Cris Cecka

Results 72 comments of Cris Cecka

This is not a multiplication operator, this is a unary dereference operator. Although it is not likely used (yet), pointers can be constexpr as well. ```cpp #include using namespace cute;...

If this helps, then sure. For symmetry change line 224 to > 3. ` = ((5,1):(16,4), (2,2):(80,4))`. Composed Layout as concatenation of sublayouts.

Correct, it's not accurate for negative strides (but the implementation does actually account for them). We've left that out for simplicity. In general, negative strides are rare and inconsistently supported...

There will be in just a few days :-)

You're right about that one too, thanks! Here's the edited image to include in this MR so we can keep everything together. ![divide2](https://github.com/NVIDIA/cutlass/assets/999912/57c7c61d-a244-479f-91b3-afddf72539e8)

Looks like that last line describing `divide2.png` also has an error. It should read > Note that the first mode of each mode of the result is the sublayout (3,(2,4)):(177,(13,2))...

Gah, good finds. Can you edit those to: ```cpp Tensor rmem_4x8_pad = make_tensor(Shape < _4,_8>{}, Stride{}); Tensor rmem_4x8_like = make_tensor_like(rmem_4x8_pad); ``` and ```cpp rmem_4x8_pad : ptr[32b](0x7ff1c8fff920) o (_4,_8):(_32,_2) rmem_4x8_like :...

EDIT: Let's actually keep the references to and uses of `composition(Tensor, Layout)` the way they are and just add an implementation for completeness purposes.

For explanation purposes, I would rather keep it as the simple `composition`. There are other examples that use a `*_divide`.