mirrors/tinygrad
2
0
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-06-24 02:14:17 +00:00
Code Issues Projects Releases Packages Wiki Activity

unaligned

Browse source
This commit is contained in:
George Hotz 2025-03-24 15:13:57 +08:00
commit 1b61cc6ec3
1 changed files with 5 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

6
tinygrad/runtime/ops_dsp.py
View file

@ -134,8 +134,12 @@ dsp_pm_late = PatternMatcher([
# prefetch L1
(UPat(Ops.LOAD, dtype=(dtypes.uchar.vec(4), dtypes.uchar.vec(8)), name="ld"), prefetch_l1),
# 64 -> 128
(UPat(Ops.LOAD, dtype=dtypes.uchar.vec(64), src=(UPat(Ops.CAST, src=(UPat(Ops.INDEX, name="idx"),)),)),
lambda idx: idx.cast(dtypes.uchar.vec(128).ptr(idx.dtype.size)).load(dtype=dtypes.uchar.vec(128)).gep(tuple(range(0,64)))),
# unaligned load
((UPat(Ops.LOAD, src=(UPat(Ops.CAST, src=(UPat(Ops.INDEX, src=(UPat(), UPat()+UPat.cvar("c"))),), name="ptr"),), dtype=dtypes.uchar.vec(128))),
(UPat(Ops.LOAD, src=(UPat(Ops.CAST, src=(UPat(Ops.INDEX, src=(UPat(), UPat()+UPat.cvar("c"))),), name="ptr"),), dtype=dtypes.uchar.vec(128)),
lambda c,ptr: UOp(Ops.CUSTOM, dtype=dtypes.uchar.vec(128), src=(ptr,), arg='vmemu({0})')),
# __builtin_HEXAGON_V6_vrmpybus_acc_128B