1 // RUN: %clang_cc1 -fenable-matrix -triple x86_64-apple-darwin %s -emit-llvm -disable-llvm-passes -o - | FileCheck %s
2
3 typedef double dx5x5_t __attribute__((matrix_type(5, 5)));
4 typedef float fx2x3_t __attribute__((matrix_type(2, 3)));
5 typedef int ix9x3_t __attribute__((matrix_type(9, 3)));
6 typedef unsigned long long ullx4x2_t __attribute__((matrix_type(4, 2)));
7
8 // Floating point matrix/scalar additions.
9
add_matrix_matrix_double(dx5x5_t a,dx5x5_t b,dx5x5_t c)10 void add_matrix_matrix_double(dx5x5_t a, dx5x5_t b, dx5x5_t c) {
11 // CHECK-LABEL: define void @add_matrix_matrix_double(<25 x double> %a, <25 x double> %b, <25 x double> %c)
12 // CHECK: [[B:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
13 // CHECK-NEXT: [[C:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
14 // CHECK-NEXT: [[RES:%.*]] = fadd <25 x double> [[B]], [[C]]
15 // CHECK-NEXT: store <25 x double> [[RES]], <25 x double>* {{.*}}, align 8
16
17 a = b + c;
18 }
19
add_matrix_matrix_float(fx2x3_t a,fx2x3_t b,fx2x3_t c)20 void add_matrix_matrix_float(fx2x3_t a, fx2x3_t b, fx2x3_t c) {
21 // CHECK-LABEL: define void @add_matrix_matrix_float(<6 x float> %a, <6 x float> %b, <6 x float> %c)
22 // CHECK: [[B:%.*]] = load <6 x float>, <6 x float>* {{.*}}, align 4
23 // CHECK-NEXT: [[C:%.*]] = load <6 x float>, <6 x float>* {{.*}}, align 4
24 // CHECK-NEXT: [[RES:%.*]] = fadd <6 x float> [[B]], [[C]]
25 // CHECK-NEXT: store <6 x float> [[RES]], <6 x float>* {{.*}}, align 4
26
27 a = b + c;
28 }
29
add_matrix_scalar_double_float(dx5x5_t a,float vf)30 void add_matrix_scalar_double_float(dx5x5_t a, float vf) {
31 // CHECK-LABEL: define void @add_matrix_scalar_double_float(<25 x double> %a, float %vf)
32 // CHECK: [[MATRIX:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
33 // CHECK-NEXT: [[SCALAR:%.*]] = load float, float* %vf.addr, align 4
34 // CHECK-NEXT: [[SCALAR_EXT:%.*]] = fpext float [[SCALAR]] to double
35 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <25 x double> undef, double [[SCALAR_EXT]], i32 0
36 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <25 x double> [[SCALAR_EMBED]], <25 x double> undef, <25 x i32> zeroinitializer
37 // CHECK-NEXT: [[RES:%.*]] = fadd <25 x double> [[MATRIX]], [[SCALAR_EMBED1]]
38 // CHECK-NEXT: store <25 x double> [[RES]], <25 x double>* {{.*}}, align 8
39
40 a = a + vf;
41 }
42
add_matrix_scalar_double_double(dx5x5_t a,double vd)43 void add_matrix_scalar_double_double(dx5x5_t a, double vd) {
44 // CHECK-LABEL: define void @add_matrix_scalar_double_double(<25 x double> %a, double %vd)
45 // CHECK: [[MATRIX:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
46 // CHECK-NEXT: [[SCALAR:%.*]] = load double, double* %vd.addr, align 8
47 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <25 x double> undef, double [[SCALAR]], i32 0
48 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <25 x double> [[SCALAR_EMBED]], <25 x double> undef, <25 x i32> zeroinitializer
49 // CHECK-NEXT: [[RES:%.*]] = fadd <25 x double> [[MATRIX]], [[SCALAR_EMBED1]]
50 // CHECK-NEXT: store <25 x double> [[RES]], <25 x double>* {{.*}}, align 8
51
52 a = a + vd;
53 }
54
add_matrix_scalar_float_float(fx2x3_t b,float vf)55 void add_matrix_scalar_float_float(fx2x3_t b, float vf) {
56 // CHECK-LABEL: define void @add_matrix_scalar_float_float(<6 x float> %b, float %vf)
57 // CHECK: [[MATRIX:%.*]] = load <6 x float>, <6 x float>* {{.*}}, align 4
58 // CHECK-NEXT: [[SCALAR:%.*]] = load float, float* %vf.addr, align 4
59 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <6 x float> undef, float [[SCALAR]], i32 0
60 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <6 x float> [[SCALAR_EMBED]], <6 x float> undef, <6 x i32> zeroinitializer
61 // CHECK-NEXT: [[RES:%.*]] = fadd <6 x float> [[MATRIX]], [[SCALAR_EMBED1]]
62 // CHECK-NEXT: store <6 x float> [[RES]], <6 x float>* {{.*}}, align 4
63
64 b = b + vf;
65 }
66
add_matrix_scalar_float_double(fx2x3_t b,double vd)67 void add_matrix_scalar_float_double(fx2x3_t b, double vd) {
68 // CHECK-LABEL: define void @add_matrix_scalar_float_double(<6 x float> %b, double %vd)
69 // CHECK: [[MATRIX:%.*]] = load <6 x float>, <6 x float>* {{.*}}, align 4
70 // CHECK-NEXT: [[SCALAR:%.*]] = load double, double* %vd.addr, align 8
71 // CHECK-NEXT: [[SCALAR_TRUNC:%.*]] = fptrunc double [[SCALAR]] to float
72 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <6 x float> undef, float [[SCALAR_TRUNC]], i32 0
73 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <6 x float> [[SCALAR_EMBED]], <6 x float> undef, <6 x i32> zeroinitializer
74 // CHECK-NEXT: [[RES:%.*]] = fadd <6 x float> [[MATRIX]], [[SCALAR_EMBED1]]
75 // CHECK-NEXT: store <6 x float> [[RES]], <6 x float>* {{.*}}, align 4
76
77 b = b + vd;
78 }
79
80 // Integer matrix/scalar additions
81
add_matrix_matrix_int(ix9x3_t a,ix9x3_t b,ix9x3_t c)82 void add_matrix_matrix_int(ix9x3_t a, ix9x3_t b, ix9x3_t c) {
83 // CHECK-LABEL: define void @add_matrix_matrix_int(<27 x i32> %a, <27 x i32> %b, <27 x i32> %c)
84 // CHECK: [[B:%.*]] = load <27 x i32>, <27 x i32>* {{.*}}, align 4
85 // CHECK-NEXT: [[C:%.*]] = load <27 x i32>, <27 x i32>* {{.*}}, align 4
86 // CHECK-NEXT: [[RES:%.*]] = add <27 x i32> [[B]], [[C]]
87 // CHECK-NEXT: store <27 x i32> [[RES]], <27 x i32>* {{.*}}, align 4
88 a = b + c;
89 }
90
add_matrix_matrix_unsigned_long_long(ullx4x2_t a,ullx4x2_t b,ullx4x2_t c)91 void add_matrix_matrix_unsigned_long_long(ullx4x2_t a, ullx4x2_t b, ullx4x2_t c) {
92 // CHECK-LABEL: define void @add_matrix_matrix_unsigned_long_long(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c)
93 // CHECK: [[B:%.*]] = load <8 x i64>, <8 x i64>* {{.*}}, align 8
94 // CHECK-NEXT: [[C:%.*]] = load <8 x i64>, <8 x i64>* {{.*}}, align 8
95 // CHECK-NEXT: [[RES:%.*]] = add <8 x i64> [[B]], [[C]]
96 // CHECK-NEXT: store <8 x i64> [[RES]], <8 x i64>* {{.*}}, align 8
97
98 a = b + c;
99 }
100
add_matrix_scalar_int_short(ix9x3_t a,short vs)101 void add_matrix_scalar_int_short(ix9x3_t a, short vs) {
102 // CHECK-LABEL: define void @add_matrix_scalar_int_short(<27 x i32> %a, i16 signext %vs)
103 // CHECK: [[MATRIX:%.*]] = load <27 x i32>, <27 x i32>* [[MAT_ADDR:%.*]], align 4
104 // CHECK-NEXT: [[SCALAR:%.*]] = load i16, i16* %vs.addr, align 2
105 // CHECK-NEXT: [[SCALAR_EXT:%.*]] = sext i16 [[SCALAR]] to i32
106 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <27 x i32> undef, i32 [[SCALAR_EXT]], i32 0
107 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <27 x i32> [[SCALAR_EMBED]], <27 x i32> undef, <27 x i32> zeroinitializer
108 // CHECK-NEXT: [[RES:%.*]] = add <27 x i32> [[MATRIX]], [[SCALAR_EMBED1]]
109 // CHECK-NEXT: store <27 x i32> [[RES]], <27 x i32>* [[MAT_ADDR]], align 4
110
111 a = a + vs;
112 }
113
add_matrix_scalar_int_long_int(ix9x3_t a,long int vli)114 void add_matrix_scalar_int_long_int(ix9x3_t a, long int vli) {
115 // CHECK-LABEL: define void @add_matrix_scalar_int_long_int(<27 x i32> %a, i64 %vli)
116 // CHECK: [[MATRIX:%.*]] = load <27 x i32>, <27 x i32>* [[MAT_ADDR:%.*]], align 4
117 // CHECK-NEXT: [[SCALAR:%.*]] = load i64, i64* %vli.addr, align 8
118 // CHECK-NEXT: [[SCALAR_TRUNC:%.*]] = trunc i64 [[SCALAR]] to i32
119 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <27 x i32> undef, i32 [[SCALAR_TRUNC]], i32 0
120 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <27 x i32> [[SCALAR_EMBED]], <27 x i32> undef, <27 x i32> zeroinitializer
121 // CHECK-NEXT: [[RES:%.*]] = add <27 x i32> [[MATRIX]], [[SCALAR_EMBED1]]
122 // CHECK-NEXT: store <27 x i32> [[RES]], <27 x i32>* [[MAT_ADDR]], align 4
123
124 a = a + vli;
125 }
126
add_matrix_scalar_int_unsigned_long_long(ix9x3_t a,unsigned long long int vulli)127 void add_matrix_scalar_int_unsigned_long_long(ix9x3_t a, unsigned long long int vulli) {
128 // CHECK-LABEL: define void @add_matrix_scalar_int_unsigned_long_long(<27 x i32> %a, i64 %vulli)
129 // CHECK: [[MATRIX:%.*]] = load <27 x i32>, <27 x i32>* [[MAT_ADDR:%.*]], align 4
130 // CHECK-NEXT: [[SCALAR:%.*]] = load i64, i64* %vulli.addr, align 8
131 // CHECK-NEXT: [[SCALAR_TRUNC:%.*]] = trunc i64 [[SCALAR]] to i32
132 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <27 x i32> undef, i32 [[SCALAR_TRUNC]], i32 0
133 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <27 x i32> [[SCALAR_EMBED]], <27 x i32> undef, <27 x i32> zeroinitializer
134 // CHECK-NEXT: [[RES:%.*]] = add <27 x i32> [[MATRIX]], [[SCALAR_EMBED1]]
135 // CHECK-NEXT: store <27 x i32> [[RES]], <27 x i32>* [[MAT_ADDR]], align 4
136
137 a = a + vulli;
138 }
139
add_matrix_scalar_long_long_int_short(ullx4x2_t b,short vs)140 void add_matrix_scalar_long_long_int_short(ullx4x2_t b, short vs) {
141 // CHECK-LABEL: define void @add_matrix_scalar_long_long_int_short(<8 x i64> %b, i16 signext %vs)
142 // CHECK: [[SCALAR:%.*]] = load i16, i16* %vs.addr, align 2
143 // CHECK-NEXT: [[SCALAR_EXT:%.*]] = sext i16 [[SCALAR]] to i64
144 // CHECK-NEXT: [[MATRIX:%.*]] = load <8 x i64>, <8 x i64>* {{.*}}, align 8
145 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <8 x i64> undef, i64 [[SCALAR_EXT]], i32 0
146 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <8 x i64> [[SCALAR_EMBED]], <8 x i64> undef, <8 x i32> zeroinitializer
147 // CHECK-NEXT: [[RES:%.*]] = add <8 x i64> [[SCALAR_EMBED1]], [[MATRIX]]
148 // CHECK-NEXT: store <8 x i64> [[RES]], <8 x i64>* {{.*}}, align 8
149
150 b = vs + b;
151 }
152
add_matrix_scalar_long_long_int_int(ullx4x2_t b,long int vli)153 void add_matrix_scalar_long_long_int_int(ullx4x2_t b, long int vli) {
154 // CHECK-LABEL: define void @add_matrix_scalar_long_long_int_int(<8 x i64> %b, i64 %vli)
155 // CHECK: [[SCALAR:%.*]] = load i64, i64* %vli.addr, align 8
156 // CHECK-NEXT: [[MATRIX:%.*]] = load <8 x i64>, <8 x i64>* {{.*}}, align 8
157 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <8 x i64> undef, i64 [[SCALAR]], i32 0
158 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <8 x i64> [[SCALAR_EMBED]], <8 x i64> undef, <8 x i32> zeroinitializer
159 // CHECK-NEXT: [[RES:%.*]] = add <8 x i64> [[SCALAR_EMBED1]], [[MATRIX]]
160 // CHECK-NEXT: store <8 x i64> [[RES]], <8 x i64>* {{.*}}, align 8
161
162 b = vli + b;
163 }
164
add_matrix_scalar_long_long_int_unsigned_long_long(ullx4x2_t b,unsigned long long int vulli)165 void add_matrix_scalar_long_long_int_unsigned_long_long(ullx4x2_t b, unsigned long long int vulli) {
166 // CHECK-LABEL: define void @add_matrix_scalar_long_long_int_unsigned_long_long
167 // CHECK: [[SCALAR:%.*]] = load i64, i64* %vulli.addr, align 8
168 // CHECK-NEXT: [[MATRIX:%.*]] = load <8 x i64>, <8 x i64>* %0, align 8
169 // CHECK-NEXT: [[SCALAR_EMBED:%.*]] = insertelement <8 x i64> undef, i64 [[SCALAR]], i32 0
170 // CHECK-NEXT: [[SCALAR_EMBED1:%.*]] = shufflevector <8 x i64> [[SCALAR_EMBED]], <8 x i64> undef, <8 x i32> zeroinitializer
171 // CHECK-NEXT: [[RES:%.*]] = add <8 x i64> [[SCALAR_EMBED1]], [[MATRIX]]
172 // CHECK-NEXT: store <8 x i64> [[RES]], <8 x i64>* {{.*}}, align 8
173 b = vulli + b;
174 }
175
176 // Tests for matrix multiplication.
177
multiply_matrix_matrix_double(dx5x5_t b,dx5x5_t c)178 void multiply_matrix_matrix_double(dx5x5_t b, dx5x5_t c) {
179 // CHECK-LABEL: @multiply_matrix_matrix_double(
180 // CHECK: [[B:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
181 // CHECK-NEXT: [[C:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
182 // CHECK-NEXT: [[RES:%.*]] = call <25 x double> @llvm.matrix.multiply.v25f64.v25f64.v25f64(<25 x double> [[B]], <25 x double> [[C]], i32 5, i32 5, i32 5)
183 // CHECK-NEXT: [[A_ADDR:%.*]] = bitcast [25 x double]* %a to <25 x double>*
184 // CHECK-NEXT: store <25 x double> [[RES]], <25 x double>* [[A_ADDR]], align 8
185 // CHECK-NEXT: ret void
186 //
187
188 dx5x5_t a;
189 a = b * c;
190 }
191
192 typedef int ix3x9_t __attribute__((matrix_type(3, 9)));
193 typedef int ix9x9_t __attribute__((matrix_type(9, 9)));
194 // CHECK-LABEL: @multiply_matrix_matrix_int(
195 // CHECK: [[B:%.*]] = load <27 x i32>, <27 x i32>* {{.*}}, align 4
196 // CHECK-NEXT: [[C:%.*]] = load <27 x i32>, <27 x i32>* {{.*}}, align 4
197 // CHECK-NEXT: [[RES:%.*]] = call <81 x i32> @llvm.matrix.multiply.v81i32.v27i32.v27i32(<27 x i32> [[B]], <27 x i32> [[C]], i32 9, i32 3, i32 9)
198 // CHECK-NEXT: [[A_ADDR:%.*]] = bitcast [81 x i32]* %a to <81 x i32>*
199 // CHECK-NEXT: store <81 x i32> [[RES]], <81 x i32>* [[A_ADDR]], align 4
200 // CHECK-NEXT: ret void
201 //
multiply_matrix_matrix_int(ix9x3_t b,ix3x9_t c)202 void multiply_matrix_matrix_int(ix9x3_t b, ix3x9_t c) {
203 ix9x9_t a;
204 a = b * c;
205 }
206
207 // CHECK-LABEL: @multiply_double_matrix_scalar_float(
208 // CHECK: [[A:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
209 // CHECK-NEXT: [[S:%.*]] = load float, float* %s.addr, align 4
210 // CHECK-NEXT: [[S_EXT:%.*]] = fpext float [[S]] to double
211 // CHECK-NEXT: [[VECINSERT:%.*]] = insertelement <25 x double> undef, double [[S_EXT]], i32 0
212 // CHECK-NEXT: [[VECSPLAT:%.*]] = shufflevector <25 x double> [[VECINSERT]], <25 x double> undef, <25 x i32> zeroinitializer
213 // CHECK-NEXT: [[RES:%.*]] = fmul <25 x double> [[A]], [[VECSPLAT]]
214 // CHECK-NEXT: store <25 x double> [[RES]], <25 x double>* {{.*}}, align 8
215 // CHECK-NEXT: ret void
216 //
multiply_double_matrix_scalar_float(dx5x5_t a,float s)217 void multiply_double_matrix_scalar_float(dx5x5_t a, float s) {
218 a = a * s;
219 }
220
221 // CHECK-LABEL: @multiply_double_matrix_scalar_double(
222 // CHECK: [[A:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
223 // CHECK-NEXT: [[S:%.*]] = load double, double* %s.addr, align 8
224 // CHECK-NEXT: [[VECINSERT:%.*]] = insertelement <25 x double> undef, double [[S]], i32 0
225 // CHECK-NEXT: [[VECSPLAT:%.*]] = shufflevector <25 x double> [[VECINSERT]], <25 x double> undef, <25 x i32> zeroinitializer
226 // CHECK-NEXT: [[RES:%.*]] = fmul <25 x double> [[A]], [[VECSPLAT]]
227 // CHECK-NEXT: store <25 x double> [[RES]], <25 x double>* {{.*}}, align 8
228 // CHECK-NEXT: ret void
229 //
multiply_double_matrix_scalar_double(dx5x5_t a,double s)230 void multiply_double_matrix_scalar_double(dx5x5_t a, double s) {
231 a = a * s;
232 }
233
234 // CHECK-LABEL: @multiply_float_matrix_scalar_double(
235 // CHECK: [[S:%.*]] = load double, double* %s.addr, align 8
236 // CHECK-NEXT: [[S_TRUNC:%.*]] = fptrunc double [[S]] to float
237 // CHECK-NEXT: [[MAT:%.*]] = load <6 x float>, <6 x float>* [[MAT_ADDR:%.*]], align 4
238 // CHECK-NEXT: [[VECINSERT:%.*]] = insertelement <6 x float> undef, float [[S_TRUNC]], i32 0
239 // CHECK-NEXT: [[VECSPLAT:%.*]] = shufflevector <6 x float> [[VECINSERT]], <6 x float> undef, <6 x i32> zeroinitializer
240 // CHECK-NEXT: [[RES:%.*]] = fmul <6 x float> [[VECSPLAT]], [[MAT]]
241 // CHECK-NEXT: store <6 x float> [[RES]], <6 x float>* [[MAT_ADDR]], align 4
242 // CHECK-NEXT: ret void
243 //
multiply_float_matrix_scalar_double(fx2x3_t b,double s)244 void multiply_float_matrix_scalar_double(fx2x3_t b, double s) {
245 b = s * b;
246 }
247
248 // CHECK-LABEL: @multiply_int_matrix_scalar_short(
249 // CHECK: [[S:%.*]] = load i16, i16* %s.addr, align 2
250 // CHECK-NEXT: [[S_EXT:%.*]] = sext i16 [[S]] to i32
251 // CHECK-NEXT: [[MAT:%.*]] = load <27 x i32>, <27 x i32>* [[MAT_ADDR:%.*]], align 4
252 // CHECK-NEXT: [[VECINSERT:%.*]] = insertelement <27 x i32> undef, i32 [[S_EXT]], i32 0
253 // CHECK-NEXT: [[VECSPLAT:%.*]] = shufflevector <27 x i32> [[VECINSERT]], <27 x i32> undef, <27 x i32> zeroinitializer
254 // CHECK-NEXT: [[RES:%.*]] = mul <27 x i32> [[VECSPLAT]], [[MAT]]
255 // CHECK-NEXT: store <27 x i32> [[RES]], <27 x i32>* [[MAT_ADDR]], align 4
256 // CHECK-NEXT: ret void
257 //
multiply_int_matrix_scalar_short(ix9x3_t b,short s)258 void multiply_int_matrix_scalar_short(ix9x3_t b, short s) {
259 b = s * b;
260 }
261
262 // CHECK-LABEL: @multiply_int_matrix_scalar_ull(
263 // CHECK: [[MAT:%.*]] = load <27 x i32>, <27 x i32>* [[MAT_ADDR:%.*]], align 4
264 // CHECK-NEXT: [[S:%.*]] = load i64, i64* %s.addr, align 8
265 // CHECK-NEXT: [[S_TRUNC:%.*]] = trunc i64 [[S]] to i32
266 // CHECK-NEXT: [[VECINSERT:%.*]] = insertelement <27 x i32> undef, i32 [[S_TRUNC]], i32 0
267 // CHECK-NEXT: [[VECSPLAT:%.*]] = shufflevector <27 x i32> [[VECINSERT]], <27 x i32> undef, <27 x i32> zeroinitializer
268 // CHECK-NEXT: [[RES:%.*]] = mul <27 x i32> [[MAT]], [[VECSPLAT]]
269 // CHECK-NEXT: store <27 x i32> [[RES]], <27 x i32>* [[MAT_ADDR]], align 4
270 // CHECK-NEXT: ret void
271 //
multiply_int_matrix_scalar_ull(ix9x3_t b,unsigned long long s)272 void multiply_int_matrix_scalar_ull(ix9x3_t b, unsigned long long s) {
273 b = b * s;
274 }
275
276 // CHECK-LABEL: @multiply_float_matrix_constant(
277 // CHECK-NEXT: entry:
278 // CHECK-NEXT: [[A_ADDR:%.*]] = alloca [6 x float], align 4
279 // CHECK-NEXT: [[MAT_ADDR:%.*]] = bitcast [6 x float]* [[A_ADDR]] to <6 x float>*
280 // CHECK-NEXT: store <6 x float> [[A:%.*]], <6 x float>* [[MAT_ADDR]], align 4
281 // CHECK-NEXT: [[MAT:%.*]] = load <6 x float>, <6 x float>* [[MAT_ADDR]], align 4
282 // CHECK-NEXT: [[RES:%.*]] = fmul <6 x float> [[MAT]], <float 2.500000e+00, float 2.500000e+00, float 2.500000e+00, float 2.500000e+00, float 2.500000e+00, float 2.500000e+00>
283 // CHECK-NEXT: store <6 x float> [[RES]], <6 x float>* [[MAT_ADDR]], align 4
284 // CHECK-NEXT: ret void
285 //
multiply_float_matrix_constant(fx2x3_t a)286 void multiply_float_matrix_constant(fx2x3_t a) {
287 a = a * 2.5;
288 }
289
290 // CHECK-LABEL: @multiply_int_matrix_constant(
291 // CHECK-NEXT: entry:
292 // CHECK-NEXT: [[A_ADDR:%.*]] = alloca [27 x i32], align 4
293 // CHECK-NEXT: [[MAT_ADDR:%.*]] = bitcast [27 x i32]* [[A_ADDR]] to <27 x i32>*
294 // CHECK-NEXT: store <27 x i32> [[A:%.*]], <27 x i32>* [[MAT_ADDR]], align 4
295 // CHECK-NEXT: [[MAT:%.*]] = load <27 x i32>, <27 x i32>* [[MAT_ADDR]], align 4
296 // CHECK-NEXT: [[RES:%.*]] = mul <27 x i32> <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>, [[MAT]]
297 // CHECK-NEXT: store <27 x i32> [[RES]], <27 x i32>* [[MAT_ADDR]], align 4
298 // CHECK-NEXT: ret void
299 //
multiply_int_matrix_constant(ix9x3_t a)300 void multiply_int_matrix_constant(ix9x3_t a) {
301 a = 5 * a;
302 }
303
304 // Tests for the matrix type operators.
305
306 typedef double dx5x5_t __attribute__((matrix_type(5, 5)));
307 typedef float fx2x3_t __attribute__((matrix_type(2, 3)));
308
309 // Check that we can use matrix index expression on different floating point
310 // matrixes and indices.
insert_double_matrix_const_idx_ll_u_double(dx5x5_t a,double d,fx2x3_t b,float e,int j,unsigned k)311 void insert_double_matrix_const_idx_ll_u_double(dx5x5_t a, double d, fx2x3_t b, float e, int j, unsigned k) {
312 // CHECK-LABEL: @insert_double_matrix_const_idx_ll_u_double(
313 // CHECK: [[D:%.*]] = load double, double* %d.addr, align 8
314 // CHECK-NEXT: [[MAT:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
315 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <25 x double> [[MAT]], double [[D]], i64 5
316 // CHECK-NEXT: store <25 x double> [[MATINS]], <25 x double>* {{.*}}, align 8
317 // CHECK-NEXT: ret void
318
319 a[0ll][1u] = d;
320 }
321
insert_double_matrix_const_idx_i_u_double(dx5x5_t a,double d)322 void insert_double_matrix_const_idx_i_u_double(dx5x5_t a, double d) {
323 // CHECK-LABEL: @insert_double_matrix_const_idx_i_u_double(
324 // CHECK: [[D:%.*]] = load double, double* %d.addr, align 8
325 // CHECK-NEXT: [[MAT:%.*]] = load <25 x double>, <25 x double>* [[MAT_ADDR:%.*]], align 8
326 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <25 x double> [[MAT]], double [[D]], i64 21
327 // CHECK-NEXT: store <25 x double> [[MATINS]], <25 x double>* [[MAT_ADDR]], align 8
328 // CHECK-NEXT: ret void
329
330 a[1][4u] = d;
331 }
332
insert_float_matrix_const_idx_ull_i_float(fx2x3_t b,float e)333 void insert_float_matrix_const_idx_ull_i_float(fx2x3_t b, float e) {
334 // CHECK-LABEL: @insert_float_matrix_const_idx_ull_i_float(
335 // CHECK: [[E:%.*]] = load float, float* %e.addr, align 4
336 // CHECK-NEXT: [[MAT:%.*]] = load <6 x float>, <6 x float>* [[MAT_ADDR:%.*]], align 4
337 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <6 x float> [[MAT]], float [[E]], i64 3
338 // CHECK-NEXT: store <6 x float> [[MATINS]], <6 x float>* [[MAT_ADDR]], align 4
339 // CHECK-NEXT: ret void
340
341 b[1ull][1] = e;
342 }
343
insert_float_matrix_idx_i_u_float(fx2x3_t b,float e,int j,unsigned k)344 void insert_float_matrix_idx_i_u_float(fx2x3_t b, float e, int j, unsigned k) {
345 // CHECK-LABEL: @insert_float_matrix_idx_i_u_float(
346 // CHECK: [[E:%.*]] = load float, float* %e.addr, align 4
347 // CHECK-NEXT: [[J:%.*]] = load i32, i32* %j.addr, align 4
348 // CHECK-NEXT: [[J_EXT:%.*]] = sext i32 [[J]] to i64
349 // CHECK-NEXT: [[K:%.*]] = load i32, i32* %k.addr, align 4
350 // CHECK-NEXT: [[K_EXT:%.*]] = zext i32 [[K]] to i64
351 // CHECK-NEXT: [[IDX1:%.*]] = mul i64 [[K_EXT]], 2
352 // CHECK-NEXT: [[IDX2:%.*]] = add i64 [[IDX1]], [[J_EXT]]
353 // CHECK-NEXT: [[MAT:%.*]] = load <6 x float>, <6 x float>* [[MAT_ADDR:%.*]], align 4
354 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <6 x float> [[MAT]], float [[E]], i64 [[IDX2]]
355 // CHECK-NEXT: store <6 x float> [[MATINS]], <6 x float>* [[MAT_ADDR]], align 4
356 // CHECK-NEXT: ret void
357
358 b[j][k] = e;
359 }
360
insert_float_matrix_idx_s_ull_float(fx2x3_t b,float e,short j,unsigned long long k)361 void insert_float_matrix_idx_s_ull_float(fx2x3_t b, float e, short j, unsigned long long k) {
362 // CHECK-LABEL: @insert_float_matrix_idx_s_ull_float(
363 // CHECK: [[E:%.*]] = load float, float* %e.addr, align 4
364 // CHECK-NEXT: [[J:%.*]] = load i16, i16* %j.addr, align 2
365 // CHECK-NEXT: [[J_EXT:%.*]] = sext i16 [[J]] to i64
366 // CHECK-NEXT: [[K:%.*]] = load i64, i64* %k.addr, align 8
367 // CHECK-NEXT: [[IDX1:%.*]] = mul i64 [[K]], 2
368 // CHECK-NEXT: [[IDX2:%.*]] = add i64 [[IDX1]], [[J_EXT]]
369 // CHECK-NEXT: [[MAT:%.*]] = load <6 x float>, <6 x float>* [[MAT_ADDR:%.*]], align 4
370 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <6 x float> [[MAT]], float [[E]], i64 [[IDX2]]
371 // CHECK-NEXT: store <6 x float> [[MATINS]], <6 x float>* [[MAT_ADDR]], align 4
372 // CHECK-NEXT: ret void
373
374 (b)[j][k] = e;
375 }
376
377 // Check that we can can use matrix index expressions on integer matrixes.
378 typedef int ix9x3_t __attribute__((matrix_type(9, 3)));
insert_int_idx_expr(ix9x3_t a,int i)379 void insert_int_idx_expr(ix9x3_t a, int i) {
380 // CHECK-LABEL: @insert_int_idx_expr(
381 // CHECK: [[I1:%.*]] = load i32, i32* %i.addr, align 4
382 // CHECK-NEXT: [[I2:%.*]] = load i32, i32* %i.addr, align 4
383 // CHECK-NEXT: [[I2_ADD:%.*]] = add nsw i32 4, [[I2]]
384 // CHECK-NEXT: [[ADD_EXT:%.*]] = sext i32 [[I2_ADD]] to i64
385 // CHECK-NEXT: [[IDX2:%.*]] = add i64 18, [[ADD_EXT]]
386 // CHECK-NEXT: [[MAT:%.*]] = load <27 x i32>, <27 x i32>* [[MAT_ADDR:%.*]], align 4
387 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <27 x i32> [[MAT]], i32 [[I1]], i64 [[IDX2]]
388 // CHECK-NEXT: store <27 x i32> [[MATINS]], <27 x i32>* [[MAT_ADDR]], align 4
389 // CHECK-NEXT: ret void
390
391 a[4 + i][1 + 1u] = i;
392 }
393
394 // Check that we can can use matrix index expressions on FP and integer
395 // matrixes.
396 typedef int ix9x3_t __attribute__((matrix_type(9, 3)));
insert_float_into_int_matrix(ix9x3_t * a,int i)397 void insert_float_into_int_matrix(ix9x3_t *a, int i) {
398 // CHECK-LABEL: @insert_float_into_int_matrix(
399 // CHECK: [[I:%.*]] = load i32, i32* %i.addr, align 4
400 // CHECK-NEXT: [[MAT_ADDR1:%.*]] = load [27 x i32]*, [27 x i32]** %a.addr, align 8
401 // CHECK-NEXT: [[MAT_ADDR2:%.*]] = bitcast [27 x i32]* [[MAT_ADDR1]] to <27 x i32>*
402 // CHECK-NEXT: [[MAT:%.*]] = load <27 x i32>, <27 x i32>* [[MAT_ADDR2]], align 4
403 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <27 x i32> [[MAT]], i32 [[I]], i64 13
404 // CHECK-NEXT: store <27 x i32> [[MATINS]], <27 x i32>* [[MAT_ADDR2]], align 4
405 // CHECK-NEXT: ret void
406
407 (*a)[4][1] = i;
408 }
409
410 // Check that we can use overloaded matrix index expressions on matrixes with
411 // matching dimensions, but different element types.
412 typedef double dx3x3_t __attribute__((matrix_type(3, 3)));
413 typedef float fx3x3_t __attribute__((matrix_type(3, 3)));
insert_matching_dimensions1(dx3x3_t a,double i)414 void insert_matching_dimensions1(dx3x3_t a, double i) {
415 // CHECK-LABEL: @insert_matching_dimensions1(
416 // CHECK: [[I:%.*]] = load double, double* %i.addr, align 8
417 // CHECK-NEXT: [[MAT:%.*]] = load <9 x double>, <9 x double>* [[MAT_ADDR:%.*]], align 8
418 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <9 x double> [[MAT]], double [[I]], i64 5
419 // CHECK-NEXT: store <9 x double> [[MATINS]], <9 x double>* [[MAT_ADDR]], align 8
420 // CHECK-NEXT: ret void
421
422 a[2u][1u] = i;
423 }
424
insert_matching_dimensions(fx3x3_t b,float e)425 void insert_matching_dimensions(fx3x3_t b, float e) {
426 // CHECK-LABEL: @insert_matching_dimensions(
427 // CHECK: [[E:%.*]] = load float, float* %e.addr, align 4
428 // CHECK-NEXT: [[MAT:%.*]] = load <9 x float>, <9 x float>* [[MAT_ADDR:%.*]], align 4
429 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <9 x float> [[MAT]], float [[E]], i64 7
430 // CHECK-NEXT: store <9 x float> [[MATINS]], <9 x float>* [[MAT_ADDR]], align 4
431 // CHECK-NEXT: ret void
432
433 b[1u][2u] = e;
434 }
435
extract_double(dx5x5_t a)436 double extract_double(dx5x5_t a) {
437 // CHECK-LABEL: @extract_double(
438 // CHECK: [[MAT:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
439 // CHECK-NEXT: [[MATEXT:%.*]] = extractelement <25 x double> [[MAT]], i64 12
440 // CHECK-NEXT: ret double [[MATEXT]]
441
442 return a[2][3 - 1u];
443 }
444
extract_float(fx3x3_t b)445 double extract_float(fx3x3_t b) {
446 // CHECK-LABEL: @extract_float(
447 // CHECK: [[MAT:%.*]] = load <9 x float>, <9 x float>* {{.*}}, align 4
448 // CHECK-NEXT: [[MATEXT:%.*]] = extractelement <9 x float> [[MAT]], i64 5
449 // CHECK-NEXT: [[TO_DOUBLE:%.*]] = fpext float [[MATEXT]] to double
450 // CHECK-NEXT: ret double [[TO_DOUBLE]]
451
452 return b[2][1];
453 }
454
extract_int(ix9x3_t c,unsigned long j)455 int extract_int(ix9x3_t c, unsigned long j) {
456 // CHECK-LABEL: @extract_int(
457 // CHECK: [[J1:%.*]] = load i64, i64* %j.addr, align 8
458 // CHECK-NEXT: [[J2:%.*]] = load i64, i64* %j.addr, align 8
459 // CHECK-NEXT: [[MAT:%.*]] = load <27 x i32>, <27 x i32>* {{.*}}, align 4
460 // CHECK-NEXT: [[IDX1:%.*]] = mul i64 [[J2]], 9
461 // CHECK-NEXT: [[IDX2:%.*]] = add i64 [[IDX1]], [[J1]]
462 // CHECK-NEXT: [[MATEXT:%.*]] = extractelement <27 x i32> [[MAT]], i64 [[IDX2]]
463 // CHECK-NEXT: ret i32 [[MATEXT]]
464
465 return c[j][j];
466 }
467
468 typedef double dx3x2_t __attribute__((matrix_type(3, 2)));
469
test_extract_matrix_pointer1(dx3x2_t ** ptr,unsigned j)470 double test_extract_matrix_pointer1(dx3x2_t **ptr, unsigned j) {
471 // CHECK-LABEL: @test_extract_matrix_pointer1(
472 // CHECK: [[J:%.*]] = load i32, i32* %j.addr, align 4
473 // CHECK-NEXT: [[J_EXT:%.*]] = zext i32 [[J]] to i64
474 // CHECK-NEXT: [[PTR:%.*]] = load [6 x double]**, [6 x double]*** %ptr.addr, align 8
475 // CHECK-NEXT: [[PTR_IDX:%.*]] = getelementptr inbounds [6 x double]*, [6 x double]** [[PTR]], i64 1
476 // CHECK-NEXT: [[PTR2:%.*]] = load [6 x double]*, [6 x double]** [[PTR_IDX]], align 8
477 // CHECK-NEXT: [[PTR2_IDX:%.*]] = getelementptr inbounds [6 x double], [6 x double]* [[PTR2]], i64 2
478 // CHECK-NEXT: [[MAT_ADDR:%.*]] = bitcast [6 x double]* [[PTR2_IDX]] to <6 x double>*
479 // CHECK-NEXT: [[MAT:%.*]] = load <6 x double>, <6 x double>* [[MAT_ADDR]], align 8
480 // CHECK-NEXT: [[IDX:%.*]] = add i64 3, [[J_EXT]]
481 // CHECK-NEXT: [[MATEXT:%.*]] = extractelement <6 x double> [[MAT]], i64 [[IDX]]
482 // CHECK-NEXT: ret double [[MATEXT]]
483
484 return ptr[1][2][j][1];
485 }
486
test_extract_matrix_pointer2(dx3x2_t ** ptr)487 double test_extract_matrix_pointer2(dx3x2_t **ptr) {
488 // CHECK-LABEL: @test_extract_matrix_pointer2(
489 // CHECK-NEXT: entry:
490 // CHECK: [[PTR:%.*]] = load [6 x double]**, [6 x double]*** %ptr.addr, align 8
491 // CHECK-NEXT: [[PTR_IDX:%.*]] = getelementptr inbounds [6 x double]*, [6 x double]** [[PTR]], i64 4
492 // CHECK-NEXT: [[PTR2:%.*]] = load [6 x double]*, [6 x double]** [[PTR_IDX]], align 8
493 // CHECK-NEXT: [[PTR2_IDX:%.*]] = getelementptr inbounds [6 x double], [6 x double]* [[PTR2]], i64 6
494 // CHECK-NEXT: [[MAT_ADDR:%.*]] = bitcast [6 x double]* [[PTR2_IDX]] to <6 x double>*
495 // CHECK-NEXT: [[MAT:%.*]] = load <6 x double>, <6 x double>* [[MAT_ADDR]], align 8
496 // CHECK-NEXT: [[MATEXT:%.*]] = extractelement <6 x double> [[MAT]], i64 5
497 // CHECK-NEXT: ret double [[MATEXT]]
498
499 return (*(*(ptr + 4) + 6))[2][1 * 3 - 2];
500 }
501
insert_extract(dx5x5_t a,fx3x3_t b,unsigned long j,short k)502 void insert_extract(dx5x5_t a, fx3x3_t b, unsigned long j, short k) {
503 // CHECK-LABEL: @insert_extract(
504 // CHECK: [[K:%.*]] = load i16, i16* %k.addr, align 2
505 // CHECK-NEXT: [[K_EXT:%.*]] = sext i16 [[K]] to i64
506 // CHECK-NEXT: [[MAT:%.*]] = load <9 x float>, <9 x float>* [[MAT_ADDR:%.*]], align 4
507 // CHECK-NEXT: [[IDX1:%.*]] = mul i64 [[K_EXT]], 3
508 // CHECK-NEXT: [[IDX2:%.*]] = add i64 [[IDX1]], 0
509 // CHECK-NEXT: [[MATEXT:%.*]] = extractelement <9 x float> [[MAT]], i64 [[IDX]]
510 // CHECK-NEXT: [[J:%.*]] = load i64, i64* %j.addr, align 8
511 // CHECK-NEXT: [[IDX3:%.*]] = mul i64 [[J]], 3
512 // CHECK-NEXT: [[IDX4:%.*]] = add i64 [[IDX3]], 2
513 // CHECK-NEXT: [[MAT2:%.*]] = load <9 x float>, <9 x float>* [[MAT_ADDR]], align 4
514 // CHECK-NEXT: [[MATINS:%.*]] = insertelement <9 x float> [[MAT2]], float [[MATEXT]], i64 [[IDX4]]
515 // CHECK-NEXT: store <9 x float> [[MATINS]], <9 x float>* [[MAT_ADDR]], align 4
516 // CHECK-NEXT: ret void
517
518 b[2][j] = b[0][k];
519 }
520
insert_compound_stmt(dx5x5_t a)521 void insert_compound_stmt(dx5x5_t a) {
522 // CHECK-LABEL: define void @insert_compound_stmt(<25 x double> %a)
523 // CHECK: [[A:%.*]] = load <25 x double>, <25 x double>* [[A_PTR:%.*]], align 8
524 // CHECK-NEXT: [[EXT:%.*]] = extractelement <25 x double> [[A]], i64 17
525 // CHECK-NEXT: [[SUB:%.*]] = fsub double [[EXT]], 1.000000e+00
526 // CHECK-NEXT: [[A2:%.*]] = load <25 x double>, <25 x double>* [[A_PTR]], align 8
527 // CHECK-NEXT: [[INS:%.*]] = insertelement <25 x double> [[A2]], double [[SUB]], i64 17
528 // CHECK-NEXT: store <25 x double> [[INS]], <25 x double>* [[A_PTR]], align 8
529 // CHECK-NEXT: ret void
530
531 a[2][3] -= 1.0;
532 }
533
534 struct Foo {
535 fx2x3_t mat;
536 };
537
insert_compound_stmt_field(struct Foo * a,float f,unsigned i,unsigned j)538 void insert_compound_stmt_field(struct Foo *a, float f, unsigned i, unsigned j) {
539 // CHECK-LABEL: define void @insert_compound_stmt_field(%struct.Foo* %a, float %f, i32 %i, i32 %j)
540 // CHECK: [[I:%.*]] = load i32, i32* %i.addr, align 4
541 // CHECK-NEXT: [[I_EXT:%.*]] = zext i32 [[I]] to i64
542 // CHECK-NEXT: [[J:%.*]] = load i32, i32* %j.addr, align 4
543 // CHECK-NEXT: [[J_EXT:%.*]] = zext i32 [[J]] to i64
544 // CHECK-NEXT: [[IDX1:%.*]] = mul i64 [[J_EXT]], 2
545 // CHECK-NEXT: [[IDX2:%.*]] = add i64 [[IDX1]], [[I_EXT]]
546 // CHECK-NEXT: [[MAT_PTR:%.*]] = bitcast [6 x float]* %mat to <6 x float>*
547 // CHECK-NEXT: [[MAT:%.*]] = load <6 x float>, <6 x float>* [[MAT_PTR]], align 4
548 // CHECK-NEXT: [[EXT:%.*]] = extractelement <6 x float> [[MAT]], i64 [[IDX2]]
549 // CHECK-NEXT: [[SUM:%.*]] = fadd float [[EXT]], {{.*}}
550 // CHECK-NEXT: [[MAT2:%.*]] = load <6 x float>, <6 x float>* [[MAT_PTR]], align 4
551 // CHECK-NEXT: [[INS:%.*]] = insertelement <6 x float> [[MAT2]], float [[SUM]], i64 [[IDX2]]
552 // CHECK-NEXT: store <6 x float> [[INS]], <6 x float>* [[MAT_PTR]], align 4
553 // CHECK-NEXT: ret void
554
555 a->mat[i][j] += f;
556 }
557
matrix_as_idx(ix9x3_t a,int i,int j,dx5x5_t b)558 void matrix_as_idx(ix9x3_t a, int i, int j, dx5x5_t b) {
559 // CHECK-LABEL: define void @matrix_as_idx(<27 x i32> %a, i32 %i, i32 %j, <25 x double> %b)
560 // CHECK: [[I1:%.*]] = load i32, i32* %i.addr, align 4
561 // CHECK-NEXT: [[I1_EXT:%.*]] = sext i32 [[I1]] to i64
562 // CHECK-NEXT: [[J1:%.*]] = load i32, i32* %j.addr, align 4
563 // CHECK-NEXT: [[J1_EXT:%.*]] = sext i32 [[J1]] to i64
564 // CHECK-NEXT: [[A:%.*]] = load <27 x i32>, <27 x i32>* %0, align 4
565 // CHECK-NEXT: [[IDX1_1:%.*]] = mul i64 [[J1_EXT]], 9
566 // CHECK-NEXT: [[IDX1_2:%.*]] = add i64 [[IDX1_1]], [[I1_EXT]]
567 // CHECK-NEXT: [[MI1:%.*]] = extractelement <27 x i32> [[A]], i64 [[IDX1_2]]
568 // CHECK-NEXT: [[MI1_EXT:%.*]] = sext i32 [[MI1]] to i64
569 // CHECK-NEXT: [[J2:%.*]] = load i32, i32* %j.addr, align 4
570 // CHECK-NEXT: [[J2_EXT:%.*]] = sext i32 [[J2]] to i64
571 // CHECK-NEXT: [[I2:%.*]] = load i32, i32* %i.addr, align 4
572 // CHECK-NEXT: [[I2_EXT:%.*]] = sext i32 [[I2]] to i64
573 // CHECK-NEXT: [[A2:%.*]] = load <27 x i32>, <27 x i32>* {{.*}}, align 4
574 // CHECK-NEXT: [[IDX2_1:%.*]] = mul i64 [[I2_EXT]], 9
575 // CHECK-NEXT: [[IDX2_2:%.*]] = add i64 [[IDX2_1]], [[J2_EXT]]
576 // CHECK-NEXT: [[MI2:%.*]] = extractelement <27 x i32> [[A2]], i64 [[IDX2_2]]
577 // CHECK-NEXT: [[MI3:%.*]] = add nsw i32 [[MI2]], 2
578 // CHECK-NEXT: [[MI3_EXT:%.*]] = sext i32 [[MI3]] to i64
579 // CHECK-NEXT: [[IDX3_1:%.*]] = mul i64 [[MI3_EXT]], 5
580 // CHECK-NEXT: [[IDX3_2:%.*]] = add i64 [[IDX3_1]], [[MI1_EXT]]
581 // CHECK-NEXT: [[B:%.*]] = load <25 x double>, <25 x double>* [[B_PTR:%.*]], align 8
582 // CHECK-NEXT: [[INS:%.*]] = insertelement <25 x double> [[B]], double 1.500000e+00, i64 [[IDX3_2]]
583 // CHECK-NEXT: store <25 x double> [[INS]], <25 x double>* [[B_PTR]], align 8
584 b[a[i][j]][a[j][i] + 2] = 1.5;
585 }
586