Masked compare and floating point classifications #2427
Conversation
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
g3doc/quick_reference.md
Outdated
| equivalent to, and potentially more efficient than, `And(m, Eq(a, b));` etc. | ||
|
|
||
| * `V`: `{f}` \ | ||
| <code>M **MaskedIsNaN**(V v)</code>: returns mask indicating whether `v[i]` |
There was a problem hiding this comment.
Let's add the mask argument to the documentation :)
There was a problem hiding this comment.
Apologies, have corrected.
g3doc/quick_reference.md
Outdated
| #### Masked floating-point classification | ||
|
|
||
| All ops in this section return `false` for `mask=false` lanes. These are | ||
| equivalent to, and potentially more efficient than, `And(m, Eq(a, b));` etc. |
There was a problem hiding this comment.
Very, true. Have corrected.
g3doc/quick_reference.md
Outdated
| All ops in this section return `false` for `mask=false` lanes. These are | ||
| equivalent to, and potentially more efficient than, `And(m, Eq(a, b));` etc. | ||
|
|
||
| * <code>M **MaskedCompEq**(M m, V a, V b)</code>: returns `a[i] == b[i]` or |
There was a problem hiding this comment.
Should we just call it MaskedEq for consistency with the usual naming convention, which is only to prepend Masked to the existing Eq?
hwy/ops/arm_sve-inl.h
Outdated
| } | ||
|
|
||
| template <class V, class M, class D = DFromV<V>> | ||
| HWY_API MFromD<D> MaskedIsInf(const M m, const V v) { |
There was a problem hiding this comment.
Do we ever plan to provide a faster implementation of MaskIfInf/IsFinite, or can those be removed?
Apologies, CLA issue has stuck around despite us signing it multiple times. Team was on holiday, now we are back and will resolve CLA and your comments soon. |
wbb-ccl
force-pushed
the
cc_up_masked_compare
branch
from
19ff43c to
bfe4014
Compare
Fix docs for masked floating-point classification Rename MaskedComp* ops to Masked* Remove redundant wrapper functions
wbb-ccl
force-pushed
the
cc_up_masked_compare
branch
from
bfe4014 to
e8a840c
Compare
Introduces:
MaskedIsNaN(v): returns mask indicating whether
v[i]is "not a number" (unordered) orfalseifm[i]is false.MaskedIsInf(v): returns mask indicating whether
v[i]is positive or negative infinity orfalseifm[i]is false.MaskedIsFinite(v): returns mask indicating whether
v[i]is neither NaN nor infinity, i.e. normal, subnormal or zero orfalseifm[i]is false. Equivalent toNot(Or(IsNaN(v), IsInf(v))).MaskedCompEq(m, a, b), MaskedCompNe(m, a, b), MaskedCompLt(m, a, b), MaskedCompGt(m, a, b), MaskedCompLe(m, a, b), MaskedCompGe(m, a, b), returning the output for all versions of
==, !=, <, >, >=, <=, orfalseifm[i]is false.All operations are written for both
arm_sve-inl.handgeneric_ops-inl.hand, equally, tests are included for all operations.