/external/tensorflow/tensorflow/compiler/xla/service/ |
hlo_matchers_test.cc | 82 EXPECT_THAT(call.get(), op::CustomCall()); 83 EXPECT_THAT(call.get(), op::CustomCall(c1.get(), c2.get())); 84 EXPECT_THAT(call.get(), op::CustomCall("foo_target")); 85 EXPECT_THAT(call.get(), op::CustomCall("foo_target", c1.get(), c2.get())); 86 EXPECT_THAT(call.get(), op::CustomCall(::testing::StartsWith("foo"))); 88 op::CustomCall(::testing::Not(::testing::StartsWith("bar")))); 91 EXPECT_THAT(call.get(), ::testing::Not(op::CustomCall(c1.get()))); 95 ::testing::Not(op::CustomCall(::testing::StartsWith("bar")))); 97 EXPECT_THAT(Explain(call.get(), op::CustomCall("bar")), 99 EXPECT_THAT(DescribeHloMatcher(op::CustomCall("foo_target")) [all...] |
hlo_matchers.h | 204 // - CustomCall(T, operand1, ..., operandN) matches a CustomCall with call 207 // - CustomCall(operand1, ..., operandN) matches any CustomCall HLO with the 210 // - CustomCall() matches any CustomCall HLO at all. 212 inline ::testing::Matcher<const ::xla::HloInstruction*> CustomCall( 217 // This overload of CustomCall(A, B, C, ...) exists iff A is not convertible to 224 inline ::testing::Matcher<const ::xla::HloInstruction*> CustomCall( 229 inline ::testing::Matcher<const ::xla::HloInstruction*> CustomCall() { [all...] |
hlo_element_type_converter_test.cc | 38 ENTRY CustomCall {
|
hlo_element_type_converter.cc | 134 // We cannot change a CustomCall since we have no way of adjusting the
|
layout_assignment.h | 312 // the given CustomCall. It's up to the backend to set one in
|
/external/tensorflow/tensorflow/compiler/xla/service/cpu/ |
custom_call_target_registry.h | 29 // The CPU JIT compiler uses this registry to resolve symbolic CustomCall 30 // targets; so when using the CPU JIT, CustomCall targets need to be registered 31 // here with the symbol name used in the CustomCall. 35 // implemented in C++) matches up with the symbolic name used in the CustomCall.
|
parallel_task_assignment.cc | 120 // HLOs like CustomCall are not yet implemented in the HloCostAnalysis).
|
/external/tensorflow/tensorflow/compiler/xla/service/gpu/ |
cudnn_batchnorm_rewriter.h | 28 // CustomCall HLO with custom_call_target equal to one of 34 // A CustomCall created by this pass has the same operands corresponding 37 // CustomCall's operands list.
|
ir_emission_utils.h | 36 // A call to cuDNN for batch normalization is represented as CustomCall HLO with 55 // This returns true if `hlo` is a CustomCall HLO with a call target equal to 58 // sequence of generic HLOs or to a cuDNN CustomCall. 62 // is represented as a CustomCall HLO with a call target equal to one of these 94 // This returns true if `hlo` is a CustomCall HLO with a call target equal to 99 // Creates a CustomCall for a cudnn forward/backward-input/backward-filter conv.
|
cudnn_convolution_rewriter.h | 26 // backwards-input convolutions into CustomCall HLOs that call into cuDNN.
|
cudnn_convolution_rewriter_test.cc | 120 op::CustomCall(kCudnnConvBackwardFilterCallTarget), 0)); 148 op::CustomCall(kCudnnConvBackwardFilterCallTarget), 0)); 178 op::CustomCall(kCudnnConvBackwardFilterCallTarget), 0)); 208 op::CustomCall(kCudnnConvBackwardFilterCallTarget), 0)); 236 op::CustomCall(kCudnnConvBackwardFilterCallTarget), 0)); 287 op::CustomCall(kCudnnConvBackwardInputCallTarget), 0)); 331 op::CustomCall(kCudnnConvBackwardInputCallTarget), 0)); 363 op::GetTupleElement(op::CustomCall(kCudnnConvForwardCallTarget), 0)); 416 op::CustomCall(kCudnnConvBackwardInputCallTarget), 0)); 463 op::GetTupleElement(op::CustomCall(kCudnnConvForwardCallTarget), 0)) [all...] |
pad_insertion.cc | 176 // The conv CustomCall returns a tuple (conv_result, scratch_buffer). Extract 261 // The shape of the backward_conv CustomCall is a tuple (conv_result, 287 // The backward_conv CustomCall returns a tuple (conv_result, scratch_memory). 333 // The CustomCall created above returns a tuple (conv_result, scratch_memory).
|
gpu_compiler.cc | 213 // customcall = (f32[...], f32[0]) 214 // return gte(customcall, 0) 217 // increases the scratch space. It replaces customcall with new_tuple, 228 // assignment, fusion would already have run, and the gte(customcall, 0)
|
ir_emitter.cc | [all...] |
ir_emission_utils.cc | 148 // Our CustomCall takes three arguments: The conv lhs and rhs, and the cudnn
|
cudnn_convolution_algorithm_picker.cc | 307 // Replace instr with a new CustomCall which has the correct algorithm, and
|
cudnn_convolution_rewriter.cc | 436 // The CustomCall returns a tuple (conv_result, scratch_memory). Extract out
|
/external/tensorflow/tensorflow/compiler/tf2xla/kernels/ |
index_ops_cpu.cc | 73 "ArgMax implementation requires a CustomCall on CPU")); 97 output = b.CustomCall("argmax_float_1d_xla_impl", args, xla_shape); 100 output = b.CustomCall("argmax_float_2d_xla_impl", args, xla_shape);
|
/external/tensorflow/tensorflow/compiler/xla/tests/ |
custom_call_test.cc | 134 // When using the client API, CustomCall targets can't begin with '$' -- these 138 auto call = builder.CustomCall("$illegal", /*operands=*/{},
|
local_client_aot_test_helper.cc | 51 auto sum = builder.CustomCall("SumStructElements", {opaque_param}, r0f32);
|
/external/tensorflow/tensorflow/compiler/tf2xla/ |
xla_context.h | 126 // Allow ops to emit CustomCall operations for CPU.
|
xla_compiler.h | 241 // If 'allow_cpu_custom_calls' is true, kernels may make use of CustomCall()
|
/external/tensorflow/tensorflow/compiler/xla/tools/parser/ |
hlo_parser_test.cc | 829 // CustomCall 831 "CustomCall", 834 ENTRY CustomCall { [all...] |
/external/tensorflow/tensorflow/compiler/tests/ |
BUILD | 86 # ArgMax needs CustomCall on CPU, which is not available in normal
|
/external/tensorflow/tensorflow/docs_src/performance/xla/ |
operation_semantics.md | 601 ## CustomCall 604 [`ComputationBuilder::CustomCall`](https://www.tensorflow.org/code/tensorflow/compiler/xla/client/computation_builder.h). 608 <b> `CustomCall(target_name, args..., shape)` </b> 626 For example, if CustomCall is used as follows: 632 CustomCall("myfunc", {x, y}, f32[3x3]) 660 > native XLA ops whenever possible; only use CustomCall as a last resort. [all...] |