- 17 3月, 2022 1 次提交
-
-
由 TFLM-bot 提交于
-
- 16 3月, 2022 1 次提交
-
-
由 deqiangc 提交于
Add broadcast_to and broadcast_args into AllOpsResolver and also add missing broadcast_args from Makefile BUG=https://b/222716314
-
- 15 3月, 2022 2 次提交
- 11 3月, 2022 3 次提交
-
-
由 deqiangc 提交于
* Rename AllocateTail * Rename GetHeadBuffer, GetHeadUsedBytes and GetTailUsedBytes Co-authored-by: NTing Yan <94130036+tingyan19@users.noreply.github.com>
-
由 deqiangc 提交于
* copy operator BROADCAST_ARGS kernel from lite This is a copy without modification of the kernel and test for operator FILL from tensorflow/lite/kernels. Adaptations to micro and addition to the micro build to follow. This follows the step in main/tensorflow/lite/micro/docs/porting_reference_ops.md#5-port-the-op-from-lite-to-micro-pr3 * remove lite-specific code from copy of BROADCAST_ARGS Remove the bulk of lite-specific code from the micro implementation of operator BROADCAST_ARGS. - Flatten the namespace - Don't resize output tensors - Don't use gtest * Implement BROADCAST_ARGS operator for micro and add tests * change 2021 to 2022
-
由 TFLM-bot 提交于
-
- 10 3月, 2022 2 次提交
-
-
由 deqiangc 提交于
* copy operator BROADCAST_TO kernel from lite This is a copy without modification of the kernel and test for operator FILL from tensorflow/lite/kernels. Adaptations to micro and addition to the micro build to follow. This follows the step in https://github.com/tensorflow/tflite-micro/blob/main/tensorflow/lite/micro/docs/porting_reference_ops.md#5-port-the-op-from-lite-to-micro-pr3 BUG= * remove lite-specific code from copy of BROADCAST_TO Remove the bulk of lite-specific code from the micro implementation of operator BROADCAST_TO. - Flatten the namespace - Don't resize output tensors - Don't use gtest * Implement BROADCAST_TO operator for micro and add tests * Add BROADCAST_TO related files to Make based build system * Add two more tests * Fix bluepill test failure by explicity specify int32_t type instead of int * Update based on review comments * Remove BroadcastToContext struct * Completely remove BroadcastToContext * Ensure prepare and eval have unique names Prefer to have unique names for Prepare and Eval. See http://b/205184077 for background Co-authored-by: Nmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
-
由 TFLM-bot 提交于
-
- 09 3月, 2022 4 次提交
-
-
由 Fredrik Knutsson 提交于
Change-Id: I27a2d540b4733be5bb3e63deb18128c38c63d19a Co-authored-by: NTing Yan <94130036+tingyan19@users.noreply.github.com>
-
由 deqiangc 提交于
This mirrors internal cl/432564758. Rename the generic SetHeadBufferSize() to two separate functions ResizeBuffer() and ReserveNonPersistentOverlayMemory(). This is to highlight the real purpose of SetHeadBufferSize() in different places and remove the reference to Head to abstract out the implementation detail. This is the first step to abstract out the interface function from SimpleMemoryAllocator so that we can eventually support separate NonPersistent and Persistent arena, in which case,the Head is not meaningful. BUG=https://b/219833563Co-authored-by: NTing Yan <94130036+tingyan19@users.noreply.github.com>
-
由 TFLM-bot 提交于
-
- 08 3月, 2022 3 次提交
- 06 3月, 2022 1 次提交
-
-
由 TFLM-bot 提交于
-
- 05 3月, 2022 1 次提交
-
-
由 TFLM-bot 提交于
-
- 03 3月, 2022 2 次提交
-
-
由 deqiangc 提交于
-
由 Jens Elofsson 提交于
* Increase stack size for Arm Corstone-300 target. Increase stack size for to 0x50000 to make the micro_interpreter_test work. * Enable the micro_interpreter_test for Arm Corstone-300 * Update copyright header with the correct year. Co-authored-by: NAdvait Jain <advaitjain@users.noreply.github.com>
-
- 02 3月, 2022 5 次提交
-
-
由 Nat Jeffries 提交于
* Add method to reset resource variables. Call this method from ResetVariableTensors. * handle case where resource_variables_ is nullptr in micro_graph
-
由 Nat Jeffries 提交于
Walk the graph with a monotonically increasing allocation context count, entering all possible subgraphs invoked by each control flow operator. Increase allocation context count each time new node or subgraph is entered. Clean up and refactor some tensor lifetime code. This change increases arena requirements for some models, especially those with >1 subgraph. This is because the old subgraph memory allocation scheme was unsafe, with tensors's lifetime overlapping in a dangerous way between subgraphs. For models where each subgraph is only invoked from a single control flow operator (this is true for all graphs currently supported by TFLM), this new allocation scheme is optimal while maintaining lifetime guarantees across all tensors in each model.
-
由 deqiangc 提交于
* Ensure all allocate temp must be paired with decallocate After this CL, there will be an error if not all temp buffers are freed before ResetTempAllocations(). This is a way to enforce the TFLM's new paradigm: all AllocateTempXxx must be paired with DeallocateTemp() for better tracking of buffers. See go/tflm-pairing-allocatetemp-with-deallocatetemp for more info * Enable kernel runner check Co-authored-by: NNat Jeffries <natmjeffries@gmail.com>
-
由 psharath1 提交于
* Added support for int8 conv on Vision P6 * Change VISIONP6 to VISION_P6 in conv.cc * Change VISIONP6 to VISION_P6 in conv_vision.cc * Change VISIONP6 to VISION_P6 in xtensa.h * Change VISIONP6 to VISION_P6 in xtensa_conv.h * Change visionp6 to vision_p6 in P6 test script * Change visionp6 to vision_p6 in xtensa.inc * Change visionp6 to vision_p6 in xtensa_download.sh * Formatting changes for conv.cc Apply clang-format to format file * Formatting changes for conv_vision.cc Apply clang-format to format file. * Format xtensa.h with clang-format * Format xtensa_conv.h with clang-format * Correct typo in test_xtensa_vision_p6.sh * Increase arena size for person detection * Replace GetInput/GetOutput with micro context API * Remove trailing space Co-authored-by: NTing Yan <94130036+tingyan19@users.noreply.github.com> Co-authored-by: NTing Yan <tingyan@google.com>
-
由 TFLM-bot 提交于
-
- 01 3月, 2022 1 次提交
-
-
由 TFLM-bot 提交于
-
- 23 2月, 2022 1 次提交
-
-
由 rockyrhodes 提交于
-
- 22 2月, 2022 1 次提交
-
-
由 dependabot[bot] 提交于
Bumps [tensorflow](https://github.com/tensorflow/tensorflow) from 2.5.2 to 2.5.3. - [Release notes](https://github.com/tensorflow/tensorflow/releases) - [Changelog](https://github.com/tensorflow/tensorflow/blob/master/RELEASE.md) - [Commits](https://github.com/tensorflow/tensorflow/compare/v2.5.2...v2.5.3) --- updated-dependencies: - dependency-name: tensorflow dependency-type: direct:production ... Signed-off-by: Ndependabot[bot] <support@github.com> Co-authored-by: Ndependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
-
- 20 2月, 2022 1 次提交
-
-
由 TFLM-bot 提交于
-
- 19 2月, 2022 2 次提交
- 18 2月, 2022 2 次提交
-
-
由 deqiangc 提交于
This PR: 1. Update the tflite_files.txt with two additional files to be pulled in 2. Add build rules to include context_util, builtin_op_data.h and builtin_ops.h 3. Sync from upstream manually BUG=observe errors in automatically sync from upstream Co-authored-by: Nmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
-
由 deqiangc 提交于
This corresonds to internal cl/429336602 TFLM is replacing GetInput/GetOutput api with more informative micro_context-> AllocateTempXxx/DeallocateTempXxx api for better readability and better temporary buffer auditing. See go/tflm-pairing-allocatetemp-with-deallocatetemp for more info BUG=http://b/217818824Co-authored-by: NTing Yan <94130036+tingyan19@users.noreply.github.com>
-
- 17 2月, 2022 4 次提交
-
-
由 TFLM-bot 提交于
Co-authored-by: Nmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
-
由 deqiangc 提交于
common.c is renamed to common.cc in upstream github.com/tensorflow/tensorflow/commit/848e84c5a3acda31847457cbd52742436e024984 Hence, change the name to sync it properly BUG=observe error in sync from tf upstream
-
由 Ting Yan 提交于
-
由 Shlomi Regev 提交于
exist) BUG=217301557
-
- 16 2月, 2022 3 次提交
-
-
由 Advait Jain 提交于
BUG=#407
-
由 Ting Yan 提交于
* Add tensor_utils header files from TensorFlow Lite * Removed tensor_utils_common.h and added new file lstm_cell.h tensor_utils_common.h is absorbed into portable_tensor_utils.h in http://cl/428845263 and lstm_cell.h is separated from reference_ops.h in http://cl/428639888 and refactored in http://cl/428808334 for TFLM.
-
由 deqiangc 提交于
* Support building custom ops This is backport of internal cl/24778552 in order to support building custom op from tflite_micro by default. BUG=https://b/278725942 * Remove empty changes Co-authored-by: NTing Yan <94130036+tingyan19@users.noreply.github.com>
-