- Sort Score
- Result 10 results
- Languages All
Results 1 - 5 of 5 for TPURewritePass (0.15 sec)
-
tensorflow/compiler/mlir/tensorflow/transforms/host_runtime/tpu_rewrite_pass.cc
namespace { #define GEN_PASS_DEF_TPUREWRITEPASS #include "tensorflow/compiler/mlir/tensorflow/transforms/host_runtime/runtime_passes.h.inc" struct TPURewritePass : public impl::TPURewritePassBase<TPURewritePass> { explicit TPURewritePass(llvm::StringRef _module_name) : module_name(_module_name) {} void runOnOperation() override; llvm::StringRef module_name; };
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Tue Apr 30 21:25:12 UTC 2024 - 29.7K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/transforms/sparsecore/sparsecore_passes.td
tf_device.launch() ops. This is because the embedding op is head outside compiled and the compile op is wrapped in launch to execute on host during TPURewritePass. For example, the tf.OpA with the `mini_batch_splits` attribute will be moved after _TPUCompileMlir and the first input will use the _TPUCompileMlir program output: ```mlir
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Mar 28 23:42:09 UTC 2024 - 3.9K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/transforms/replicate_invariant_op_hoisting.cc
if (!read_var_op) return; // TODO(lyandy): Check if resource (first replica or replicate block arg) // shape has not changed in replicate prior to read. Currently after both // ResourceOpLiftingPass and TPURewritePass, there should not be any updates // to resources prior to their respective ReadVariableOp. if (auto block_arg = mlir::dyn_cast<BlockArgument>(read_var_op.getResource())) {
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Thu Apr 25 16:01:03 UTC 2024 - 7.5K bytes - Viewed (0) -
tensorflow/compiler/mlir/tensorflow/transforms/host_runtime/runtime_passes.td
See the License for the specific language governing permissions and limitations under the License. ==============================================================================*/ include "mlir/Pass/PassBase.td" def TPURewritePass : Pass<"tf-tpu-rewrite", "mlir::ModuleOp"> { let summary = "Rewrites a `tf_device.cluster_func` on TPUs into TPU runtime operations."; let description = [{
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Jan 10 18:58:57 UTC 2024 - 10.7K bytes - Viewed (0) -
tensorflow/compiler/mlir/g3doc/_includes/tf_passes.md
Both the embedding op and compile op are assumed to be wrapped in separate tf_device.launch() ops. This is because the embedding op is head outside compiled and the compile op is wrapped in launch to execute on host during TPURewritePass. For example, the tf.OpA with the `mini_batch_splits` attribute will be moved after _TPUCompileMlir and the first input will use the _TPUCompileMlir program output: ```mlir "tf_device.launch"() ({
Registered: Sun Jun 16 05:45:23 UTC 2024 - Last Modified: Wed Aug 02 02:26:39 UTC 2023 - 96.4K bytes - Viewed (0)