Initial Implementation for the new WASM based instruction set #952
Conversation
|
Now if I understand correctly as WASM support for CI has been introduced, the doctests would be run in a wasm runtime (for all archs including the wasm one ?) . Hence we would need all essential function ( or rather functions that don't have a fallback implementation in our How do we plan to test the new instruction set out if that's the case. We could surely implement everything in this pr itself if we would like to. Not sure if we could maybe stop it here as we already have some operations implemented here and take care of other operations ( in a bunch) in subsequent prs! |
| { | ||
| assert(false && "unsupported arch/op combination"); | ||
| return {}; | ||
| } |
There was a problem hiding this comment.
This funciton and the previous one have duplicated implementations, maybe a common "impl" mehtod accepting the underlying WASM registers could be used to avoid code duplication?
There was a problem hiding this comment.
Sure I'll try refactoring any code duplication or repetition in the subsequent PR.
| } | ||
| else XSIMD_IF_CONSTEXPR(sizeof(T) == 4) | ||
| { | ||
| return from_mask(batch_bool<float, A> {}, mask, wasm {}); |
There was a problem hiding this comment.
I think we would like to convert the batch_bool<float, A> value returned here to a batch_bool<int32_t, A> but I am not sure we have an intrinsic or any operation already for it !
1) Bitwise: bitwise_rshift, bitwise_not, bitwise_and, bitwise_or, bitwise_lshift, bitwise_xor, bitwise_andnot 2) Logical: gt, lt, eq, neq, all, any, ge, le 3) Arithmetic: add, sub, mul, div, neg, reciprocal 4) Math: abs, sqrt, rsqrt, max, min 5) Roudning: floor, ceil , trunc 6) Memory: store_aligned, store_unaligned, load_aligned, load_unaligned, set 7) Complex: isnan 7) Misc: mask, select, broadcast, insert
|
Hey @anutosh491 what about formatting this so the CI can be green and then we can merge and iterate in the next PR? This would ease the review process. |
|
Yes that sounds like a plan. I've made a commit which should hopefully fix the style format error. |
1) arithmetic: sadd, ssub, hadd, haddp 2) batch manipulation: zip_lo, zip_hi, slide_left, slide_right 3) math: reduce_add 4) memory: store_complex, load_complex 5) misc: from_mask
|
TODO
|
|
Do not hesitate to open an issue to track them, and check the boxes as you implement them. |
|
Let me know if you would like me to squash all commits into a couple/few meaningful commits with good commit messages :) |
|
Indeed, that would nice. Thank you! |
|
I'll also get rid of the test file I had added just for reference to get started ( |
anutosh491
force-pushed
the
Introducing_wasm_based_IS
branch
from
3d98a7d to
1402e0e
Compare
|
Hey @JohanMabille I've squashed the commits and hopefully they look more meaningful now. I do think we can take up all refactoring and other things up in subsequent PRs. For now if you don't find any obvious errors, this might be ready. |
|
Congrats @anutosh491, this is a great milestone. |
No description provided.