Age | Commit message (Collapse) | Author |
|
With recent changes for Tile and Reshape operators some input
tensors are now required to have a "shape" type but it's not
supported by the ModelRunner. This patch adds ability to set
input tensors of type TOSA_REF_TYPE_SHAPE.
Signed-off-by: Dmitrii Agibov <dmitrii.agibov@arm.com>
Change-Id: I3b721e933548d0dd72319aeb2de01ae249ccb6a6
|
|
Update serialization_lib to store SHAPE as INT64.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ie589cd6670dc79b77df981c81cd7c27b982f20fa
|
|
Update data generator for main compliance values.
Add test generation support.
Fixed test set by including large 65k tests that were missing.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I8668c774e01c17e5d999aadf99c317e2dd893857
|
|
Added DEPTHWISE_CONV2D data generation.
Updated test generation for FP16 and FP32.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I0471d0a1e4e279a27233f4d285082906ceea1bff
|
|
Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I85595b1ea82e3a9db8e5f654767a911fa58c842f
|
|
Signed-off-by: evacha01 <evan.chandler@arm.com>
Change-Id: Ib9d0abe0bd8826ca5ac051a5fe6ba18e1ba727f1
|
|
Signed-off-by: Won Jeon <won.jeon@arm.com>
Change-Id: Id37100ba8bc2ac64b1f54788c6f765fedfab0816
|
|
* Int8/UInt8 reference outputs were previously saved to INT32
* Save those in their native dtypes and updated other affected code
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: Id65fc8773150d3c56bc2c72789a6a0d3c78cd363
|
|
Signed-off-by: Won Jeon <won.jeon@arm.com>
Change-Id: Ie80570146601c470a3be7c04a9d6e1016a7c547c
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Id0030fbd1b3b7a526dc17e947a433b4957dbf5e4
|
|
Also fix the dimension mask out logic that only set batch dimension
to unknown but others won't.
Change-Id: I9e1d2c3bb1d24cba1242103aa2c7609ef0c2c0b3
Signed-off-by: TatWai Chong <tatwai.chong@arm.com>
|
|
Dynamic shape model cannot directly run on the refenence model as
the concrete size of tensor is unknown therefore the volume of tensor
is not able to be allocated. Furthemore, the operators also expect
the input model is static-shaped.
This change turns dynamic model to static model prior to execution.
- Add `ifm_dynamic` field into json description to indicate whether
the model has dynamic shape or not.
- Add the shape inference pass into the compilation pipeline, firstly
legalize the dynamic tf/tfl model to dynamic tosa model with unknown
shapes, and then run the shape inference pass with static shapes
input argument to resolve unknown dimensions.
Change-Id: I5d2ffd452becc562dc30546789705bd01dd7a0b0
Signed-off-by: TatWai Chong <tatwai.chong@arm.com>
|
|
Signed-off-by: Won Jeon <won.jeon@arm.com>
Change-Id: Iaf59472cb32e03a92bad87ae9dba1b7548f20268
|
|
- fixed up reshape conformance tests to use shape input instead of attribute
- fixed up tile conformance tests to use shape input instead of attribute
- fixed output and output rank of dim op
- allow rank 0 and rank 1 tensors for tosa.shape values (for shape = {})
- added initialization of rank 0 const_shape tensors (for shape = {})
- Update conformance tests to use new rescale attributes
Signed-off-by: Tai Ly <tai.ly@arm.com>
Signed-off-by: Won Jeon <won.jeon@arm.com>
Change-Id: I6cce0d2a9ab066fe20a2abf9d2cfde3eb3d8c18b
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I7f7449c652844097fec6226bcc32ca765b40bc04
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: If08c560eac171124641dac66d7de9d5053db72df
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I7276f2db39e67314c950e972cc1a97b7796dcd18
|
|
Add support for Fp64 tensors in the eager runner's
helper functions, when precise mode is enabled.
Signed-off-by: Fabrizio Indirli <Fabrizio.Indirli@arm.com>
Change-Id: Ib737c0d18fb1c7ac40ce6ea03a4fbcefae88ba5c
|
|
int8/int16 should be used for the clamping, not int32_t
Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: I18209ca76cc83d95dc61f20f88344aafdbd72033
|
|
Update the minimum cmake version to 3.16 and set policy CMP0135
Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: Idd96c7a318f9458e39d38cd07cf29b9ac6ee63ad
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I466c1a8cb7c4cb3a8f6f18cf9885e7eb5e9819dd
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I7b0b14475fc4397baa404d57e03d6ca2f73ea746
|
|
Update and fix REDUCE_PRODUCT compliance verify lib support.
Added compliance test generation with data range to not cause infs.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I3b3004c6caa80d97e330a6393f435f5270b56e21
|
|
Add lower bound to ABS ERROR checks to allow for cancellation of
small values in error bounds checking.
Re-adjust the error bounds multiplier to match the specification.
Fix up naming of verify library info structs.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I3e178c3d7d59fef9c3696178646b23ed2a3ffc61
|
|
The operators API generated by the script is no longer used
and could be removed from the project.
Signed-off-by: Dmitrii Agibov <dmitrii.agibov@arm.com>
Change-Id: Ia611b069463b3aded7d6546987c2323674184673
|
|
- Found the error for framework test generator when it's not generating
dynamic_shaped tests
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: I54bf1422026f7f962d3f6ed0be9bd2e4dade14c5
|
|
- For Casting from Float to Integers, if the input float is greater
than INT_MAX, an overflow will happen when calling rint which causes the
clipplings to be ineffectives
- Moved all the range checks and clippings before rint to avoid this
issue
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: Ic189d59685b6d36464e3ef26766665148a660a14
|
|
- With input of 2147483648.00, the output overflows to -2147483648
- The root cause is the following:
- std::rint still returns float, the existing implementation is
forcing a cast from that float to int32_t
- when the input is over INT32_MAX, the output right after rint will
overflow which casues the clipplings later to be ineffective
- Instead, perform the range check before rint
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: Ib5a8cfd98aea17e326f8b11097beeb2d2b3efac9
|
|
- Added a basic infrastructure for allowing generate network with
dynamic_shapes
- Added tests cases for
- batch_to_space
- depth_to_space, space_to_depth
- linear
Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: Ie3f13231a74485df64b852f554cfe65e995f0d03
|
|
Test generation changed to only produce static reshape tests
Reference model changed to produce ERROR_IF on inferred shapes
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I92c92a40e7c0e457961bc654630040dff79a750b
|
|
Added indices shuffling and random INT32 support to generate lib
with testing of these new random generator modes
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I058d8b092470228075e8fe69c2ededa639163003
|
|
Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Change-Id: Ife703534594713616a992c460d201a8666c7b770
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I35d14e3e9f80198c1da3d267f12bc7a9a055e698
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I3555e7216d403d436bf6e39d4b16bb000645c4bb
|
|
Operator coverage restored/improved:
* REDUCE_SUM
* AVG_POOL2D
* FULLY_CONNECTED
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ie622b2f97b5c34971fdde5b20df3d616171ac11f
|
|
Some operators (e.g. Concat) expect list of tensor as an input
parameter. Currently operators API does not support passing
such parameters from the client code.
In order to enable it:
- Add new type tensor_list_t
- Update operators API generation script to support new type
- Add unit test for operator Concat
Signed-off-by: Dmitrii Agibov <dmitrii.agibov@arm.com>
Change-Id: Ib2f61bcea5e5ecabf56ce031d905cb46a4cc68ea
|
|
Implement support of int16_t inputs through int32_t underlying storage
buffers. Values are upcasted/downcasted as needed through explicit copy.
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I3752d98531c859002539bf1cb65413ceeff05e95
|
|
Use platform agnostic format specifiers for `int64_t`.
Change-Id: I002d94c1a0c0431ec09fc165a584a8f4b3ddc17d
Signed-off-by: Jack Frankland <jack.frankland@arm.com>
|
|
Support relative desc.json paths in refmodel of just "desc.json".
Catch when generate and verify library paths are None and provide
help in tosa_verif_check_result.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ie52ac399ea002e5fcdcc1eec3d9df6153a778e88
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Icf04afc7fdae8f506ba4710aaa085d6ea53bb5bf
|
|
FP16 support for all existing operators for compliance:
* DOT_PRODUCT
* ULP
* EXACT
* ABS_ERROR
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I8d25448a793375b53880da3787d8f839767f02cf
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I750ef5d3c95daca10e7717437ed375a81b434582
|
|
Limit CAST input tensor to maximums of output type to avoid
saturation and infinity.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I33350a4ce0ec828da7d2e7aa8cd3183a89a97431
|
|
POW - there are now 3 test sets to cover random ranges. Also added ROUND
mode to data generator to force integer exponent values.
LOG, EXP, RSQRT, REDUCE_SUM & FULLY_CONNECTED - have had their ranges
reduced for each test.
Fix generate library configuration defaults and checks.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ie5d3bd78f690cc787a2ca4eb9b4bd6808bd9238c
|
|
* Change ignore list in generate_api.py to generate operators
information.
* Fix serialization attributes mapping for operator FFT and RFFT
* Add a unit test for Fft2d and Rfft2d operator
Change-Id: I3ad7a77a3c46aa586834188bab42cbdcc423e834
Signed-off-by: Dhruv Chauhan <dhruv.chauhan@arm.com>
|
|
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Id6229cfaccad866b110630119eb045dbf6453bf5
|
|
Increase exp2 allowed range to account for denormals.
Minor adjustments to verify to match spec updates for pseudo code.
Set ranges of activation function inputs to match spec.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I6fcf665932ac2c9080e284b865da8f7746801f59
|
|
Make test_verif_result_check output to only say result files
are non-compliant as it may not be implmenetation result issues.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I14d10a7ee3aeff7802b6b173280d9f2054760a08
|
|
Fix test set 4 to have more symmetrical data set for AVG_POOL2D.
Increased size of AVG_POOL2D test shapes to meet MIN_DOT_PRODUCTS.
Print out sign of bias magnitude in verify library to help debugging.
Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Icf139cf5788cbc68f2c537567bdfb4cfe31274df
|
|
The bias operand of the fully connected operator must be a 1D tensor
either equal to the output channel size or of size 1. Previously we
asserted the former case, we now include the second case.
Signed-off-by: Jack Frankland <jack.frankland@arm.com>
Change-Id: I07dbc8a3aa1650703e5c50e1e7f36bb9539fd5db
|