Fix snorm #2033
Fix snorm #2033
Conversation
When comparing scanlines for SNORM images, take into account that -1.0 can be exactly represented in two different ways.
test_common/harness/imageHelpers.cpp
Outdated
| break; | ||
|
|
||
| case CL_SNORM_INT16: { | ||
| cl_short aPixel = *(cl_short *)aPtr; |
There was a problem hiding this comment.
This needs to be ushort
Does this only apply to the Embedded Profile or also to the Full Profile? |
|
Can you please explain a bit more exactly what scenario this change is trying to fix? Does the problem happen when copying SNORM data from one image to the other? If so, does the multiple representations for -1.0 cause problems for the source data, or the destination data, or both? The reason why I ask is because the spec is pretty clear how the conversions between SNORM <-> float should occur: I think I have an idea how this still could cause a problem (specifically with the two source representations for -1?) but I'd like to be sure I understand it - thanks! |
When comparing scanlines for SNORM images, take into account that -1.0 can be exactly represented in two different ways.