Now an 8-bit stencil buffer is requested when using a 24-bit depth buffer (the default).
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1209 af15c1b1-3010-417e-b628-4374ebc0bcbd
mh suggest that not doing this may cause performance
issues: http://forums.inside3d.com/viewtopic.php?t=2408
It's also mentioned in a "Depth In-Depth" document from AMD.
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1203 af15c1b1-3010-417e-b628-4374ebc0bcbd
When cvar vid_desktopfullscreen is set to 1, fullscreen mode ignores vid_width/vid_height and uses the desktop resolution. SDL docs recommend using this, it's supposed to offer better OS integration.
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1164 af15c1b1-3010-417e-b628-4374ebc0bcbd
Avoids general bugginess of hardware gamma, doesn't affect the whole screen in windowed mode like HW gamma. Should fix http://sourceforge.net/p/quakespasm/bugs/5/
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1157 af15c1b1-3010-417e-b628-4374ebc0bcbd
In GL_MakeAliasModelDisplayLists, after saving the standerd triangle strips and fans onto the hunk, if the system is GL2 capable, we also save a second version of the mdl on the hunk designed to be loaded into a VBO (GL_MakeAliasModelDisplayLists_VBO). In R_NewMap, and on video mode changes, we call GLMesh_LoadVertexBuffers which loops over all precached mdl's and loads the data into a pair of VBO's (vertices and vertex indices).
Finally, in R_DrawAliasModel, assuming no rendering options are disabling the fast-path (r_drawflat 1, r_lightmap 1, or r_fullbright 1 would disable it), we call GL_DrawAliasFrame_GLSL, which sets up all of the bindings and draws the (possibly lerped) mdl in one glDrawElements call.
Special thanks to MH for some of the code from RMQEngine and the general concept.
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1151 af15c1b1-3010-417e-b628-4374ebc0bcbd
This moves some of the OpenGL implementation to a background thread, and makes draw calls block for a shorter amount of time. AFAIK, this is the default mode on most other OS'es, not sure why you have to explicitly ask for it on OS X.
N.B. the kCGLCEMPEngine constant is available on 10.4+ so I don't bother checking the OS X version.
More info: https://developer.apple.com/library/mac/technotes/tn2085/_index.html
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1137 af15c1b1-3010-417e-b628-4374ebc0bcbd
1. Remove Key_IgnoreTextInput(), and simple always send a Char_Event when
we receive TEXTINPUT (SDL2) or the unicode field of a KEYDOWN event is
filled (SDL1.).
2. Remove handling of K_KP_* in the menu/console, since they cannot be
relied on to not also send text (see issue described above). The handling
of K_KP_ENTER can stay, since we do know that it never sends text.
3. Remove the interpretation hack for the numpad, since it will no longer
be needed (and doesn't currently work for SDL2 anyway); if a numpad key
generates text we will handle it (because of "1"), if not then it will
simply be if ignored (because of "2”).
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1110 af15c1b1-3010-417e-b628-4374ebc0bcbd
On my old laptop, when running "./quakespasm -fsaa 2", quakespasm would error out with a "Couldn't create window" message. Our "no FSAA available" fallback was applied at OpenGL context creation time, but according to the SDL2 wiki FSAA settings should be done before creation of the window (see https://wiki.libsdl.org/SDL_GLattr#multisample). Moved it there.
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1084 af15c1b1-3010-417e-b628-4374ebc0bcbd
R_DrawTextureChains_Multitexture: revert to the way it was before VBO support was added.
gl_texmgr: expose GL_SelectTexture. make the implementation less convoluted and support 3 TMUs.
gl_vidsdl: check GL_MAX_TEXTURE_UNITS
r_brush: only create VBOs if 3 TMUs available
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1035 af15c1b1-3010-417e-b628-4374ebc0bcbd
This seems to help fps a fair bit on maps with a lot of world polys like jam2_tronyn. Tried on a few computers with intel and nvidia gpus, windows, mac os, linux, and there's always at least some fps improvement. Best case was 70fps -> 96fps on jam2_tronyn, on OS X + nvidia 650gt.
Interested to hear how this works for amd gpu's, just do a timedemo with and without "-novbo".
Only downside is I had to disable the fast path in Vid_Toggle_f() because at least with SDL1, the vbo no longer works after a toggle. So as a result, fullscreen toggles with alt-enter are slightly slower.
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@1018 af15c1b1-3010-417e-b628-4374ebc0bcbd
It caused a dramatic slowdown on my machine on windows (50% slower)
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@983 af15c1b1-3010-417e-b628-4374ebc0bcbd
The glBegin(GL_POLYGON), for() loop with GL_MTexCoord2fFunc()/glVertex3fv() calls, and glEnd () per polygon is replaced by a memcpy() and a glDrawArrays() call per poly.
My guess is this only saves function call overhead, but it gives a decent ~20% speedup (36 -> 42fps) on my machine on orl’s map.
Note: glDrawArrays() is in OpenGL 1.1, but the glClientActiveTextureARB function is part of the GL_ARB_multitexture extension but not GL_SGIS_multitexture. Instead of splitting into two code paths, I just dropped support for GL_SGIS_multitexture.
git-svn-id: svn://svn.code.sf.net/p/quakespasm/code/trunk/quakespasm@981 af15c1b1-3010-417e-b628-4374ebc0bcbd
