1# RUN: %PYTHON -m mlir.dialects.linalg.opdsl.dump_oplib --file %s | FileCheck %s
2
3from mlir.dialects.linalg.opdsl.lang import *
4
5
6# CHECK: ---
7# CHECK-LABEL: matmul
8# CHECK: args:
9# CHECK:     name: A
10# CHECK:     kind: input_tensor
11# CHECK:     type_var: T
12# CHECK:     shape_map: affine_map<()[s0, s1, s2] -> (s0, s1)>
13# CHECK:     name: B
14# CHECK:     kind: input_tensor
15# CHECK:     type_var: T
16# CHECK:     shape_map: affine_map<()[s0, s1, s2] -> (s1, s2)>
17# CHECK:     name: C
18# CHECK:     kind: output_tensor
19# CHECK:     type_var: U
20# CHECK:     shape_map: affine_map<()[s0, s1, s2] -> (s0, s2)>
21# CHECK:     name: bfn
22# CHECK:     kind: binary_fn_attr
23# CHECK:     default_fn: mul
24# CHECK:     name: ufn
25# CHECK:     kind: unary_fn_attr
26# CHECK:     default_fn: exp
27# CHECK:     name: cast
28# CHECK:     kind: type_fn_attr
29# CHECK:     default_fn: cast_signed
30@linalg_structured_op
31def matmul(
32    A=TensorDef(T, S.M, S.K),
33    B=TensorDef(T, S.K, S.N),
34    C=TensorDef(U, S.M, S.N, output=True),
35    bfn=BinaryFnAttrDef(default=BinaryFn.mul),
36    ufn=UnaryFnAttrDef(default=UnaryFn.exp),
37    cast=TypeFnAttrDef(default=TypeFn.cast_signed)):
38  C[D.m, D.n] += bfn(cast(U, A[D.m, D.k]), cast(U, B[D.k, D.n]))
39
40
41# CHECK: ---
42# CHECK-LABEL: fill
43# CHECK: args:
44# CHECK:     name: value
45# CHECK:     kind: scalar
46# CHECK-NOT: shape_map:
47# CHECK:     type_var: T
48@linalg_structured_op
49def fill(value=ScalarDef(T), O=TensorDef(T, S.M, S.K, output=True)):
50  O[D.m, D.n] = value
51
52
53# CHECK: ---
54# CHECK-LABEL: strided_copy
55# CHECK: args:
56# CHECK:     name: I
57# CHECK:     kind: input_tensor
58# CHECK:     type_var: T
59# CHECK:     shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s0, s1)>
60# CHECK:     name: O
61# CHECK:     kind: output_tensor
62# CHECK:     type_var: T
63# CHECK:     shape_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s2, s3)>
64# CHECK:     name: strides
65# CHECK:     kind: index_attr
66# CHECK:     index_attr_map: affine_map<()[s0, s1, s2, s3, s4, s5] -> (s4, s5)>
67# CHECK:     default_indices:
68# CHECK:     - 1
69# CHECK:     - 2
70@linalg_structured_op
71def strided_copy(
72    I=TensorDef(T, S.IH, S.IW),
73    O=TensorDef(T, S.OH, S.OW, output=True),
74    strides=IndexAttrDef(S.SH, S.SW, default=[1, 2])):
75  O[D.oh, D.ow] = I[D.oh * S.SH, D.ow * S.SW]
76