1 /* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
2 
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6 
7     http://www.apache.org/licenses/LICENSE-2.0
8 
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15 
16 #include "tensorflow/compiler/xla/service/gpu/reduction_degenerate_dim_remover.h"
17 
18 #include <algorithm>
19 
20 #include "absl/algorithm/container.h"
21 #include "absl/strings/str_join.h"
22 #include "tensorflow/compiler/xla/service/dfs_hlo_visitor_with_default.h"
23 #include "tensorflow/compiler/xla/service/gpu/ir_emission_utils.h"
24 #include "tensorflow/compiler/xla/service/hlo_casting_utils.h"
25 #include "tensorflow/compiler/xla/service/hlo_instructions.h"
26 #include "tensorflow/compiler/xla/service/hlo_opcode.h"
27 #include "tensorflow/compiler/xla/service/pattern_matcher.h"
28 #include "tensorflow/compiler/xla/shape_util.h"
29 #include "tensorflow/compiler/xla/status_macros.h"
30 #include "tensorflow/compiler/xla/statusor.h"
31 #include "tensorflow/core/lib/core/errors.h"
32 #include "tensorflow/stream_executor/lib/statusor.h"
33 
34 namespace xla {
35 namespace gpu {
36 
37 class ReductionDegenerateDimRemoverVisitor : public DfsHloRewriteVisitor {
38  public:
HandleReduce(HloInstruction * hlo)39   Status HandleReduce(HloInstruction *hlo) override {
40     auto instr = Cast<HloReduceInstruction>(hlo);
41     absl::InlinedVector<HloInstruction *, 2> input_reshapes;
42     absl::InlinedVector<Shape, 2> canonical_reduce_shapes;
43 
44     int idx = -1;
45     std::vector<int64_t> updated_reduced_dimensions;
46     for (HloInstruction *reduced_op : instr->inputs()) {
47       idx++;
48       const Shape &input_shape = reduced_op->shape();
49       const Shape &reduce_shape = instr->shape().IsTuple()
50                                       ? instr->shape().tuple_shapes(idx)
51                                       : instr->shape();
52 
53       if (!ShapeUtil::HasDegenerateDimensions(reduced_op->shape())) {
54         return OkStatus();
55       }
56       Shape canonical_input_shape =
57           ShapeUtil::DropDegenerateDimensions(input_shape);
58 
59       Shape canonical_reduce_shape =
60           ShapeUtil::DropDegenerateDimensions(reduce_shape);
61 
62       auto reduced_dimensions = instr->dimensions();
63       int64_t shift = 0;
64 
65       for (int dim = 0; dim < input_shape.rank(); dim++) {
66         if (input_shape.dimensions(dim) == 1) {
67           shift++;
68         } else {
69           if (absl::c_linear_search(reduced_dimensions, dim) && idx == 0) {
70             // Only populate on first iteration.
71             updated_reduced_dimensions.push_back(dim - shift);
72           }
73         }
74       }
75 
76       if (updated_reduced_dimensions.empty()) {
77         std::unique_ptr<HloInstruction> reshape =
78             HloInstruction::CreateBitcast(reduce_shape, reduced_op);
79         return ReplaceWithNewInstruction(instr, std::move(reshape));
80       }
81 
82       input_reshapes.push_back(instr->parent()->AddInstruction(
83           HloInstruction::CreateBitcast(canonical_input_shape, reduced_op)));
84       canonical_reduce_shapes.push_back(canonical_reduce_shape);
85     }
86 
87     Shape canonical_reduce_shape =
88         ShapeUtil::MakeMaybeTupleShape(canonical_reduce_shapes);
89     const Shape &orig_reduce_shape = instr->shape();
90     std::unique_ptr<HloInstruction> new_reduce = HloInstruction::CreateReduce(
91         canonical_reduce_shape, input_reshapes, instr->init_values(),
92         updated_reduced_dimensions, instr->to_apply());
93 
94     if (canonical_reduce_shape != instr->shape()) {
95       HloInstruction *wrapped_reduce =
96           instr->parent()->AddInstruction(std::move(new_reduce));
97       absl::InlinedVector<HloInstruction *, 2> out;
98       if (!canonical_reduce_shape.IsTuple()) {
99         new_reduce =
100             HloInstruction::CreateBitcast(orig_reduce_shape, wrapped_reduce);
101       } else {
102         for (int oidx = 0; oidx < instr->input_count(); oidx++) {
103           HloInstruction *gte = instr->parent()->AddInstruction(
104               HloInstruction::CreateGetTupleElement(wrapped_reduce, oidx));
105           out.push_back(
106               instr->parent()->AddInstruction(HloInstruction::CreateBitcast(
107                   orig_reduce_shape.tuple_shapes(oidx), gte)));
108         }
109         new_reduce = HloInstruction::CreateTuple(out);
110       }
111     }
112 
113     return ReplaceWithNewInstruction(instr, std::move(new_reduce));
114   }
115 };
116 
Run(HloModule * module,const absl::flat_hash_set<absl::string_view> & execution_threads)117 StatusOr<bool> ReductionDegenerateDimRemover::Run(
118     HloModule *module,
119     const absl::flat_hash_set<absl::string_view> &execution_threads) {
120   TF_ASSIGN_OR_RETURN(bool changed,
121                       ReductionDegenerateDimRemoverVisitor().RunOnModule(
122                           module, execution_threads));
123   return changed;
124 }
125 
126 }  // namespace gpu
127 }  // namespace xla
128