Conversation
|
I didn't look yet, but it'd be useful anyways. Could you explain why this is a breaking change in the description? And, I suppose we need to add smth like BREAKING CHANGE (?) somewhere |
The size check implies a tensor size restriction to 2^31-1 bytes. Kernel configurations larger than that will no longer validate. Resolves: COMPMID-8697 Signed-off-by: Andreas Flöjt <andreas.floejt@arm.com> Change-Id: I54f73ade5cb4a0d34d831505d83d1d7ef526b5db
andflo-Arm
force-pushed
the
pr/tensor-size-checks
branch
from
493638b to
773dc2f
Compare
Done. The exclamation mark serves the purpose of conveying breaking changes. |
| const size_t one_channel = 1u; | ||
| ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN( | ||
| input, one_channel, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::S32, DataType::F16, DataType::F32); | ||
| ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, one_channel, DataType::S32, DataType::S64); |
There was a problem hiding this comment.
This is a bit weird. I know it existed before but the comment header of the operator says it's either U32 or S32. I'm not sure why it's checking S64 here.
There was a problem hiding this comment.
We added support for S64 a while back, maybe we forgot to update the header. Check MLCE-1089
There was a problem hiding this comment.
validate() says otherwise :D
3 participants
The size check implies a tensor size restriction to 2^31-1 bytes. Kernel
configurations larger than that will no longer validate.
Resolves: COMPMID-8697
Change-Id: I54f73ade5cb4a0d34d831505d83d1d7ef526b5db