- this commit fixes the grid-process assignment logic robust
when the number of grids cannot be exactly divided among
the number of MPI ranks (pieces in this code)
- this commit also adds MPI awareness to an existing unit test.
- the unit test TestAMReXParticlesReader is run with MPI, when enabled,
on 3 nodes. It passes only when the sum of the number of points across
all nodes is equal to a known value.
- without the change in vtkAMReXParticlesReader, the unit test will faill with MPI

(cherry picked from commit de94dced)

- this commit should allow opening files larger than 2GB on windows.
- it uses `size_t`/`vtkIdType` instead of `long` when doing stream
  operations like `seekg`, `tellg`. earlier, it passed `long` (32-bit in
  windows) to `seekg`. As a result, arrays beyond the 2GB offset could
  not be read into the buffer.
- see user report https://github.com/AMReX-Codes/amrex/issues/4167
 and https://discourse.paraview.org/t/potential-length-overflow-for-large-data/15516

(cherry picked from commit d9d40248)

- address paraview/paraview#22835

(cherry picked from commit 2f2329d0)

- address paraview/paraview#22733
- This commit fixes the logic in the
  `vtkPlotBar::GetBounds(double*, bool unscaled)` method
  to scale the bounds only if `unscaled` is false. Previously,
  it returned unscaled bounds even when `unscaled` was false and
  returned scaled bounds when `unscaled` was true.

(cherry picked from commit 6d90695f)

- address paraview/paraview#22594
- address paraview/paraview#21832
- Thick lines were drawn by expanding a line to 2 triangles stacked on their longest side using a geometry shader. When coloring by cell scalars, this worked everywhere except on apple silicon.
- There seems to be a bug related to geometry shaders in Apple's OpenGL driver that caused thick lines to not correctly pick up cell colors.
- It looks like the value written to `gl_PrimitiveID` is overwritten implicitly by `EmitVertex` and incremented each time a vertex is emitted from the geometry shader.
- This commit works around it by fixing the usage of `gl_PrimitiveID` in fragment shader when drawing thick lines.

(cherry picked from commit 62466099)

This reverts commit ff11f345.

This commit caused issue with Qt event processing, especially
in ParaView: vtk/vtk!11788 (comment 1611317)

There was an issue with Fides crashing because it attempted to allocate
memory with HIP before HIP was initialized. (See
fides#26.) The problem was
actually with VTK-m. (Initializing VTK-m is supposed to be optional,
but HIP was not getting initialized in time if it was not.) To fix
the Fides issue, update VTK-m to the changes provided by
vtk-m!3289.

Common cases:
- Cell array only contains triangles
- Cell array only contains triangles, and connectivity type is uint32

- this commit fixes the grid-process assignment logic robust
when the number of grids cannot be exactly divided among
the number of MPI ranks (pieces in this code)
- this commit also adds MPI awareness to an existing unit test.
- the unit test TestAMReXParticlesReader is run with MPI, when enabled,
on 3 nodes. It passes only when the sum of the number of points across
all nodes is equal to a known value.
- without the change in vtkAMReXParticlesReader, the unit test will faill with MPI

Jaswant Panchumarti (Kitware) authored 3 months ago and Jaswant Panchumarti (Kitware) committed 3 months ago

- this commit should allow opening files larger than 2GB on windows.
- it uses `size_t`/`vtkIdType` instead of `long` when doing stream
  operations like `seekg`, `tellg`. earlier, it passed `long` (32-bit in
  windows) to `seekg`. As a result, arrays beyond the 2GB offset could
  not be read into the buffer.
- see user report https://github.com/AMReX-Codes/amrex/issues/4167
 and https://discourse.paraview.org/t/potential-length-overflow-for-large-data/15516

SetTimer is not precise. It's replace by CreateTimerQueueTimer.
This new API is also supporting one-shot timers so it's not necessary
to kill the timer anymore.

- address paraview/paraview#22835

I am bumping the required JAVA SDK to 11 since it simplifies its usage
which results in improved maintanability. It also eliminates the need to
use bootclasspath.

- Merge VTK_JAVA_TARGET_VERSION and VTK_JAVA_SOURCE_VERSION to
  VTK_JAVA_RELEASE_VERSION.
- RENAME VTK_JAVA_INSTALL to VTK_JAVA_BUILD_MAVEN_PKG.
- Update Java build instructions

Co-authored-by: Ben Boeckel <ben.boeckel@kitware.com>

- address paraview/paraview#22733
- This commit fixes the logic in the
  `vtkPlotBar::GetBounds(double*, bool unscaled)` method
  to scale the bounds only if `unscaled` is false. Previously,
  it returned unscaled bounds even when `unscaled` was false and
  returned scaled bounds when `unscaled` was true.

- address paraview/paraview#22594
- address paraview/paraview#21832
- Thick lines were drawn by expanding a line to 2 triangles stacked on their longest side using a geometry shader. When coloring by cell scalars, this worked everywhere except on apple silicon.
- There seems to be a bug related to geometry shaders in Apple's OpenGL driver that caused thick lines to not correctly pick up cell colors.
- It looks like the value written to `gl_PrimitiveID` is overwritten implicitly by `EmitVertex` and incremented each time a vertex is emitted from the geometry shader.
- This commit works around it by fixing the usage of `gl_PrimitiveID` in fragment shader when drawing thick lines.

Jaswant Panchumarti (Kitware) authored 1 year ago and Dan Lipsa committed 3 months ago

Includes the following MRs squashed for easy of rebasing:

commit 695023a3
Author: Dan Lipsa <dan.lipsa@kitware.com>
Date:   Thu Oct 31 13:56:35 2024 -0400

    No need for memorySpace parameter

commit cbd3489d
Author: Dan Lipsa <dan.lipsa@kitware.com>
Date:   Thu Oct 31 13:12:53 2024 -0400

    Use IsDevicePointer instead of IsDirectAccessPossible

commit 6e749aff
Author: Dan Lipsa <dan.lipsa@kitware.com>
Date:   Thu Oct 24 20:29:18 2024 -0400

    Add error logging

commit 144af51f
Author: Dan Lipsa <dan.lipsa@kitware.com>
Date:   Thu Oct 24 16:18:51 2024 -0400

    Use cudaPointerGetAttributes to ask if device or host pointer

commit 9a46db52
Author: Dan Lipsa <dan.lipsa@kitware.com>
Date:   Fri Sep 27 09:47:55 2024 -0400

    No conversion for vtkmDataArray

commit b52093c5
Author: Dan Lipsa <dan.lipsa@kitware.com>
Date:   Thu Sep 26 10:37:21 2024 -0400

    vtkErrorMacro is not available in a static method

commit c318c5bf
Author: Dan Lipsa <dan.lipsa@kitware.com>
Date:   Thu Sep 5 11:26:15 2024 -0400

    Add vtkConduitArrayUtilities to classes of IOCatalystConduit

commit f84ef340
Author: Dan Lipsa <dan.lipsa@kitware.com>
Date:   Thu Sep 5 09:24:36 2024 -0400

    Remove not needed link with catalyst

commit 908563b1
Author: Jaswant Panchumarti <jaswant.panchumarti@kitware.com>
Date:   Tue Dec 19 17:45:09 2023 -0500

    vtkConduitSource: Support mesh data from external memory spaces

feat: add touch event processing flag to QVTKOpenGLWindow

chore: reformat code

Stephen Hogarth authored 6 years ago and Spiros Tsalikis committed 3 months ago

This mapper uses GPU texture acceleration to draw labels at much
higher frame rates. When hundreds or thousands of labels are on
screen at one time, existing mappers can drop below 1 fps. This
acceleration is designed to render the same number of labels at
over 60 fps so a user can label many points while retaining
interactive rotations and animations.

vtkExplodeDataSet creates a vtkPartitionedDataSetCollection from any input dataset according to a cell scalar isovalue criteria.
Each partition of the output contains cells that share the same value for the given cell array.

It comes in replacement of vtkSplitByCellScalarFilter to use
vtkPartitionedDataSetCollection instead of vtkMultiBlockDataSet as output.
Also it relies on the threaded vtkExtractCells filter.

- This commit removes a left over print statement in
  QVTKRenderWindowAdapter.cxx
- The message is already handled correctly by the opengl
debug logger mechanism in VTK. It is caught by vtkOpenGLMessageHandler
in vtkOpenGLRenderWindow and sent to vtkLog macro. There is no need to print it to console.
- This was not caught until 5d4edb2c
enabled debug OpenGL context in Qt. The surrounding code never executed at all.
- added a release note for OpenGL debug output feature in VTK.

David Thompson authored 4 months ago and Ben Boeckel committed 4 months ago

This also forces all inclusions of the `Singleton.h` header
to use angle brackets (since in theory third-party libraries
may be in system paths and even when provided by VTK should
use `-isystem` instead of `-I`).

(cherry picked from commit 615c9f21)

- Adapts TestBandedContourFilter to use SetInputArrayToProcess
in order to verify that the provided array is correctly processed.
- Explicitly turn off copying for the `inScalars` array in addition to
generally turning off copy for all scalars to maintain previous behavior
of the filter.

This also forces all inclusions of the `Singleton.h` header
to use angle brackets (since in theory third-party libraries
may be in system paths and even when provided by VTK should
use `-isystem` instead of `-I`).