[MLIR][XeGPU] Refactor XeGPU layout propagation: passing lane_layout/lane_data with inst_data #203156
[mlir][xegpu] Refactor reduction/bitcast/interleave layout setup
77c58e0c
remove isChunkedStore/Load handling in the setUpGenericLoad/store rule
ece64f99
refactor generic load/store setup rule
5c19d85d
add lane layout to dpas, refactor load rules to be more generic. stil…
edd49273
add setupStoreNdAnchorLayout and setupPrefetchNdAnchorLayout. shared …
d9178532
add setupLoadNdAnchorLayout. validates consumer inst_data/sg_layout a…
e3b6e229
complete inst_data-only anchor layouts on scatter ops; preserve lane …
e671f4a1
honor consumer lane_layout/lane_data in nd anchor setup. dpas operand…
557b76b6
lane-kind nd anchor setup skips block-WHC lookup. lets sub-byte float…
d8a007c9
write completed scatter/store-matrix anchor layout back to the op so …
7dba091e
test: update CHECK lines in propagate-layout-inst-data.mlir to expect…
8a816db7
load_gather setup: take consumer's inst_data as-is and consumer's lan…
9b857eb6
createScaleLayout: cap scale lane_layout by inst_data and derive lane…
63f6d96b
test: update case 19 (dpas_mx_f4e2m1) check lines. load_nd anchors ca…
78daa3e7
test: update case 6 (load_gather_with_coalesce_chunksize) check lines…
5225fe7f
test: update case 12 (vector_shape_cast_expand_and_merge) check lines…
d10c559f
test: update case 18 (dpas_mx_f8e5m2) check lines. load_nd anchors ca…
3159743d
setupGenericNdAnchorLayout: cap default lane_data and lane_layout by …
a2098b60
test: fix case 6 (load_gather_with_coalesce_chunksize). source memref…
349bd34b
test: fix case 7 (scatter_ops_coalesce_chunksize). source memref to 1…
e8bbbf36
remove debug print and polish
7177f328
refactor common function to compute default layout, and make computeC…
8bd0d2c5
polish setupLoadNdAnchorLayout
1bbcee36
fix dpas_mx_f8e5m2 inst_data
5d1bfb97
passing tests, adjust uArch for subByte
8fed9248
passing all, refactor getNumSg
7d12d0a6
refactor computeReductionLaneLayoutAndData, passing all tests
4093160a
polish
b07274f9
Jianhui-Li
marked this pull request as ready for review 38 days ago
fix multireduction 4d to 3d case
ea2c3a07
polish
423a33fc
rearrange the code location, put utitlies/helpers togehter
5ed1bbeb
add comments and reorder functions
736c9b23
polish setup rules
5a069045
polish completeScatterIOLaneLayoutFromInstData; add BlockIOInstructio…
dedd802d
refactor buildInstDataLayoutWithLane
8e9e69aa
add full complete lane layouts for user specified inst_data layout, a…
05a09d2a
add missing parameters in convert_layout propagation, some unit test …
57e6ba49
fixed bugs in load_nd layout seting and convert_layout handling
a7c192ba
git merge main
d27d174e
fix format
63325893
polish
045976a9
fix uArch table
07ce0a98
address claude review comments
e5fb64f6
removing unused test
9e1fc0b7
nbpatel
approved these changes
on 2026-06-22
git merge main
70057ad1
address feedbacK
58d53851
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub