Caleb James DeLisle
Caleb James DeLisle
It might be worth considering trying to target SPIR-V, either directly or through the LLVMSPIR-V converter ( https://github.com/KhronosGroup/SPIRV-LLVM ) because SPIR-V is an intermediate which can be used to represent...
> I think these type of discussions are better had in the mailing list I will register on the list so i can stay up to date. > If there's...
> It might work if clBuildProgram() detects the blocks as kernels and treats them as ones and then you can refer to it by that magic name from the enqueue_kernel()...
A couple of comments about `posit_add_exact`: * If adding a + b can be represented perfectly in 1 posit, the result will be [sum, 0] * The first output of...
The first reason is because I expect hardware to settle on posit, posit, posit and posit so if you convert a posit to a posit that's not super useful because...
Good questions, I always consider that multiplication doubles the required bits. My reasoning about why this should work is: * max mantissa size is nbits - 3 - es +...
Can you push that in a branch so I can see what I'm looking at in the code ? I'm trying to make out just where the problems are...
Ahh I see, but I think it should be ok to declare that maxpos and minpos remain the maximum and minimum values, even when using add_exact(). I'm still working on...
Thinking further, it strikes me that in no case should add_exact fail, if it can do nothing else, it can just return its arguments, right ?