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

enable float16 and unaligned vector load/store

Browse source
This commit is contained in:
ttomsa 2026-02-23 01:01:57 +00:00
commit dacfa01c0d
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
tinygrad/codegen/late/devectorizer.py
View file

@ -157,8 +157,8 @@ def split_load_store(ctx:Renderer|None, ls:UOp, idx:UOp):
elif isinstance(buf.dtype, ImageDType):
lengths = [4]
elif ctx is not None and CPU_X86:
lengths = [4,2] if buf.dtype.base == dtypes.float32 else []
#must_divide = False
lengths = [8,4,2] if buf.dtype.base == dtypes.half else [4,2]
must_divide = False
elif ctx is not None and ctx.supports_float4:
# TODO: a better way to get this than ctx
lengths = [8,4,2] if buf.dtype.base == dtypes.half and getenv("ALLOW_HALF8") else ([16,8,4,2] if AMX else [4,2])