CatsAreFluffy
CatsAreFluffy
Well, I'm kinda lazy, so i just used Wine :P On Thu, Sep 21, 2017 at 8:49 AM, Vittorio Romeo wrote: > I don't own a Mac, but I see...
This is currently being tracked as #319.
NPUs are generally specialized for stuff like FMAs ([like the Apple Neural Engine](https://github.com/tinygrad/tinygrad/tree/d0e752003da3fc023fa85094d7f5b65b47dd5091/extra/accel/ane)), so I don't think it's possible to use them to run Bend/HVM.
Any updates?
> Will it be merged? O'Rear hasn't been active on GitHub for over a year, so probably not. > does this result still directly search for Con(ZF) or is it...
There are no longer any mentions of SRAI64, SRLI64, or SLLI64 in the manual. Close?
`(*slice.device_ptr()) as *const T` works.
If I disable the REF-DUP optimization, I get a very different result (`((@List/Cons/tag (((1 (1 (((1 (* (((1 (* (((0 vce) vce) vc4))) vc4) vb3))) vb3) va2))) va2) (@List/Nil vfc)))...
uint64 to float conversion is incorrect for large inputs, eg: ```python from tinygrad import Tensor, dtypes x = Tensor([1 1).cast(dtypes.int64).cast(c.dtype) * 2 + (c.src[0] & 1).cast(dtypes.int32).cast(c.dtype)),` Also, uint8/int8 mod is...
> Since Robinson Arithmetic has fewer axioms than ZF, it should have (far?) fewer states than the current ZF record of 643 Robinson arithmetic has a more complicated language than...