Age | Commit message (Collapse) | Author |
|
Change-Id: I20fa06c560213d572c7a5f7d5940d43aea4822a6
|
|
|
|
If the driver can't handle the call, don't use it.
Bug: 67310896
Test: System tests and VTS tests
Change-Id: Ia44b324a0f6116395967812d05c22b99b131608d
|
|
Change-Id: Icc95a2df808220c564540cb4c41e1bdc90ae93fe
|
|
- The failing tests incorrectly expect the data to be copied right
away when calling setOperandValue on large tensors.
- Making the affected tensors static so that they are still valid till
Model.finish() is called.
Bug: 67603060
Test: NeuralNetworksTest pass
Change-Id: I2ad627b9022eae6144d95e06d310ea3d8a2ec42e
|
|
Test: ran NeuralNetworksTest on marlin" into oc-mr1-dev
|
|
Bug: 67331808
Test: ran NeuralNetworksTest on marlin
Change-Id: I4b879f8a67bc366bffb12601dba7f021659a6921
|
|
Change-Id: I4108aef1f08312bfaa33ed06c3a6a8394096396e
|
|
Because of a limitation in HIDL, large tensor values (e.g. weights)
need to pass in shared memory rather than as HIDL arguments. This
prevented large models from running on a driver.
Separated the handling of memory pools so that request and model
pools are not comingled.
Also improve logging so we see more details about the models when
logging.
Bug: 67603060
Test: Ran system tests and VTS tests.
Change-Id: I760e31275699f9306c4b21945935dc3a4ca68754
|
|
|
|
|
|
|
|
|
|
Bug: 67427635
These are auto-generated tests. Updates the specs so that states
are not zeros.
Test: NeuralNetworksTest
Change-Id: Id419d020a8c0abc1bf03242c99ad03b2cb192ccb
|
|
Generate MobileNet quantized and prevent the test from being
overwritten. It was with a different suffix from the one recognized by
the test geneartor script, and thus effectively it was ignored.
This CL updates the suffix of the test spec for MobileNet. Also, since
the example file is generated elsewhere, we want the generator script
not to overwrite it.
Also, skip MobileNet quantized entirely in VTS as it is known to fail
at this moment.
Test: NeuralNetworkTest on Angler
Test: VtsHalNeuralnetworksV1_0TargetTest on CPU sample driver
Bug: 67394906
Change-Id: Ia944aeaa74e9318014f34cfc2fea6dcbf3f0f1ba
|
|
Bug: 67716699
Pulled in the latest implementation of tensor utils from TF Lite,
which has enabled NEON for tensor maths.
Test: NeuralNetworksTest
Change-Id: Ic1abbb40c1a8f9c218ea20e58108c6f88e044369
|
|
Bug: 67427635
These tests got broken during the transition from the stateful API
to a stateless API.
This CL fixes them (LSTMTest, RNNTest, and SVDFTest) to correctly
use double buffering to manage states in the client, i.e., swapping
the two buffers for states before and after the ops across inferences.
Test: LSTMTest: adb shell /data/nativetest64/lstm_test/lstm_test
Test: RNNTest: adb shell /data/nativetest64/rnn_test/rnn_test
Test: SVDFTest: adb shell /data/nativetest64/svdf_test/svdf_test
Test: NeuralNetworksTest (not affected)
Change-Id: I77e34e42613174ac793410c154f9ecad86e03165
|
|
oc-mr1-dev
|
|
While debugging setValueCode, I found that I was not getting log
trace from the sample driver. It's because the vlog initialization
was done when the driver started. If you do a setprop afterwards,
nothing happened. It's better to do the call on getCapabilities.
Bug: 67724550
Test: Compiled and ran sample drivers
Change-Id: I5068851c145a2e2aaff43f3d783362b0714a385d
|
|
Change-Id: Ibcc15afba99f85d3399aba64f3563a43b48da024
|
|
We believe having a +/- 1 tolerance for quant8 types are appropriate
for the current set of tests we have, as some devices have different
rounding modes from the ones of CPU.
Bug: 67644069
Test: VtsHalNeuralnetworksV1_0TargetTest on Angler with sample driver
Test: NeuralNetworkTest on Angler
Change-Id: I2aea18d2d8cfc2a2de2e50ceb1fa4631b10f8615
|
|
Regenerate VTS tests after test generator updates in ag/3050465.
Bug: 67607461
Test: VtsHalNeuralnetworksV1_0TargetTest on Angler with sample driver
Change-Id: I23898061809fe371049337d8c103c05427a728ee
|
|
|
|
|
|
Fixed how test generator enumerate output operands. Instead of using
set(), use a list and return unique output operands on the fly when
queried.
Also updated test generator's test suite with recent fixes on FP literals,
and added new testcases.
Bug: 67607461
Test: NeuralNetworkTest on Angler
Test: tools/test_generator/tests/test.py
Change-Id: Ic519035974ea5acedef6730acc1827d4f768c21d
|
|
Change-Id: I2b2db8f065a3692266f0af409aefb7f55aff1e41
|
|
* changes:
Close file descriptors in the TestMemory test.
Fix a bad depthwise_conv test.
|
|
|
|
Bug: 67650395
Test: Loop the affected test 1000 times and verified the bug is fixed.
Change-Id: Ib4b7ed567828511769fd7cb52e51e1e5d560e454
|
|
Bug: 67650128
Test: mm
Test: NeuralNetworksTest pass
Change-Id: Ibfd68139dfb30bfadf4898de964c66fbd584f2d0
|
|
padding" into oc-mr1-dev
|
|
Bug: 63905942
Test: mma (userdebug)
ml/nn/runtime/tests (userdebug)
Change-Id: I0c673b052033b1d0f2686bca193a675dd7fd03b5
|
|
63905942 Test: ran NeuralNetworksTest on marlin" into oc-mr1-dev
|
|
Bug: 67459279
NN API only supported float tensors when this op was first added.
Now int tensors are allowed in the API and we have updated the
documentation saying the lookup indices are integers, we need
update the implementaion and tests of this op to match the
documentation.
Also reverted the input order to match TF Lite.
Test: NeuralNetworksTest
Test: adb shell
/data/nativetest64/embedding_lookup_test/embedding_lookup_test
Change-Id: Ie0db884ca772e8fb388876c3cef609016e753bf1
|
|
HASHTABLE_LOOKUP Bug: 67459279 Test: locally run cts NeuralNetworksTests on marlin" into oc-mr1-dev
|
|
Bug: 63905942
Test: ran NeuralNetworksTest on marlin
Change-Id: I11df832b5e69479b7fe4e8beff9c0add2ed9992c
|
|
HASHTABLE_LOOKUP
Bug: 67459279
Test: locally run cts NeuralNetworksTests on marlin
Change-Id: Ib0bc48b3d0a2403820989d8c7d517081f5a9b647
|
|
Change-Id: Icf7d7d45b45720a75a0e6e0a567d0b8d200e76a0
|
|
Bug: 67478959
Test: mm
Test: NerualNetworksTest pass
Change-Id: I3f959031e5d31e2b01ee613b4f1038df883b708d
|
|
- For operands with OperandLifeTime::MODEL_OUTPUT, the dimensions,
type, and other meta-data must match the output Shape calculated
from the operation preparation step.
- Fix the ill-defined tests caught by the added validation.
- Incidental changes: generated more tests from tests specs.
Bug: 67390841
Test: NeuralNetworksTests pass
Change-Id: I40d35db0f7a868feae773dbf7e12cf4bf5f5e275
|
|
|
|
|
|
This CL turn them into constant operands set at model building time.
Also fixed problems in the test generator on handling FP literals in
model paramters.
Original tests are renamed with _weights_as_inputs suffix.
Bug: 67015862
Bug: 63905942
Test: NeuralNetworksTests on Angler
Change-Id: Ic59c78e7e6fcb143fb05a0360baa281b7528050c
|
|
Fix unaddressed comments in previous CLs on test harness; rewrite macros
using templates.
Bug: 67597726
Test: NeuralNetworksTest on Angler
Change-Id: I19fe804a71554246b953471e2966335cfd9170e5
|
|
Bug: 67503562
Test: none needed
Change-Id: Ib59ce8d8dbe814b835c400081a286a0514b5f1be
|
|
Change-Id: I6e38543850bc1e7d349c4240ea9388bbff8f6c5d
|
|
- Existing DEBUG and INFO change to the new mechanism, existing ERROR and
WARNING unchanged.
- By default, all verbose logging are disabled.
- setprop debug.nn.vlog 1 enables all verbose logging.
- setprop debug.nn.vlog "tag1 tag2 ..." only enable the selected tags.
- Available tags:
all: same as 1.
model: enable logging with tag MODEL.
compilation: enable logging with tag COMPILATION.
execution: enable logging with tag EXECUTION.
cpuexe: enable logging with tag CPUEXE.
manager: enable logging with tag MANAGER.
driver: enable logging with tag DRIVER.
Bug: 63905942
Test: mm
Test: NeuralNetworksTests pass
Test: manually set property to allow individual module to logging
verbosely
Test: tested with debug.nn.partition default -> 0 -> 1 to make
sure getProp still works fine
Change-Id: Iaa0ffa62176dabcdc35058748597df720fd6e47e
|
|
|
|
Bug: 63905942
Test: mma (userdebug)
ml/nn/runtime/tests (userdebug)
(no failures, logcat looks plausible)
Change-Id: Ib9a5eb75b9e4e2c3534f392d5fb0e5ba3555197d
|
|
67381597 Test: locally run cts NeuralNetworksTests on marlin" into oc-mr1-dev
|