-
Bug
-
Resolution: Fixed
-
P2: Important
-
6.9.3, 6.10.0
-
None
-
Intel X86_64 platform.
-
Windows
-
de1aff1b8 (dev), b37d0a67b (6.10), 4347c2fbb (tqtc/lts-6.8)
In our writing tests with KImageFormat's EXR plugin, we noticed a significant conversion difference
between ARM64 and AMD64. From our tests it appears that the problem occurs on Intel x86_64 (verified on Linux and Windows).
The following QImage formats were incorrectly converted to Format_RGBA16FPx4:
- Format_ARGB32_Premultiplied
- Format_ARGB6666_Premultiplied
- Format_ARGB8555_Premultiplied
- Format_ARGB8565_Premultiplied
- Format_RGBA8888_Premultiplied
I've attached the source image used, an example of an FP16 image with the problem, and an FP32 image for comparison (unaffected).
I've also included a small program to facilitate verification.
| For Gerrit Dashboard: QTBUG-141255 | ||||||
|---|---|---|---|---|---|---|
| # | Subject | Branch | Project | Status | CR | V |
| 685257,2 | Fix alpha value after to-FP16 conversions on AVX2 | dev | qt/qtbase | Status: MERGED | +2 | 0 |
| 685347,2 | Fix alpha value after to-FP16 conversions on AVX2 | 6.10 | qt/qtbase | Status: MERGED | +2 | 0 |
| 685582,2 | Fix alpha value after to-FP16 conversions on AVX2 | tqtc/lts-6.8 | qt/tqtc-qtbase | Status: MERGED | +2 | 0 |