[train] Modify maxPool2D in training #14166
Conversation
This commit change casting to static cast. ONE-DCO-1.0-Signed-off-by: JuYoung Lee [email protected]
|
from : #14149 (comment) |
icodo98
changed the title
| assert(paddingTop < (1 << 8)); | ||
| assert(paddingLeft < (1 << 8)); | ||
| _op_params.padding_values.height = static_cast<uint8_t>(paddingTop); | ||
| _op_params.padding_values.width = static_cast<uint8_t>(paddingLeft); |
There was a problem hiding this comment.
Ah? It looks like typecasting to int16_t is necessary, not uint8_t.
ONE/compute/cker/include/cker/Types.h
Lines 81 to 83 in c82f8cb
ONE/compute/cker/include/cker/Types.h
Lines 66 to 70 in c82f8cb
There was a problem hiding this comment.
Tthere is no problem since _op_param.padding_values are filled with zero-extended value from positive numbers of equal range.
- These padding parameters are always positive number.
The positive number range ofThe positive number range of both are not equal. Sorry for the wrong information.int16_tis equal to the range ofuint8_t.
There was a problem hiding this comment.
Aha I got it! Thank you for explanation 👍
There was a problem hiding this comment.
The positive number range of int16_t is equal to the range of uint8_t.
@ragmani Sorry, But I have another question.
Does it mean that int16_t and uint8_t have a same max value?
There was a problem hiding this comment.
Sorry, it was the wrong information. They are not the same max value.
uint8_t: max value is1 << 7 - 1int16_t: max value is1 << 15 - 1
There was a problem hiding this comment.
Could you please check which type is appropriate?
#14166 (comment)
This commit updates padding type to int16 ONE-DCO-1.0-Signed-off-by: JuYoung Lee [email protected]
|
@nnfw-bot test all |
This commit change casting to static cast.
ONE-DCO-1.0-Signed-off-by: JuYoung Lee [email protected]