47 Commits

Author	SHA1	Message	Date
NiccoloN	62b0a6e19d	merge remote changes	2026-05-03 22:30:46 +02:00
NiccoloN	b605585b1f	compact spatial IR through different new operations and dedicated syntax ... fast spatial node merging with batch operations	2026-05-03 14:14:14 +02:00
NiccoloN	15e8edb9c4	better spat computes merging	2026-04-25 19:24:09 +02:00
NiccoloN	5545b0f672	fix MatMul pattern non-contiguous extract_slices	2026-04-23 14:44:30 +02:00
NiccoloN	89b3501aa8	fix weightAlways attribute in spatial	2026-04-23 10:04:47 +02:00
NiccoloN	412ca957f6	multiple-output spat computes	2026-04-23 09:28:57 +02:00
NiccoloN	85e2750d6c	faster (and refactored) DCP analysis	2026-04-21 12:33:44 +02:00
NiccoloN	95ae93e07d	promote weight inputs to actual weights in spat compute nodes	2026-04-14 19:44:35 +02:00
ilgeco	2151e322ca	Fix trivialmerge considering also weight attributes	2026-04-14 16:45:14 +02:00
NiccoloN	a7dee5b840	fix mergeTriviallyConnectedComputes	2026-04-14 15:13:45 +02:00
NiccoloN	77fe293062	Merge branch 'main' of chef.heaplab.deib.polimi.it:nnicolosi/Raptor	2026-04-14 13:45:10 +02:00
ilgeco	6727785ab7	Not functioning merging	2026-04-14 13:43:30 +02:00
NiccoloN	eade488d13	fix missed failing tests for channels ... moderate refactor	2026-04-14 12:26:41 +02:00
NiccoloN	30ee9640d4	remove spatial bufferization logic (didn't make much sense) and move channel lowering to SpatialToPim	2026-04-14 11:55:19 +02:00
NiccoloN	368e340a40	minor refactor	2026-04-14 11:21:11 +02:00
NiccoloN	e866ec6f87	convolution uses crossbar size better	2026-04-14 11:06:35 +02:00
NiccoloN	f054e66ed0	reduce spatial compile-times in convolutions using a scf.for instead of materializing a huge number of instructions	2026-04-10 18:50:25 +02:00
NiccoloN	9e0d31af50	automatic code-reformat	2026-04-09 14:27:23 +02:00
NiccoloN	1a0192d1f9	add support for softmax, resize, split, gather	2026-04-09 14:25:00 +02:00
ilgeco	3f870fb74b	MergeDCP pass all test	2026-04-08 20:39:01 +02:00
ilgeco	ca56e3d4f1	Add DCP alghoritm, partial working test	2026-04-07 22:05:39 +02:00
NiccoloN	3625edc80a	add better createSpatCompute helper	2026-03-30 16:14:26 +02:00
NiccoloN	39830be888	add support for operations: reduceMean, add, mul, div, sigmoid	2026-03-30 15:41:12 +02:00
NiccoloN	742df111e3	add relu lowering ... add relu validation add spatial compute helper minor refactors	2026-03-25 11:03:03 +01:00
NiccoloN	a4f3eed3e0	reformat code	2026-03-23 21:25:51 +01:00
NiccoloN	93e20c1dfc	standardize spatial and pim dialects ... remove old unused stuff	2026-03-23 21:21:31 +01:00
NiccoloN	da01e6d697	refactor Pim passes layout	2026-03-23 20:14:59 +01:00
NiccoloN	f869925b64	remove old unused stuff	2026-03-23 20:00:09 +01:00
NiccoloN	661170a9aa	reimplement pool lowering ... add pool validation align PIM ops/codegen/parser with the ISA move constant materialization to MLIR rename the PIM verification/materialization passes better folded-constant handling	2026-03-23 19:14:50 +01:00
NiccoloN	461bdd808d	replace helper-op cleanup with canonicalization ... clean up PIM pattern naming remove unused ValueMap.hpp	2026-03-23 17:13:54 +01:00
NiccoloN	50c545539b	clean up PIM CMake ... update README.md	2026-03-23 16:39:14 +01:00
NiccoloN	11916a2595	refactor Pim constant folding pass ... share contiguous address resolution in PimCommon group patterns in subdir for each pass with pattern files	2026-03-23 15:36:58 +01:00
NiccoloN	670d6ce94f	extend operation support for conv and gemm ... add more tests in validation	2026-03-23 14:46:08 +01:00
NiccoloN	568529ea5f	fix batched conv	2026-03-20 22:00:46 +01:00
NiccoloN	ca2e1645bb	simple convolutions now work :)	2026-03-20 21:17:02 +01:00
NiccoloN	6933804003	constant fold linalg.map (generated from tensor.pad for padding) ... refactor pim helpers in PimCommon	2026-03-20 20:51:20 +01:00
NiccoloN	dbe646ac0d	fix gemm segfault ... print exit signals on validation failure	2026-03-20 14:00:16 +01:00
NiccoloN	bb6dcd38a3	replace deprecated "rewriter.create()" pattern ... refactor PIM to Pim everywhere except for the accelerator name	2026-03-20 13:30:53 +01:00
NiccoloN	db3f52a647	conv now lowers correctly down to bufferized pim	2026-03-20 12:55:09 +01:00
NiccoloN	6e1de865bb	add constant folding and verification pass for pim host operations ... better validation scripts output big refactors	2026-03-20 12:08:12 +01:00
NiccoloN	4e50e056e3	replace old convolution support in spatial (WIP)	2026-03-13 17:46:10 +01:00
NiccoloN	1348bb1c97	generic gemm now works :)	2026-03-06 18:23:27 +01:00
NiccoloN	8ee1e5ece8	implement mem copy codgen (lmv) ... add more gemv/gemm tests refactor	2026-03-04 18:04:48 +01:00
NiccoloN	810e5e75f9	add .clang-format ... reformat all src	2026-02-26 19:16:42 +01:00
NiccoloN	8bb5385961	add host core pim codegen	2026-02-26 17:53:25 +01:00
NiccoloN	ae6e815c7b	fix output paths ... add gemm test	2026-02-25 17:24:31 +01:00
NiccoloN	a6e928bdd7	add PIM accelerator	2026-02-24 15:09:18 +01:00