Home
last modified time | relevance | path

Searched refs:parallel_execute (Results 1 – 15 of 15) sorted by relevance

/external/tensorflow/tensorflow/compiler/mlir/tensorflow/transforms/
Dtpu_parallel_execute_sink_resource_write.cc49 tf_device::ParallelExecuteOp parallel_execute, Value result) { in GetSingleUseResourceWrite() argument
59 if (resource_handle_op == parallel_execute) return nullptr; in GetSingleUseResourceWrite()
63 parallel_execute.getOperation()->getBlock() && in GetSingleUseResourceWrite()
64 parallel_execute.getOperation()->isBeforeInBlock(resource_handle_op)) in GetSingleUseResourceWrite()
74 tf_device::ParallelExecuteOp parallel_execute) { in SinkResourceWritesIntoParallelExecute() argument
76 const int num_regions = parallel_execute.getNumRegions(); in SinkResourceWritesIntoParallelExecute()
85 Block& block = parallel_execute.GetRegionBlockWithIndex(i); in SinkResourceWritesIntoParallelExecute()
86 auto results = parallel_execute.GetRegionOutputs(i); in SinkResourceWritesIntoParallelExecute()
92 GetSingleUseResourceWrite(parallel_execute, result.value()); in SinkResourceWritesIntoParallelExecute()
115 parallel_execute.GetRegionBlockWithIndex(results_to_remove.index()); in SinkResourceWritesIntoParallelExecute()
[all …]
Dtpu_merge_variables_with_execute.cc140 auto parallel_execute = llvm::dyn_cast<tf_device::ParallelExecuteOp>( in BuildVariableAccessInfo() local
143 parallel_execute ? parallel_execute.getOperation() : execute_launch; in BuildVariableAccessInfo()
241 parallel_execute in BuildVariableAccessInfo()
242 ? parallel_execute.GetRegionOutputs( in BuildVariableAccessInfo()
317 tf_device::ParallelExecuteOp parallel_execute, int start, in AppendTypes() argument
321 Block& block = parallel_execute.GetRegionBlockWithIndex(index); in AppendTypes()
331 void ReplaceParallelExecute(tf_device::ParallelExecuteOp parallel_execute, in ReplaceParallelExecute() argument
336 Operation* parallel_execute_op = parallel_execute.getOperation(); in ReplaceParallelExecute()
346 AppendTypes(&output_types, parallel_execute, 0, region_index); in ReplaceParallelExecute()
352 &output_types, parallel_execute, region_index + 1, num_regions); in ReplaceParallelExecute()
[all …]
Dtf_passes.td339 … = "Extracts TPU outside compilation computation to a separate tf_device.parallel_execute region.";
345 a tf_device.parallel_execute region. The TPU cluster is also moved to a
346 tf_device.parallel_execute region. Communication ops between device and host are
363 will become a tf_device.parallel_execute op with a CPU/host region and
368 %0 = "tf_device.parallel_execute"() ( {
693 %1 = "tf_device.parallel_execute"() ( {
/external/tensorflow/tensorflow/compiler/mlir/tensorflow/tests/
Dtpu_parallel_execute_sink_resource_write.mlir6 // CHECK: [[PARALLEL_EXECUTE:%.+]]:2 = "tf_device.parallel_execute"
7 %0:2 = "tf_device.parallel_execute"() ( {
22 // CHECK: [[PARALLEL_EXECUTE:%.+]]:2 = "tf_device.parallel_execute"
23 %0:2 = "tf_device.parallel_execute"() ( {
37 // CHECK: [[PARALLEL_EXECUTE:%.+]]:2 = "tf_device.parallel_execute"
38 %0:2 = "tf_device.parallel_execute"() ( {
51 // CHECK: [[PARALLEL_EXECUTE:%.+]]:2 = "tf_device.parallel_execute"
52 %0:2 = "tf_device.parallel_execute"() ( {
64 // CHECK: [[PARALLEL_EXECUTE:%.+]]:2 = "tf_device.parallel_execute"
65 %0:2 = "tf_device.parallel_execute"() ( {
[all …]
Dtf_device_ops_invalid.mlir200 // Check that a parallel_execute op with a single region is not allowed.
202 "tf_device.parallel_execute"() ( {
203 // expected-error@-1 {{'tf_device.parallel_execute' op must have at least two regions.}}
211 // Check that a parallel_execute op with empty regions are not allowed.
213 "tf_device.parallel_execute"() ( {
214 // expected-error@-1 {{'tf_device.parallel_execute' op region #0 ('regions') failed to verify const…
224 // Check that a parallel_execute ops with invalid number of output types are
227 "tf_device.parallel_execute"() ({
228 // expected-error@-1 {{'tf_device.parallel_execute' op number of output types (3) must match the to…
242 // Check that a parallel_execute ops with mismatching output types are not
[all …]
Dparallel_execute_to_islands.mlir7 "tf_device.parallel_execute"() ({
35 %4:2 = "tf_device.parallel_execute"() ({
68 %3:2 = "tf_device.parallel_execute"() ({
96 %1:2 = "tf_device.parallel_execute"() ({
128 %2:2 = "tf_device.parallel_execute"() ({
Dtpu_extract_outside_compilation.mlir6 // Tests that TPU cluster with no outside compilation does not generate parallel_execute.
18 // CHECK-NOT: "tf_device.parallel_execute"
24 // CHECK: "tf_device.parallel_execute"
46 // CHECK: "tf_device.parallel_execute"
71 // CHECK: "tf_device.parallel_execute"
94 // CHECK: %[[PARALLEL_EXECUTE_OUTPUT:[0-9]*]] = "tf_device.parallel_execute"
122 // CHECK: %[[PARALLEL_EXECUTE_OUTPUT:[0-9]*]]:2 = "tf_device.parallel_execute"
147 // CHECK: %[[PARALLEL_EXECUTE_OUTPUT:[0-9]*]] = "tf_device.parallel_execute"
179 // CHECK: %[[PARALLEL_EXECUTE_OUTPUT:[0-9]*]] = "tf_device.parallel_execute"
201 // CHECK: %[[PARALLEL_EXECUTE_OUTPUT:[0-9]*]] = "tf_device.parallel_execute"
[all …]
Dtpu-merge-variables-with-execute.mlir229 // with TPUExecutes in a tf_device.parallel_execute.
231 // CHECK-LABEL: func @parallel_execute
235 func @parallel_execute(
242 // CHECK: "tf_device.parallel_execute"
243 %pe:2 = "tf_device.parallel_execute"() ( {
273 // tf_device.parallel_execute that is replicated (tf_device.replicate).
295 // CHECK: "tf_device.parallel_execute"
296 %pe:2 = "tf_device.parallel_execute"() ( {
Dtf_device_ops.mlir167 "tf_device.parallel_execute"() ({
178 "tf_device.parallel_execute"() ({
192 "tf_device.parallel_execute"() ({
Dtpu-dynamic-layout-pass.mlir410 // CHECK-LABEL: func @parallel_execute
411 func @parallel_execute(%arg0: tensor<*x!tf.resource> {tf.device = "/device:CPU:0"}) {
427 // CHECK: "tf_device.parallel_execute"
428 "tf_device.parallel_execute"() ({
502 // CHECK: "tf_device.parallel_execute"
503 "tf_device.parallel_execute"() ({
Dtpu_rewrite.mlir1229 // CHECK: [[PARALLEL_EXECUTE_OUTPUT:[0-9]*]]:2 = "tf_device.parallel_execute"
1267 // CHECK: [[PARALLEL_EXECUTE_OUTPUT:[0-9]*]]:2 = "tf_device.parallel_execute"
1391 // parallel_execute.
1403 // CHECK: "tf_device.parallel_execute"
1409 %3 = "tf_device.parallel_execute"() ( {
1448 // CHECK: "tf_device.parallel_execute"
1512 // CHECK: "tf_device.parallel_execute"
1548 // CHECK: %[[PARALLEL_EXECUTE_OUTPUT:[0-9]*]] = "tf_device.parallel_execute"
1586 // CHECK: %[[PARALLEL_EXECUTE_OUTPUT:[0-9]*]] = "tf_device.parallel_execute"
1623 // CHECK: %[[PARALLEL_EXECUTE_OUTPUT:[0-9]*]]:3 = "tf_device.parallel_execute"
[all …]
Dtpu-variable-runtime-reformatting.mlir204 "tf_device.parallel_execute"() ({
/external/tensorflow/tensorflow/compiler/mlir/tensorflow/utils/
Dxla_sharding_util.cc393 mlir::tf_device::ParallelExecuteOp parallel_execute) { in GetTileShardedOutputsToMerge() argument
401 const auto output_from_logical_device = parallel_execute.GetRegionOutputs( in GetTileShardedOutputsToMerge()
413 mlir::tf_device::ParallelExecuteOp parallel_execute, in HandleTileShardedOutputs() argument
417 builder->setInsertionPointAfter(parallel_execute); in HandleTileShardedOutputs()
422 cluster_func_output_index, sharding, parallel_execute); in HandleTileShardedOutputs()
540 mlir::tf_device::ParallelExecuteOp parallel_execute, in RemapOutputsFromLogicalDevices() argument
576 parallel_execute.GetRegionOutputs( in RemapOutputsFromLogicalDevices()
584 output_index, output_sharding, parallel_execute); in RemapOutputsFromLogicalDevices()
594 cluster_func_output, parallel_execute, in RemapOutputsFromLogicalDevices()
608 const auto output_from_logical_device = parallel_execute.GetRegionOutputs( in RemapOutputsFromLogicalDevices()
Dxla_sharding_util.h65 mlir::tf_device::ParallelExecuteOp parallel_execute,
/external/tensorflow/tensorflow/compiler/mlir/tensorflow/ir/
Dtf_device_ops.td130 def TfDevice_ParallelExecuteOp : TfDevice_Op<"parallel_execute",
165 // Checks if a tf_device.parallel_execute index'th region block wraps a
190 A tf_device.parallel_execute inside the tf_device.replicate op region may be
193 concurrent execution (i.e. region) defined by tf_device.parallel_execute op.