Repository navigation
Merge chips-IT contributions to regression_tests - #29
Open
RiccardoGandolfi wants to merge 11 commits into
Open
RiccardoGandolfi wants to merge 11 commits into
RiccardoGandolfi wants to merge 11 commits into
Conversation
* Working matmul with cluster_it != 0 * Fixed header file generator and source code. * Added cluster index input in timer functions. * Fixed references to code sections. * Adapted mchan tests header files to new memory map. * Removed some mchan tests from pulp_cluster-mchan-tests.yaml because CI takes too long. * Added RedMulE test. * Cleaner HWPE test integration, added DMA-based tensors initialization. TODO: understand why the core reads Xs during matrix checks. * Updated HWPE test, restored old CI script for mchan tests * Added carfield-related regressions. * Added DMR matmul to carfield regressions. * Add test for the FPUs of the cluster cores * Rename application in makefiles * Change matmul golden. * Add YAML script for the CI. * Update headers in parallel bare tests. * Add yaml file to automate execution of Carfield tests * Added regression test for TCDM scrubber, * Fix error-checking function. * Add fault injection capabilities to selected tests * Update ECC test to inject single or multiple bitflips * Use XOR over OR in ECC test return condition When simulating multiple bitflips, it is expected that `error != 0` * Fix fault injection toggle via CLI * Add astral target. * Add N-EUREKA regression test (not fully self-checking, very minimal) This is a temporary reg-test for N-EUREKA integration. * Fix version of pulp-nnx * Use symlinks to avoid code duplication * Update carfield.yaml * Update astral target * Fix ECC scrubber test return value * Added softex tests * Remove submodule SSH. * Added link to softex test * Updated softex test * Add fault injection in redmule regression to test ECC HCI * Add fault injection in neureka regression to test ECC HCI * Add redmule_softclear, redmule_256iter tests * Remove hardcoded addresses in softex and redmule tests * Move reliability tests to their own directory * Neureka: add configuration for the pulp cluster * Add cluster icache fault injection test * Move icache_fi_conv16 tests to reliability_tests directory * Fix neureka's check_output function * Add define to redmule and neureka tests to skip ecc-related checks if they are not implemented * Update `.gitignore` * carfield: Adapt Carfield's DMR test to 8 cores * Revert "carfield: Adapt Carfield's DMR test to 8 cores" This reverts commit 26d7629. * carfield: Set DMR test to 8 cores * [WIP] carfield: Comment printfs for DMR test * [WIP] Adapt redmule software to `2.0.0` version * Remove filters from RedMulE checking functions. * Add RedMulE golden model as a dependency. * Make HWPE read/write functions volatile to prevent compiler from swapping execution order. * Bump redmule golden. * Remove unintended file * NEureka: update submodule * Added first tests for iDMA: simple_transfer and 2d_transfer + python stimuli generation * Updated idma-tests.yml file * Update to 2d transfer test * Fixed and updated 2d-transfer test for iDMA * Updated simple transfer test * Updated idma_simple and idma_2d tests: added execution on all cores, + refined stimuli randomization * Further updates on 1D and 2D iDMA tests * Pushed updated stimuli * Minor fix to 2d transfer test: added missing ext2loc parameter in dma cpy function call * Updated idma_tests dir: added 3D transfer test + stimuli randomization + minor cleanups * Added README for iDMA tests * Updated README * Added multi-core test * Few updates to iDMA tests: replaced plp_dma_wait with plp_dma_barrier; Fixed idma_multi_core test * Updated idma_multi_core test * Cleaned multi-core test for iDMA * Updated multi core test for iDMA * Some clean-up on simple 1D and 2D tests * Aligned idma_multi_core 1D to latest version by Dkeller + multicore support * Updated test for 2D transfers: multi_core and single_core modes * Updated 3D transfer test, ongoing * Updated results comparison in 3D test * Added multicore code for 3D transfer test * Few improvements to multi core tests for iDMA * Updated tests: added multi-core serial mode; Some improvements * Updated Makefile for idma_multi_core_2d test * Updated idma_multi_core_1D: added L1 <-> L1 transfers * Added L1-L1 transfers in iDMA multicore tests * Fixed remaining errors in multi core modes: added individual plp_dma_wait functions depending on direction of transfer * Removed old tests for iDMA * Updated Readme + changes in 1d stimuli generator * Few updates to iDMA tests * Updated iDMA tests: memory allocation through pi_malloc functions and overall cleanup and reorganization * Updated iMDA tests: added QUICK mode for non-random inputs testing * Minor change on idma_multi_core: moved synch_barrier out of for loop * Updated Readme.md * Updated tests: initialize test_status to NB_CORES and decrease when no errors occur * Some clean-up and minor changes * Fixed core_id in free_allocated_mem function * Updated iDMA tests: added synch_barrier for serial multicore mode and switch for standalone_cluster/pulp functionality * Add pulp_cluster target for regression * Updated on iDMA tests --> initializing dst region to different values with respect to src region * Added iDMA test for max size transfer * Added mchan_1d test * Updated iDMA tests: added performance counters + unified stimuli generator * Added mchan_2d test * Added unified stimuli generator for mchan tests + stimuli rules in 1d and 2d Makefiles * Added iDMA test for PPA evaluation * Added benchmark test for mchan --------- Co-authored-by: Yvan Tortorella <ytortorella@lagrev3.ee.ethz.ch> Co-authored-by: Riccardo Tedeschi <riccardo.tedeschi4@studio.unibo.it> Co-authored-by: Riccardo Tedeschi <riccardo.tedeschi6@unibo.it> Co-authored-by: Yvan Tortorella <48508508+yvantor@users.noreply.github.com> Co-authored-by: Francesco Conti <f.conti@unibo.it> Co-authored-by: Andrea Belano <andrea.belano@gmail.com> Co-authored-by: Luigi Ghionda <luigi.ghionda2@unibo.it> Co-authored-by: Andrea Belano <andrea.belano@studio.unibo.it> Co-authored-by: Michael Rogenmoser <michael@rogenmoser.us> Co-authored-by: aottaviano <aottaviano@iis.ee.ethz.ch> Co-authored-by: Yvan Tortorella <ytortorella@lagrev5.ee.ethz.ch> Co-authored-by: Luigi Ghionda <luigi.ghio.98@gmail.com> Co-authored-by: Riccardo Gandolfi <riccardo.gandolfi@chips.it>
* Added queue of transfers to iDMA benchmark test * Added queue of transfers to MCHAN benchmark test * Updated IDLE testcase --> transfer done by core
…ures to iDMA * [TEST IDMA CLOCK GATING] Updated idma_benchmark_test with calls to clock gating drivers * [IDMA TEST CLOCK GATING] Added sw clock gating control in iDMA tests + improved printfs
…00MHz) (#26) * [FAULT_INJECTION] Re-align time window for fault injection with updated pulp cluster clock * [FAUL INJECTION SIM] Align fault injection time windows for neureka, redmule and ecc tests
* [FPU/HWPE] Build the cluster suite with corev-gcc - fpu_tests: float16/float16alt are RI5CY-toolchain builtins. Map float16 to _Float16 on CV32 and guard the half-precision helpers; FP16ALT stays unbuildable there because fcvt.s.ah has no CORE-V encoding, so fail with a clear #error rather than an assembler one. Drop -mno-memcpy (RI5CY-only), add a profile= knob for the perf counters. - neureka: inc/pmsis.h is a stub for pulp-runtime, which has no real pmsis.h. Under pulp-sdk -Iinc made it shadow the real one, leaving the SDK's own fll-v1.c without POS_LOG_INFO; defer via #include_next, keeping the deliberate PI_L2-in-L1 override. corev-gcc <=v0.4 ICEs in maybe_postinc expanding the inlined neureka_task_set_* helpers at -O1..-O3 and -Og, so pin nnx_layer.c to -Os. - pulp_cluster-subset.yaml: pulp_cluster set + all idma and tcdm tests at cores=8. The idma entries appear once, so no directory is built twice concurrently. * [iDMA Tests] Fixes to Makefile and printfs for iDMA tests * [TCDM TESTS] Add success/fail printfs for CI check * [HWPE] Add the FC-to-cluster prologue These five tests were written for the standalone cluster and only ever listed in the cluster-only suites, so they skipped the forward that parMatrixMul, fpu_tests and the idma tests all have. On pulp-open main stays on the FC, where get_core_id() is 0 too, and they drove the accelerators' cluster registers with the cluster never opened. bench_cluster_forward() re-enters main on the cluster; it is a no-op under ARCHI_NO_FC, so pulp_cluster is unaffected. * [TCDM]Fix L2 check test, avoid writing on the text section * [HWPE TESTS] Update printfs * [iDMA TESTS] Update idma-tests.yaml * [FPU_TESTS] Add switch depending on core type for -mno-mempcpy flag in fpu_tests makefile --------- Co-authored-by: petreiuss <pierangelo.rapa@gmail.com> Co-authored-by: petreiuss <79929488+petreiuss@users.noreply.github.com>
* [CI] Add platform variable to .yaml testsets * [iDMA Tests] Update main for idma_benchmark test * [iDMA] Update max_transfer_size test with clock gating control * [CI] Add no_ecc flag for hwpe tests + launch idma_max_size test just for gvsoc
* [Tcdm tests] Fix address range for L1_control_check test * [Tcdm tests] Better parametrize number of checks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.