Escape subnormal values #1953
Escape subnormal values #1953
Conversation
|
|
|
Section 8.3.3 of the C spec specifically allows sampling to flush denorms, so this change seems to be long overdue. |
|
Reviewing |
There was a problem hiding this comment.
I think I'm OK with this, but noting for completeness: this change removes all fp32 and fp16 denorm values from the input data. If we want to have some coverage for denorm inputs still, we will want to continue to include denorm inputs, and instead we will want to handle cases where denorms are preserved and when denorms are flushed to zero.
|
A bit of spec archaeology, for the Khronos folks:
|
Currently we don't escape subnormal values when generating image data. In sampler read tests, we use `!=` to check the two values even when it is floating-point data, which requires the two values are bitwise equal. However, a sampler might flush subnormal values, causing the test case to fail. In this patch, when generating random image data, we escape subnormal values.
|
I don't have privileges to merge the PR. Can someone do it for me? Thanks! |
|
It'd be nice to get one more approval before merging - @silverclaw or @lakshmih maybe? If not, we'll aim to merge in our next teleconference, on Tuesday (June 4th). |
|
Thanks! Merging. |
Currently we don't escape subnormal values when generating image data. In sampler read tests, we use
!=to check the two values even when it is floating-point data, which requires the two values are bitwise equal. However, a sampler might flush subnormal values, causing the test case to fail.In this patch, when generating random image data, we escape subnormal values.