goals for 0.2.x
Opened this issue · 1 comments
ariasanovsky commented
features
- tensor fusion/multivariate functions
- impure functions (
&
and&mut
) - assess what to pack into kernel typestate w/ traits
- e.g., dims, rank, crate metadata, compiler flags, thread geometry, ...
- versatile thread layout
- device ordinals for
CudaSlice
initialization - fill with const
- sum/summands
- map fn w/ range entry
- refilling an existing slice
-
fn __unlabel(&self)
&unsafe fn __relabel(&self)
-
DevSlice::new
ergonomics
- better display for spindle crates
- assess the "crate" approach
- ?compile a
kernel.c
andFFI
the Rust part - goal: target different architecture + ?migrate off of experimental ptx ABI
- ?compile a
correctness
- ?does
cudarc
always 0-initialize - use
uuid
better (e.g.,rusqlite
feature) - feature to use rust
ptx_compiler
orabi_ptx
for kernels
sanitation
- dump artifacts according to the appropriate environment variables
- clean out fledgling
dbg!
andtodo!
(write tests as needed) [ ] ?smuggle traitconst
to config/typestate- ?refactor
TypeDb
to use embedded hash map/cache instead ofSQL
misc
- ?bundle smaller Issues here to avoid clutter
- refactor & test camel/snake case
- say hi 💘to the
candle
folks 🫂🤗 <-hugging
+face
emoji pun- improve pun skills
ariasanovsky commented