Clément Fournier 36c3466aef

[mlir][linalg] Fix neutral elt for softmax (#118952) ...

The decomposition of `linalg.softmax` uses `maxnumf`, but the identity
element that is used in the generated code is the one for `maximumf`.
They are not the same, as the identity for `maxnumf` is `NaN`, while the
one of `maximumf` is `-Infty`. This is wrong and prevents the maxnumf
from being folded.

Related to #114595, which fixed the folder for maxnumf.

2025-01-13 15:21:07 +08:00

..

td

[mlir] Add missing patterns to linalg.decompose_pack_unpack TD Op (#121400)

2025-01-03 09:11:38 +00:00

affine.mlir

[mlir][Linalg] NFC - Retire LinalgToLLVM pass

2023-07-26 20:26:07 +02:00

block-pack-matmul-layout.mlir

[mlir][linalg] Block pack matmul pass (#89782)

2024-05-09 18:02:44 +01:00

block-pack-matmul-padding.mlir

[mlir][linalg] Block pack matmul pass (#89782)

2024-05-09 18:02:44 +01:00

block-pack-matmul.mlir

[mlir][linalg] Avoid emitting errors in block pack matmul (#93170)

2024-05-24 09:29:28 +02:00

bubble-up-extract-slice-op.mlir

[MLIR] Generalize expand_shape to take shape as explicit input (#90040)

2024-04-30 09:28:35 -07:00

bufferize.mlir

[mlir][bufferization] Fix OneShotBufferize when defaultMemorySpaceFn is used (#91524)

2024-11-26 09:45:57 -07:00

canonicalize.mlir

[mlir][bufferization] Fix OneShotBufferize when defaultMemorySpaceFn is used (#91524)

2024-11-26 09:45:57 -07:00

collapse-dim.mlir

[MLIR] Generalize expand_shape to take shape as explicit input (#90040)

2024-04-30 09:28:35 -07:00

constant-fold.mlir

[mlir][linalg] Add linalg.transpose constant folding (#92589)

2024-05-28 10:42:32 -04:00

continuous-tiling-full.mlir

[MLIR][Transform] Consolidate result of structured.split into one list (#111171)

2024-11-15 10:53:34 +08:00

continuous-tiling-multiway-split.mlir

[MLIR][Transform] Consolidate result of structured.split into one list (#111171)

2024-11-15 10:53:34 +08:00

conv-interface-invalid.mlir

…

convert-conv2d-to-img2col.mlir

[MLIR] Generalize expand_shape to take shape as explicit input (#90040)

2024-04-30 09:28:35 -07:00

convert-elementwise-to-linalg.mlir

…

data-layout-propagation.mlir

[MLIR][Linalg] Fix DataLayoutPropagation for tensor.unpack + linalg.generic (#101755)

2024-08-05 23:04:47 +05:30

decompose-generic-by-unfolding-projected-permutation.mlir

[mlir][linalg] unfold projected permutation. (#114704)

2024-11-09 23:34:57 +00:00

decompose-ops.mlir

…

decompose-pad-tensor.mlir

[mlir][linalg] Extract GeneralizePadOpPattern into a standalone transformation (#117329)

2024-11-26 08:11:15 +00:00

decompose-tensor-pack-tile.mlir

[mlir] Rename GeneralizeOuterUnitDims{Un}PackOpPatterns (#116439)

2024-11-19 09:09:33 +00:00

decompose-tensor-pack.mlir

[mlir][linalg] Add tests for tensor.unpack decomposition (#118786)

2024-12-06 09:28:15 +00:00

decompose-tensor-unpack-tile.mlir

[mlir] Add missing patterns to linalg.decompose_pack_unpack TD Op (#121400)

2025-01-03 09:11:38 +00:00

decompose-tensor-unpack.mlir

[mlir] Add missing patterns to linalg.decompose_pack_unpack TD Op (#121400)

2025-01-03 09:11:38 +00:00

detensorize_0d.mlir

…

detensorize_br_operands.mlir

…

detensorize_entry_block.mlir

[mlir] Fix block merging (#102038)

2024-08-07 09:10:01 +01:00

detensorize_if.mlir

[mlir] Fix block merging (#102038)

2024-08-07 09:10:01 +01:00

detensorize_trivial.mlir

…

detensorize_while_impure_cf.mlir

[mlir] Fix block merging (#102038)

2024-08-07 09:10:01 +01:00

detensorize_while_pure_cf.mlir

[mlir] Fix block merging (#102038)

2024-08-07 09:10:01 +01:00

detensorize_while.mlir

[mlir] Fix block merging (#102038)

2024-08-07 09:10:01 +01:00

drop-unit-extent-dims.mlir

[MLIR][Tensor] Fix out-of-bounds FoldEmptyTensorWithDimOp crash (#112196)

2024-11-06 11:56:51 +01:00

erase-unused-operands-and-results.mlir

…

fill-interface-invalid.mlir

…

flatten-elementwise.mlir

[MLIR] Generalize expand_shape to take shape as explicit input (#90040)

2024-04-30 09:28:35 -07:00

flatten-unsupported.mlir

[mlir][transform] Emit error message with emitSilenceableFailure (#86146)

2024-03-22 12:37:39 -05:00

fold-add-into-dest.mlir

[MLIR][Linalg] Pattern to fold AddOp to accumulation via contraction op's dest (#110514)

2024-10-03 12:22:57 +02:00

forward-vector-transfers.mlir

…

fuse-with-reshape-by-collapsing.mlir

[mlir] Allow multi-result ops in reshape fusion (#108576)

2024-09-16 13:06:38 -04:00

fusion-elementwise-ops.mlir

[mlir][test] Fix filecheck annotation typos [2/n] (#93476)

2024-06-14 17:16:02 +02:00

fusion-elementwise-options.mlir

…

fusion-elementwise.mlir

[MLIR][Linalg] Fix linalg crash during elementwise op fusion (#117667)

2024-11-26 16:00:18 -08:00

fusion-multiuse-producer.mlir

…

fusion-push-reshape.mlir

[MLIR] Generalize expand_shape to take shape as explicit input (#90040)

2024-04-30 09:28:35 -07:00

generalize-named-ops.mlir

[MLIR][Linalg] Re-land linalg.matmul move to ODS. + Remove/update failing obsolete OpDSL tests. (#115319)

2024-11-07 14:51:02 +00:00

generalize-named-polymorphic-ops.mlir

Remove redundant linalg.matmul_signed (#98615)

2024-07-13 14:46:44 +01:00

hoisting.mlir

[mlir] Don't hoist transfers from potentially zero trip loops (#112752)

2024-10-18 16:11:21 -04:00

inline-scalar-operands.mlir

[mlir][linalg] Bugfix for InlineScalarOperands (#111534)

2024-10-14 15:38:35 +08:00

inlining.mlir

…

invalid.mlir

Revert "[MLIR][Arith] Add denormal attribute to binary/unary operations (#112700)"

2024-12-10 04:18:20 +01:00

library-calls.mlir

…

lit.local.cfg

[mlir] Add apply_patterns.linalg.generalize_pack_unpack TD Op (#116373)

2024-11-18 07:51:18 +00:00

loops.mlir

[mlir][linalg][nfc] Delete references to args_in/args_out (#111517)

2024-10-10 15:45:52 +01:00

match-ops-interpreter.mlir

[mlir][linalg] fix segmentation fault in isContractionBody function (#108703)

2024-09-27 09:43:57 +08:00

match-ops-invalid.mlir

[mlir][transform] Add elementwise criteria to match.structured.body (#79626)

2024-01-31 10:12:33 +01:00

match-ops.mlir

…

matmul-shared-memory-padding.mlir

Revert "[MLIR] Make OneShotModuleBufferize use OpInterface (#110322)" (#113124)

2024-10-22 13:28:44 +01:00

mesh-sharding-propagation.mlir

[mlir][mesh] Fix empty split_axes sharding annotation (#108236)

2024-09-21 08:43:12 -07:00

mesh-spmdization.mlir

[mlir][mesh] Shardingcontrol (#102598)

2024-08-12 12:20:58 +01:00

multisize-tiling-full.mlir

[MLIR][Transform] Consolidate result of structured.split into one list (#111171)

2024-11-15 10:53:34 +08:00

named-ops-fail.mlir

[MLIR][Linalg] Ternary Op & Linalg select (#91461)

2024-05-14 10:50:35 +01:00

named-ops.mlir

[mlir][linalg] Add Grouped Convolution Ops: conv_2d_nhwgc_gfhwc and conv_2d_nhwgc_gfhwc_q (#108192)

2024-11-08 09:23:17 -08:00

namedop_conversion.mlir

…

one-shot-bufferize-analysis-2fill-extract-matmul-all-perms.mlir

…

one-shot-bufferize-analysis.mlir

[mlir][bufferization] Fix bug in bufferization of elementwise ops (#97209)

2024-07-01 19:00:21 +02:00

one-shot-bufferize-empty-tensor-elimination.mlir

[mlir][linalg] Fix for invalid IR in eliminate_empty_tensors (#73513)

2024-01-01 17:12:40 +00:00

one-shot-bufferize.mlir

[mlir][bufferization] Add "bottom-up from terminators" analysis heuristic (#83964)

2024-03-21 14:16:02 +09:00

pad_fusion.mlir

[mlir][Affine] Delete duplicate code: applyMapToValues

2023-06-30 14:01:13 +02:00

pad-to-specific-memory-space.mlir

Revert "[MLIR] Make OneShotModuleBufferize use OpInterface (#110322)" (#113124)

2024-10-22 13:28:44 +01:00

parallel-loops.mlir

[mlir][SCF] scf.parallel: Make reductions part of the terminator (#75314)

2023-12-20 11:06:27 +09:00

promote.mlir

[mlir][memref] Improve memref.subview type inference (#96421)

2024-06-23 19:05:00 +02:00

promotion_options.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

rank-reduce-contraction-ops.mlir

[mlir][Linalg] Fix non-matmul linalg structured ops (#116412)

2024-11-16 08:13:10 +00:00

reshape_control_fusion.mlir

[MLIR] Generalize expand_shape to take shape as explicit input (#90040)

2024-04-30 09:28:35 -07:00

reshape_fusion.mlir

[mlir] Add reshape propagation patterns for tensor.pad (#94489)

2024-06-07 09:09:06 -04:00

resolve-shaped-type-result-dims.mlir

[MLIR] Generalize expand_shape to take shape as explicit input (#90040)

2024-04-30 09:28:35 -07:00

roundtrip-broadcast.mlir

[mlir][linalg] raise generic to named ops. (#110421)

2024-10-11 15:27:27 +01:00

roundtrip-linalg-named-ops.mlir

[MLIR][Linalg] Add pass to convert linalg.generic back to named ops (#95656)

2024-06-30 19:37:51 +01:00

roundtrip-transpose.mlir

[mlir][linalg] raise generic to named ops. (#110421)

2024-10-11 15:27:27 +01:00

roundtrip.mlir

[mlir][linalg] Add quantized conv2d operator with FCHW,NCHW order (#107740)

2024-10-19 18:25:27 +02:00

runtime-verification.mlir

[mlir][linalg] Add runtime verification for linalg ops (#89917)

2024-04-25 10:12:55 -07:00

specialize-generic-ops-fail.mlir

[mlir][linalg] raise generic to named ops. (#110421)

2024-10-11 15:27:27 +01:00

specialize-generic-ops.mlir

[MLIR][Linalg] Add pass to convert linalg.generic back to named ops (#95656)

2024-06-30 19:37:51 +01:00

standard.mlir

…

subtensor-of-padtensor.mlir

…

swap-extract-slice-with-fill.mlir

…

test-drop-unit-dims.mlir

[mlir][Linalg] Cleanup the drop unit dims pass in Linalg.

2023-07-19 17:47:18 +00:00

tile-and-fuse-tensors.mlir

[MLIR][Tensor] Add Destination style RewritePattern for DimOp. (#65780)

2023-09-09 06:01:56 -04:00

tile-conv.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

tile-indexed.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

tile-softmax.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

tile-tensors.mlir

[mlir][Linalg] Deprecate linalg::tileToForallOp and linalg::tileToForallOpUsingTileSizes (#91878)

2024-07-31 12:32:07 -07:00

tile-to-forall.mlir

[mlir][Linalg] Deprecate linalg::tileToForallOp and linalg::tileToForallOpUsingTileSizes (#91878)

2024-07-31 12:32:07 -07:00

transform-lower-pack.mlir

[MLIR] Add allow Insert/extract slice option to pack/unpack op (#117340)

2024-12-10 11:31:30 -06:00

transform-op-bufferize-to-allocation.mlir

[mlir] introduce debug transform dialect extension (#77595)

2024-01-12 13:24:02 +01:00

transform-op-compose-masked-vectorize-and-cleanups.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

transform-op-decompose.mlir

[mlir][linalg] Fix neutral elt for softmax (#118952)

2025-01-13 15:21:07 +08:00

transform-op-fuse-into-containing.mlir

[mlir] introduce debug transform dialect extension (#77595)

2024-01-12 13:24:02 +01:00

transform-op-fuse.mlir

[mlir] Add option for a cleanup pattern set to SCF tiling helper (#109554)

2024-10-04 14:42:55 -04:00

transform-op-generalize.mlir

[mlir] use transform-interpreter in test passes (#70040)

2023-10-24 16:12:34 +02:00

transform-op-gpu-map-copy-to-threads.mlir

[mlir] use transform-interpreter in test passes (#70040)

2023-10-24 16:12:34 +02:00

transform-op-hoist-pad-build-packing-loop-nest.mlir

[MLIR][Transform] Hoist Pad generates linalg.transpose (#109669)

2024-09-26 09:33:47 +01:00

transform-op-hoist-pad.mlir

[MLIR][Transform] Hoist Pad generates linalg.transpose (#109669)

2024-09-26 09:33:47 +01:00

transform-op-insert-slice-to-copy.mlir

[mlir] use transform-interpreter in test passes (#70040)

2023-10-24 16:12:34 +02:00

transform-op-interchange.mlir

[mlir] use transform-interpreter in test passes (#70040)

2023-10-24 16:12:34 +02:00

transform-op-match.mlir

[mlir] introduce debug transform dialect extension (#77595)

2024-01-12 13:24:02 +01:00

transform-op-matmul-to-outerproduct.mlir

[mlir][nfc] Update Linalg matmul -> Vector OP test (#81416)

2024-03-13 16:53:26 +00:00

transform-op-mmt4d-to-fma.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

transform-op-multitile-sizes.mlir

[mlir] introduce debug transform dialect extension (#77595)

2024-01-12 13:24:02 +01:00

transform-op-pack.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

transform-op-pad.mlir

[mlir][td] Rename pack_paddings in structured.pad (#111036)

2024-10-15 19:24:43 +01:00

transform-op-peel-and-vectorize-conv.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

transform-op-peel-and-vectorize.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

transform-op-replace.mlir

[mlir][Symbol] Add verification that symbol's parent is a SymbolTable (#80590)

2024-02-05 22:59:03 -08:00

transform-op-rewrite-in-destination-passing-style.mlir

[mlir] use transform-interpreter in test passes (#70040)

2023-10-24 16:12:34 +02:00

transform-op-scalarize.mlir

[mlir][transform] Consistent linalg transform op syntax for dynamic index lists (#90897)

2024-05-08 09:11:53 -05:00

transform-op-specialize-elemwise-binary.mlir

[MLIR][Linalg] Add pass to convert linalg.generic back to named ops (#95656)

2024-06-30 19:37:51 +01:00

transform-op-specialize-elemwise-unary.mlir

[MLIR][Linalg] Add pass to convert linalg.generic back to named ops (#95656)

2024-06-30 19:37:51 +01:00

transform-op-specialize-matmul.mlir

[MLIR][Linalg] Add pass to convert linalg.generic back to named ops (#95656)

2024-06-30 19:37:51 +01:00

transform-op-specialize.mlir

[mlir][linalg] raise generic to named ops. (#110421)

2024-10-11 15:27:27 +01:00

transform-op-split-reduction-by-scaling.mlir

[mlir] use transform-interpreter in test passes (#70040)

2023-10-24 16:12:34 +02:00

transform-op-split-reduction.mlir

[mlir][arith] Add neutral element support to arith.maxnumf/arith.minnumf (#93278)

2024-05-29 10:20:49 -04:00

transform-op-split.mlir

[MLIR][Transform] Consolidate result of structured.split into one list (#111171)

2024-11-15 10:53:34 +08:00

transform-op-tile.mlir

[mlir][Linalg] Deprecate linalg::tileToForallOp and linalg::tileToForallOpUsingTileSizes (#91878)

2024-07-31 12:32:07 -07:00

transform-op-vectorize.mlir

[mlir] use transform-interpreter in test passes (#70040)

2023-10-24 16:12:34 +02:00

transform-ops-invalid.mlir

[mlir] Improve error message when number of operands and types differ (#118488)

2024-12-04 09:36:05 +01:00

transform-ops.mlir

[MLIR][Transform] Consolidate result of structured.split into one list (#111171)

2024-11-15 10:53:34 +08:00

transform-pack-greedily.mlir

[mlir] use transform-interpreter in test passes (#70040)

2023-10-24 16:12:34 +02:00

transform-patterns.mlir

[mlir][linalg][nfc] Delete references to args_in/args_out (#111517)

2024-10-10 15:45:52 +01:00

transform-promotion.mlir

[mlir][memref] Improve memref.subview type inference (#96421)

2024-06-23 19:05:00 +02:00

transform-tile-and-fuse-pack-unpack.mlir

[MLIR] Add allow Insert/extract slice option to pack/unpack op (#117340)

2024-12-10 11:31:30 -06:00

transform-tile-and-fuse.mlir

[mlir] use transform-interpreter in test passes (#70040)

2023-10-24 16:12:34 +02:00

transform-tile-and-winograd-rewrite.mlir

[mlir][linalg] Fix for bias handling for Winograd (#110331)

2024-10-11 09:39:19 +01:00

transform-tile-reduction.mlir

[mlir][scf] Add getPartialResultTilePosition to PartialReductionOpInterface (#120465)

2024-12-27 16:52:34 +00:00

transform-tile-winograd.mlir

[mlir][linalg] Fix for bias handling for Winograd (#110331)

2024-10-11 09:39:19 +01:00

transform-winograd-conv2d.mlir

[mlir][linalg] Add transform operator for Winograd Conv2D algorithm (#96182)

2024-07-11 14:45:36 +01:00

transpose-conv2d.mlir

[MLIR][Linalg] (NFC) Drop verify-diagnostics from transpose-conv2d.mlir

2023-12-11 16:13:19 +01:00

transpose-matmul-a.mlir

[mlir][linalg] Move transpose_matmul to targeted transform op (#89717)

2024-04-23 10:52:50 +01:00

transpose-matmul-b.mlir

[mlir][linalg] Move transpose_matmul to targeted transform op (#89717)

2024-04-23 10:52:50 +01:00

transpose-matmul.mlir

[mlir][linalg] Add patterns to convert matmul to transposed variants (#89075)

2024-04-23 07:21:06 +01:00

value-bounds-op-interface-impl.mlir

[mlir][GPU] Implement ValueBoundsOpInterface for GPU ID operations (#122190)

2025-01-09 11:42:22 -08:00

vectorization-pad-patterns.mlir

[mlir][linalg] Extract GeneralizePadOpPattern into a standalone transformation (#117329)

2024-11-26 08:11:15 +00:00

vectorization-scalable.mlir

[mlir][linalg] Relax scalable vectorization restrictions (#117991)

2024-11-29 07:43:10 +00:00

vectorization-unsupported.mlir

[mlir][linalg] Split GenericPadOpVectorizationPattern into two patterns (#111349)

2024-10-29 16:57:23 +00:00

vectorization-with-patterns.mlir

[mlir][linalg] Enable Vectorization of 0-D tensor.extract (#119079)

2024-12-11 07:47:23 +00:00

vectorization.mlir

[mlir][linalg] Add a new helper hook: hasVectorizationImpl (#110708)

2024-10-04 10:06:33 +01:00

vectorize-conv-masked-and-scalable.mlir

[mlir][linalg] Mark xfers as in-bounds when masking depthwise convs (#96771)

2024-06-27 15:47:00 +01:00

vectorize-convolution-flatten.mlir

[mlir][linalg] Use vector.shuffle to flatten conv filter (#75038)

2023-12-15 17:56:59 +00:00

vectorize-convolution.mlir

[mlir][linalg] Vectorization support for convolution of i1 type (#109480)

2024-09-24 12:24:59 -05:00

vectorize-tensor-extract-masked.mlir

[mlir][linalg] Fix vectorization of tensor.extract (#118105)

2024-12-05 09:24:53 +00:00

vectorize-tensor-extract.mlir

[mlir][nfc] Update vectorize-tensor-extract.mlir (4/N) (#119697)

2024-12-16 20:38:19 +00:00

winograd-conv2d-rewrite.mlir

[mlir][linalg] Fix for bias handling for Winograd (#110331)

2024-10-11 09:39:19 +01:00

winograd-conv2d.mlir

Add missing FillOp to winograd lowering (#108181)

2024-09-13 15:48:17 +01:00