Oliver J. Woodford
Oliver J. Woodford
Thanks. The reduce algorithm implementation in xsimd is a useful baseline implementation. A useful function would be: ``` template std::array batch_sum(batch_type a, batch_type b, batch_type c, batch_type d); ``` such...
Actually, you already have the in-vector reductions I was asking about, at least for summing, `hadd` and `haddp`: https://github.com/QuantStack/xsimd/blob/dde445fe36a2bc218e170408a4c48dab40833e0f/include/xsimd/types/xsimd_sse_float.hpp#L637 However, I find that there is a significant time penalty associated...
Yes, I see that. But for the algorithms I was envisaging, you'd know the number of batches, if not the batch size. For now I'm just fixing on batch, and...
I still think there is a need for cross batch additions for a fixed number of batches, regardless of the (instruction set specific) batch length. An example algorithm which requires...
This package provides an interface to easily access web-based datasets. It follows the standard naming convention of such packages: an abbreviation of the dataset type (here VG, for visual geometry),...
Also failing: ```julia using Symbolics @variables x[1:2] s (x[1] * s) / (x[2] * s) # Works y = x .* s z = y[1] / y[2] # ERROR: MethodError:...
> Is this supposed to be listed as WIP, or just add to the tests? WIP
@shashi Done. Are you able to approve the workflows?
Shouldn't `isequal` on symbolic expressions resolve this kind of kind of problem?: `isequal(s - x[2] + x[1] - s, x[1] - x[2])` returns false In fact, if I write the...
The problem has been suggested in #704