Age | Commit message (Collapse) | Author |
|
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I0ef7607f4266296a1204c5cccdb5be36f345b5ba
|
|
Change-Id: I81097ae263a64ee8e35b8bd798598671ac089c3f
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
|
|
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: Iad035b31d5e5e83040068e6311501490765bfff7
|
|
This adds --precise_mode=1 option to tosa_referece_model,
which will cause reference model to convert all floating point tensors
to FP64 tensors and compute all operators accordingly.
Also adds optional -p arguments to test runners tosa_verif_run_tests.py
and tosa_verif_framework_compiler_runner.py to run tests in precise mode
Signed-off-by: Tai Ly <tai.ly@arm.com>
Change-Id: I156055216ad61710096497a8fa1a653be2a602a3
|
|
Operators updated: AVG_POOL2D, MAX_POOL2D, CONV2D, CONV3D,
DEPTHWISE_CONV2D & TRANSPOSE_CONV2D
tosa_verif_build_tests argument --level-8k-sizes used to
allow kernel/stride maximum boundary testing
Fixed bugs in height/width validator function meaning some
esixting avg_pool2d float tests need regening.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I7aeab82d3bd3c49d02d54708f2c9d995cd3cf2df
|
|
Add support for producing scaling 64 tests in RESIZE and make sure
default tests are not generated with scaling greater than 64.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I3e99cde7e6bec6ea7854a139c6fc2ed151af5b1e
|
|
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: I5689d7c6b902a319a68fa4628b59e0bcc23aeca4
|
|
Change-Id: I665acac9b5171efd0c5a2b68b516609048f6e187
Signed-off-by: Luke Hutton <luke.hutton@arm.com>
|
|
Split generation of conformance tests into groups to avoid a change
in existing tests selection
* A standard group is used for the normal/default conformance tests
* An 8k level group covers the extension of testing for 8k level
* Other arbitrary groups can be created and the group names will be
used to tag the tests, so that in the future they could be filtered
Group selection criteria is in a new dictionary
* A default selection config is used for generation groups
* But a different config can be used for a particular generation group
Added parameter sparsity for pad of rank6 and above, and further
parameter sparsity options to test_select to limit number of tests
selected.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ib9f1185c0ae6d0957df2116411c17e958157c6d3
|
|
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: I1e9e97ead447295e1252785106931b261df7bcea
|
|
Remove ERRORIF tests for CONST as they do not create valid
negative tests.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ia96b184f34010a452d2c4848b0653514352c751e
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ifce52ef0e082e9e6289938baa6c73a6ebd9ded62
|
|
Was causing the generated set of tests to be too big.
Change-Id: I45292e9317ab93786a1ec4670283385d06d52c0d
Signed-off-by: Eric Kunze <eric.kunze@arm.com>
|
|
in nearest mode
Note:
- align_corners and half_pixel can both be true in nearest mode for tfl
but not tf
- this is inconsistent between tf and tfl
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: Ib396f994874904e017e0d696f70370bd51d60161
|
|
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: I6fa14e9dbe7c0cd21156a5779d68fd473af322f2
|
|
Also add input size = 1 in the shape list, and extend scaling to
1x, 2x and 3x, so that the cases of broadcasting, power-of-two
scaling, no-scaling (e.g. 1x1 -> 1x1), scaling accuracy (3x) can
be tested. Since the scalar tensor is tiny, should not noticeably
impact the execution time of the framework test.
Change-Id: Iec53da3cbb60e087077d6e2d8eb205e76e6c1313
Signed-off-by: TatWai Chong <tatwai.chong@arm.com>
|
|
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: If18fc61514cbc2d31a15ed7dd87147b0d8e69f16
|
|
Signed-off-by: TatWai Chong <tatwai.chong@arm.com>
Change-Id: Ifb3ed9c448005d6d53c5d0938b5146003a26e8a3
|
|
* Restricting size of tensors due to expanded values of ofm_depth
Signed-off-by: James Ward <james.ward@arm.com>
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I582d8f0073c2d01244976f8e2e14783b506d8091
|
|
Updated to align with corresponding changes to the
spec.
In addition, some ERROR_IF tests have been updated to
match the checks specified by the spec, including:
PAD, SLICE, TILE, TRANSPOSE.
Signed-off-by: Luke Hutton <luke.hutton@arm.com>
Change-Id: Ie2c5f48e79a5610eb82739170e25057a63dac1d8
|
|
Signed-off-by: TatWai Chong <tatwai.chong@arm.com>
Change-Id: I4a49f45aa73044aff5b0a8b3dba58c1f52c1ae21
|
|
Change-Id: I4f967ce2dfb0cb44dc58307170fd9306a0ea1ed5
Signed-off-by: James Ward <james.ward@arm.com>
|
|
* Conformance regen required for tosa-bi: conv2d conv3d depthwise_conv2d transpose_conv2d
* Include bug fix for testGen.makeShape()
* Include json changes to limit size of tests which now have ofm_depth > 1
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: Ic8221b9a742f5737df523997bee7260f4dfcdef3
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I44ad6513fad831ff50fcc5ab801b46a5bf25d0a8
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I2c1e87ec586195f82e97bbc70e4e2bc18cd2cc88
|
|
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: I6c2087c94c2c024aabc2086d40cbca9c719c69dd
|
|
* Add TOSA test coverage
* Add some basic error checking
Signed-off-by: Luke Hutton <luke.hutton@arm.com>
Change-Id: Ib02df7319954a2dcb915e5f5a70699ca398fd483
|
|
Due to change in generator creating different test patterns.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I58224dc117c766b22fccd418f64004aea31905b5
|
|
Includes:
* FFT2d reference implementation
* Basic TOSA tests
Change-Id: Ie79fcb713542345d550ec013646810c1e890e388
Signed-off-by: Luke Hutton <luke.hutton@arm.com>
|
|
So they do not overwrite tosa-bi tests, ops affected:
NEGATE, TILE, REDUCE_MIN, REDUCE_MAX, RESIZE, SCATTER, GATHER
Update test-select exclude-patterns to work with ERRORIF tests.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Idb0d26512ce9a48f6d868663a630e343c8785899
|
|
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: I874c87745b929f9808a5e45490042420a57b22f7
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Iefd8490d9eedc76d199d36943b87c99e4dd97612
|
|
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: If67f503a1848967bc1671646c3011d055b622c52
|
|
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: I21b8f8139ce069d5fc6bbeabe292e3a44d99fe10
|
|
* Add exclusion regex's to conformance generation
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: I15bef7451efd5662065060242d35bd7fa3381487
|
|
Signed-off-by: Luke Hutton <luke.hutton@arm.com>
Change-Id: I11a891296a040f4e48187d0d51a4dd2c554805ac
|
|
This fixes framework test generation when the "optimized"
kernel mode option is not selected. As a result, the
optimized RFFT2d kernel is used when generating tests in
the reference kernel mode to prevent "op not found"
errors.
Signed-off-by: Luke Hutton <luke.hutton@arm.com>
Change-Id: Ice18bfa08c46e237e39c99f849622bf99d9b849c
|
|
The result comparison between Tensorflow runtime and the reference
model hasn't been checked as the sign operator is not supported by
the native TFLite runtime. That said, since the generated tosa ops
for tf.sign and tfl.sign is identical, the correctness presumably
can be proved by the result from tf.sign.
Change-Id: I72eb415df7fb6ca4dc9103f9ddc7104b0ba39234
Signed-off-by: TatWai Chong <tatwai.chong@arm.com>
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I4d4834260698db75427590c0a8b80cc3c824ef3b
|
|
Also fix the ERROR_IF conformance test generator to apply correct
operator filtering to create the tests.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I5788932e3529ec9386900623b20c22c98e4201b7
|
|
Includes:
* RFFT2d reference implementation
* TFLite framework tests
* Basic TOSA tests
* Serialization submodule upgrade with support for FFT/RFFT
Signed-off-by: Luke Hutton <luke.hutton@arm.com>
Change-Id: I2a687e9cf87fb62a26160ea52439ba9830bea36e
|
|
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: Ief60ae8fa490ded21aa77726d738565be5f8d91c
|
|
* Existing float attributes now serialized as bytes
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: I415276706b9daf0893e3a59189f387f872ff07c2
|
|
Signed-off-by: Luke Hutton <luke.hutton@arm.com>
Change-Id: I342a162e252c8527d5b318ceaf5f1de70fa5dc34
|
|
Rationale for making this change:
- In the original design, for control flow operators like WhileOp,
child blocks couldn't read the tensor variables (global consts) in the root level block,
this patch added the machanism for child blocks to access their parent
level block's tensors.
- This change also relies on another serialization change on adding
another layer of abtraction called Region:
- Serialization patch: [region] Add TosaSerializationRegion to serialization_lib
- Updated the corresponding python version of the serialization code: TosaSerializerRegion to python version of serialization_lib
- This change also relies on the TOSA MLIR Translator change: Add RegionBuilder to TOSA MLIR Translator
- Added the WhileOp related test cases: While, LSTM, GRU, RNN
- Other related fixes
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: I13ae33628ad07e41d248e88652ce1328654694ab
|
|
Add option to control negative tests dim range to stop ERROR_IF
test clashes between profiles.
Change-Id: I1f806934480cc32b95e80613c66fecb1e6ff5426
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I8d478b7bd13559ea37e27f77b05ee5db56c8f9e3
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I7276dc686d8d18ba44663b73e35ceca2a1cbaadf
|
|
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: I440590fa59433b9ddc1934188f2b34b377129c1f
|
|
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: I75d6280ecb5219d50c91d14029e90ff3a2f6eb06
|