Shouldn't SwapBuffers be fast?

#21 What happens if you turn shadow casting off Sean?

#22 Empty room has nothing to do with it. It's deferred lighting. Change your resolution. If the framerate goes up, you're fillrate bound.

#23 I'll also repeat what I said in other threads: for *any* Intel IGPs, or low end GeForce 7XXX / ATI X1XXX cards or older, force T3D to use to 2.0 shaders:

$pref::Video::forcePixVersion = true;
$pref::Video::forcedPixVersion = 2.0;

These are pref values. The best way for forcing them as defaults for specific video cards is creating vendor/card-specific profile scripts in the profile folder (an Intel.cs script would be a nice place to force stuff off on all Intel cards, as example).

#24

Quote:
Empty room has nothing to do with it. It's deferred lighting. Change your resolution. If the framerate goes up, you're fillrate bound.

I'm talking about Basic Lighting. I assumed that was clear from my post. Does Basic Lighting use deferred lighting as well?

On the deathball desert level, in BASIC LIGHTING, in windowed mode, 800x600 resolution, when I run the profiler I'm getting 66% time spent in swapbuffers. When I go to full-screen, it goes down to 10%.

Again, this is BASIC LIGHTING mode, not deferred lighting. I switched AL off long ago and never looked back. Obviously, being in windowed mode is making swapbuffers do more work. This makes sense, but these percentages are still way off what they should be.

For comparison, when I run TGEA 1.8.2 in windowed mode, 800x600 resolution, I get about 1% frametime spent in swapbuffers. When I switch to the highest resolution, in windowed mode, I get about 1% frametime spent in swapbuffers. swapbuffers is taking up much less frame time in TGEA. Even in windowed mode, in TGEA swapbuffers only takes about 1%-2% frame time.

it's easy to say it's just a video card or resolution issue, but when you take into consideration the comparison of my results in TGEA to my results in T3D BASIC LIGHTING, the numbers just don't make sense to me. I thought TGEA was comparable to T3D basic lighting.

#25 *points to Manoel's post*

Several newer threads have solved their issues using his suggested fix.

Afaik Gideon uses SM 3 shaders so there will be a bit of fixing involved.

#26 I appreciate your help Jacob, but that change had no effect on my results:

==>echo($pref::Video::forcePixVersion);
1
==>echo($pref::Video::forcedPixVersion);
2
==>profilerdump();
Profiler Data Dump:
Ordered by non-sub total time -
%NSTime  % Time  Invoke #  Name
 62.534  62.534     2522 SwapBuffers
  9.931  11.422   129457 GFXDevice_drawTextN
  3.913  33.005     2522 CanvasRenderControls
  2.515   4.939    10088 RenderMeshMgr_render
  2.371  99.922     2522 MainLoop
  2.115   2.147     5044 TSSkinMesh_UpdateSkin
  1.309   2.026   390491 GFXDevice_updateStates
  1.200   5.895     2522 ProjectedShadow_RenderToTexture
  0.979   1.926     2522 RenderTerrainMgr_Render
  0.872   0.872 10833150 GFont_getCharInfo
  0.716   0.716    89135 GFXD3D9StateBlock_Activate
  0.709   0.965     2522 AdvanceClientTime
  0.696   0.696     5044 TSMesh_CreateVBIB
  0.600   0.621     2522 DecalManager_RenderDecals_RenderBatch
  0.561   0.805    17654 RenderObjectMgr_render
  0.478   0.478  2330328 GenericConstBufferLayout_set
  0.439   0.439    45396 terrCheck
  0.433   3.860    47918 TSShapeInstance_Render
  0.413   0.413     7566 ContainerCastRay
  0.371   0.591   216892 LightManager_Update4LightConsts
  0.338   0.693   388388 GFXD3D9ShaderConstBuffer_activate
  0.335   0.335   224458 SceneRenderPassManager_addInst
  0.311   0.582    68094 TSMesh_InnerRender
  0.310   0.465     3652 Player_PhysicsSection
  0.298   0.325     2522 RenderGlowMgr_Render
  0.254   0.360   171496 MatInstance_setTransforms
  0.233   0.234   134937 convertUTF8toUTF16
  0.233   2.535    50440 treeTraverseVisit_prepRenderImage
  0.214   0.439   141232 ProcessedShaderMaterial_SetupPass
  0.206   0.206    35308 Frustum_UpdatePlanes
  0.202   0.202     2522 ContainerCastRayRendered
  0.179   0.263     2522 TerrainBlock_RenderBlock
  0.165   3.557     5044 ShapeBase_PrepRenderImage
  0.164   0.396    85748 TerrainCellMaterial_SetupPass
  0.153   0.193    42874 TerrainCellMaterial_SetTransformAndEye
  0.148   0.154     6174 ContainerFindObjects_Box
  0.146   0.234   108446 GFXD3D9ShaderConstBuffer_activate_dirty_check_1
  0.142   0.142     2522 GFXEndScene
  0.135   8.166     5044 RenderPassManager_Render
  0.128   0.174    73138 ProcessedShaderMaterial_SetShaderConstants
  0.128   0.639     3652 AdvanceObjects
  0.126   0.327     2522 LightFlareData_prepRender
  0.122   0.122   368212 GenericConstBuffer_getDirtyBuffer
  0.112   0.122     2522 CanvasPreRender
  0.109   0.109     5044 RenderPassManager_Sort
  0.105   0.797   388388 GFXD3D9Device_setShaderConstBufferInternal
  0.100   1.152     2522 ClientProcess
  0.091   6.052     2522 SceneGraphRender_PreRenderSignal
  0.088   0.088   433784 GenericConstBuffer_isEqual
  0.082   0.673   216892 BasicLightManager_SetLightInfo
  0.078   0.562   153842 MatInstance_SetupPass
  0.076   0.076     4354 SimFindObject
  0.071   3.246     2522 BuildSceneTree
  0.070  16.690     2522 SceneGraphRender
  0.066   0.066     2522 GFXBeginScene
  0.060   0.060     2522 SFXXAudioDevice_Update
  0.057   0.057    95836 String_char_constructor
  0.055   0.155     3652 Player_UpdatePos
  ...

If I were to take a guess at what the problem may be, I'd say either T3D is creating multiple swap chains in Basic Lighting for some reason, or it's using 1 swap chain but allocating more than 3 buffers degrading performance.

#27 @Logan - turning off shadows doesn't seem to make much of a difference. swapbuffers went from 63% to 59%.

switching resolution while in windowed mode doesnt seem to do much either. I switched to 640x480 and swapbuffers is still taking about 50% frame time while running at 50 fps in deathball. Is this a false metric from the profiler tool? is the profiler in TGEA giving me a false metric for swapbuffers?

#28 *update*

I just ran Forge Demo in TGEA yesterday to see what swapbuffers is doing there. swapbuffers was at 75% which means TGEA is showing this same discrepancy in the profiler, depending on the mission.

this leads me to conclude that T3D has simply inherited this from TGEA, which makes sense.

I'm now inclined to think that this actually is simply a side effect of the CPU stalling while waiting for the GPU to finish up it's work as someone posted earlier.

This should be addressed since, at the very least, this is making the profiler show inaccurate results.

AFAIC, this issue is resolved.