- Sort Score
- Result 10 results
- Languages All
Results 1 - 10 of 11 for XlaLaunch (0.12 sec)
-
tensorflow/compiler/jit/encapsulate_xla_computations_pass.cc
// the arguments into the order expected by XlaLaunch computations: // 1) arguments // 2) resource variable arguments // See the documentation of EncapsulateSubgraphsInFunctions for the meaning // of the arguments. // // TODO(b/113166435): Ordering constraints on XlaLaunch op can be relaxed. Status RewriteSubgraph(const std::vector<OutputTensor>& arg_source_tensors,
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue Mar 12 06:33:33 UTC 2024 - 15.1K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/transforms/tf_device_passes.td
This pass rewrites `tf.PartitionedCall` and `tf.StatefulPartitionedCall` operations with `_xla_compile_device_type` attribute in a `tf_device.cluster` into `tf.XlaLaunch` operations. This makes the attached function execute with XLA. `tf.XlaLaunch` requires resource-type arguments come at the end, so this pass rewrites the called function if necessary. This pass assumes there are no nested `tf_device.cluster`s so we don't end
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Apr 17 18:52:57 UTC 2024 - 12.5K bytes - Viewed (0) -
tensorflow/compiler/jit/flags.h
public: // Allow using Device API (PjRt) for `device_type` in the XlaLaunch op. // Please note that `enabled_for_xla_launch_` needs to be true in addition // to the `device_type` being allowed in order to use the Device API for // single device compilation and execution in the XlaLaunch op. void AllowForDeviceInXlaLaunch(const DeviceType& device_type) {
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Apr 17 18:52:57 UTC 2024 - 14.5K bytes - Viewed (0) -
tensorflow/compiler/jit/kernels/xla_ops.cc
if (ctx->has_input(i) || ctx->has_input(++i)) { ctx->set_output(0, ctx->input(i)); } } REGISTER_KERNEL_BUILDER(Name("XlaLaunch").Device(DEVICE_CPU), XlaLocalLaunchOp); REGISTER_KERNEL_BUILDER(Name("XlaLaunchV2").Device(DEVICE_CPU), XlaLaunchV2Op); REGISTER_KERNEL_BUILDER(Name("XlaLaunch") .Device(DEVICE_GPU) .HostMemory("constants")
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Fri May 17 22:46:36 UTC 2024 - 41.4K bytes - Viewed (0) -
tensorflow/compiler/mlir/tfrt/tests/mlrt/tf_to_mlrt.mlir
%unused = "tf.TestAsyncIdentity"(%x) {__op_key = 0: i32, T = i32} : (tensor<i32>) -> tensor<i32> // CHECK: mlrt.await_all_control [[unused]] return %x : tensor<i32> } // ----- // Test for XlaLaunch func.func private @xla_func_0(%arg0: tensor<1x3xf32>, %arg1: tensor<1x3xf32>) -> tensor<1x3xf32> attributes {tf._XlaMustCompile = true, tf._noinline = true, tf._original_func_name = "should_not_be_used"} {
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Fri May 31 20:44:15 UTC 2024 - 24.7K bytes - Viewed (0) -
tensorflow/compiler/jit/xla_launch_util.h
int device_ordinal, bool allocate_xla_tensors, bool use_multiple_streams); // Builds a XlaCompiler::Argument vector from the arguments to an XlaLaunch // op. // Precondition: variables in `variable_args` are locked. static absl::StatusOr<std::vector<XlaCompiler::Argument>> BuildXlaCompilerArguments(absl::Span<int const> must_be_constant_idxs,
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Feb 21 09:53:30 UTC 2024 - 11.8K bytes - Viewed (0) -
tensorflow/compiler/jit/encapsulate_xla_computations_pass_test.cc
auto v = ops::Placeholder(scope.WithOpName("V"), DT_RESOURCE); auto w = ops::Placeholder(scope.WithOpName("W"), DT_RESOURCE); NameAttrList function; function.set_name("launch0"); auto launch = ops::XlaLaunch( scope.WithOpName("launch0").WithDevice("/gpu:0"), std::initializer_list<Input>{}, std::initializer_list<Input>{a, b, c, d}, std::initializer_list<Input>{u, v, w},
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Mon Oct 16 18:03:15 UTC 2023 - 14.7K bytes - Viewed (0) -
tensorflow/compiler/jit/compilability_check_util.h
const tensorflow::FunctionBody* fbody, absl::Span<int const> constant_arg_indices, absl::Span<int const> resource_arg_indices); // Returns output memory types. // // XlaLaunch kernel keeps all outputs (including constants, which it copies), // in device memory except for resources. tensorflow::MemoryTypeVector GetOutputMemoryTypes( const tensorflow::FunctionBody* fbody);
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Sep 06 19:12:29 UTC 2023 - 14.9K bytes - Viewed (0) -
tensorflow/compiler/mlir/tf2xla/internal/passes/clustering_passes.td
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue Apr 30 02:01:13 UTC 2024 - 19.8K bytes - Viewed (0) -
tensorflow/compiler/mlir/tfrt/tests/hoist_invariant_ops.mlir
attributes {tf_saved_model.exported_names = ["main"]} { %0 = "tf.VarHandleOp"() {device = "/device:CPU:0", container = "", shared_name = "variable"} : () -> tensor<!tf_type.resource<tensor<1x3xf32>>> %1 = "tf.XlaLaunch"(%arg0, %0) {device = "/device:GPU:0", function = @xla_func, operandSegmentSizes = array<i32: 0, 2, 0>} : (tensor<1x3xf32>, tensor<!tf_type.resource<tensor<1x3xf32>>>) -> tensor<*xf32> func.return %1 : tensor<*xf32> }
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Mon Apr 01 23:54:14 UTC 2024 - 18.3K bytes - Viewed (0)