add 'swapEndian' intrinsics to bitops & endians module . #5898
Conversation
Parashurama
force-pushed
the
bitops_use_endians
branch
from
b48be21 to
8512599
Compare
|
I wouldn't mind losing the proc swapEndian*[T:SomeInteger](x: T): T? |
|
I think that the Int suffix was to prevent name confusion (no name collision) with |
|
Any progress? |
this optimizsation can lead to undefined behaviour since strict aliasing rule is broken.
Parashurama
force-pushed
the
bitops_use_endians
branch
from
8512599 to
10d4555
Compare
|
issue #7587 as well as any sign extension bug should now be fixed. |
Parashurama
changed the title
| template toUint32[T](x: T): uint32 = | ||
| when sizeof(x) == 1: cast[uint8](x).uint32 | ||
| elif sizeof(x) == 2: cast[uint16](x).uint32 | ||
| else: cast[uint32](x) |
There was a problem hiding this comment.
I think this should be elif sizeof(x) == 4: cast[uint32](x) else: {.error: "invalid size".} and likewise for the other parts where this pattern is used.
There was a problem hiding this comment.
not relevant as of 1.3.5 278b458
There was a problem hiding this comment.
oups, didn't realize PR was closed but not merged
|
we already have https://github.com/nim-lang/Nim/blob/devel/lib/pure/endians.nim |
EDIT: The failing test is caused by differences in type conversion between the different platforms & the VM.
Will go back to this later, once this issue is solved.
EDIT: needs integer cast support in VM (PR #5908 waiting for review)