[SYCL] Remove direct initialization constructor from bfloat16 class #4989
Conversation
vladimirlaz
changed the title
There was a problem hiding this comment.
LGTM. Is this something that should be reflected in the extension document?
|
@AlexeySotkin, can you please answer the question above before merging of this PR? |
in fact, the spec does not contain the removed constructor. So nothing to be changed. |
The latest change contains a new method which should be added to the spec. Also I'd like to hear feedback from @GHGmc2. Marking this PR as a draft for now. |
|
This PR fixes #4871 |
|
@AlexeySotkin Thanks, I'll have have try and give feedback soon. |
|
@AlexeySotkin ambiguous error when cast bfloat16 to long "error: conversion from 'bfloat16' (aka 'sycl::ext::intel::experimental::bfloat16') to 'long' is ambiguous": #include <sycl/ext/intel/experimental/bfloat16.hpp>
using bfloat16 = sycl::ext::intel::experimental::bfloat16;
int main() {
long ret = bfloat16(3.14f);
return 0;
} |
This constructor is being removed to resolve ambiguity when user trying initialize bfloat16 with int, like auto a = bfloat16(0);
AlexeySotkin
dismissed stale reviews from vladimirlaz and steffenlarsen
via
1f1147a
AlexeySotkin
force-pushed
the
bf16_remove_ctor
branch
4 times, most recently
from
bca94e6 to
3b3a4a7
Compare
|
@AlexeySotkin could you please fix the CI failure? |
Add conversion operator for sycl::half Conversion operator for storage_t is replaced by 'raw()' method Signed-off-by: Alexey Sotkin <[email protected]>
AlexeySotkin
force-pushed
the
bf16_remove_ctor
branch
from
3b3a4a7 to
46f81d6
Compare
Yes, but I'd like to have this PR approved first. |
|
LGTM |
Align the document with changes made in intel#4989
This constructor is being removed to resolve ambiguity when user trying
initialize bfloat16 with int, like
auto a = bfloat16(0);Fixes #4871.