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
|
// RUN: mlir-opt %s -convert-vector-to-scf -lower-affine -convert-scf-to-cf -convert-vector-to-llvm="enable-amx" -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
// RUN: mlir-translate -mlir-to-llvmir | \
// RUN: %lli --entry-function=entry --mattr="+amx-tile,+amx-int8,+amx-bf16" --dlopen=%mlir_c_runner_utils | \
// RUN: FileCheck %s
// Note: To run this test, your CPU must support AMX.
func.func @tilezero(%arg0: memref<?x?xi32>, %i: index, %j: index) {
%1 = amx.tile_zero : vector<16x16xi32>
amx.tile_store %arg0[%i, %j], %1 : memref<?x?xi32>, vector<16x16xi32>
return
}
func.func @entry() -> i32 {
%i0 = arith.constant 0: i32
%i1 = arith.constant 1: i32
%c0 = arith.constant 0: index
%c1 = arith.constant 1: index
%c3 = arith.constant 3: index
%c19 = arith.constant 19: index
// Set up memory.
%a = memref.alloc(%c19, %c19) : memref<?x?xi32>
scf.for %i = %c0 to %c19 step %c1 {
scf.for %j = %c0 to %c19 step %c1 {
memref.store %i1, %a[%i, %j] : memref<?x?xi32>
}
}
// Call kernel.
call @tilezero(%a, %c1, %c1) : (memref<?x?xi32>, index, index) -> ()
// Print and verify that the tilezero is correctly strided within
// the enveloping 19x19 buffer.
//
// CHECK: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
// CHECK-NEXT: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
//
scf.for %i = %c0 to %c19 step %c1 {
%av = vector.transfer_read %a[%i, %c0], %i0: memref<?x?xi32>, vector<19xi32>
vector.print %av : vector<19xi32>
}
// Call kernel with different indices.
call @tilezero(%a, %c0, %c3) : (memref<?x?xi32>, index, index) -> ()
// Print and verify that the tilezero is again correctly strided
// within the enveloping 19x19 buffer.
//
// CHECK-NEXT: ( 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
// CHECK-NEXT: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
// CHECK-NEXT: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
//
scf.for %i = %c0 to %c19 step %c1 {
%av = vector.transfer_read %a[%i, %c0], %i0: memref<?x?xi32>, vector<19xi32>
vector.print %av : vector<19xi32>
}
// Release resources.
memref.dealloc %a : memref<?x?xi32>
return %i0 : i32
}
|
