1/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
2
3Licensed under the Apache License, Version 2.0 (the "License");
4you may not use this file except in compliance with the License.
5You may obtain a copy of the License at
6
7    http://www.apache.org/licenses/LICENSE-2.0
8
9Unless required by applicable law or agreed to in writing, software
10distributed under the License is distributed on an "AS IS" BASIS,
11WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12See the License for the specific language governing permissions and
13limitations under the License.
14==============================================================================*/
15
16#include "tensorflow/compiler/tf2xla/xla_helpers.h"
17#include "tensorflow/compiler/tf2xla/xla_op_kernel.h"
18#include "tensorflow/compiler/tf2xla/xla_op_registry.h"
19
20namespace tensorflow {
21namespace {
22
23class CrossOp : public XlaOpKernel {
24 public:
25  explicit CrossOp(OpKernelConstruction* context) : XlaOpKernel(context) {}
26
27  void Compile(XlaOpKernelContext* ctx) override {
28    TensorShape in0_shape = ctx->InputShape(0);
29    TensorShape in1_shape = ctx->InputShape(1);
30    OP_REQUIRES(ctx, in0_shape == in1_shape,
31                errors::InvalidArgument("Both inputs must be of same shape: ",
32                                        in0_shape.DebugString(), " vs. ",
33                                        in1_shape.DebugString()));
34    OP_REQUIRES(ctx, in0_shape.dims() >= 1,
35                errors::InvalidArgument("Input must be at least 1D",
36                                        in0_shape.DebugString()));
37
38    auto inner_dim = in0_shape.dim_size(in0_shape.dims() - 1);
39    OP_REQUIRES(ctx, inner_dim == 3,
40                errors::FailedPrecondition(
41                    "Cross-products are only defined for 3-element vectors."));
42
43    // in0 is a [...,X,Y,Z,3]
44    // in1 is the same shape as in0
45    // So slice 0 is: in0[...,:,:,:,0:1]
46    // So slice 1 is: in0[...,:,:,:,1:2]
47    // So slice 2 is: in0[...,:,:,:,2:3]
48
49    std::vector<int64> starts(in0_shape.dims(), 0);
50    std::vector<int64> limits;
51    for (auto dim_size : in0_shape.dim_sizes()) {
52      limits.push_back(dim_size);
53    }
54    std::vector<int64> strides(in0_shape.dims(), 1);
55
56    xla::ComputationBuilder* b = ctx->builder();
57    auto in0 = ctx->Input(0);
58    auto in1 = ctx->Input(1);
59    starts.back() = 0;
60    limits.back() = 1;
61    auto u1 = b->Slice(in0, starts, limits, strides);
62    auto v1 = b->Slice(in1, starts, limits, strides);
63    starts.back() = 1;
64    limits.back() = 2;
65    auto u2 = b->Slice(in0, starts, limits, strides);
66    auto v2 = b->Slice(in1, starts, limits, strides);
67    starts.back() = 2;
68    limits.back() = 3;
69    auto u3 = b->Slice(in0, starts, limits, strides);
70    auto v3 = b->Slice(in1, starts, limits, strides);
71
72    auto s1 = b->Sub(b->Mul(u2, v3), b->Mul(u3, v2));
73    auto s2 = b->Sub(b->Mul(u3, v1), b->Mul(u1, v3));
74    auto s3 = b->Sub(b->Mul(u1, v2), b->Mul(u2, v1));
75    auto output = b->ConcatInDim({s1, s2, s3}, in0_shape.dims() - 1);
76
77    ctx->SetOutput(0, output);
78  }
79
80 private:
81  TF_DISALLOW_COPY_AND_ASSIGN(CrossOp);
82};
83
84REGISTER_XLA_OP(Name("Cross"), CrossOp);
85
86}  // namespace
87}  // namespace tensorflow
88