| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
shared GLContext, removing resource restrictions
In a use case with hundreds of GLMediaPlayer instances,
this causes the application to bail out due to running out of resources.
+++
GLMediaPlayer exposes resource restrictions and locking related
with the created off-thread shared GLContext
due to its newly created NativeWindow *GraphicsDevice instance (on X11).
On the X11 platform, the *GraphicsDevice actually uses a native resource (X11 Display handle) and hence creating such device is costly and limited.
To operate an off-thread GLContext w/o actual X11 input handling,
it is *NOT* required to use a new instance.
+++
Further more, the device is using locking.
To operate an off-thread GLContext, it is *NOT* required to use locking on it
as it does not perform actual X11 input handling etc.
All operations are performed on the shared GL context.
+++
Solution is to create a shared non-locking device clone
compatible with the source.
A share counter shall determine that the last one actually gets destructed.
The usual ..
|
|
|
|
|
|
|
|
|
| |
[initStreamImpl - destruction]
initStreamImpl() calls ffmpeg natives.setStream0(..), which in turn callsback to the GLMediaPlayerImpl
and FFMPEGPlayer and hence requires a valid moviePtr. In total, it covers a longer time period.
This patch uses a moviePtrLock object avoiding destruction while within initStreamImpl.
|
|
|
|
| |
DefaultGraphicsDevice.swap{Device->}HandleAndOwnership()
|
|
|
|
|
|
|
|
|
|
| |
DefaultGraphicsDevice: Move ownership (Object) code into base class ensuring same code
NativeWindowFactory.createDevice(..) w/ unitID
- Allows cloning a device instance with same parameter.
DefaultGraphicsDevice: Move ownership (Object) code final into base class ensuring same code
- Rename DefaultGraphicsDevice.swapDeviceHandleAndOwnership() -> swapHandleAndOwnership()
|
|
|
|
|
|
|
|
|
|
| |
bit-shift expressions and simplify it
commit 1dcfdf71c09c6d774ac47012c05e09da4a085d7b
- still used the 'hash code' bit shift pattern, not necessary -> simplified
- the value as not ensured to be long, hence conversion occured
This caused Graph's MSAA not being picked up properly using the shaderKey.
|
| |
|
|
|
|
| |
oops
|
|
|
|
| |
actual tc_w_1 = (float)getWidth() / (float)texWidth value as hardcoded within the shader.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
[0-31] bit values and state, [32-63] colorTexSeqHash
This leaves only room for a key collision on the 32-bit colorTexSeqHash value
and hence should be save within our shader-code environment.
+ // # | s |
+ // 0 | 1 | isTwoPass
+ // 1 | 1 | pass1
+ // 2 | 5 | ShaderModeSelector1
+ // 7 | 1 | hasFrustumClipping
+ // 8 | 1 | hasColorChannel
+ // 9 | 1 | hasColorTexture
+ // 32 | 32 | colorTexSeqHash
+ long hash = ( isTwoPass ? 1 : 0 );
+ hash = ( hash << 1 ) | ( pass1 ? 1 : 0 ) ;
+ hash = ( hash << 1 ) | sms.ordinal(); // incl. pass2Quality + sampleCount
+ hash = ( hash << 5 ) | ( hasFrustumClipping ? 1 : 0 );
+ hash = ( hash << 1 ) | ( hasColorChannel ? 1 : 0 );
+ hash = ( hash << 1 ) | ( hasColorTexture ? 1 : 0 );
+ hash = ( hash << 1 ) | ( ( colorTexSeqHash & 0xFFFFFFL ) << 32 );
|
|
|
|
|
|
| |
using 'static final boolean useShaderPrograms0 = true'
For now, let's leave the dead shaderPrograms1 code path inside the class for further consideration.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
-> PMVMatrix4f
Shape.setTransformMv() is called for each renderer frame and for each shape,
involving 6 Matrix4f.mul() and set*() operations.
Since mutation of shape's position, rotation or scale is less frequent
than rendering one frame (for all shapes),
it is more efficient to maintain a local Matrix4f and update it
on such single mutations.
Rendering then only needs to perform one Matrix4f.mul() operation
using this internal matrix.
+++
Also changes name from setTransformMv(PMVMatrix4f) to applyMatToMv(PMVMatrix4f),
as its name might be misleading.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
and throw exception on shaderKey collision
Commit 6363ae5fb6975a6f2e7c1093ce81f25b699e3e61 changed
RegionRenderer.useShaderProgram()'s shader mapping using a new ShaderKey instance.
Such ShaderKey instance is created every time @ RegionRenderer.useShaderProgram()
to retrieve the ShaderProgram from the HashMap<ShaderKey, Shader Program>.
While this is most correct, creating the ShaderKey instance causes
a lot of temp objects.
ShaderKey also uses the optional colorTexSeq shader code for equality test
in case of hash-collisions.
Previous code simply ignored hash-collisions and used a 1:1 hashCode -> ShaderProgram mapping using our IntObjectHashMap.
However, there was no test whether collision occur.
+++
Solution would be either
1- Revert fully to the previous code just using an IntObjectHashMap,
but throwing a RuntimeException in case of hashCode collisions.
In case of a collisions, we would need to produce a better hashCode.
This is possible, as the underlying data is fully internal .. etc.
2- Use the IntObjectHashMap as long there is no hashCode collision,
then revert back to HashMap<ShaderKey, Shader Program>.
+++
This patch implements variant (1), so far no exception has been thrown on
multiple demos w/ and w/o diff color-textures.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Frustum mapping + GraphUI Support
AABBox clipping naturally couldn't be transformed into 3D Model-View (Mv) Space,
as it is axis aligned and only provided 2 points (min/max).
Therefor we map the Group's AABBox to a 8-point Cube,
perform the Mv-transformation and then produce the 6-plane Frustum.
As before, we cull fully outside shapes within the Group's draw method
and perform fragment clipping with same Frustum planes in the shader.
With clipping enabled, the 3D z-axis getBounds() depth
will be slightly increased for functional Frustum operation.
This is also done for setFixedSize(Vec2f).
The Frustum planes are copied to the Graph shader
via float[4*6] -> uniform vec4 gcu_ClipFrustum[6]; // L, R, B, T, N, F each {n.x, n.y, n.z, d}
+++
Concludes related work of below commits
- 1040bed4ecc6f4598ea459f1073a9240583fc3c3
- 5cca51e32999a882e2a5f00cb45ecafc824ffd86
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
well as to extract planes for float[] vec4-shader uniforms.
commit 1040bed4ecc6f4598ea459f1073a9240583fc3c3 added AABBox -> Cube -> Frustum mapping (incomplete)
and requires Frustum.Plane.set(..) by normal and point-on-plane for distance.
Frustum.isOutside(Cube) has been added, testing all its 8-points similar to AABBox.
Further all 6 Frustum.Plane shall be extracted to Vec4f and float[],
the latter to pass the whole float[4*6] as a vec4[6] uniform array to the shader.
+++
Constructor, setter and getter have been adjusted accordingly.
Most of the loops have been unrolled.
+++
Method names to query Frustum, i.e. 'is*Outside(<Type>)'
have been reduced to 'isOutside(<Type>)'
where <Type> uniquely indenticates the purpose.
Hence only 'isSphereOutside()' is left over.
|
|
|
|
| |
AABBox into model-view Cube to Frustum.Plane for culling (cpu) and clipping (gpu)
|
|
|
|
|
|
|
|
|
|
|
| |
orientation in API doc
br, tr wasn't sufficient as in commit d778889f36bd6bee999ceb502c5f0ce265b014bf
while working on Frustum, as it doesn't properly reflect axis order not z.
Hence going back to 'low' and 'high' semantics,
but using same length identifier and emphasizing far (lo) < near (hi)
of our model-view coordinate system.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
RegionRenderer's RenderState usually rarely set from top of user API, reducing complexity.
Discussion:
Alternative was to pass AA-Quality same as SampleCount from the top (e.g. GraphUI Scene),
however, this convolutes the API even more.
Both parameter modify the resulting shader code in pass2 rendering (only).
The used 'renderMode' is still maintained within the Region,
since it contains more dynamic states individual to each Region instance (color-texture, ..).
This despite 'renderMode' also changes the RenderState's shader program.
In the end, it really doesn't matter and is a choice of frequency - the pipeline is
usually rendering from on OpenGL rendering thread sequentially.
AA-Quality and SampleCount simply usually don't change that often
and are set only once.
|
|
|
|
| |
and scale2(float, float, float)
|
|
|
|
| |
Unroll getRayIntersection()'s 'find candidate planes'
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
clipping funs for aaQuality/sampleCount; TextRegionUtil: Pass quality parameter in draw-functions
Region.DEFAULT_AA_QUALITY defaults to MAX_AA_QUALITY still
- TODO: AA shader is subject to change ..
Region.draw(..) clips the quality param (save)
TextRegionUtil: Pass quality parameter in draw-functions
- Allowing to select the AA shader
GraphUI Scene and some demos add the AA-quality param
to the status line or screenshot-filename.
- See Region.getRenderModeString(..)
+++
TestTextRendererNEWT20 and TestTextRendererNEWT21
now iterate through all fonts, AA-quality shader and sample-sizes.
Most demos and some more tests take AA-quality into acount,
demos via CommandlineOptions.graphAAQuality
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Pass2-AA, revise Pass2 AA Quality parameter ..
Misc:
- Graph VBORegion2PVBAAES2: Drop unused FBO rescale
- Move MIN/MAX QUALITY/SAMPLE from GraphUI Scene -> Graph Region
+++
Quality -> Pass2 AA Quality
- Drop quality field in region
- Pass quality to GLRegion.draw(..) similar to sampleCount
for dynamic shader and switch
- TODO: Pass quality parameter in TextRegionUtil's functions
Fix RegionRenderer Shader Mapping
- Use ShaderKey class to properly implement the hash value and equals method
- For this, TextureSequence.getTextureFragmentShaderHashID() has been added
to provide actual shader-snippet for the equals function
- All required criterias are included in the hash value and equals method
Fix AABBox Clipping for Pass-2 AA
- Clipping in pass2-AA must happen in pass2 on actual gcu_PMVMatrix01 (not ortho)
+++
GraphUI GraphShape
- Rename: [get,set]{->AA}Quality()
GraphUI Scene
- Rename: mark{All->}ShapesDirty(), set{AllShapes->}Sharpness(), set{AllShapes->AA}Quality()
- Fix setSampleCount(..), i.e. markStatesDirty() not markShapesDirty()
- Fix setAAQuality(), markShapesDirty() and markStatesDirty(): Use forAll(..) to traverse through all shapes and groups.
GraphUI Group
- Add setFixedSize()
- Add setClipOnBox()
- Document setRelayoutOnDirtyShapes(), isShapeDirty()
|
|
|
|
| |
FBO attachments (GL_COLOR_ATTACHMENTi)
|
| |
|
|
|
|
| |
use winHeight (typo)
|
|
|
|
| |
expansion
|
|
|
|
|
|
| |
- GLSL vertex shader sets smooth varying 'gcv_ClipBBoxCoord' w/ Mv multiplied vertex-coord
- RegionRenderer.setClipBBox(AABBox) expects a pre-multiplied Mv AABBox covering an independent area, not per Shape/Region.
- This works as expected with moving/scaling of each Shape/Region etc
|
|
|
|
|
|
|
|
|
|
|
|
| |
(Convenient using Graph/GraphUI produced AABBox)
Simple demo, setting clip-bbox manually:
- src/demos/com/jogamp/opengl/demos/graph/ui/UIShapeClippingDemo00.java
TODO:
- GLSL: Add missing Mv-multiplication of vertex-position -> gcv_ClipBBoxCoord
-- AABBox min/max should be set pre-multiplied w/ Mv covering an independent area, not per Shape/Region.
-- This to properly work with moving/scaling of each Shape/Region etc
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
min/max vec3 as convenient using Graph/GraphUI produced AABBox
USE_AABBOX_CLIPPING
- Conditional compilation w/ macro 'USE_AABBOX_CLIPPING'
- gcv_ClipBBoxCoord smooth varying setup in vertex shader
- fragment shader clips via branch
if( is_inside(gcv_ClipBBoxCoord, gcu_ClipBBox[0], gcu_ClipBBox[1]) < 0.5 ) { CLIP }
- clipping via discard or alpha=0 in case of buggy-discard.
Other optimization:
- Drop gcv_ColorTexExt, fragment-shader uses gcu_ColorTexBBox[2] directly (flat)
- Simplified gcv_ColorTexCoord smooth varying equation in vertex shader.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
'and'/'or' semantic (swapped); Add EPSILON in clip_coord(..) and add is_inside(..) function
Complete overload vec2 and vec3 variants
Fix 'and'/'or' semantic (swapped)
- 'and' uses multiplication, i.e. all arguments must be > 0 (ideally 1)
- 'or' uses addition, i.e. only one arguments must be > 0 (ideally 1)
- both uses clamp [0..1]
Add EPSILON in clip_coord(..)
- Only 'coord > high+EPSILON' is outside
Add is_inside(..) function
- Similar to clip_coord(..) but returns float 0 or 1
instead of selecting color.
|
| |
|
|
|
|
|
|
|
|
| |
com.jogamp.common.av.PTS.millisToTimeStr(..)
Chapter metadata is now supported via our FFMPEGMediaPlayer implementation.
Added public method: 'Chapters[] GLMediaPlayer.getChapters()'
|
|
|
|
| |
d99c2d8b28470d335ab5b30124ef8b0607b3b90f
|
|
|
|
| |
usable w/ fix for Bug 1472 (Enhance A/V Sync)
|
|
|
|
| |
information (if any), FFMPEG* utilizes NativeLibrary.get[Native]LibraryPath()
|
|
|
|
| |
'curverenderer01-pass1.vp' (commit 297c48f4fefd1ab59800524ea5f0dd56684d6786)
|
| |
|
|
|
|
| |
library path, support throughout DynamicLibraryBundle[Info]
|
|
|
|
| |
misleading and wrong info (e.g. on mute)
|
|
|
|
| |
disruption after pause(flush:=false)
|
|
|
|
| |
i.e. ignore audio_queued_ms in such case (Fixes video sync on SCR w/o audio)
|
|
|
|
| |
direct user-exposure and potential data-race
|
|
|
|
| |
w/o video-frame (fixes 'massive' seek'ing)
|
|
|
|
| |
Left behind hacking for AV sync ..
|
|
|
|
|
|
| |
!hasVideoFrame upfront, simplifying sync-block
Fix playSpeed > 1: dt_v (keep playSpeed factor), allow drop frame on no-frames w/ playSpeed > 2
|
|
|
|
| |
stream is not yet initialized or playing, cached clipped value and set at initGL(..)
|
|
|
|
| |
user from selecting video or audio PTS.
|
|
|
|
| |
USE_GLSL_VERTICAL_FLIP, changed names in commit d6cf89f22a5926b437c4430eb166972d90fd92be
|