Skip to content

Releases: alibaba/heterogeneity-aware-lowering-and-optimization

IPU_STABLE_SDK_2.3.1_v8

24 May 07:30
1c52231
Compare
Choose a tag to compare

ODLA-POPART

  • Fixed a bug in multi-thread initialization.
  • Fixed a bug in QManager initialization.

CMB Demo v0.2.4

11 May 06:53
Compare
Choose a tag to compare
  1. Update vODLA Adapter to avoid redundant file transformation.

v0.8.0

24 Apr 19:09
Compare
Choose a tag to compare

This release contains the following major changes since v0.7.2:

  • Enhance dynamic shape and dynamic batch support
  • Support TF switch/merge control flow
  • Support serialization of IR to protobuf format
  • Improve fp16 support
  • More ops support for Caffe, ONNX & Tensorflow
  • More optimization on fusion and instruction simplification
  • Bug fixes

ODLA:

  • Add Cambrian MagicMind support
  • Enhance ODLA/Popart support
  • Upgrade ODLA/TRT to support cuda 11.4.2 / TensorRT 8.2.0

IPU_STABLE_SDK_2.3.1_v7_2

19 Apr 12:22
c6b6bc5
Compare
Choose a tag to compare

This release contains the following major changes since IPU_STABLE_SDK_2.3.1_v7

  • Add tid print in vODLA Adapter

  • Modified vODLA Adapter error message in odla_ExecuteComputation.

  • Reset the QManager status when create the new queue to fit potential error handling scenario.

IPU_STABLE_SDK_2.3.1_v7

30 Mar 06:21
f702ef3
Compare
Choose a tag to compare
  • Check the execution mode when do the async computation

  • Make all resources used in thread intialized before start the thread

  • debug core dump problem with vodla device reuse

  • Add e.what() in print when catch exception

IPU_STABLE_SDK_2.3.1_v4_2

30 Mar 03:25
0e005de
Compare
Choose a tag to compare
  • add tid print in vODLA Adapter odla_ExecuteComputation

IPU_STABLE_SDK_2.3.1_v4_1

16 Mar 23:23
Compare
Choose a tag to compare
  • add dynamic shape

IPU_STABLE_SDK_2.3.1_v6

14 Mar 11:03
33093f1
Compare
Choose a tag to compare
  • callback function definition changes, and handle the error raised by poplar SDK

  • Move computation init to odla_BindToArgument to avoid early call odla_CreteContext before compuation graph constructed in ASYN call

  • Fixed include sequence error when using makefile.

  • Modified typedef vodla_context to keep consistence with vodh_sdk, and fixed several bug for odla_context.

  • callback function definition changes, and handle the error raised by poplar SDK

  • Move computation init to odla_BindToArgument to avoid early call odla_CreteContext before compuation graph constructed in ASYN call

  • Add the mode string for pipeline_async

  • Add more debug information for async context

  • return the error by callback

  • Use pipeline_async to create context when execution mode is UNKNOWN

  • fixed a bug about adapter in debug mode

  • fix for wrong simplify_preprocess option

  • fix vODLA Adapter data reuse bug.

IPU_STABLE_SDK_2.3.1_v5

01 Mar 10:27
0f52340
Compare
Choose a tag to compare
  • Implemented vODLA adapter in parallel setting.

  • Updated callback function and corresponding error handler.

  • callback function definition changes, and handle the error raised by poplar SDK

  • Move computation init to odla_BindToArgument to avoid early call odla_CreteContext before compuation graph constructed in ASYN call

IPU_STABLE_SDK_2.3.1_v4

20 Jan 06:29
fc97a57
Compare
Choose a tag to compare
  • Add information to check the cache sdk version and the working sdk version

  • Set the log level to info for test

  • Add environment variable for error injection test

  • format the code for lint check

  • Add information to show the ENGION OPTIONS set

  • Make it recompile when cache version mismatch

  • Remote the environment for error simulator to make it set by environment variable

  • Add error injector for temporarily test

  • Fix the problem when there's not the error injection config file

  • Move the error injector to the odla_CreateComputation to make sure it will be called for each inference request

  • Add try catch to computation create and init

  • make the error handling code at one place

  • Use config file setting for batch per step when it's inconsistent with ODLA set

  • set the log level when error injector used

  • pass the error code to caller

  • Make sdk version mismatch cache recompile when error injector set

  • Make the popart log level take POPART_LOG_LEVEL firstly

  • change format to pass lint

Co-authored-by: tianboh [email protected]