.overflow.i16(i16 undef, i16 undef)
7127bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.smul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
7137bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 232 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
7147bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 624 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.smul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
71564cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I8 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 undef, i8 undef)
7167bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.smul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
7177bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 360 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.smul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
7187bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 880 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.smul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
719f2675ab4SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
72064cfb8a8SSam Parker;
72164cfb8a8SSam Parker; V8M-SIZE-LABEL: 'smul'
7227bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %I64 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 undef, i64 undef)
7237bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.smul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
7247bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.smul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
7257bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.smul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
7267bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %I32 = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 undef, i32 undef)
727309f1e4aSDavid Green; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
728309f1e4aSDavid Green; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.smul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
729309f1e4aSDavid Green; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
730e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call { i16, i1 } @llvm.smul.with.overflow.i16(i16 undef, i16 undef)
731e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.smul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
732e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
733e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.smul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
734e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I8 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 undef, i8 undef)
735e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.smul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
736e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.smul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
737e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 195 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.smul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
73864cfb8a8SSam Parker; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
73964cfb8a8SSam Parker;
74064cfb8a8SSam Parker; NEON-SIZE-LABEL: 'smul'
741e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %I64 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 undef, i64 undef)
7427bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.smul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
7437bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.smul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
7447bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.smul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
745e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I32 = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 undef, i32 undef)
7467bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
7477bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.smul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
7487bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
749e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call { i16, i1 } @llvm.smul.with.overflow.i16(i16 undef, i16 undef)
750e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.smul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
751e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
752e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.smul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
753e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I8 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 undef, i8 undef)
754e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.smul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
755e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.smul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
756e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.smul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
75764cfb8a8SSam Parker; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
75864cfb8a8SSam Parker;
75964cfb8a8SSam Parker; MVE-SIZE-LABEL: 'smul'
7607bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %I64 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 undef, i64 undef)
7617bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.smul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
7627bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 109 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.smul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
7637bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 213 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.smul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
7647bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %I32 = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 undef, i32 undef)
7657bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 150 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
7667bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 328 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.smul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
7677bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 652 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
768e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call { i16, i1 } @llvm.smul.with.overflow.i16(i16 undef, i16 undef)
7697bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.smul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
7707bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 142 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
7717bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 282 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.smul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
772e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I8 = call { i8, i1 } @llvm.smul.with.overflow.i8(i8 undef, i8 undef)
7737bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.smul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
7747bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 270 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.smul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
7757bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 538 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.smul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
77664cfb8a8SSam Parker; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
7770ef62fc2SSam Parker;
7780ef62fc2SSam Parker  %I64 = call {i64, i1} @llvm.smul.with.overflow.i64(i64 undef, i64 undef)
7790ef62fc2SSam Parker  %V2I64 = call {<2 x i64>, <2 x i1>} @llvm.smul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
7800ef62fc2SSam Parker  %V4I64 = call {<4 x i64>, <4 x i1>} @llvm.smul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
7810ef62fc2SSam Parker  %V8I64 = call {<8 x i64>, <8 x i1>} @llvm.smul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
7820ef62fc2SSam Parker
7830ef62fc2SSam Parker  %I32 = call {i32, i1} @llvm.smul.with.overflow.i32(i32 undef, i32 undef)
7840ef62fc2SSam Parker  %V4I32  = call {<4 x i32>, <4 x i1>}  @llvm.smul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
7850ef62fc2SSam Parker  %V8I32  = call {<8 x i32>, <8 x i1>}  @llvm.smul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
7860ef62fc2SSam Parker  %V16I32 = call {<16 x i32>, <16 x i1>} @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
7870ef62fc2SSam Parker
7880ef62fc2SSam Parker  %I16 = call {i16, i1} @llvm.smul.with.overflow.i16(i16 undef, i16 undef)
7890ef62fc2SSam Parker  %V8I16  = call {<8 x i16>, <8 x i1>}  @llvm.smul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
7900ef62fc2SSam Parker  %V16I16 = call {<16 x i16>, <16 x i1>} @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
7910ef62fc2SSam Parker  %V32I16 = call {<32 x i16>, <32 x i1>} @llvm.smul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
7920ef62fc2SSam Parker
7930ef62fc2SSam Parker  %I8 = call {i8, i1} @llvm.smul.with.overflow.i8(i8 undef, i8 undef)
7940ef62fc2SSam Parker  %V16I8 = call {<16 x i8>, <16 x i1>} @llvm.smul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
7950ef62fc2SSam Parker  %V32I8 = call {<32 x i8>, <32 x i1>} @llvm.smul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
7960ef62fc2SSam Parker  %V64I8 = call {<64 x i8>, <64 x i1>} @llvm.smul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
7970ef62fc2SSam Parker
7980ef62fc2SSam Parker  ret i32 undef
7990ef62fc2SSam Parker}
8000ef62fc2SSam Parker
8010ef62fc2SSam Parkerdeclare {i64, i1}              @llvm.umul.with.overflow.i64(i64, i64)
8020ef62fc2SSam Parkerdeclare {<2 x i64>, <2 x i1>}  @llvm.umul.with.overflow.v2i64(<2 x i64>, <2 x i64>)
8030ef62fc2SSam Parkerdeclare {<4 x i64>, <4 x i1>}  @llvm.umul.with.overflow.v4i64(<4 x i64>, <4 x i64>)
8040ef62fc2SSam Parkerdeclare {<8 x i64>, <8 x i1>}  @llvm.umul.with.overflow.v8i64(<8 x i64>, <8 x i64>)
8050ef62fc2SSam Parker
8060ef62fc2SSam Parkerdeclare {i32, i1}               @llvm.umul.with.overflow.i32(i32, i32)
8070ef62fc2SSam Parkerdeclare {<4 x i32>, <4 x i1>}   @llvm.umul.with.overflow.v4i32(<4 x i32>, <4 x i32>)
8080ef62fc2SSam Parkerdeclare {<8 x i32>, <8 x i1>}   @llvm.umul.with.overflow.v8i32(<8 x i32>, <8 x i32>)
8090ef62fc2SSam Parkerdeclare {<16 x i32>, <16 x i1>} @llvm.umul.with.overflow.v16i32(<16 x i32>, <16 x i32>)
8100ef62fc2SSam Parker
8110ef62fc2SSam Parkerdeclare {i16, i1}               @llvm.umul.with.overflow.i16(i16, i16)
8120ef62fc2SSam Parkerdeclare {<8 x i16>,  <8 x i1>}  @llvm.umul.with.overflow.v8i16(<8 x i16>, <8 x i16>)
8130ef62fc2SSam Parkerdeclare {<16 x i16>, <16 x i1>} @llvm.umul.with.overflow.v16i16(<16 x i16>, <16 x i16>)
8140ef62fc2SSam Parkerdeclare {<32 x i16>, <32 x i1>} @llvm.umul.with.overflow.v32i16(<32 x i16>, <32 x i16>)
8150ef62fc2SSam Parker
8160ef62fc2SSam Parkerdeclare {i8, i1}                @llvm.umul.with.overflow.i8(i8, i8)
8170ef62fc2SSam Parkerdeclare {<16 x i8>, <16 x i1>}  @llvm.umul.with.overflow.v16i8(<16 x i8>, <16 x i8>)
8180ef62fc2SSam Parkerdeclare {<32 x i8>, <32 x i1>}  @llvm.umul.with.overflow.v32i8(<32 x i8>, <32 x i8>)
8190ef62fc2SSam Parkerdeclare {<64 x i8>, <64 x i1>}  @llvm.umul.with.overflow.v64i8(<64 x i8>, <64 x i8>)
8200ef62fc2SSam Parker
8210ef62fc2SSam Parkerdefine i32 @umul(i32 %arg) {
82264cfb8a8SSam Parker; V8M-RECIP-LABEL: 'umul'
8237bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %I64 = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 undef, i64 undef)
8247bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.umul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
8257bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 156 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.umul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
8267bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 312 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.umul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
8277bd097fdSSimon Pilgrim; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %I32 = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 undef, i32 undef)
828309f1e4aSDavid Green; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.umul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
829309f1e4aSDavid Green; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.umul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
830309f1e4aSDavid Green; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
83164cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 undef, i16 undef)
83264cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.umul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
83364cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
83464cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.umul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
83564cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 undef, i8 undef)
83664cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.umul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
83764cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.umul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
83864cfb8a8SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.umul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
8392596da31SSam Parker; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
8400ef62fc2SSam Parker;
84164cfb8a8SSam Parker; NEON-RECIP-LABEL: 'umul'
8427bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %I64 = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 undef, i64 undef)
8437bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.umul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
8447bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 154 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.umul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
8457bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 308 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.umul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
8467bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %I32 = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 undef, i32 undef)
8477bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.umul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
8487bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.umul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
8497bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
85064cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 undef, i16 undef)
8517bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.umul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
8527bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
8537bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.umul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
85464cfb8a8SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 undef, i8 undef)
8557bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.umul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
8567bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.umul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
8577bd097fdSSimon Pilgrim; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.umul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
858f2675ab4SSam Parker; NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
8590ef62fc2SSam Parker;
86064cfb8a8SSam Parker; MVE-RECIP-LABEL: 'umul'
8617bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %I64 = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 undef, i64 undef)
8627bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 472 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.umul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
8637bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 944 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.umul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
8647bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1888 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.umul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
8657bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %I32 = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 undef, i32 undef)
8667bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 186 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.umul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
8677bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 484 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.umul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
8687bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1288 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
86964cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 undef, i16 undef)
8707bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.umul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
8717bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
8727bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 616 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.umul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
87364cfb8a8SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 undef, i8 undef)
8747bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.umul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
8757bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 356 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.umul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
8767bd097fdSSimon Pilgrim; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 872 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.umul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
877f2675ab4SSam Parker; MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
87864cfb8a8SSam Parker;
87964cfb8a8SSam Parker; V8M-SIZE-LABEL: 'umul'
8807bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %I64 = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 undef, i64 undef)
8817bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.umul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
8827bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.umul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
8837bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 69 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.umul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
8847bd097fdSSimon Pilgrim; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %I32 = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 undef, i32 undef)
885309f1e4aSDavid Green; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.umul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
886309f1e4aSDavid Green; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.umul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
887309f1e4aSDavid Green; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
888e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 undef, i16 undef)
889e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.umul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
890e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
891e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.umul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
892e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 undef, i8 undef)
893e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.umul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
894e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.umul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
895e32bd351SSanjay Patel; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.umul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
89664cfb8a8SSam Parker; V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
89764cfb8a8SSam Parker;
89864cfb8a8SSam Parker; NEON-SIZE-LABEL: 'umul'
899e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %I64 = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 undef, i64 undef)
9007bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.umul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
9017bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.umul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
9027bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.umul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
903e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I32 = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 undef, i32 undef)
9047bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.umul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
9057bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.umul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
9067bd097fdSSimon Pilgrim; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
907e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 undef, i16 undef)
908e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.umul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
909e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
910e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.umul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
911e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 undef, i8 undef)
912e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.umul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
913e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.umul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
914e32bd351SSanjay Patel; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.umul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
91564cfb8a8SSam Parker; NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
91664cfb8a8SSam Parker;
91764cfb8a8SSam Parker; MVE-SIZE-LABEL: 'umul'
9187bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %I64 = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 undef, i64 undef)
9197bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V2I64 = call { <2 x i64>, <2 x i1> } @llvm.umul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
9207bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.umul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
9217bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 69 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.umul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
9227bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %I32 = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 undef, i32 undef)
9237bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 149 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.umul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
9247bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 326 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.umul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
9257bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 648 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
926e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 undef, i16 undef)
9277bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.umul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
9287bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 140 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
9297bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 278 for instruction: %V32I16 = call { <32 x i16>, <32 x i1> } @llvm.umul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
930e32bd351SSanjay Patel; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I8 = call { i8, i1 } @llvm.umul.with.overflow.i8(i8 undef, i8 undef)
9317bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %V16I8 = call { <16 x i8>, <16 x i1> } @llvm.umul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
9327bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %V32I8 = call { <32 x i8>, <32 x i1> } @llvm.umul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
9337bd097fdSSimon Pilgrim; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 534 for instruction: %V64I8 = call { <64 x i8>, <64 x i1> } @llvm.umul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
93464cfb8a8SSam Parker; MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
9350ef62fc2SSam Parker;
9360ef62fc2SSam Parker  %I64 = call {i64, i1} @llvm.umul.with.overflow.i64(i64 undef, i64 undef)
9370ef62fc2SSam Parker  %V2I64 = call {<2 x i64>, <2 x i1>} @llvm.umul.with.overflow.v2i64(<2 x i64> undef, <2 x i64> undef)
9380ef62fc2SSam Parker  %V4I64 = call {<4 x i64>, <4 x i1>} @llvm.umul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
9390ef62fc2SSam Parker  %V8I64 = call {<8 x i64>, <8 x i1>} @llvm.umul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
9400ef62fc2SSam Parker
9410ef62fc2SSam Parker  %I32 = call {i32, i1} @llvm.umul.with.overflow.i32(i32 undef, i32 undef)
9420ef62fc2SSam Parker  %V4I32  = call {<4 x i32>, <4 x i1>}  @llvm.umul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
9430ef62fc2SSam Parker  %V8I32  = call {<8 x i32>, <8 x i1>}  @llvm.umul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
9440ef62fc2SSam Parker  %V16I32 = call {<16 x i32>, <16 x i1>} @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
9450ef62fc2SSam Parker
9460ef62fc2SSam Parker  %I16 = call {i16, i1} @llvm.umul.with.overflow.i16(i16 undef, i16 undef)
9470ef62fc2SSam Parker  %V8I16  = call {<8 x i16>, <8 x i1>}  @llvm.umul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
9480ef62fc2SSam Parker  %V16I16 = call {<16 x i16>, <16 x i1>} @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
9490ef62fc2SSam Parker  %V32I16 = call {<32 x i16>, <32 x i1>} @llvm.umul.with.overflow.v32i16(<32 x i16> undef, <32 x i16> undef)
9500ef62fc2SSam Parker
9510ef62fc2SSam Parker  %I8 = call {i8, i1} @llvm.umul.with.overflow.i8(i8 undef, i8 undef)
9520ef62fc2SSam Parker  %V16I8 = call {<16 x i8>, <16 x i1>} @llvm.umul.with.overflow.v16i8(<16 x i8> undef, <16 x i8> undef)
9530ef62fc2SSam Parker  %V32I8 = call {<32 x i8>, <32 x i1>} @llvm.umul.with.overflow.v32i8(<32 x i8> undef, <32 x i8> undef)
9540ef62fc2SSam Parker  %V64I8 = call {<64 x i8>, <64 x i1>} @llvm.umul.with.overflow.v64i8(<64 x i8> undef, <64 x i8> undef)
9550ef62fc2SSam Parker
9560ef62fc2SSam Parker  ret i32 undef
9570ef62fc2SSam Parker}
958