Vulkan dump resources: Add support for R32G32B32A32_SFLOAT images #1768
Conversation
|
CI gfxreconstruct build queued with queue ID 264775. |
|
CI gfxreconstruct build # 4904 running. |
|
CI gfxreconstruct build # 4904 passed. |
| float float_b = float_vals[x + 2]; | ||
| float float_a = float_vals[x + 3]; | ||
|
|
||
| uint8_t r = static_cast<uint8_t>(float_r * 255.0f); |
There was a problem hiding this comment.
If the user wants the range [0.0f,1.0f] to be evenly distributed among uint8_t values, I think 255.0f should be std::nextafter(256.0f, 0.0f) (which will equate to 255.99...f). Also, SFLOAT isn't normalized so you probably need to clamp the source values here to [0.0f, 1.0f]. Maybe some day in the future we can go back and alter the other case statements to match but not in this PR.
There was a problem hiding this comment.
To prevent this feedback from blocking this PR when it applies to all other cases in image_writer.cpp and not just this new addition, I've moved this feedback to #1790 and will remove my feedback on this PR. Note that values will not be distributed uniformly from [0,1] to [0,255] and values outside the range [0,1] will wrap.
bradgrantham-lunarg
added
the
P1
|
Submitted #1799 instead |
No description provided.