xref: /llvm-project/mlir/test/Dialect/Affine/loop-tiling-parametric.mlir (revision 96ff0255f2ecd0bd5d1aba650a99e046a4bd5ec6)
1// RUN: mlir-opt %s -split-input-file -test-affine-parametric-tile -verify-diagnostics | FileCheck %s
2// Test cases to test the utility introduced to tile affine for loops using
3// SSA values as tiling parameters(tile sizes). The tile sizes are expected
4// to be passed as input arguments(before any other argument) to the function
5// enclosing the loop nest. Currently hyper-rectangular loop nests with constant
6// lower bounds are supported.
7
8// -----
9
10// CHECK-DAG: [[LBI:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0)>
11// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0 + s0, 256)>
12// CHECK-DAG: [[UBI1:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0 + s0, 512)>
13// CHECK-DAG: [[UBI2:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0 + s0, 1024)>
14// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<()[s0] -> (256 ceildiv s0)>
15// CHECK-DAG: [[UBO1:#map[0-9]*]] = affine_map<()[s0] -> (512 ceildiv s0)>
16// CHECK-DAG: [[UBO2:#map[0-9]*]] = affine_map<()[s0] -> (1024 ceildiv s0)>
17
18// CHECK: func @loop_tiling_3d([[ARG0:%arg[0-9]+]]: index, [[ARG1:%arg[0-9]+]]: index, [[ARG2:%arg[0-9]+]]: index)
19// CHECK-NEXT:   affine.for [[ARG3:%arg[0-9]+]] = 0 to [[UBO0]](){{.*}}[[ARG0]]
20// CHECK-NEXT:     affine.for [[ARG4:%arg[0-9]+]] = 0 to [[UBO1]](){{.*}}[[ARG1]]
21// CHECK-NEXT:       affine.for [[ARG5:%arg[0-9]+]] = 0 to [[UBO2]](){{.*}}[[ARG2]]
22// CHECK-NEXT:         affine.for %[[I:.*]] = [[LBI]]{{.*}}[[ARG3]]{{.*}}[[ARG0]]{{.*}} to min [[UBI0]]{{.*}}[[ARG3]]{{.*}}[[ARG0]]
23// CHECK-NEXT:          affine.for %[[J:.*]] = [[LBI]]{{.*}}[[ARG4]]{{.*}}[[ARG1]]{{.*}} to min [[UBI1]]{{.*}}[[ARG4]]{{.*}}[[ARG1]]
24// CHECK-NEXT:            affine.for %[[K:.*]] = [[LBI]]{{.*}}[[ARG5]]{{.*}}[[ARG2]]{{.*}} to min [[UBI2]]{{.*}}[[ARG5]]{{.*}}[[ARG2]]
25// CHECK-NEXT:              "test.foo"(%[[I]], %[[J]], %[[K]])
26func.func @loop_tiling_3d(%t0 : index, %t1 : index, %t2 : index) {
27  affine.for %i = 0 to 256 {
28    affine.for %j = 0 to 512 {
29      affine.for %k = 0 to 1024 {
30        "test.foo"(%i, %j, %k) : (index, index, index) -> ()
31      }
32    }
33  }
34  return
35}
36
37// -----
38
39// CHECK-DAG: [[LBI:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0)>
40// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0 + s0 * 4, 256)>
41// CHECK-DAG: [[UBI1:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0 + s0 * 3, 512)>
42// CHECK-DAG: [[UBI2:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0 + s0 * 2, 1024)>
43// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<()[s0] -> (256 ceildiv s0)>
44// CHECK-DAG: [[UBO1:#map[0-9]*]] = affine_map<()[s0] -> (512 ceildiv s0)>
45// CHECK-DAG: [[UBO2:#map[0-9]*]] = affine_map<()[s0] -> (1024 ceildiv s0)>
46
47// CHECK: func @loop_tiling_non_unit_step([[ARG0:%arg[0-9]+]]: index, [[ARG1:%arg[0-9]+]]: index, [[ARG2:%arg[0-9]+]]: index)
48// CHECK-NEXT:  affine.for [[ARG3:%arg[0-9]+]] = 0 to [[UBO0]](){{.*}}[[ARG0]]{{.*}}step 4
49// CHECK-NEXT:    affine.for [[ARG4:%arg[0-9]+]] = 0 to [[UBO1]](){{.*}}[[ARG1]]{{.*}} step 3
50// CHECK-NEXT:       affine.for [[ARG5:%arg[0-9]+]] = 0 to [[UBO2]](){{.*}}[[ARG2]]{{.*}} step 2
51// CHECK-NEXT:         affine.for %[[I:.*]] = [[LBI]]{{.*}}[[ARG3]]{{.*}}[[ARG0]]{{.*}} to min [[UBI0]]{{.*}}[[ARG3]]{{.*}}[[ARG0]]{{.*}} step 4
52// CHECK-NEXT:          affine.for %[[J:.*]] = [[LBI]]{{.*}}[[ARG4]]{{.*}}[[ARG1]]{{.*}} to min [[UBI1]]{{.*}}[[ARG4]]{{.*}}[[ARG1]]{{.*}} step 3
53// CHECK-NEXT:            affine.for %[[K:.*]] = [[LBI]]{{.*}}[[ARG5]]{{.*}}[[ARG2]]{{.*}} to min [[UBI2]]{{.*}}[[ARG5]]{{.*}}[[ARG2]]{{.*}} step 2
54// CHECK-NEXT:              "test.foo"(%[[I]], %[[J]], %[[K]])
55func.func @loop_tiling_non_unit_step(%t0: index, %t1: index, %t2: index){
56  affine.for %i = 0 to 256 step 4 {
57    affine.for %j = 0 to 512  step 3 {
58      affine.for %k = 0 to 1024 step 2 {
59        "test.foo"(%i, %j, %k) : (index, index, index) -> ()
60      }
61    }
62  }
63  return
64}
65
66// -----
67
68// CHECK-DAG: [[LBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0)>
69// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0)[s0, s1, s2] -> (d0 * s2 + s2, s0, 4096 floordiv s1)>
70// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<()[s0, s1, s2] -> (s0 ceildiv s2, (4096 floordiv s1) ceildiv s2)>
71
72// CHECK: func @tile_loop_with_div_in_upper_bound([[ARG0:%arg[0-9]+]]: index, %{{.*}}: memref<?xi32>, %{{.*}}: index, %{{.*}}: index)
73#ub = affine_map<()[s0, s1] -> (s0, 4096 floordiv s1)>
74func.func @tile_loop_with_div_in_upper_bound(%t5 : index, %A : memref<? x i32>, %L : index, %U : index) {
75  %c0 = arith.constant 0 : index
76  %M = memref.dim %A, %c0 : memref<? x i32>
77  affine.for %i = 0 to min #ub()[%M, %U] {
78    arith.addi %i, %i : index
79  }
80  // CHECK:  affine.for [[ARG1:%arg[0-9]+]] = 0 to min [[UBO0]]()[%{{.*}}, %{{.*}}, [[ARG0]]]
81  // CHECK-NEXT:    affine.for %[[I:.*]] = [[LBI0]]([[ARG1]]){{.*}}[[ARG0]]{{.*}} to min [[UBI0]]({{.*}})[{{.*}}, {{.*}}, [[ARG0]]]
82  // CHECK-NEXT:      arith.addi %[[I]], %[[I]]
83  return
84}
85
86// -----
87
88// CHECK-DAG: [[LBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0)>
89// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0)[s0, s1, s2] -> (d0 * s2 + s2 * 4, s0, 4096 floordiv s1)>
90// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<()[s0, s1, s2] -> (s0 ceildiv s2, (4096 floordiv s1) ceildiv s2)>
91
92// CHECK: func @tile_loop_with_div_in_upper_bound_non_unit_step([[ARG0:%arg[0-9]+]]: index, %{{.*}}: memref<?xi32>, %{{.*}}: index, %{{.*}}: index)
93#ub = affine_map<()[s0, s1] -> (s0, 4096 floordiv s1)>
94func.func @tile_loop_with_div_in_upper_bound_non_unit_step(%t5 : index, %A : memref<? x i32>, %L : index, %U : index) {
95  %c0 = arith.constant 0 : index
96  %M = memref.dim %A, %c0 : memref<? x i32>
97  affine.for %i = 0 to min #ub()[%M, %U] step 4 {
98    arith.addi %i, %i : index
99  }
100  // CHECK: affine.for [[ARG1:%arg[0-9]+]] = 0 to min [[UBO0]]()[%{{.*}}, %{{.*}}, [[ARG0]]]{{.*}} step 4{{.*}}
101  // CHECK-NEXT:    affine.for %[[I:.*]] = [[LBI0]]([[ARG1]]){{.*}}[[ARG0]]{{.*}} to min [[UBI0]]({{.*}})[{{.*}}, {{.*}}, [[ARG0]]]{{.*}} step 4{{.*}}
102  // CHECK-NEXT:      arith.addi %[[I]], %[[I]]
103  return
104}
105
106// -----
107
108// CHECK-DAG: [[LBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> ((d0 - 8) * s0 + 8)>
109// CHECK-DAG: [[UBI2:#map[0-9]*]] = affine_map<(d0)[s0, s1] -> ((d0 - 8) * s1 + s1 * 4 + 8, s0 + 16)>
110// CHECK-DAG: [[UBI1:#map[0-9]*]] = affine_map<(d0)[s0, s1] -> ((d0 - 8) * s1 + s1 + 8, s0 + 16)>
111// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> ((d0 - 8) * s0 + s0 + 8, 256)>
112// CHECK-DAG: [[UBO1:#map[0-9]*]] = affine_map<()[s0, s1] -> ((s0 + 8) ceildiv s1 + 8)>
113// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<()[s0] -> (248 ceildiv s0 + 8)>
114
115// CHECK: func @tile_loop_with_non_zero_lb([[ARG0:%arg[0-9]+]]: index, [[ARG1:%arg[0-9]+]]: index, [[ARG2:%arg[0-9]+]]: index, %{{.*}}: index)
116// CHECK-NEXT:  affine.for [[ARG3:%arg[0-9+]]] = 8 to [[UBO0]]{{.*}}[[ARG0]]{{.*}}
117// CHECK-NEXT:    affine.for [[ARG4:%arg[0-9+]]] = 8 to [[UBO1]]{{.*}}[[ARG1]]{{.*}}
118// CHECK-NEXT:      affine.for [[ARG5:%arg[0-9+]]] = 8 to [[UBO1]]{{.*}}[[ARG2]]{{.*}} step 4
119// CHECK-NEXT:        affine.for %[[I:.*]] = [[LBI0]]([[ARG3]]){{.*}}[[ARG0]]{{.*}} to min [[UBI0]]([[ARG3]]){{.*}}[[ARG0]]{{.*}}
120// CHECK-NEXT:          affine.for %[[J:.*]] = [[LBI0]]([[ARG4]]){{.*}}[[ARG1]]{{.*}} to min [[UBI1]]([[ARG4]]){{.*}}[[ARG1]]{{.*}}
121// CHECK-NEXT:            affine.for %[[K:.*]] = [[LBI0]]([[ARG5]]){{.*}}[[ARG2]]{{.*}} to min [[UBI2]]([[ARG5]]){{.*}}[[ARG2]]{{.*}}step 4{{.*}}
122// CHECK-NEXT:              "test.foo"(%[[I]], %[[J]], %[[K]]) : (index, index, index) -> ()
123#ubi = affine_map<()[s0] -> (s0 + 16)>
124func.func @tile_loop_with_non_zero_lb(%t0: index, %t1: index, %t2: index, %U: index){
125  affine.for %i = 8 to 256 {
126    affine.for %j = 8 to #ubi()[%U] {
127      affine.for %k = 8 to #ubi()[%U] step 4 {
128        "test.foo"(%i, %j, %k) : (index, index, index) -> ()
129      }
130    }
131  }
132  return
133}
134
135// -----
136
137// CHECK-DAG: [[LBI:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0)>
138// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0 + s0, 256)>
139// CHECK-DAG: [[UBI1:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0 + s0, 250)>
140// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<()[s0] -> (256 ceildiv s0)>
141// CHECK-DAG: [[UBO1:#map[0-9]*]] = affine_map<()[s0] -> (250 ceildiv s0)>
142
143// CHECK: func @simple_matmul([[ARG0:%arg[0-9]+]]: index, [[ARG1:%arg[0-9]+]]: index, [[ARG2:%arg[0-9]+]]: index{{.*}})
144// CHECK-NEXT:   affine.for [[ARG3:%arg[0-9]+]] = 0 to [[UBO0]](){{.*}}[[ARG0]]{{.*}}
145// CHECK-NEXT:     affine.for [[ARG4:%arg[0-9]+]] = 0 to [[UBO0]](){{.*}}[[ARG1]]{{.*}}
146// CHECK-NEXT:       affine.for [[ARG5:%arg[0-9]+]] = 0 to [[UBO1]](){{.*}}[[ARG2]]{{.*}}
147// CHECK-NEXT:         affine.for %[[I:.*]] = [[LBI]]{{.*}}[[ARG3]]{{.*}}[[ARG0]]{{.*}} to min [[UBI0]]{{.*}}[[ARG3]]{{.*}}[[ARG0]]{{.*}}
148// CHECK-NEXT:          affine.for %[[J:.*]] = [[LBI]]{{.*}}[[ARG4]]{{.*}}[[ARG1]]{{.*}} to min [[UBI0]]{{.*}}[[ARG4]]{{.*}}[[ARG1]]{{.*}}
149// CHECK-NEXT:            affine.for %[[K:.*]] = [[LBI]]{{.*}}[[ARG5]]{{.*}}[[ARG2]]{{.*}} to min [[UBI1]]{{.*}}[[ARG5]]{{.*}}[[ARG2]]{{.*}}
150// CHECK-NEXT:                 affine.load %{{.*}}[%[[I]], %[[K]]]
151// CHECK-NEXT:                 affine.load %{{.*}}[%[[K]], %[[J]]]
152// CHECK-NEXT:                 affine.load %{{.*}}[%[[I]], %[[J]]]
153// CHECK-NEXT:                 arith.mulf %{{.*}}
154// CHECK-NEXT:                 arith.addf %{{.*}}
155// CHECK-NEXT:                 affine.store %{{.*}}[%[[I]], %[[J]]]
156func.func @simple_matmul(%t6 : index, %t7 : index, %t8 : index, %arg0: memref<256x256xvector<64xf32>>, %arg1: memref<256x256xvector<64xf32>>, %arg2: memref<256x256xvector<64xf32>>) -> memref<256x256xvector<64xf32>> {
157  affine.for %i = 0 to 256 {
158    affine.for %j = 0 to 256 {
159      affine.for %k = 0 to 250 {
160        %l = affine.load %arg0[%i, %k] : memref<256x256xvector<64xf32>>
161        %r = affine.load %arg1[%k, %j] : memref<256x256xvector<64xf32>>
162        %o = affine.load %arg2[%i, %j] : memref<256x256xvector<64xf32>>
163        %m = arith.mulf %l, %r : vector<64xf32>
164        %a = arith.addf %o, %m : vector<64xf32>
165        affine.store %a, %arg2[%i, %j] : memref<256x256xvector<64xf32>>
166      }
167    }
168  }
169  return %arg2 : memref<256x256xvector<64xf32>>
170}
171
172// -----
173
174// CHECK-DAG: [[LBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0)>
175// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0)[s0, s1] -> (d0 * s1 + s1, s0)>
176// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<()[s0, s1] -> (s0 ceildiv s1)>
177
178// CHECK: func @tile_using_symbolic_loop_upper_bounds([[ARG0:%arg[0-9]+]]: index, [[ARG1:%arg[0-9]+]]: index{{.*}}){{.*}}
179// CHECK:        affine.for [[ARG2:%arg[0-9]+]] = 0 to [[UBO0]](){{.*}}[[ARG0]]{{.*}}
180// CHECK-NEXT:     affine.for [[ARG3:%arg[0-9]+]] = 0 to [[UBO0]](){{.*}}[[ARG1]]{{.*}}
181// CHECK-NEXT:       affine.for %[[I0:.*]] = [[LBI0]]{{.*}}[[ARG2]]{{.*}}[[ARG0]]{{.*}} to min [[UBI0]]{{.*}}[[ARG2]]{{.*}}[[ARG0]]{{.*}}
182// CHECK-NEXT:         affine.for %[[I1:.*]] = [[LBI0]]{{.*}}[[ARG3]]{{.*}}[[ARG1]]{{.*}} to min [[UBI0]]{{.*}}[[ARG3]]{{.*}}[[ARG1]]{{.*}}
183// CHECK-NEXT:           affine.store %{{.*}}, %{{.*}}[%[[I0]], %[[I1]]] : memref<?x?xf32>
184// CHECK-NEXT:           affine.for %[[I2:.*]] = 0 to %{{.*}} {
185// CHECK-NEXT:             affine.load %{{.*}}%[[I0]], %[[I2]]
186// CHECK-NEXT:             affine.load %{{.*}}%[[I2]], %[[I1]]
187// CHECK-NEXT:             arith.mulf
188// CHECK-NEXT:             affine.load %{{.*}}%[[I0]], %[[I1]]
189// CHECK-NEXT:             arith.addf
190// CHECK-NEXT:             affine.store %{{.*}}%[[I0]], %[[I1]]
191func.func @tile_using_symbolic_loop_upper_bounds(%t9 : index, %t10: index, %arg0: memref<?x?xf32>, %arg1: memref<?x?xf32>, %arg2: memref<?x?xf32>) {
192  %cst = arith.constant 0.000000e+00 : f32
193  %c0 = arith.constant 0 : index
194  %0 = memref.dim %arg0, %c0 : memref<?x?xf32>
195  affine.for %i0 = 0 to %0 {
196    affine.for %i1 = 0 to %0 {
197      affine.store %cst, %arg2[%i0, %i1] : memref<?x?xf32>
198      affine.for %i2 = 0 to %0 {
199        %1 = affine.load %arg0[%i0, %i2] : memref<?x?xf32>
200        %2 = affine.load %arg1[%i2, %i1] : memref<?x?xf32>
201        %3 = arith.mulf %1, %2 : f32
202        %4 = affine.load %arg2[%i0, %i1] : memref<?x?xf32>
203        %5 = arith.addf %4, %3 : f32
204        affine.store %5, %arg2[%i0, %i1] : memref<?x?xf32>
205      }
206    }
207  }
208  return
209}
210
211// -----
212
213// CHECK-DAG: [[LBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0)>
214// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0)[s0, s1, s2] -> (d0 * s2 + s2, s0 + s1)>
215// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<()[s0, s1, s2] -> ((s0 + s1) ceildiv s2)>
216
217// CHECK: func @tile_using_loop_upper_bounds_in_two_symbols([[ARG0:%arg[0-9]+]]: index{{.*}}){{.*}}
218func.func @tile_using_loop_upper_bounds_in_two_symbols(%t11 : index, %arg0: memref<?xf32>, %limit: index) {
219  %c0 = arith.constant 0 : index
220  %dim0 = memref.dim %arg0, %c0 : memref<?xf32>
221  affine.for %i0 = 0 to affine_map<()[s0, s1] -> (s0 + s1)> ()[%dim0, %limit] {
222    %v0 = affine.load %arg0[%i0] : memref<?xf32>
223  }
224  // CHECK:  affine.for [[ARG1:%arg[0-9]+]] = 0 to [[UBO0]]()[%{{.*}}, %{{.*}}, [[ARG0]]]
225  // CHECK-NEXT:    affine.for %[[I:.*]] = [[LBI0]]([[ARG1]]){{.*}}[[ARG0]]{{.*}} to min [[UBI0]]([[ARG1]])[{{.*}}, {{.*}}, [[ARG0]]]
226  // CHECK-NEXT:      affine.load %{{.*}}[%[[I]]]
227  return
228}
229
230// -----
231
232// CHECK-DAG: [[LBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0)>
233// CHECK-DAG: [[UBI1:#map[0-9]*]] = affine_map<(d0, d1)[s0, s1] -> (d1 * s1 + s1, d0 + s0 + 4)>
234// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0, d1)[s0, s1] -> (d1 * s1 + s1, d0 + s0 + 2)>
235// CHECK-DAG: [[UBO1:#map[0-9]*]] = affine_map<(d0)[s0, s1] -> ((d0 + s0 + 4) ceildiv s1)>
236// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<(d0)[s0, s1] -> ((d0 + s0 + 2) ceildiv s1)>
237
238// CHECK: func @tile_using_upper_bounds_in_dimensions_and_symbols([[ARG0:%arg[0-9]+]]: index, [[ARG1:%arg[0-9]+]]: index, [[ARG2:%arg[0-9]+]]: index, [[ARG3:%arg[0-9]+]]: index{{.*}}){{.*}}
239// CHECK-NEXT: affine.for [[ARG4:%arg[0-9]+]] = 0 to [[UBO0]]({{.*}}){{.*}}[[ARG0]]
240// CHECK-NEXT:   affine.for [[ARG5:%arg[0-9]+]] = 0 to [[UBO1]]({{.*}}){{.*}}[[ARG1]]
241// CHECK-NEXT:     affine.for {{.*}} = [[LBI0]]([[ARG4]]){{.*}}[[ARG0]]{{.*}} to min [[UBI0]]({{.*}}, [[ARG4]]){{.*}}[[ARG0]]{{.*}}
242// CHECK-NEXT:       affine.for {{.*}} = [[LBI0]]([[ARG5]]){{.*}}[[ARG1]]{{.*}} to min [[UBI1]]({{.*}}, [[ARG5]]){{.*}}[[ARG1]]{{.*}}
243func.func @tile_using_upper_bounds_in_dimensions_and_symbols(%t12 : index, %t13 :index, %M: index, %N:  index, %K: index) {
244  affine.for %i = 0 to affine_map<(d0)[s0] -> (d0 + s0 + 2)>(%M)[%K] {
245    affine.for %j = 0 to affine_map<(d0)[s0] -> (d0 + s0 + 4)>(%N)[%K] {
246      "test.foo" () : () -> ()
247    }
248  }
249  return
250}
251
252// -----
253
254// CHECK-DAG: [[LBI0:#map[0-9]*]] = affine_map<(d0)[s0] -> (d0 * s0)>
255// CHECK-DAG: [[UBI1:#map[0-9]*]] = affine_map<(d0, d1)[s0, s1] -> (d1 * s1 + s1 * 4, d0 + s0 + 4)>
256// CHECK-DAG: [[UBI0:#map[0-9]*]] = affine_map<(d0, d1)[s0, s1] -> (d1 * s1 + s1 * 2, d0 + s0 + 2)>
257// CHECK-DAG: [[UBO1:#map[0-9]*]] = affine_map<(d0)[s0, s1] -> ((d0 + s0 + 4) ceildiv s1)>
258// CHECK-DAG: [[UBO0:#map[0-9]*]] = affine_map<(d0)[s0, s1] -> ((d0 + s0 + 2) ceildiv s1)>
259
260// CHECK: func @tile_using_upper_bounds_in_dimensions_and_symbols_non_unit_steps
261// CHECK-SAME: ([[ARG0:%arg[0-9]+]]: index, [[ARG1:%arg[0-9]+]]: index, [[ARG2:%arg[0-9]+]]: index, [[ARG3:%arg[0-9]+]]: index{{.*}}){{.*}}
262// CHECK-NEXT: affine.for [[ARG4:%arg[0-9]+]] = 0 to [[UBO0]]({{.*}}){{.*}}[[ARG0]]{{.*}} step 2{{.*}}
263// CHECK-NEXT:   affine.for [[ARG5:%arg[0-9]+]] = 0 to [[UBO1]]({{.*}}){{.*}}[[ARG1]]{{.*}} step 4{{.*}}
264// CHECK-NEXT:     affine.for {{.*}} = [[LBI0]]([[ARG4]]){{.*}}[[ARG0]]{{.*}} to min [[UBI0]]({{.*}}, [[ARG4]]){{.*}}[[ARG0]]{{.*}} step 2{{.*}}
265// CHECK-NEXT:       affine.for {{.*}} = [[LBI0]]([[ARG5]]){{.*}}[[ARG1]]{{.*}} to min [[UBI1]]({{.*}}, [[ARG5]]){{.*}}[[ARG1]]{{.*}} step 4{{.*}}
266func.func @tile_using_upper_bounds_in_dimensions_and_symbols_non_unit_steps(%t12 : index, %t13 :index, %M: index, %N :  index, %K: index) {
267  affine.for %i = 0 to affine_map<(d0)[s0] -> (d0 + s0 + 2)>(%M)[%K] step 2 {
268    affine.for %j = 0 to affine_map<(d0)[s0] -> (d0 + s0 + 4)>(%N)[%K] step 4 {
269      "test.foo" () : () -> ()
270    }
271  }
272  return
273}
274
275// -----
276
277func.func @too_few_tile_size_params() {
278  // expected-error@+1 {{too few tile sizes provided in the argument list of the function which contains the current band}}
279  affine.for %i = 0 to 256 {
280    affine.for %j = 0 to 512 {
281      affine.for %k = 0 to 1024 {
282        "test.foo"(%i, %j, %k) : (index, index, index) -> ()
283      }
284    }
285  }
286  return
287}
288
289// -----
290
291func.func @invalid_type_for_tile_size_params(%arg0: f32, %arg1: f32, %arg2: f32) {
292  // expected-error@+1 {{expected tiling parameters to be of index type}}
293  affine.for %i = 0 to 256 {
294    affine.for %j = 0 to 512 {
295      affine.for %k = 0 to 1024 {
296        "test.foo"(%i, %j, %k) : (index, index, index) -> ()
297      }
298    }
299  }
300  return
301}
302