Bug report #21405
Raster Calculator wrong results
| Status: | Closed | ||
|---|---|---|---|
| Priority: | High | ||
| Assignee: |  Alessandro Pasotti | ||
| Category: | Raster Calculator | ||
| Affected QGIS version: | 3.6.0 | Regression?: | No |
| Operating System: | Easy fix?: | No | |
| Pull Request or Patch supplied: | Yes | Resolution: | fixed/implemented |
| Crashes QGIS or corrupts data: | Yes | Copied to github as #: | 29222 |
Description
In Qgis 3.6.0 the Raster Calculator delivers wrong results.
Example:
two geotiffs Sentinel-2
expression: 0.5*((2*"B08@1"+1)-sqrt((2*"B08@1"+1)^2-8*("B08@1"-"B04@1")))
3.6.0: 0.5*((2*0.4544+1)-sqrt((2*0.4544+1)^2-8*(0.4544-"0.0514")))= nan
3.4.5: 0.5*((2*0.4544+1)-sqrt((2*0.4544+1)^2-8*(0.4544-"0.0514")))= 0.630549
Excel: 0.5*((2*0.4544+1)-sqrt((2*0.4544+1)^2-8*(0.4544-"0.0514")))= 0.630549
3.6.0: 0.5*((2*0.2768+1)-sqrt((2*0.2768+1)^2-8*(0.2768-"0.0448")))= 0.883769
3.4.5: 0.5*((2*0.2768+1)-sqrt((2*0.2768+1)^2-8*(0.2768-"0.0448")))= 0.4034125
Excel: 0.5*((2*0.2768+1)-sqrt((2*0.2768+1)^2-8*(0.2768-"0.0448")))= 0.4034125
Associated revisions
[opencl] Fix raster calculator operator precedence
With test
Possibly fixes #21405 (not yet sure if the user was using OpenCL)
History
#1
Updated by Alessandro Pasotti over 5 years ago
- Assignee set to Alessandro Pasotti
#2
Updated by Alessandro Pasotti over 5 years ago
Can you share a small portion of the rasters where we can reproduce the issue?
Also, can you check if OpenCL acceleration is enabled in the QGIS settings? If it is enabled please check if disabling it does fix the issue.
#3
Updated by Alessandro Pasotti over 5 years ago
- Status changed from Open to Feedback
#4
Updated by Alessandro Pasotti over 5 years ago
- Pull Request or Patch supplied changed from No to Yes
PR https://github.com/qgis/QGIS/pull/9305
I've found an issue in the OpenCL implementation of the calculator, but I'm not sure if that is the same issue.
#5
Updated by Alessandro Pasotti over 5 years ago
- % Done changed from 0 to 100
- Status changed from Feedback to Closed
Applied in changeset qgis|88a96122b80475dcae53965d78f4e7c27c374a60.
#6
Updated by monokultur - over 5 years ago
- File Result_3.4.5_no_opencl.tif added
- File Result_3.6.0_opencl_AMD_RX580.tif added
- File RT_T33UVV_A008628_20181031T101140_B08_clip.tif added
- File RT_T33UVV_A008628_20181031T101140_B04_clip.tif added
- File Result_3.6.0_opencl_NVIDIA_1000M.tif added
- File Result_3.6.0_no_opencl.tif added
- File Result_3.4.5_opencl_AMD_RX580.tif added
Yes, it's an issue in the OpenCL implementation of the calculator.
I've tested some variants:
- 3.4.5 with OpenCL (AMD RX580) - OK
- 3.4.5 without OpenCL (AMD RX580) - OK
- 3.6.0 with OpenCL (AMD RX580) - wrong result
- 3.6.0 without OpenCL (AMD RX580) - OK
- 3.6.0 with OpenCL (NVIDIA 1000M) - wrong result
- 3.6.0 without OpenCL (NVIDIA 1000M) - OK
#7
Updated by Giovanni Manghi over 5 years ago
- Crashes QGIS or corrupts data changed from No to Yes
- Easy fix? changed from Yes to No
- Resolution set to fixed/implemented
#8
Updated by monokultur - over 5 years ago
- Status changed from Closed to Reopened
Version 3.6.1, 3.6.2, 3.6.3 > Problem still there.
#9
Updated by Alessandro Pasotti over 5 years ago
The patch was never backported, can you please check if current master works with OpenCL? If it does I can backport the fixes to 3.6 in time for the next point release.
#10
Updated by monokultur - over 5 years ago
- Status changed from Reopened to Closed
Ah, OK I thought it should be fixed since 3.6.1. My mistake.
The current master works with OpenCL.
Next point release is 3.8.0. So I think its no nessesary to backport to 3.6.