1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2021-2024 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; RUN: %opt %use_old_pass_manager% -GenXPatternMatch -march=genx64 -mcpu=Gen9 \
; RUN: -mtriple=spir64-unknown-unknown -S < %s | FileCheck %s
; CHECK-LABEL: @test_inverse
define <16 x float> @test_inverse(<16 x float> %val) {
%sqrt = call <16 x float> @llvm.genx.sqrt.v16f32(<16 x float> %val)
; CHECK: [[INVERSE_SQRT:%.*]] = call <16 x float> @llvm.genx.rsqrt.v16f32(<16 x float> %val)
; CHECK-NEXT: ret <16 x float> [[INVERSE_SQRT]]
%inv = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %sqrt)
ret <16 x float> %inv
}
; CHECK-LABEL: @test_inverse2
define <16 x float> @test_inverse2(<16 x float> %val1, <16 x float> %val2) {
; CHECK: [[INVERSE_SQRT21:%.*]] = call <16 x float> @llvm.genx.rsqrt.v16f32(<16 x float> %val1)
; CHECK-NEXT: [[INVERSE_SQRT22:%.*]] = call <16 x float> @llvm.genx.rsqrt.v16f32(<16 x float> %val2)
; CHECK-NEXT: [[FADD:%.*]] = fadd <16 x float> [[INVERSE_SQRT21]], [[INVERSE_SQRT22]]
; CHECK-NEXT: ret <16 x float> [[FADD]]
%sqrt1 = call fast <16 x float> @llvm.sqrt.v16f32(<16 x float> %val1)
%inv1 = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %sqrt1)
%sqrt2 = call <16 x float> @llvm.genx.sqrt.v16f32(<16 x float> %val2)
%inv2 = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %sqrt2)
%add = fadd <16 x float> %inv1, %inv2
ret <16 x float> %add
}
; CHECK-LABEL: @test_inverse_scalar
define float @test_inverse_scalar(float %val) {
%sqrt = call float @llvm.genx.sqrt.f32(float %val)
; CHECK: [[INVERSE_SQRT_SCALAR:%.*]] = call float @llvm.genx.rsqrt.f32(float %val)
; CHECK-NEXT: ret float [[INVERSE_SQRT_SCALAR]]
%inv = call float @llvm.genx.inv.f32(float %sqrt)
ret float %inv
}
; CHECK-LABEL: @test_inverse_fast
define <16 x float> @test_inverse_fast(<16 x float> %val) {
%sqrt = call fast <16 x float> @llvm.sqrt.v16f32(<16 x float> %val)
; CHECK: [[INVERSE_SQRT_FAST:%.*]] = call <16 x float> @llvm.genx.rsqrt.v16f32(<16 x float> %val)
; CHECK-NEXT: ret <16 x float> [[INVERSE_SQRT_FAST]]
%inv = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %sqrt)
ret <16 x float> %inv
}
; CHECK-LABEL: @test_inverse_not_fast
define <16 x float> @test_inverse_not_fast(<16 x float> %src) {
; CHECK: @llvm.genx.rsqrt.v16f32(<16 x float> %src)
%sqrt = call <16 x float> @llvm.sqrt.v16f32(<16 x float> %src)
%inv = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %sqrt)
ret <16 x float> %inv
}
; CHECK-LABEL: @test_inverse_double
define <16 x double> @test_inverse_double(<16 x double> %val_double) {
%sqrt = call <16 x double> @llvm.sqrt.v16f64(<16 x double> %val_double)
; CHECK: call <16 x double> @llvm.genx.rsqrt.v16f64(<16 x double> %val_double)
%inv = call <16 x double> @llvm.genx.inv.v16f64(<16 x double> %sqrt)
ret <16 x double> %inv
}
; CHECK-LABEL: @test_not_inverse_multiple_uses
define <16 x float> @test_not_inverse_multiple_uses(<16 x float> %val) {
; CHECK: call <16 x float> @llvm.sqrt.v16f32(<16 x float> %val)
; CHECK: call <16 x float> @llvm.genx.rsqrt.v16f32(<16 x float> %val)
; CHECK: fadd <16 x float>
%sqrt = call <16 x float> @llvm.sqrt.v16f32(<16 x float> %val)
%inv = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %sqrt)
%add = fadd <16 x float> %inv, %sqrt
ret <16 x float> %add
}
; CHECK-LABEL: @test_inverse3
define <16 x float> @test_inverse3(<16 x float> %val) {
%sqrt = call <16 x float> @llvm.genx.sqrt.v16f32(<16 x float> %val)
; CHECK: [[INVERSE_SQRT_FDIV:%.*]] = call <16 x float> @llvm.genx.rsqrt.v16f32(<16 x float> %val)
; CHECK-NEXT: ret <16 x float> [[INVERSE_SQRT_FDIV]]
%inv = fdiv fast <16 x float> <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>, %sqrt
ret <16 x float> %inv
}
; CHECK-LABEL: @test_inverse_of_not_sqrt
define <16 x float> @test_inverse_of_not_sqrt(<16 x float> %val, <16 x float> %tmp) {
; CHECK-NEXT: fsub <16 x float>
; CHECK-NEXT: call <16 x float> @llvm.genx.inv.v16f32
%sub = fsub <16 x float> %val, %tmp
%inv = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %sub)
ret <16 x float> %inv
}
; CHECK-LABEL: @test_sqrt_multiple_use
define <16 x float> @test_sqrt_multiple_use(<16 x float> %val) {
; CHECK-NEXT: call <16 x float> @llvm.genx.sqrt.v16f32(<16 x float> %val)
; CHECK-NEXT: call <16 x float> @llvm.genx.rsqrt.v16f32(<16 x float> %val)
; CHECK-NEXT: fadd <16 x float>
%sqrt = call <16 x float> @llvm.genx.sqrt.v16f32(<16 x float> %val)
%inv = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %sqrt)
%add = fadd <16 x float> %sqrt, %inv
ret <16 x float> %add
}
; CHECK-LABEL: @test_sqrt_inv_constant
define float @test_sqrt_inv_constant() {
; CHECK: ret float 0x3FB99999A0000000
%sqrt = call float @llvm.genx.sqrt.f32(float 100.0)
%inv = call float @llvm.genx.inv.f32(float %sqrt)
ret float %inv
}
; CHECK-LABEL: @test_sqrt_div_constant
define float @test_sqrt_div_constant() {
; CHECK: ret float 0x3FB99999A0000000
%sqrt = call float @llvm.genx.sqrt.f32(float 100.0)
%inv = fdiv arcp float 1.0, %sqrt
ret float %inv
}
; CHECK-LABEL: @test_sqrt_fast_inv_constant
define float @test_sqrt_fast_inv_constant() {
; CHECK: ret float 0x3FB99999A0000000
%sqrt = call afn float @llvm.sqrt.f32(float 100.0)
%inv = call float @llvm.genx.inv.f32(float %sqrt)
ret float %inv
}
; CHECK-LABEL: @test_sqrt_inv_constant_neg
define float @test_sqrt_inv_constant_neg() {
; CHECK: ret float 0x7FF8000000000000
%sqrt = call float @llvm.genx.sqrt.f32(float -100.0)
%inv = call float @llvm.genx.inv.f32(float %sqrt)
ret float %inv
}
; CHECK-LABEL: @test_sqrt_inv_constant_negzero
define float @test_sqrt_inv_constant_negzero() {
; CHECK: ret float 0xFFF0000000000000
%sqrt = call float @llvm.genx.sqrt.f32(float -0.0)
%inv = call float @llvm.genx.inv.f32(float %sqrt)
ret float %inv
}
; CHECK-LABEL: @test_sqrt_inv_constant_zero
define float @test_sqrt_inv_constant_zero() {
; CHECK: ret float 0x7FF0000000000000
%sqrt = call float @llvm.genx.sqrt.f32(float 0.0)
%inv = call float @llvm.genx.inv.f32(float %sqrt)
ret float %inv
}
; CHECK-LABEL: @test_sqrt_inv_constant_inf
define float @test_sqrt_inv_constant_inf() {
; CHECK: ret float 0.000000e+00
%sqrt = call float @llvm.genx.sqrt.f32(float 0x7FF0000000000000)
%inv = call float @llvm.genx.inv.f32(float %sqrt)
ret float %inv
}
; CHECK-LABEL: @test_sqrt_inv_constant_neginf
define float @test_sqrt_inv_constant_neginf() {
; CHECK: ret float 0x7FF8000000000000
%sqrt = call float @llvm.genx.sqrt.f32(float 0xFFF0000000000000)
%inv = call float @llvm.genx.inv.f32(float %sqrt)
ret float %inv
}
; CHECK-LABEL: @test_sqrt_inv_constant_undef
define float @test_sqrt_inv_constant_undef() {
; CHECK: ret float undef
%sqrt = call float @llvm.genx.sqrt.f32(float undef)
%inv = call float @llvm.genx.inv.f32(float %sqrt)
ret float %inv
}
; CHECK-LABEL: @test_sqrt_inv_constant_vector
define <2 x float> @test_sqrt_inv_constant_vector() {
; CHECK: ret <2 x float> <float 0x3FC99999A0000000, float 5.000000e-01>
%sqrt = call <2 x float> @llvm.genx.sqrt.v2f32(<2 x float> <float 25.0, float 4.0>)
%inv = call <2 x float> @llvm.genx.inv.v2f32(<2 x float> %sqrt)
ret <2 x float> %inv
}
; CHECK-LABEL: @test_sqrt_inv_constant_vector_zero
define <2 x float> @test_sqrt_inv_constant_vector_zero() {
; CHECK: ret <2 x float> <float 0x7FF0000000000000, float 0x7FF0000000000000>
%sqrt = call <2 x float> @llvm.genx.sqrt.v2f32(<2 x float> zeroinitializer)
%inv = call <2 x float> @llvm.genx.inv.v2f32(<2 x float> %sqrt)
ret <2 x float> %inv
}
; CHECK-LABEL: @test_inv_sqrt_1
define <16 x float> @test_inv_sqrt_1(<16 x float> %val) {
; CHECK: call <16 x float> @llvm.genx.rsqrt.v16f32(<16 x float> %val)
%inv = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %val)
%sqrt = call <16 x float> @llvm.genx.sqrt.v16f32(<16 x float> %inv)
ret <16 x float> %sqrt
}
; CHECK-LABEL: @test_inv_sqrt_2
define <16 x float> @test_inv_sqrt_2(<16 x float> %val) {
; CHECK: call <16 x float> @llvm.genx.rsqrt.v16f32(<16 x float> %val)
%inv = call <16 x float> @llvm.genx.inv.v16f32(<16 x float> %val)
%sqrt = call <16 x float> @llvm.sqrt.v16f32(<16 x float> %inv)
ret <16 x float> %sqrt
}
; CHECK-LABEL: @test_inv_sqrt_3
define float @test_inv_sqrt_3(float %val) {
; CHECK: call float @llvm.genx.rsqrt.f32(float %val)
%inv = fdiv float 1.0, %val
%sqrt = call float @llvm.genx.sqrt.f32(float %inv)
ret float %sqrt
}
; CHECK-LABEL: @test_inv_sqrt_4
define float @test_inv_sqrt_4(float %val) {
; CHECK: call float @llvm.genx.rsqrt.f32(float %val)
%inv = fdiv float 1.0, %val
%sqrt = call afn float @llvm.sqrt.f32(float %inv)
ret float %sqrt
}
; CHECK-LABEL: @test_inv_sqrt_5
define float @test_inv_sqrt_5(float %val) {
; CHECK: call float @llvm.genx.rsqrt.f32(float %val)
%inv = fdiv arcp float 1.0, %val
%sqrt = call float @llvm.sqrt.f32(float %inv)
ret float %sqrt
}
; CHECK-LABEL: @test_inv_sqrt_6
define float @test_inv_sqrt_6(float %val) {
; CHECK-NOT: call float @llvm.genx.rsqrt.f32(float %val)
%inv = fdiv float 1.0, %val
%sqrt = call float @llvm.sqrt.f32(float %inv)
ret float %sqrt
}
; CHECK-LABEL: @test_inverse_double_2
define <16 x double> @test_inverse_double_2(<16 x double> %val_double) {
%sqrt = call <16 x double> @llvm.genx.sqrt.v16f64(<16 x double> %val_double)
%div = fdiv <16 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>, %sqrt
; CHECK: call <16 x double> @llvm.genx.rsqrt.v16f64(<16 x double> %val_double)
ret <16 x double> %div
}
; CHECK-LABEL: @test_inverse_double_3
define <16 x double> @test_inverse_double_3(<16 x double> %val_double) {
%div = fdiv arcp <16 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>, %val_double
%sqrt = call <16 x double> @llvm.genx.sqrt.v16f64(<16 x double> %div)
; CHECK: call <16 x double> @llvm.genx.rsqrt.v16f64(<16 x double> %val_double)
ret <16 x double> %sqrt
}
declare float @llvm.sqrt.f32(float)
declare float @llvm.genx.sqrt.f32(float)
declare float @llvm.genx.inv.f32(float)
declare <2 x float> @llvm.genx.sqrt.v2f32(<2 x float>)
declare <2 x float> @llvm.genx.inv.v2f32(<2 x float>)
declare <16 x float> @llvm.sqrt.v16f32(<16 x float>)
declare <16 x double> @llvm.sqrt.v16f64(<16 x double>)
declare <16 x float> @llvm.genx.sqrt.v16f32(<16 x float>)
declare <16 x double> @llvm.genx.sqrt.v16f64(<16 x double>)
declare <16 x float> @llvm.genx.inv.v16f32(<16 x float>)
declare <16 x double> @llvm.genx.inv.v16f64(<16 x double>)
|
