This project is mirrored from https://review.mlplatform.org/tosa/reference_model.
Pull mirroring updated .
- Feb 19, 2025
-
-
Missed being updated in recent RESCALE change. Signed-off-by:
Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I4ab4d93dfc74033191ec6fe2f07d54ba7b114c24
-
- Feb 14, 2025
-
-
Aligns avg_pool2d zero points to v1.0 of the sepc. Also add CTC tests for new zero point inputs. Signed-off-by:Luke Hutton <luke.hutton@arm.com> Change-Id: I5540b561c03fb660d89820e324bd2926e23e7bb9
-
- Feb 13, 2025
-
-
TatWai Chong authored
Create an arith right shift helper function based on the implementation reference in the spec. Explicitly use int32_t with sign extend and the helper above for ARS operation. Change-Id: Ie465ae212e97cbeab091084403edec25f9fac133 Signed-off-by:TatWai Chong <tatwai.chong@arm.com>
-
- Feb 12, 2025
-
-
Change-Id: I16ea5a6b056eb8bae5f340a502ce9e13f289bbdb Signed-off-by:Hampus Linse <hampus.linse@arm.com>
-
Jeremy Johnson authored
Update to remove erroneous support for unsigned INT48 and INT32. Add new ERROR_IFs to match specification. Fix the test data ranges for unsigned data, after signless patch incorrectly truncated the data. Rename the ERROR_IF tests within the generator to be Rescale specific. Enable CTS random const and lazy data gen modes. Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I93f9dbec008b1a9a0c86837aa812439f4101f29d
-
- Feb 11, 2025
-
-
Signed-off-by:Udaya Ranga <udaya.ranga@arm.com> Change-Id: I6b044c0295bd15b86c3ecba5a77398f6ef2691a5
-
- Feb 10, 2025
-
-
Tai Ly authored
Changed to use ERROR_IF instead of printNodeValidationError to align with spec. Signed-off-by:Tai Ly <tai.ly@arm.com> Change-Id: Ibeec6fcca63c3115560771e26cb6605ffb198ec1
-
Fix REDUCE_MIN, REDUCE_MAX, and MAX_POOL2D issue where reductions in which all input values where NaNs would wrongly yield the initialisation value when the nan_mode was set to IGNORE. Add utilities getApplyMinPadding and getApplyMaxPadding to standardise the way we get padding/initialisation values and updates MAX_POOL2D and ARGMAX to use those. Add unit tests to catch regressions and other wrong behaviour for FP_SPECIAL REDUCE_MIN and REDUCE_MAX. Signed-off-by:Ian Tayler Lessa <ian.taylerlessa@arm.com> Change-Id: I9644cc82c5b7390186f80dc3dfe14c5e062eaad7
-
Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I43c26cbf9a54eb22fcd9df56712d930b30d8ebe3 -
Fix "warning: explicit instantiation of 'writeValue' that occurs after an explicit specialization has no effect" Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ib111d089a62e1e92cc02476f534300cbdd9faa80 -
- Update the output tensor directly within the computation loop. - Apply the bias inside the loop instead of initializing it at the beginning. Signed-off-by:Eirene Vlassi Pandi <eirene.vlassipandi@arm.com> Change-Id: I5c4260356b74a8be30aee1d89a802e278052e1f8
-
- Update eval functions of AVG_POOL2D operator to use acc_t, consistently in intermediate operations, including subtracting the zero point from input in int8. - add unit tests to catch regressions in accumulator behaviour in AVG_POOL2D. Signed-off-by: -
Peng Sun authored
Switch RESCALE op zero point from attributes to operands * update serialization library submodule Change-Id: Icab0204f192f674a666bbe8011ca6ac37a60dbf3 Signed-off-by:
Jeremy Johnson <jeremy.johnson@arm.com>
Signed-off-by: Peng Sun <peng.sun@arm.com>
-
- Feb 05, 2025
-
-
Ian Tayler Lessa authored
Signed-off-by:
-
- Jan 31, 2025
-
-
Ian Tayler Lessa authored
Add intTrunc utility which is also used in CAST. Signed-off-by:
-
- Jan 29, 2025
-
-
CAST was not truncating values when downcasting integers, and table was silently allowing values outsides of the input range when that should be taken as an internal reference model error. Add unit tests for CAST checking that the reference model has the proper truncation behaviour. Signed-off-by:
-
- Jan 28, 2025
-
-
Complete re-write of RESCALE test generation to match spec Input files changed from UINT8/UINT16 to "signless" INT8/INT16 types Does not support UINT48 as input. Remove all UINT8/UINT16 dtype support from data generator library. Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ia1a11020dd398e3ed6c023115f2a97ecdd96983c -
Change conformance gen output when performing --list-tests-to-file operation to refer it to "Listing" instead of "Creating" Fix a problem when generating CTS tests using specific profile and extension options that won't produce ERROR_IF tests - this causes a generation failure - now the check is ignored if not producing all tests Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Icf2a9cec0253f328a7a95161b1d98282b7cad5f9 -
Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ibbfe2bf1de3f5a13758d1f15d3ecf60f65bfc999
-
- Jan 27, 2025
-
-
- Update eval functions of matmul operator to use acc_t, which in reference model captures the output type, consistently in intermediate operations, including subtracting the zero point from inputs and weights in int8. - add unit tests to catch regressions in accumulator behaviour in matmul. Signed-off-by: -
We were failing to find compiled dynamic libraries on macOS as we were assuming .so suffix Signed-off-by: -
Signed-off-by: -
macOS port: resolve ambiguous calls to std::isnan/std::signbit and friends when called using a softfloat Signed-off-by:
-
- Jan 22, 2025
-
-
Luke Hutton authored
This commit changes the zero point attribute to an input to align with the 1.0 spec. Also marks both as CTC (Compile Time Constants) and enables EXT-DYNAMIC tests. Change-Id: I175c45a0f25470fbef5d2c2087bf882fb7fcf46d Signed-off-by:
Jeremy Johnson <jeremy.johnson@arm.com>
Co-authored-by: Jeremy Johnson <jeremy.johnson@arm.com>
-
- Jan 21, 2025
-
-
Change from using double_round attribute to rounding_mode Add support for EXT-DOUBLEROUND extension Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I75dab269450a928966723dfe95fa27794b590ce1 -
Match the updated specification to allow the magnitude bound for dot product calculations to not be so strict. Also rename meta data "ks" to be "ksb" to match values passed via desc.json in tests. Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I0c77ec5bffd1ad7b4666483af8b652add0dddd06
-
- Jan 16, 2025
-
-
Eirene Vlassi Pandi authored
Signed-off-by:
-
- Jan 14, 2025
-
-
Enable EXT-DYNAMIC testing for RESCALE, RESIZE & MATMUL Improve notes about CTC support and shape_t inputs Add TODOs for future updates when shape_t can be expressed as something other than CONST_SHAPE for RESHAPE, SLICE & TILE Add TODOs for ops missing CTC support as waiting on zero point updates for AVG_POOL2D, NEGATE & PAD Clean up test data set lists - remove unused Add missing int48 RESCALE integer special test Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ifd85f46f0a117f5c5dc12013ed08ed2949dd8056 -
Jack Frankland authored
Make use of the explicit `ct::compat::cast` helper to avoid inncorect ambiguous constructor calls that result from a compiler bug in gcc-8.3. Signed-off-by:Jack Frankland <jack.frankland@arm.com> Change-Id: Iad2de6bda88e691ad5e6647550c8bbbde1622403
-
- Jan 13, 2025
-
-
Tai Ly authored
This patch adapts reference model to use new tosa serialization library schema where there is one attribute per operator Signed-off-by: -
Switch to using the generate library for all data generation. Also clean up compliance to always use verify library apart from ERROR_IF tests as all data types are supported. Add BOOL and INT48 support to fixed data generator. Alias SHAPE type support to INT32 for other data generators. Improve unsupported data type reporting from data generator. Updated fixed_data unit tests to test bool and other types. Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I45dd4df39aa8e88c3d84f5268ff3e1c8c79ca6d5 -
- Add input_unsigned and output_unsigned to conformance testing Signed-off-by:Won Jeon <won.jeon@arm.com> Change-Id: I339eb9685a14d22fc5dcb59bfe0d50726623be6f
-
- Jan 10, 2025
-
-
Also update the CTS generator to match and the serialization lib submodule. Signed-off-by: -
Signed-off-by:
-
- Jan 09, 2025
-
-
Jeremy Johnson authored
Include tests that mirror the ERF, SIGMOID and TANH table generation in the specification using FULL range testing to access all values. Add generate unit test for TABLE full range. Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I1a0d263bd13953d0383d437e7197cc5cf94f4142 -
- Add REQUIRE statements for OpPow for range checking. - Remove test set 2 with negative base inputs. - Remove FP special test data. - Refactor a to x, b to y, c to z to match with the specification. Signed-off-by:
Jerry Ge <jerry.ge@arm.com>
Change-Id: Ib89a2dba7b030f196af083c7160fe5ce28393fcf
-
- Jan 08, 2025
-
-
Jeremy Johnson authored
Align shift input to specification - rank 1 not rank 0. Change to using const as the default for the shift input. Align MUL reference code to pseudo code in spec. Update example LSTM test. Fix rank 0 data output to be rank 0 (instead of rank 1). Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ic9cdb0e50f0b01c9e7c446b2642586fd74e85ca0
-
- Jan 07, 2025
-
-
Jeremy Johnson authored
Enable integer special tests for CAST and RESIZE Enable generate unit tests Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I64bf8fe1f8718f20a9124ec23722211c43af6202
-
- Jan 06, 2025
-
-
Use data generator to produce valid test data using simplified value range Add special tests to check range of rescale Added UINT8 and UINT16 support to generate library Add unit tests for RESCALE and UINT8/16 Update test selection to include all input types, ranks and output types NOTE: Changes test names of some existing tests as it will use updated shape instead of first generated shape - POW, MUL, MINIMUM, GREATER, LOGICAL_LEFT/RIGHT_SHIFT, BITWISE_AND/XOR are affected Signed-off-by:Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Iea402dfbf5a2a741947606636ee18329b797885d
-
- Dec 30, 2024
-
-
Jack Frankland authored
Extend a linker fix for all gcc-8 versions requires executables/libraries to link against the C++ filesystem library explicitly. Signed-off-by:
-
