Challenge ProMode Arena

CNQ3 1.52 Changelog

Content

Changelog for CNQ3 1.52

Additions

  • FreeBSD x64 support

  • r_backend <GL2|GL3|D3D11> (default: D3D11 on Windows, GL3 otherwise) selects the rendering back-end

    • GL2 - OpenGL 2.0 minimum, OpenGL 3+ features used for r_msaa
    • GL3 - OpenGL 3.2 minimum, OpenGL 4+ features used for faster geometry upload, compute shaders, etc
    • D3D11 - Direct3D 11 hardware feature level 10.1 minimum, feature level 11.0 for compute shaders
    Feature / r_backend GL2 GL3 D3D11
    MSAA GL 3.0 always always
    MSAA centroid sampling never always always
    depth fade never always always
    alpha to coverage AA never always always
    dithering never always always
    GPU mip-map generation never GL 4.3 feature level 11.0

  • r_depthFade <0|1> (default: 1) enables depth fade rendering

    • this helps explosions, smoke, blood, etc. not having "sharp edges" when intersecting geometry

  • r_alphaToCoverage <0|1> (default: 1) enables alpha to coverage anti-aliasing

    • A2C AA can only work if MSAA is enabled (i.e. r_msaa > 1)
    • it is used to super-sample all fragments of alpha-tested surfaces

  • r_dither <0|1> (default: 0) reduces color banding artefacts by adding noise

  • r_ditherStrength <0.125 to 8.0> (default: 1.0) controls the amount of noise added by dithering

  • r_gpuMipGen <0|1> (default: 1) enables GPU-side mip-map generation

    • this helps improve map load time (less to upload) and image quality (see r_mipGenGamma, r_mipGenFilter)

  • r_mipGenGamma <1.0 to 3.0> (default: 1.8) specifies the gamma space in which the textures are stored

    • r_gpuMipGen 1 is required
    • it should be closer to 2.4 but defaults lower to not break the looks of some things (e.g. cpm3a tele)
    • while 1.0 will preserve the original engine's look, higher values will preserve contrast better

  • r_mipGenFilter <filter> (default: "L4") specified which filter window is used to generate mips

    • r_gpuMipGen 1 is required
    • the numbers specify the "radius" of the window filters
    • r_mipGenFilter L4 = Lanczos
    • r_mipGenFilter L3 = Lanczos
    • r_mipGenFilter BL = Bi-linear
    • r_mipGenFilter MN2 = Mitchell-Netravali (B = 1/3, C = 1/3)
    • r_mipGenFilter BH4 = 3-term Blackman-Harris
    • r_mipGenFilter BH3 = 3-term Blackman-Harris
    • r_mipGenFilter BH2 = 3-term Blackman-Harris
    • r_mipGenFilter T2 = Tent (1/3 2/3) -> same as the CPU version (r_gpuMipGen 0)

  • r_frameSleep <0|1|2> (default: 2) selects the frame capping mode

    • r_frameSleep 0 = forced OFF (when V-Sync is enabled)
    • r_frameSleep 1 = forced ON (when V-Sync is disabled)
    • r_frameSleep 2 = automatic (detects V-Sync status)
    • this only applies to OpenGL rendering back-ends
    • for Direct3D 11, the behavior is always equivalent to "automatic"

  • r_gl3_geoStream <0 to 4> (default: 0) sets the geometry upload strategy

    • r_gl3_geoStream 0 = decide automatically
    • r_gl3_geoStream 1 = BufferSubData
    • r_gl3_geoStream 2 = Map Unsychronized
    • r_gl3_geoStream 3 = Map Persistent when possible, else automatic
    • r_gl3_geoStream 4 = AMD pinned memory when available, else automatic
    • this only applies to the GL3 rendering back-end

  • r_d3d11_syncOffsets <0|1|2> (default: 2) synchronizes vertex buffer offsets

    • r_d3d11_syncOffsets 0 = split buffer offsets (more data transferred)
    • r_d3d11_syncOffsets 1 = sync'd buffer offsets (more API calls)
    • r_d3d11_syncOffsets 2 = decide automatically
    • this only applies to the D3D11 rendering back-end

  • r_d3d11_presentMode <0|1> (default: 0) sets the Direct3D presentation mode

    • r_d3d11_presentMode 0 = blit + discard
    • r_d3d11_presentMode 1 = flip + discard when available, else blit + discard
    • about the 'flip + discard' presentation mode:
    • Windows 10 is required and tearing must be supported
    • it should help performance by sharing buffer data (less copying)
    • it allows for V-Sync to be disabled with variable refresh rate displays
    • performance issues were observed on some machines, so it's off by default

  • r_khr_debug <0|1|2> (default: 2) decides whether to create an OpenGL debug context

    • r_khr_debug 0 = forced OFF
    • r_khr_debug 1 = forced ON
    • r_khr_debug 2 = only in debug builds

  • r_rtColorFormat <0|1|2> (default: 0) selects the color format for render targets

    • r_rtColorFormat 0 = R8G8B8A8
    • r_rtColorFormat 1 = R10G10B10A2
    • r_rtColorFormat 2 = R16G16B16A16
    • higher color bit depths help reduce color banding

  • r_lightmapGreyscale <0.0 to 1.0> (default: 0) controls the level of color desaturation in the lightmap

    • r_lightmapGreyscale 0 = full color (nothing done)
    • r_lightmapGreyscale 1 = completely monochrome

  • r_transpSort <0|1> (default: 0) to select the transparency sorting mode

    • r_transpSort 0 = sort dynamic transparent surfaces
    • r_transpSort 1 = sort all transparent surfaces
    • the drawback of r_transpSort 1 is that gameplay legibility might suffer and results might look
    • quite inconsistent based on the view angles and surface dimensions
    • example: dropped items in the cpm18r acid with cg_simpleItems 1

  • global shader directive q3map_cnq3_depthFade enables depth fade rendering

    • scale is the depth range over which to fade the currently rendered surface
    • bias is a value added to the currently rendered surface's depth, positive goes towards the camera
    • whether this is enabled is decided by r_depthFade (must be 1) and r_backend (can't be GL2)
    • depth writes must be disabled in all stages (no shader stage can use the depthWrite directive)
    • the blend mode (blendFunc) must be the same for all stages and must be one of:
    • (one, one) (src_alpha, one) (one_minus_src_alpha, one) (dst_color, zero) (zero, src_color)
    • (src_alpha, one_minus_src_alpha) (one, one_minus_src_alpha)

  • r_mapBrightness now works on q3map2's external lightmap atlas images

  • the renderer can now batch surfaces with different (but sufficiently similar) shaders

    • this new optimization works with the output of "Frozen Sand Particle Studio"

  • /modellist /skinlist /imagelist /shaderlist can now filter results with pattern matching

Changes

  • improved the map download initiation logic and updated the CNQ3 map server addresses

  • increased the small zone allocator's size from 512 KB to 4 MB (1 MB in the dedicated executable)

  • dynamic lights now use a softer fall-off curve to create softer edges

  • reverted an old change to the Escape key handling for cgame

    • this fixes demo playback behavior in CPMA, which enables custom binds after pressing Escape once

  • searching for valid sample counts for MSAA in GL2 and GL3 instead of failing right away

  • switched away from stb_image for .tga file loading and now using an updated version of the old code

    • this lowers the chances of a fatal error from the zone memory allocator when reading large images

  • SSE2 instruction set support is now required

  • removed FreeType 2 and the unused R_REGISTERFONT syscalls that were using it

  • removed light flares and all related cvars (r_flare*)

  • r_textureMode <filter> (default: "best") is now latched and only supports 2 modes:

    • r_textureMode GL_NEAREST = LEGO(R) mode (nearest-neighbor filtering for a subset of surfaces)
    • r_textureMode anything else = normal mode (the engine picks the most appropriate filters)

  • removed the following CVars: r_measureOverdraw, r_nobind, r_showImages

  • r_speeds is no longer cheat-protected

    • it now also reports more precise timings, V-Sync status, GPU time and the selected back-end

  • using the roundss SSE4.1 instruction for inlining floor and ceil syscalls when possible

  • improved face and grid processing performance with SSE2

  • r_lightmap is now latched again

    • the implementation is now more correct and will apply to all surfaces referencing a lightmap
    • it now also supports surfaces referencing q3map2's "external lightmap" atlas images

  • r_fullbright is now latched again

    • it now also supports surfaces referencing q3map2's "external lightmap" atlas images
    • r_mapBrightness is now used by r_fullbright for lightmapped surfaces when r_vertexLight is 0

  • negative r_swapInterval values will request adaptive V-Sync when using an OpenGL back-end

Fixes

  • the engine could crash due to render command list overflows

  • made more keys bindable on Linux/FreeBSD and /bind now accepts uppercase hex digits

  • removing CVars through unset/cvar_trim/cvar_restart could leave data in an invalid state

  • r_detailTextures 0 would mess up shaders where "detail" was used in a stage that isn't the last one

  • dynamic lights could incorrectly stop applying to moving entities (e.g. cpm25 elevator)

  • crash due to lack of memory for the sound system

  • incorrect shader state when loading a map with no lighting

  • fog-only non-sky shaders were incorrectly handled as sky shaders, which broke fog on many maps

  • r_monitor on Linux/FreeBSD is now 0-based and the value doesn't change incorrectly on its own anymore

  • when a latched cvar's value is set to its default, the cvar is no longer marked as changed

  • broken face normals get fixed up on map load, which helps with dynamic lights (e.g. industrial)

  • classes of transparent surfaces (sprites, beams, etc) now get drawn in the right order

  • dynamic lights can now affect transparent surfaces and double-sided surfaces on both sides

  • dynamic lights no longer "reveal" the diffuse texture when r_lightmap is 1

  • incorrect image flags for various things (clamp mode, mip generation, anisotropic filtering, …)

Known issues

  • OBS captures go blank after doing a video restart with the D3D11 back-end

    • Explanation: device/swap chain re-creation isn't handled by OBS' capture plug-in
    • User work-around: use the GL3 back-end
    • Development status: won't fix

  • Once the GL2 back-end is used, performance drops when switching back to the GL3 back-end

    • Config: nVidia GeForce GTX 1070 - Windows 7 SP1 x64 - drivers 430.64 and 445.87
    • User work-around: don't switch back-ends or upgrade drivers
    • Development status: can't fix (root cause unknown)

  • With GL3 and r_gpuMipGen 1, performance collapses big time

    • Config: AMD Radeon HD 6950 - Windows 10 Pro version 10.0.16299 - drivers 15.201.1151.1008
    • Explanation: confirmed: whenever glTexStorage2D is called (even when using that function is the only change)
    • User work-around: use D3D11 or upgrade drivers
    • Development status: won't fix

  • With GL3 and r_gpuMipGen 1, the GPU-generated mips are corrupted

    • Config: AMD Radeon R7 360 - Windows 7 SP1 x64 - drivers 14.502.0.0
    • Explanation: not confirmed: broken barrier implementation?
    • User work-around: use D3D11 or upgrade drivers
    • Development status: can't fix (root cause unknown)

  • With GL3, depth faded surfaces break anti-aliasing

    • Explanation: can't fetch from the right sample index because render buffers are used as FBO attachments instead of textures
    • User work-around: use D3D11
    • Development status: will get fixed

  • Strong gamma correction breaks anti-aliasing

    • Explanation: MSAA resolve happens before the post-process pass
    • Development status: will get fixed for D3D11 and GL3

  • D3D11 full-screen toggles with the 32-bit client freeze myT's PC :-(

    • Config: nVidia GeForce GTX 1070 - Windows 7 SP1 x64 - drivers 445.87
    • User work-around: use the 64-bit client
    • Development status: can't fix (root cause unknown)