xref: /llvm-project/llvm/test/Analysis/CostModel/X86/reduce-fadd.ll (revision 6c80267d0ff445c0c47c6ddb283da5a8bc4feb64)
1; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
2; RUN: opt < %s -passes="print<cost-model>" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
3; RUN: opt < %s -passes="print<cost-model>" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
4; RUN: opt < %s -passes="print<cost-model>" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE41
5; RUN: opt < %s -passes="print<cost-model>" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
6; RUN: opt < %s -passes="print<cost-model>" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx | FileCheck %s --check-prefixes=AVX1
7; RUN: opt < %s -passes="print<cost-model>" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
8; RUN: opt < %s -passes="print<cost-model>" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512
9; RUN: opt < %s -passes="print<cost-model>" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512
10; RUN: opt < %s -passes="print<cost-model>" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512
11
12define void @reduce_f64(double %arg) {
13; SSE2-LABEL: 'reduce_f64'
14; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1 = call double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
15; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2 = call double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
16; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4 = call double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
17; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V8 = call double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
18; SSE2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V16 = call double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
19; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
20;
21; SSSE3-LABEL: 'reduce_f64'
22; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1 = call double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
23; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2 = call double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
24; SSSE3-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4 = call double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
25; SSSE3-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V8 = call double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
26; SSSE3-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V16 = call double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
27; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
28;
29; SSE41-LABEL: 'reduce_f64'
30; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1 = call double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
31; SSE41-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2 = call double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
32; SSE41-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4 = call double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
33; SSE41-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V8 = call double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
34; SSE41-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V16 = call double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
35; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
36;
37; SSE42-LABEL: 'reduce_f64'
38; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V1 = call double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
39; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2 = call double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
40; SSE42-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4 = call double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
41; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8 = call double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
42; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16 = call double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
43; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
44;
45; AVX1-LABEL: 'reduce_f64'
46; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V1 = call double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
47; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2 = call double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
48; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4 = call double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
49; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8 = call double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
50; AVX1-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16 = call double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
51; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
52;
53; AVX2-LABEL: 'reduce_f64'
54; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V1 = call double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
55; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2 = call double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
56; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4 = call double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
57; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8 = call double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
58; AVX2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16 = call double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
59; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
60;
61; AVX512-LABEL: 'reduce_f64'
62; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V1 = call double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
63; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2 = call double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
64; AVX512-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4 = call double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
65; AVX512-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8 = call double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
66; AVX512-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V16 = call double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
67; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
68;
69  %V1  = call double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
70  %V2  = call double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
71  %V4  = call double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
72  %V8  = call double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
73  %V16 = call double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
74  ret void
75}
76
77define void @reduce_f32(float %arg) {
78; SSE2-LABEL: 'reduce_f32'
79; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1 = call float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
80; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2 = call float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
81; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V4 = call float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
82; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8 = call float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
83; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V16 = call float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
84; SSE2-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V32 = call float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
85; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
86;
87; SSSE3-LABEL: 'reduce_f32'
88; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1 = call float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
89; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2 = call float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
90; SSSE3-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V4 = call float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
91; SSSE3-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8 = call float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
92; SSSE3-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V16 = call float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
93; SSSE3-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V32 = call float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
94; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
95;
96; SSE41-LABEL: 'reduce_f32'
97; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1 = call float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
98; SSE41-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2 = call float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
99; SSE41-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V4 = call float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
100; SSE41-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8 = call float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
101; SSE41-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V16 = call float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
102; SSE41-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V32 = call float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
103; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
104;
105; SSE42-LABEL: 'reduce_f32'
106; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V1 = call float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
107; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2 = call float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
108; SSE42-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4 = call float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
109; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8 = call float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
110; SSE42-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16 = call float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
111; SSE42-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V32 = call float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
112; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
113;
114; AVX1-LABEL: 'reduce_f32'
115; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V1 = call float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
116; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2 = call float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
117; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4 = call float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
118; AVX1-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8 = call float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
119; AVX1-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V16 = call float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
120; AVX1-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V32 = call float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
121; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
122;
123; AVX2-LABEL: 'reduce_f32'
124; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V1 = call float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
125; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2 = call float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
126; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4 = call float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
127; AVX2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8 = call float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
128; AVX2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V16 = call float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
129; AVX2-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V32 = call float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
130; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
131;
132; AVX512-LABEL: 'reduce_f32'
133; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V1 = call float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
134; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2 = call float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
135; AVX512-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4 = call float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
136; AVX512-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8 = call float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
137; AVX512-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V16 = call float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
138; AVX512-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V32 = call float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
139; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
140;
141  %V1  = call float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
142  %V2  = call float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
143  %V4  = call float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
144  %V8  = call float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
145  %V16 = call float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
146  %V32 = call float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
147  ret void
148}
149
150; Fast Reductions
151
152define void @reduce_f64_fast(double %arg) {
153; SSE2-LABEL: 'reduce_f64_fast'
154; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
155; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
156; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4 = call fast double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
157; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8 = call fast double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
158; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16 = call fast double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
159; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
160;
161; SSSE3-LABEL: 'reduce_f64_fast'
162; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
163; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
164; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4 = call fast double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
165; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8 = call fast double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
166; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16 = call fast double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
167; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
168;
169; SSE41-LABEL: 'reduce_f64_fast'
170; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
171; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
172; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4 = call fast double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
173; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8 = call fast double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
174; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16 = call fast double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
175; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
176;
177; SSE42-LABEL: 'reduce_f64_fast'
178; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
179; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
180; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V4 = call fast double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
181; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V8 = call fast double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
182; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V16 = call fast double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
183; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
184;
185; AVX1-LABEL: 'reduce_f64_fast'
186; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
187; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
188; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V4 = call fast double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
189; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V8 = call fast double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
190; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V16 = call fast double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
191; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
192;
193; AVX2-LABEL: 'reduce_f64_fast'
194; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
195; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
196; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V4 = call fast double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
197; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8 = call fast double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
198; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16 = call fast double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
199; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
200;
201; AVX512-LABEL: 'reduce_f64_fast'
202; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
203; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
204; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V4 = call fast double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
205; AVX512-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V8 = call fast double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
206; AVX512-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V16 = call fast double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
207; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
208;
209  %V1  = call fast double @llvm.vector.reduce.fadd.v1f64(double %arg, <1 x double> undef)
210  %V2  = call fast double @llvm.vector.reduce.fadd.v2f64(double %arg, <2 x double> undef)
211  %V4  = call fast double @llvm.vector.reduce.fadd.v4f64(double %arg, <4 x double> undef)
212  %V8  = call fast double @llvm.vector.reduce.fadd.v8f64(double %arg, <8 x double> undef)
213  %V16 = call fast double @llvm.vector.reduce.fadd.v16f64(double %arg, <16 x double> undef)
214  ret void
215}
216
217define void @reduce_f32_fast(float %arg) {
218; SSE2-LABEL: 'reduce_f32_fast'
219; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
220; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
221; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4 = call fast float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
222; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V8 = call fast float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
223; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V16 = call fast float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
224; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V32 = call fast float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
225; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
226;
227; SSSE3-LABEL: 'reduce_f32_fast'
228; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
229; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
230; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4 = call fast float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
231; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V8 = call fast float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
232; SSSE3-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V16 = call fast float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
233; SSSE3-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V32 = call fast float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
234; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
235;
236; SSE41-LABEL: 'reduce_f32_fast'
237; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
238; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
239; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4 = call fast float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
240; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V8 = call fast float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
241; SSE41-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V16 = call fast float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
242; SSE41-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V32 = call fast float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
243; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
244;
245; SSE42-LABEL: 'reduce_f32_fast'
246; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
247; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
248; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4 = call fast float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
249; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V8 = call fast float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
250; SSE42-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V16 = call fast float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
251; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V32 = call fast float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
252; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
253;
254; AVX1-LABEL: 'reduce_f32_fast'
255; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
256; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
257; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V4 = call fast float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
258; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8 = call fast float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
259; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16 = call fast float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
260; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V32 = call fast float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
261; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
262;
263; AVX2-LABEL: 'reduce_f32_fast'
264; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
265; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
266; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V4 = call fast float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
267; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8 = call fast float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
268; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V16 = call fast float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
269; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V32 = call fast float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
270; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
271;
272; AVX512-LABEL: 'reduce_f32_fast'
273; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1 = call fast float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
274; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2 = call fast float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
275; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V4 = call fast float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
276; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8 = call fast float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
277; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16 = call fast float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
278; AVX512-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V32 = call fast float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
279; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
280;
281  %V1  = call fast float @llvm.vector.reduce.fadd.v1f32(float %arg, <1 x float> undef)
282  %V2  = call fast float @llvm.vector.reduce.fadd.v2f32(float %arg, <2 x float> undef)
283  %V4  = call fast float @llvm.vector.reduce.fadd.v4f32(float %arg, <4 x float> undef)
284  %V8  = call fast float @llvm.vector.reduce.fadd.v8f32(float %arg, <8 x float> undef)
285  %V16 = call fast float @llvm.vector.reduce.fadd.v16f32(float %arg, <16 x float> undef)
286  %V32 = call fast float @llvm.vector.reduce.fadd.v32f32(float %arg, <32 x float> undef)
287  ret void
288}
289
290declare double @llvm.vector.reduce.fadd.v1f64(double, <1 x double>)
291declare double @llvm.vector.reduce.fadd.v2f64(double, <2 x double>)
292declare double @llvm.vector.reduce.fadd.v4f64(double, <4 x double>)
293declare double @llvm.vector.reduce.fadd.v8f64(double, <8 x double>)
294declare double @llvm.vector.reduce.fadd.v16f64(double, <16 x double>)
295
296declare float @llvm.vector.reduce.fadd.v1f32(float, <1 x float>)
297declare float @llvm.vector.reduce.fadd.v2f32(float, <2 x float>)
298declare float @llvm.vector.reduce.fadd.v4f32(float, <4 x float>)
299declare float @llvm.vector.reduce.fadd.v8f32(float, <8 x float>)
300declare float @llvm.vector.reduce.fadd.v16f32(float, <16 x float>)
301declare float @llvm.vector.reduce.fadd.v32f32(float, <32 x float>)
302