Builders.cpp
11.1 KB
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
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
//===- Builders.cpp - MLIR Declarative Linalg Builders --------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#include "mlir/IR/Builders.h"
#include "mlir/Dialect/Affine/EDSC/Intrinsics.h"
#include "mlir/Dialect/Linalg/EDSC/Builders.h"
#include "mlir/Dialect/Linalg/EDSC/Intrinsics.h"
#include "mlir/Dialect/SCF/EDSC/Builders.h"
#include "mlir/Dialect/StandardOps/EDSC/Intrinsics.h"
#include "mlir/Dialect/Utils/StructuredOpsUtils.h"
#include "mlir/IR/AffineExpr.h"
using namespace mlir;
using namespace mlir::edsc;
using namespace mlir::edsc::intrinsics;
using namespace mlir::linalg;
using namespace mlir::scf;
Operation *mlir::edsc::makeGenericLinalgOp(
ArrayRef<IteratorType> iteratorTypes, ArrayRef<StructuredIndexed> inputs,
ArrayRef<StructuredIndexed> outputs,
function_ref<void(ValueRange)> regionBuilder, ArrayRef<Value> otherValues,
ArrayRef<Attribute> otherAttributes) {
for (unsigned i = 0, e = outputs.size(); i + 1 < e; ++i)
assert(!(outputs[i].getType().isa<RankedTensorType>() &&
outputs[i + 1].getType().isa<MemRefType>()) &&
"output tensors must be passed after output buffers");
auto &builder = edsc::ScopedContext::getBuilderRef();
auto *ctx = builder.getContext();
unsigned nInputs = inputs.size();
unsigned nOutputs = outputs.size();
SmallVector<SmallVector<AffineExpr, 4>, 4> exprsList;
exprsList.reserve(nInputs + nOutputs);
for (auto structuredIndexed : inputs)
exprsList.emplace_back(structuredIndexed.getExprs().begin(),
structuredIndexed.getExprs().end());
for (auto structuredIndexed : outputs)
exprsList.emplace_back(structuredIndexed.getExprs().begin(),
structuredIndexed.getExprs().end());
auto maps = AffineMap::inferFromExprList(exprsList);
unsigned nViews = nInputs + nOutputs;
SmallVector<Value, 4> values;
values.reserve(nViews);
values.append(inputs.begin(), inputs.end());
std::copy_if(outputs.begin(), outputs.end(), std::back_inserter(values),
[](StructuredIndexed s) { return s.hasValue(); });
SmallVector<Type, 4> types;
std::copy_if(outputs.begin(), outputs.end(), std::back_inserter(types),
[](StructuredIndexed s) { return !s.hasValue(); });
auto iteratorStrTypes =
llvm::to_vector<8>(llvm::map_range(iteratorTypes, toString));
// clang-format off
auto *op =
edsc::ScopedContext::getBuilderRef()
.create<linalg::GenericOp>(
edsc::ScopedContext::getLocation(),
types,
values,
IntegerAttr::get(IntegerType::get(64, ctx), nInputs),
IntegerAttr::get(IntegerType::get(64, ctx), nOutputs),
builder.getAffineMapArrayAttr(maps),
builder.getStrArrayAttr(iteratorStrTypes),
StringAttr() /*doc*/,
StringAttr() /*library_call*/
/* TODO: other attributes in op */
)
.getOperation();
// clang-format on
using namespace edsc;
SmallVector<Type, 4> blockTypes;
blockTypes.reserve(values.size());
for (auto it : llvm::enumerate(values))
blockTypes.push_back((it.index() < nViews)
? getElementTypeOrSelf(it.value())
: it.value().getType());
assert(op->getNumRegions() == 1);
assert(op->getRegion(0).empty());
OpBuilder opBuilder(op);
ScopedContext scope(opBuilder, op->getLoc());
buildInNewBlock(op->getRegion(0), blockTypes, regionBuilder);
assert(llvm::hasSingleElement(op->getRegion(0)));
return op;
}
void mlir::edsc::ops::mulRegionBuilder(ValueRange args) {
using edsc::op::operator+;
using edsc::op::operator*;
assert(args.size() == 2 && "expected 2 block arguments");
Value a(args[0]), b(args[1]);
linalg_yield(a * b);
}
void mlir::edsc::ops::macRegionBuilder(ValueRange args) {
using edsc::op::operator+;
using edsc::op::operator*;
assert(args.size() == 3 && "expected 3 block arguments");
Value a(args[0]), b(args[1]), c(args[2]);
linalg_yield(c + a * b);
}
Operation *mlir::edsc::ops::linalg_generic_pointwise(
UnaryPointwiseOpBuilder unaryOp, StructuredIndexed I, StructuredIndexed O) {
SmallVector<IteratorType, 4> iterTypes(O.getExprs().size(),
IteratorType::Parallel);
if (O.getType().isa<RankedTensorType>()) {
auto fun = [&unaryOp](ValueRange args) {
assert(args.size() == 1 && "expected 1 block arguments");
Value a(args[0]);
linalg_yield(unaryOp(a));
};
return makeGenericLinalgOp(iterTypes, {I}, {O}, fun);
}
auto fun = [&unaryOp](ValueRange args) {
assert(args.size() == 2 && "expected 2 block arguments");
Value a(args[0]);
linalg_yield(unaryOp(a));
};
return makeGenericLinalgOp(iterTypes, {I}, {O}, fun);
}
Operation *mlir::edsc::ops::linalg_generic_pointwise_tanh(StructuredIndexed I,
StructuredIndexed O) {
UnaryPointwiseOpBuilder unOp([](Value a) -> Value { return std_tanh(a); });
return linalg_generic_pointwise(unOp, I, O);
}
/// Binary pointwise operation (with broadcast) entry point.
Operation *mlir::edsc::ops::linalg_generic_pointwise(
BinaryPointwiseOpBuilder binaryOp, StructuredIndexed I1,
StructuredIndexed I2, StructuredIndexed O) {
SmallVector<IteratorType, 4> iterTypes(O.getExprs().size(),
IteratorType::Parallel);
if (O.getType().isa<RankedTensorType>()) {
auto fun = [&binaryOp](ValueRange args) {
assert(args.size() == 2 && "expected 2 block arguments");
Value a(args[0]), b(args[1]);
linalg_yield(binaryOp(a, b));
};
return makeGenericLinalgOp(iterTypes, {I1, I2}, {O}, fun);
}
auto fun = [&binaryOp](ValueRange args) {
assert(args.size() == 3 && "expected 3 block arguments");
Value a(args[0]), b(args[1]);
linalg_yield(binaryOp(a, b));
};
return makeGenericLinalgOp(iterTypes, {I1, I2}, {O}, fun);
}
Operation *mlir::edsc::ops::linalg_generic_pointwise_add(StructuredIndexed I1,
StructuredIndexed I2,
StructuredIndexed O) {
using edsc::op::operator+;
BinaryPointwiseOpBuilder binOp(
[](Value a, Value b) -> Value { return a + b; });
return linalg_generic_pointwise(binOp, I1, I2, O);
}
Operation *mlir::edsc::ops::linalg_generic_pointwise_max(StructuredIndexed I1,
StructuredIndexed I2,
StructuredIndexed O) {
BinaryPointwiseOpBuilder binOp([](Value a, Value b) -> Value {
using edsc::op::sgt;
return std_select(sgt(a, b), a, b);
});
return linalg_generic_pointwise(binOp, I1, I2, O);
}
Operation *
mlir::edsc::ops::linalg_generic_matmul(Value vA, Value vB, Value vC,
MatmulRegionBuilder regionBuilder) {
// clang-format off
AffineExpr m, n, k;
bindDims(ScopedContext::getContext(), m, n, k);
StructuredIndexed A(vA), B(vB), C(vC);
return makeGenericLinalgOp(
{IteratorType::Parallel, IteratorType::Parallel, IteratorType::Reduction},
{A({m, k}), B({k, n})},
{C({m, n})},
regionBuilder);
// clang-format on
}
Operation *
mlir::edsc::ops::linalg_generic_matmul(Value vA, Value vB, RankedTensorType tC,
MatmulRegionBuilder regionBuilder) {
// clang-format off
AffineExpr m, n, k;
bindDims(ScopedContext::getContext(), m, n, k);
StructuredIndexed A(vA), B(vB), C(tC);
return makeGenericLinalgOp(
{IteratorType::Parallel, IteratorType::Parallel, IteratorType::Reduction},
{A({m, k}), B({k, n})},
{C({m, n})},
regionBuilder);
// clang-format on
}
Operation *
mlir::edsc::ops::linalg_generic_matmul(Value vA, Value vB, Value vC,
RankedTensorType tD,
MatmulRegionBuilder regionBuilder) {
// clang-format off
AffineExpr m, n, k;
bindDims(ScopedContext::getContext(), m, n, k);
StructuredIndexed A(vA), B(vB), C(vC), D(tD);
return makeGenericLinalgOp(
{IteratorType::Parallel, IteratorType::Parallel, IteratorType::Reduction},
{A({m, k}), B({k, n}), C({m, n})},
{D({m, n})},
regionBuilder);
// clang-format on
}
Operation *mlir::edsc::ops::linalg_generic_conv_nhwc(Value vI, Value vW,
Value vO,
ArrayRef<int> strides,
ArrayRef<int> dilations) {
MLIRContext *ctx = ScopedContext::getContext();
// TODO: some template magic to make everything rank-polymorphic.
assert((dilations.empty() || dilations.size() == 2) && "only 2-D conv atm");
assert((strides.empty() || strides.size() == 2) && "only 2-D conv atm");
// Some short names.
auto par = IteratorType::Parallel;
auto red = IteratorType::Reduction;
auto s = strides;
auto d = dilations;
AffineExpr b, f, h, w, kh, kw, c;
bindDims(ctx, b, f, h, w, kh, kw, c);
unsigned numDims = c.cast<AffineDimExpr>().getPosition() + 1;
StructuredIndexed I(vI), W(vW), O(vO);
// clang-format off
return makeGenericLinalgOp(
{par, par, par, par, red, red, red}, {
I({b,
// Roundtrip to flattened form to serve as canonicalization and ensure
// consistent ordering of subexpressions.
simplifyAffineExpr(s[0] * h + d[0] * kh, numDims, 0),
simplifyAffineExpr(s[1] * w + d[1] * kw, numDims, 0),
c}),
W({kh, kw, c, f})}, {
O({b, h, w, f})},
macRegionBuilder);
// clang-format on
}
Operation *mlir::edsc::ops::linalg_generic_dilated_conv_nhwc(
Value vI, Value vW, Value vO, int depth_multiplier, ArrayRef<int> strides,
ArrayRef<int> dilations) {
MLIRContext *ctx = ScopedContext::getContext();
// TODO: some template magic to make everything rank-polymorphic.
assert((dilations.empty() || dilations.size() == 2) && "only 2-D conv atm");
assert((strides.empty() || strides.size() == 2) && "only 2-D conv atm");
// Some short names.
auto par = IteratorType::Parallel;
auto red = IteratorType::Reduction;
auto s = strides;
auto d = dilations;
// clang-format off
AffineExpr b, dm, c, h, w, kh, kw;
bindDims(ctx, b, dm, c, h, w, kh, kw);
unsigned numDims = kw.cast<AffineDimExpr>().getPosition() + 1;
StructuredIndexed I(vI), W(vW), O(vO);
return makeGenericLinalgOp(
{par, par, par, par, par, red, red}, {
I({b,
// Roundtrip to flattened form to serve as canonicalization and ensure
// consistent ordering of subexpressions.
simplifyAffineExpr(s[0] * h + d[0] * kh, numDims, 0),
simplifyAffineExpr(s[1] * w + d[1] * kw, numDims, 0),
c}),
W({kh, kw, c, dm})}, {
O({b, h, w, simplifyAffineExpr(c * depth_multiplier + dm, numDims, 0)})},
macRegionBuilder);
// clang-format on
}