qf/quakeforge
0
0
Fork 0
mirror of https://git.code.sf.net/p/quake/quakeforge synced 2025-02-25 13:11:00 +00:00
Code Issues Projects Releases Packages Wiki Activity
quakeforge/libs/video/renderer/vulkan/vulkan_lighting.c

2886 lines
84 KiB
C
Raw Normal View History

[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
/*
vulkan_lighting.c
Vulkan lighting pass pipeline
Copyright (C) 2021 Bill Currie <bill@taniwha.org>
Author: Bill Currie <bill@taniwha.org>
Date: 2021/2/23
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
as published by the Free Software Foundation; either version 2
of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to:
Free Software Foundation, Inc.
59 Temple Place - Suite 330
Boston, MA 02111-1307, USA
*/
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
#ifdef HAVE_STRING_H
# include <string.h>
#endif
#ifdef HAVE_STRINGS_H
# include <strings.h>
#endif
#include <stdlib.h>
#include "qfalloca.h"
[vulkan] Add some more debug stack contexts
2021-12-02 13:48:50 +00:00
#include "QF/cvar.h"
[vulkan] Load the map's lights into an array It's not used yet as work needs to be done to better support generic entities, but this is the next step to real-time lighting (though, to be honest, I expect it will be too slow to be usable).
2021-03-03 09:14:16 +00:00
#include "QF/dstring.h"
[vulkan] Resurrect the shadow map resource creation However, this time it doesn't modify the light array when it sorts the lights by size since the lights are now located before the renderer gets to see them, and having the fix up the light leafs array would be too painful (and probably the completely wrong thing to do anyway: the light array should be treated as constant by the renderer). 1.6GB of memory for gmsp3v2's lights (a little better than marcher: more smaller lights?). For reference: gmsp3v2: shadow maps: 8330 layers in 29 images: 1647706112 marcher: shadow maps: 2440 layers in 11 images: 2358575104
2022-05-10 02:15:52 +00:00
#include "QF/heapsort.h"
[util] Rename qfplist.[ch] The name is a hold-over from before the current quakeforge tree and the QF include directory.
2021-03-21 07:13:03 +00:00
#include "QF/plist.h"
[vulkan] Load the map's lights into an array It's not used yet as work needs to be done to better support generic entities, but this is the next step to real-time lighting (though, to be honest, I expect it will be too slow to be usable).
2021-03-03 09:14:16 +00:00
#include "QF/progs.h"
#include "QF/script.h"
[model] Remove 64k limit on visible leafs Modern maps can have many more leafs (eg, ad_tears has 98983 leafs). Using set_t makes dynamic leaf counts easy to support and the code much easier to read (though set_is_member and the iterators are a little slower). The main thing to watch out for is the novis set and the set returned by Mod_LeafPVS never shrink, and may have excess elements (ie, indicate that nonexistent leafs are visible).
2021-07-26 02:15:51 +00:00
#include "QF/set.h"
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
#include "QF/sys.h"
[vulkan] Implement deferred lighting for dlights Static lights are yet to come (so the screen is black most of the time), but dynamic lights work very nicely (and look very good) despite the falloff being incorrect.
2021-02-25 06:51:54 +00:00
#include "QF/va.h"
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
[vulkan] Move non-specific lighting code out of Vulkan The parsing of light data from maps is now in the client library, and basic light management is in scene. Putting the light loading code into the Vulkan renderer was a mistake I've wanted to correct for a while. The client code still needs a bit of cleanup, but the basics are working nicely.
2022-05-05 05:58:47 +00:00
#include "QF/scene/scene.h"
[vulkan] Show surfaces for selected light leaf This has resulted in some rather interesting information: it seems the surfaces (and thus, presumably bounding boxes) for leafs have little to do with the actual leaf node's volume.
2023-08-08 16:23:13 +00:00
#include "QF/ui/imui.h"
#define IMUI_context imui_ctx
[vulkan] Move non-specific lighting code out of Vulkan The parsing of light data from maps is now in the client library, and basic light management is in scene. Putting the light loading code into the Vulkan renderer was a mistake I've wanted to correct for a while. The client code still needs a bit of cleanup, but the basics are working nicely.
2022-05-05 05:58:47 +00:00
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
#include "QF/Vulkan/qf_bsp.h"
[vulkan] Add light count display Enabled by 'developer lighting'. It was good for confirming that the lights in ad_e1m1 (Doom Hangar 16) were actually being output (over 600 of them sometimes, ouch). Turned out to be the color scale ambiguity.
2022-05-27 01:19:15 +00:00
#include "QF/Vulkan/qf_draw.h"
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
#include "QF/Vulkan/qf_lighting.h"
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
#include "QF/Vulkan/qf_matrices.h"
[vulkan] Start work on shadows This gets the shaders needed for creating shadow maps, and the changes to the lighting pipeline for binding the shadow maps, but no generation or reading is done yet. It feels like parts of various systems are getting a little big for their britches and I need to do an audit of various things.
2021-04-24 01:40:39 +00:00
#include "QF/Vulkan/qf_texture.h"
[vulkan] Use OIT to visualize the light hulls Using the translucency pass made it easy to have depth-tested translucent "solid" light volumes instead of always visible lines (which are still an option as that's useful too). Most importantly, being able to see the surfaces helped no end in figuring out that my hulls were created with counter-clockwise windings instead of quake's usual clockwise windings and thus my hulls were being rendered inside-out in the occlusion pass.
2023-12-17 01:56:59 +00:00
#include "QF/Vulkan/qf_translucent.h"
[vulkan] Clean up buffer barriers a bit This even fixes a couple of minor issues that snuck past validation.
2021-04-24 06:47:31 +00:00
#include "QF/Vulkan/barrier.h"
[vulkan] Implement deferred lighting for dlights Static lights are yet to come (so the screen is black most of the time), but dynamic lights work very nicely (and look very good) despite the falloff being incorrect.
2021-02-25 06:51:54 +00:00
#include "QF/Vulkan/buffer.h"
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
#include "QF/Vulkan/debug.h"
#include "QF/Vulkan/device.h"
[vulkan] Use the descriptor set manager This cleans up a lot of ugly code that I always thought was rather dumb.
2023-06-24 12:42:46 +00:00
#include "QF/Vulkan/dsmanager.h"
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
#include "QF/Vulkan/image.h"
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
#include "QF/Vulkan/instance.h"
#include "QF/Vulkan/projection.h"
[vulkan] Register task functions for the pipelines They're currently just stubs, but this gets the render info loading working without any errors. The next step is to connect up pipelines and create the image resources, then implementing the task functions will have meaning.
2023-02-13 15:55:22 +00:00
#include "QF/Vulkan/render.h"
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
#include "QF/Vulkan/resource.h"
[vulkan] Implement deferred lighting for dlights Static lights are yet to come (so the screen is black most of the time), but dynamic lights work very nicely (and look very good) despite the falloff being incorrect.
2021-02-25 06:51:54 +00:00
#include "QF/Vulkan/staging.h"
#include "compat.h"
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
#include "r_internal.h"
#include "vid_vulkan.h"
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
#include "vkparse.h"
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
[vulkan] Change shadow quanta to 128 pixels Interestingly, this caused a reduction in memory use for some maps (but did increase marcher's again, but not as much as the bogus rounding did). The idea was to use sparse bindings to remap shadow map layers, but it turns out sparse bindings are insanely slow (beyond unusable). However, the reduction in the number of shadow map images seems to be worth it.
2023-12-14 00:50:49 +00:00
#define shadow_quanta 128
[vulkan] Make near and far clip explicit parameters This improves the projection API in that near clip is a parameter rather than being taken directly from the cvar, and a far clip (ie, finite far plane) version is available (necessary for cascaded shadow maps as it's rather hard to fit a box to an infinite frustum). Also, the orthographic projection matrix is now reversed as per the perspective matrix (and the code tidied up a little), and a version that takes min and max vectors is available.
2023-08-13 08:25:17 +00:00
#define lnearclip 4
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
#define num_cascade 4
[vulkan] Create render passes for up to 32 views Actually, only 29 are used because nvidia's drivers segfault when there are more than 29 views (regardless of the exact bit pattern in the view mask). This will allow rendering shadow maps in large batches, which should make for better GPU utilization.
2023-12-14 00:56:53 +00:00
#define max_views 29 // FIXME should be 32 (or really, maxMultiviewViewCount,
// but there are other problems there), but nvidia's
// drivers segfault for > 29
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
static vec4f_t ref_direction = { 1, 0, 0, 0 };
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
#define ico_verts 12
#define cone_verts 7
static int ico_inds[] = {
0, 4, 6, 9, 2, 8, 4, -1,
3, 1, 10, 5, 7, 11, 1, -1,
1, 11, 6, 4, 10, -1,
9, 6, 11, 7, 2, -1,
5, 10, 8, 2, 7, -1,
4, 8, 10,
};
#define num_ico_inds (sizeof (ico_inds) / sizeof (ico_inds[0]))
static int cone_inds[] = {
0, 1, 2, 3, 4, 5, 6, 1, -1,
1, 6, 5, 4, 3, 2,
};
#define num_cone_inds (sizeof (cone_inds) / sizeof (cone_inds[0]))
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
#define dynlight_max 32
static int dynlight_size;
static cvar_t dynlight_size_cvar = {
.name = "dynlight_size",
.description =
"Effective radius of dynamic light shadow maps. Needs map reload to "
"take effect",
[vulkan] Change shadow quanta to 128 pixels Interestingly, this caused a reduction in memory use for some maps (but did increase marcher's again, but not as much as the bogus rounding did). The idea was to use sparse bindings to remap shadow map layers, but it turns out sparse bindings are insanely slow (beyond unusable). However, the reduction in the number of shadow map images seems to be worth it.
2023-12-14 00:50:49 +00:00
.default_value = "256",
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
.flags = CVAR_NONE,
.value = { .type = &cexpr_int, .value = &dynlight_size },
};
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
static const light_t *
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
get_light (entity_t ent)
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
{
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
return Ent_GetComponent (ent.id, ent.base + scene_light, ent.reg);
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
}
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
static const dlight_t *
get_dynlight (entity_t ent)
{
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
return Ent_GetComponent (ent.id, ent.base + scene_dynlight, ent.reg);
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
}
static bool
has_dynlight (entity_t ent)
{
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
return Ent_HasComponent (ent.id, ent.base + scene_dynlight, ent.reg);
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
}
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
static uint32_t
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
get_lightstyle (entity_t ent)
{
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
return *(uint32_t *) Ent_GetComponent (ent.id, ent.base + scene_lightstyle, ent.reg);
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
}
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
static uint32_t
get_lightleaf (entity_t ent)
{
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
return *(uint32_t *) Ent_GetComponent (ent.id, ent.base + scene_lightleaf, ent.reg);
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
}
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
static uint32_t
get_lightid (entity_t ent)
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
{
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
return *(uint32_t *) Ent_GetComponent (ent.id, ent.base + scene_lightid, ent.reg);
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
}
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
static void
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
set_lightid (entity_t ent, uint32_t id)
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
{
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
Ent_SetComponent (ent.id, ent.base + scene_lightid, ent.reg, &id);
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
}
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
static void
[vulkan] Rename bsp aux pass to shadow Much better name.
2023-08-08 09:04:58 +00:00
lighting_setup_shadow (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
{
[build] Add support for building with Tracy Tracy is a frame profiler: https://github.com/wolfpld/tracy This uses Tracy's C API to instrument the code (already added in several places). It turns out there is something very weird with the fence behavior between the staging buffers and render commands as the inter-frame delay occurs in a very strangle place (in the draw code's packet acquisition rather than the fence waiter that's there for that purpose). I suspect some tangled dependencies.
2023-11-28 06:54:55 +00:00
qfZoneNamed (zone, true);
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto lctx = ctx->lighting_context;
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
auto lframe = &lctx->frames.a[ctx->curFrame];
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
if (!lctx->ldata) {
return;
}
[vulkan] Show surfaces for selected light leaf This has resulted in some rather interesting information: it seems the surfaces (and thus, presumably bounding boxes) for leafs have little to do with the actual leaf node's volume.
2023-08-08 16:23:13 +00:00
auto pass = Vulkan_Bsp_GetPass (ctx, QFV_bspShadow);
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
auto brush = pass->brush;
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
[vulkan] Implement direction shadow lighting Finally, even suns cast shadows :)
2023-08-13 17:33:19 +00:00
if (brush->submodels) {
lctx->world_mins = loadvec3f (brush->submodels[0].mins);
lctx->world_maxs = loadvec3f (brush->submodels[0].maxs);
} else {
// FIXME better bounds
lctx->world_mins = (vec4f_t) { -512, -512, -512, 0 };
lctx->world_maxs = (vec4f_t) { 512, 512, 512, 0 };
}
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
set_t leafs = SET_STATIC_INIT (brush->modleafs, alloca);
set_empty (&leafs);
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
for (int i = 0; i < ST_COUNT; i++) {
auto q = lframe->light_queue[i];
for (uint32_t j = 0; j < q.count; j++) {
uint32_t leafnum = lframe->id_radius[q.start + j].leafnum;
if (leafnum != ~0u) {
set_add (&leafs, leafnum);
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
}
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
}
}
set_t pvs = SET_STATIC_INIT (brush->visleafs, alloca);
auto iter = set_first (&leafs);
if (!iter) {
return;
}
if (iter->element == 0) {
set_assign (&pvs, lctx->ldata->sun_pvs);
} else {
Mod_LeafPVS_set (brush->leafs + iter->element, brush, 0, &pvs);
}
for (iter = set_next (iter); iter; iter = set_next (iter)) {
Mod_LeafPVS_mix (brush->leafs + iter->element, brush, 0, &pvs);
}
visstate_t visstate = {
.node_visframes = pass->node_frames,
.leaf_visframes = pass->leaf_frames,
.face_visframes = pass->face_frames,
.visframecount = pass->vis_frame,
.brush = pass->brush,
};
R_MarkLeavesPVS (&visstate, &pvs);
pass->vis_frame = visstate.visframecount;
}
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
static VkFramebuffer
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
create_framebuffer (vulkan_ctx_t *ctx, int size,
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
VkImageView view, VkRenderPass renderpass)
{
auto device = ctx->device;
auto dfunc = device->funcs;
VkFramebuffer framebuffer;
dfunc->vkCreateFramebuffer (device->dev,
&(VkFramebufferCreateInfo) {
.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
.renderPass = renderpass,
.attachmentCount = 1,
.pAttachments = &view,
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
.width = size,
.height = size,
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
.layers = 1,
}, 0, &framebuffer);
return framebuffer;
}
static void
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
copy_maps (uint32_t start, uint32_t count, int stage_id,
lightingframe_t *lframe, lightingctx_t *lctx, vulkan_ctx_t *ctx,
qfv_taskctx_t *taskctx)
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
{
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
qfZoneScoped (true);
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
auto device = ctx->device;
auto dfunc = device->funcs;
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
int num_regions = 0;
int num_copies = 0;
uint32_t last = ~0u;
for (uint32_t j = 0; j < count; j++) {
auto tgt = lframe->stage_targets[start + j];
// if the map id is different or if the layers aren't sequential
if (tgt != last + 0x20) {
num_regions++;
if ((tgt & 0x1f) != (last & 0x1f)) {
num_copies++;
}
}
last = tgt;
}
auto ib = imageBarriers[qfv_LT_TransferDst_to_ShaderReadOnly];
ib.barrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
ib.barrier.subresourceRange.levelCount = VK_REMAINING_MIP_LEVELS;
ib.barrier.subresourceRange.layerCount = 0;
VkImageMemoryBarrier barriers[num_regions];
VkImageCopy2 regions[num_regions];
VkCopyImageInfo2 copies[num_copies];
num_regions = 0;
num_copies = 0;
last = ~0u;
for (uint32_t j = 0; j < count; j++) {
auto tgt = lframe->stage_targets[start + j];
int ind = num_regions - 1;
// if the map id is different or if the layers aren't sequential
if (tgt != last + 0x20) {
int cpind = num_copies - 1;
if ((tgt & 0x1f) != (last & 0x1f)) {
cpind = num_copies++;
copies[cpind] = (VkCopyImageInfo2) {
.sType = VK_STRUCTURE_TYPE_COPY_IMAGE_INFO_2,
.srcImage = lctx->stage_images[stage_id],
.srcImageLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
.dstImage = lctx->map_images[tgt & 0x1f],
.dstImageLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
.pRegions = &regions[num_regions],
};
}
copies[cpind].regionCount++;
ind = num_regions++;
barriers[ind] = ib.barrier;
barriers[ind].subresourceRange.baseArrayLayer = tgt >> 5;
barriers[ind].image = lctx->map_images[tgt & 0x1f];
regions[ind] = (VkImageCopy2) {
.sType = VK_STRUCTURE_TYPE_IMAGE_COPY_2,
.srcSubresource = {
.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
.baseArrayLayer = j,
},
.dstSubresource = {
.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
.baseArrayLayer = tgt >> 5,
},
.extent = {
.width = (stage_id + 1) * shadow_quanta,
.height = (stage_id + 1) * shadow_quanta,
.depth = 1,
},
};
}
barriers[ind].subresourceRange.layerCount++;
regions[ind].srcSubresource.layerCount++;
regions[ind].dstSubresource.layerCount++;
last = tgt;
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
auto cmd = QFV_GetCmdBuffer (ctx, false);
dfunc->vkBeginCommandBuffer (cmd, &(VkCommandBufferBeginInfo) {
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
});
{
qftVkScopedZoneC (taskctx->frame->qftVkCtx, cmd, "copy stage",
0xa0c060);
for (int i = 0; i < num_copies; i++) {
dfunc->vkCmdCopyImage2 (cmd, &copies[i]);
}
dfunc->vkCmdPipelineBarrier (cmd, ib.srcStages, ib.dstStages,
0, 0, 0, 0, 0, num_regions, barriers);
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
dfunc->vkEndCommandBuffer (cmd);
QFV_AppendCmdBuffer (ctx, cmd);
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
}
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
static void
lighting_draw_shadow_maps (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
{
[build] Add support for building with Tracy Tracy is a frame profiler: https://github.com/wolfpld/tracy This uses Tracy's C API to instrument the code (already added in several places). It turns out there is something very weird with the fence behavior between the staging buffers and render commands as the inter-frame delay occurs in a very strangle place (in the draw code's packet acquisition rather than the fence waiter that's there for that purpose). I suspect some tangled dependencies.
2023-11-28 06:54:55 +00:00
qfZoneNamed (zone, true);
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto lctx = ctx->lighting_context;
auto shadow = QFV_GetStep (params[0], ctx->render_context->job);
auto render = shadow->render;
auto lframe = &lctx->frames.a[ctx->curFrame];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
if (!lctx->num_maps) {
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
return;
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
uint32_t id_base = 0;
for (int i = 0; i < LIGHTING_STAGES; i++) {
auto queue = &lframe->stage_queue[i];
int count;
for (int remaining = queue->count; remaining > 0; remaining -= count) {
count = min (remaining, max_views);
int rpind = count - 1;
auto renderpass = &render->renderpasses[rpind];
auto fbuffer = lctx->stage_framebuffers[rpind][i];
uint32_t size = (i + 1) * shadow_quanta;
auto bi = &renderpass->beginInfo;
if (!fbuffer) {
auto view = lctx->stage_views[i];
fbuffer = create_framebuffer (ctx, size, view, bi->renderPass);
lctx->stage_framebuffers[rpind][i] = fbuffer;
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
bi->framebuffer = fbuffer;
QFV_RunRenderPass (ctx, renderpass, size, size, &id_base);
bi->framebuffer = 0;
copy_maps (id_base, count, i, lframe, lctx, ctx, taskctx);
id_base += count;
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
}
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
}
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
}
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
[vulkan] Clean up make_id() Passing the control struct and a flag for style/nostyle makes the calls much easier to read.
2023-12-17 04:14:43 +00:00
typedef enum : uint32_t {
style_enable,
style_disable = 0x80000000,
} style_e;
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
static uint32_t
[vulkan] Clean up make_id() Passing the control struct and a flag for style/nostyle makes the calls much easier to read.
2023-12-17 04:14:43 +00:00
make_id (const light_control_t *cont, style_e style)
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
{
[vulkan] Clean up make_id() Passing the control struct and a flag for style/nostyle makes the calls much easier to read.
2023-12-17 04:14:43 +00:00
uint32_t matrix_index = cont->matrix_id;
uint32_t map_index = cont->map_index;
uint32_t layer = cont->layer;
uint32_t type = cont->mode;
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
if (type == ST_CUBE) {
[vulkan] Increase matrix id bits I had missed that upping max lights to 2048 meant that up to 12288 matrices are needed for all the possible lights. This made it so the light type could not be encoded in id_data, but the shaders never used it anyway. This leaves one bit free.
2023-12-13 03:20:10 +00:00
// on the GPU, layer is the cube layer, and one cube layer is 6
// flat image layers
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
layer /= 6;
}
[vulkan] Increase matrix id bits I had missed that upping max lights to 2048 meant that up to 12288 matrices are needed for all the possible lights. This made it so the light type could not be encoded in id_data, but the shaders never used it anyway. This leaves one bit free.
2023-12-13 03:20:10 +00:00
return ((matrix_index & 0x3fff) << 0)
| ((map_index & 0x1f) << 14)
[vulkan] Clean up make_id() Passing the control struct and a flag for style/nostyle makes the calls much easier to read.
2023-12-17 04:14:43 +00:00
| ((layer & 0x7ff) << 19)
| style;
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
}
static void
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
cube_mats (mat4f_t *mat, vec4f_t position)
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
{
mat4f_t view;
mat4fidentity (view);
view[3] = -position;
view[3][3] = 1;
mat4f_t proj;
[vulkan] Make near and far clip explicit parameters This improves the projection API in that near clip is a parameter rather than being taken directly from the cvar, and a far clip (ie, finite far plane) version is available (necessary for cascaded shadow maps as it's rather hard to fit a box to an infinite frustum). Also, the orthographic projection matrix is now reversed as per the perspective matrix (and the code tidied up a little), and a version that takes min and max vectors is available.
2023-08-13 08:25:17 +00:00
QFV_PerspectiveTan (proj, 1, 1, lnearclip);
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
for (int j = 0; j < 6; j++) {
mat4f_t side_view;
mat4f_t rotinv;
mat4ftranspose (rotinv, qfv_box_rotations[j]);
mmulf (side_view, rotinv, view);
mmulf (side_view, qfv_z_up, side_view);
mmulf (mat[j], proj, side_view);
}
}
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
static void
frustum_corners (vec4f_t corners[8], float minz, float maxz,
const mat4f_t invproj)
{
for (int i = 0; i < 8; i++) {
vec4f_t p = {
(i & 1) ? 1 : -1,
(i & 2) ? 1 : -1,
(i & 4) ? maxz : minz,
1
};
p = mvmulf (invproj, p);
corners[i] = p / p[3];
}
}
static vec4f_t
vec_select (vec4i_t c, vec4f_t a, vec4f_t b)
{
return (vec4f_t) ((c & (vec4i_t) a) | (~c & (vec4i_t) b));
}
static void
transform_corners (vec4f_t *mins, vec4f_t *maxs, const vec4f_t corners[8],
const mat4f_t lightview)
{
*mins = (vec4f_t) { INFINITY, INFINITY, INFINITY, INFINITY };
*maxs = (vec4f_t) { -INFINITY, -INFINITY, -INFINITY, -INFINITY };
for (int i = 0; i < 8; i++) {
vec4f_t p = mvmulf (lightview, corners[i]);
vec4i_t tmin = p <= *mins;
vec4i_t tmax = p >= *maxs;
*mins = vec_select (tmin, p, *mins);
*maxs = vec_select (tmax, p, *maxs);
}
}
static void
cascade_mats (mat4f_t *mat, vec4f_t position, vulkan_ctx_t *ctx)
{
auto mctx = ctx->matrix_context;
auto lctx = ctx->lighting_context;
mat4f_t invproj;
QFV_InversePerspectiveTanFar (invproj, mctx->fov_x, mctx->fov_y,
r_nearclip, r_farclip);
mat4f_t inv_z_up;
mat4ftranspose (inv_z_up, qfv_z_up);
mmulf (invproj, inv_z_up, invproj);
mmulf (invproj, r_refdef.camera, invproj);
mat4f_t lightview;
// position points towards the light, but the view needs to be from the
// light.
mat4fquat (lightview, qrotf (-position, ref_direction));
// Find the world corner closest to the light in order to ensure the
// entire world between the focal point and the light is included in the
// depth range
vec4i_t d = position >= (vec4f_t) {0, 0, 0, 0};
vec4f_t lightcorner = vec_select (d, lctx->world_maxs, lctx->world_mins);
// assumes position is a unit vector (which it will be for directional
// lights loaded from quake maps)
float corner_dist = dotf (lightcorner-r_refdef.camera[3], position)[0];
// Pre-swizzle the light view so it's done only once (and so the
// frustum bounds get swizzled correctly for creating the orthographic
// projection matrix).
// Also, no need (actually, undesirable) to include any translation in
// the light view matrix as the orthographic matrix setup includes
// translation based on the bounds.
mmulf (lightview, qfv_z_up, lightview);
vec2f_t z_range[] = {
{ r_nearclip / 32, 1 },
{ r_nearclip / 256, r_nearclip / 32 },
{ r_nearclip / 1024, r_nearclip / 256 },
{ 0, r_nearclip / 1024 },
};
for (int i = 0; i < num_cascade; i++) {
vec4f_t corners[8];
vec4f_t fmin, fmax;
frustum_corners (corners, z_range[i][0], z_range[i][1], invproj);
transform_corners (&fmin, &fmax, corners, lightview);
// ensure evertything between the light and the far frustum corner
// is in the depth range
fmin[2] = min(fmin[2], -corner_dist);
QFV_OrthographicV (mat[i], fmin, fmax);
mmulf (mat[i], mat[i], lightview);
}
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
static uint16_t
make_target (uint16_t map_index, uint16_t layer)
{
return (map_index & 0x1f) | ((layer & 0x7ff) << 5);
}
static void
enqueue_map (uint32_t *ids, lightingframe_t *lframe, light_control_t *r)
{
auto q = &lframe->stage_queue[r->stage_index];
auto tgt = lframe->stage_targets;
for (uint32_t i = 0; i < r->numLayers; i++) {
ids[q->start + q->count + i] = r->matrix_id + i;
tgt[q->start + q->count + i] = make_target (r->map_index, r->layer + i);
}
q->count += r->numLayers;
}
static void
transition_shadow_targets (lightingframe_t *lframe, vulkan_ctx_t *ctx)
{
auto device = ctx->device;
auto dfunc = device->funcs;
auto lctx = ctx->lighting_context;
int num_barriers = 0;
uint32_t last = ~0u;
for (int i = 0; i < LIGHTING_STAGES; i++) {
auto q = lframe->stage_queue[i];
for (uint32_t j = 0; j < q.count; j++) {
auto tgt = lframe->stage_targets[q.start + j];
// if the map id is different or if the layers aren't sequential
if (tgt != last + 0x20) {
num_barriers++;
}
last = tgt;
}
}
auto ib = imageBarriers[qfv_LT_ShaderReadOnly_to_TransferDst];
ib.barrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
ib.barrier.subresourceRange.levelCount = VK_REMAINING_MIP_LEVELS;
ib.barrier.subresourceRange.layerCount = 0;
VkImageMemoryBarrier barriers[num_barriers];
num_barriers = 0;
last = ~0u;
for (int i = 0; i < LIGHTING_STAGES; i++) {
auto q = lframe->stage_queue[i];
for (uint32_t j = 0; j < q.count; j++) {
auto tgt = lframe->stage_targets[q.start + j];
int ind = num_barriers - 1;
// if the map id is different or if the layers aren't sequential
if (tgt != last + 0x20) {
barriers[num_barriers++] = ib.barrier;
ind = num_barriers - 1;
barriers[ind].subresourceRange.baseArrayLayer = tgt >> 5;
barriers[ind].image = lctx->map_images[tgt & 0x1f];
}
barriers[ind].subresourceRange.layerCount++;
last = tgt;
}
}
auto cmd = QFV_GetCmdBuffer (ctx, false);
dfunc->vkBeginCommandBuffer (cmd, &(VkCommandBufferBeginInfo) {
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
});
dfunc->vkCmdPipelineBarrier (cmd, ib.srcStages, ib.dstStages,
0, 0, 0, 0, 0, num_barriers, barriers);
dfunc->vkEndCommandBuffer (cmd);
QFV_AppendCmdBuffer (ctx, cmd);
}
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
static float
light_radius (const light_t *l)
{
return l->attenuation[3] > 0 ? 1 / l->attenuation[3]
: l->attenuation[0] > 0 ? sqrt(abs(l->color[3]/l->attenuation[0]))
: l->attenuation[1] > 0 ? abs(l->color[3]/l->attenuation[1])
// FIXME ambient lights. not right, but at least it will render
: sqrt(abs(l->color[3]));
}
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
static void
lighting_update_lights (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
{
[build] Add support for building with Tracy Tracy is a frame profiler: https://github.com/wolfpld/tracy This uses Tracy's C API to instrument the code (already added in several places). It turns out there is something very weird with the fence behavior between the staging buffers and render commands as the inter-frame delay occurs in a very strangle place (in the draw code's packet acquisition rather than the fence waiter that's there for that purpose). I suspect some tangled dependencies.
2023-11-28 06:54:55 +00:00
qfZoneNamed (zone, true);
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto lctx = ctx->lighting_context;
auto lframe = &lctx->frames.a[ctx->curFrame];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
memset (lframe->light_queue, 0, sizeof (lframe->light_queue));
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
if (!lctx->scene || !lctx->scene->lights) {
return;
}
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto bb = &bufferBarriers[qfv_BB_TransferWrite_to_UniformRead];
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
uint32_t light_ids[ST_COUNT][MaxLights];
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
float light_radii[ST_COUNT][MaxLights];
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
uint32_t light_leafs[ST_COUNT][MaxLights];
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
vec4f_t light_positions[ST_COUNT][MaxLights];
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
entity_t entids[ST_COUNT][MaxLights];
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
uint32_t light_count = 0;
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
auto queue = lframe->light_queue;
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
entity_t dynamic_light_entities[MaxLights];
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
uint32_t dynamic_light_leafs[MaxLights];
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
const dlight_t *dynamic_lights[MaxLights];
int ndlight = 0;
auto entqueue = r_ent_queue; //FIXME fetch from scene
for (size_t i = 0; i < entqueue->ent_queues[mod_light].size; i++) {
entity_t ent = entqueue->ent_queues[mod_light].a[i];
if (has_dynlight (ent)) {
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
dynamic_light_entities[ndlight] = ent;
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
dynamic_light_leafs[ndlight] = get_lightleaf (ent);
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
dynamic_lights[ndlight] = get_dynlight (ent);
ndlight++;
continue;
}
auto ls = get_lightstyle (ent);
if (!d_lightstylevalue[ls]) {
continue;
}
light_count++;
uint32_t id = lctx->light_control.a[get_lightid (ent)].light_id;
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
auto r = &lctx->light_control.a[id];
int mode = r->mode;
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
auto light = get_light (ent);
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
uint32_t ind = queue[mode].count++;
light_ids[mode][ind] = id;
light_radii[mode][ind] = light_radius (light);
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
light_leafs[mode][ind] = get_lightleaf (ent);
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
light_positions[mode][ind] = light->position;
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
entids[mode][ind] = ent;
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
}
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
[vulkan] Add some comments to lighting code
2023-12-12 12:13:14 +00:00
size_t packet_size = 0;
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
packet_size += sizeof (vec4f_t[NumStyles]);
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
if (queue[ST_CASCADE].count) {
uint32_t mat_count = queue[ST_CASCADE].count * num_cascade;
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
packet_size += sizeof (mat4f_t[mat_count]);
}
if (ndlight) {
packet_size += sizeof (mat4f_t[ndlight * 6]);
packet_size += sizeof (light_t[ndlight]);
packet_size += sizeof (qfv_light_render_t[ndlight]);
}
if (light_count) {
// light ids
packet_size += sizeof (uint32_t[light_count]);
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
// light radii
packet_size += sizeof (float[light_count]);
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
// ent ids
packet_size += sizeof (uint32_t[light_count]);
}
auto packet = QFV_PacketAcquire (ctx->staging);
byte *packet_start = QFV_PacketExtend (packet, packet_size);
byte *packet_data = packet_start;
qfv_scatter_t style_scatter = {
.srcOffset = 0,
.dstOffset = 0,
.length = sizeof (vec4f_t[NumStyles]),
};
auto styles = (vec4f_t *) packet_data;
packet_data += style_scatter.length;
for (int i = 0; i < NumStyles; i++) {
styles[i] = (vec4f_t) { 1, 1, 1, d_lightstylevalue[i] / 65536.0};
}
QFV_PacketScatterBuffer (packet, lframe->style_buffer,
1, &style_scatter, bb);
if (queue[ST_CASCADE].count) {
uint32_t mat_count = queue[ST_CASCADE].count * num_cascade;
auto mats = (mat4f_t *) packet_data;
auto base = packet_data - packet_start;
packet_data += sizeof (mat4f_t[mat_count]);
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
qfv_scatter_t scatter[queue[ST_CASCADE].count];
for (uint32_t i = 0; i < queue[ST_CASCADE].count; i++) {
auto r = &lctx->light_control.a[light_ids[ST_CASCADE][i]];
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
auto light = get_light (entids[ST_CASCADE][i]);
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
cascade_mats (&mats[i * num_cascade], light->position, ctx);
scatter[i] = (qfv_scatter_t) {
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
.srcOffset = base + sizeof (mat4f_t[i * num_cascade]),
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
.dstOffset = sizeof (mat4f_t[r->matrix_id]),
.length = sizeof (mat4f_t[num_cascade]),
};
}
QFV_PacketScatterBuffer (packet, lframe->shadowmat_buffer,
queue[ST_CASCADE].count, scatter, bb);
}
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
if (ndlight) {
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
light_count += ndlight;
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
auto mats = (mat4f_t *) packet_data;
qfv_scatter_t mat_scatter = {
.srcOffset = packet_data - packet_start,
.dstOffset = sizeof (mat4f_t[lctx->dynamic_matrix_base]),
.length = sizeof (mat4f_t[ndlight * 6]),
};
[vulkan] Fix broken dynamic light shadows They weren't rendering properly at all due to the matrix updates getting overwritten by the light data (I'd forgotten to advance the packet data pointer).
2023-12-14 14:17:14 +00:00
packet_data += mat_scatter.length;
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
for (int i = 0; i < ndlight; i++) {
cube_mats (&mats[i * 6], dynamic_lights[i]->origin);
}
QFV_PacketScatterBuffer (packet, lframe->shadowmat_buffer,
1, &mat_scatter, bb);
auto lights = (light_t *) packet_data;
qfv_scatter_t light_scatter = {
.srcOffset = packet_data - packet_start,
.dstOffset = sizeof (light_t[lctx->dynamic_base]),
.length = sizeof (light_t[ndlight]),
};
packet_data += light_scatter.length;
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
for (int i = 0; i < ndlight; i++) {
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
light_t light = {
.color = {
VectorExpand (dynamic_lights[i]->color),
// dynamic lights seem a tad faint, so 16x map lights
dynamic_lights[i]->radius / 16,
},
// dlights are local point sources
.position = { VectorExpand (dynamic_lights[i]->origin), 1 },
// full sphere, normal light (not ambient)
.direction = { 0, 0, 1, 1 },
.attenuation = { 0, 0, 1, 1/dynamic_lights[i]->radius },
};
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
uint32_t id = lctx->dynamic_base + i;
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
set_lightid (dynamic_light_entities[i], id);
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
uint32_t ind = queue[ST_CUBE].count++;
light_ids[ST_CUBE][ind] = id;
light_radii[ST_CUBE][ind] = light_radius (&light);
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
light_leafs[ST_CUBE][ind] = dynamic_light_leafs[i];
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
light_positions[ST_CUBE][ind] = light.position;
entids[ST_CUBE][ind] = dynamic_light_entities[i];
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
lights[i] = light;
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
}
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
QFV_PacketScatterBuffer (packet, lframe->light_buffer,
1, &light_scatter, bb);
auto render = (qfv_light_render_t *) packet_data;
qfv_scatter_t render_scatter = {
.srcOffset = packet_data - packet_start,
.dstOffset = sizeof (qfv_light_render_t[lctx->dynamic_base]),
.length = sizeof (qfv_light_render_t[ndlight]),
};
packet_data += render_scatter.length;
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
for (int i = 0; i < ndlight; i++) {
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
auto r = &lctx->light_control.a[lctx->dynamic_base + i];
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
render[i] = (qfv_light_render_t) {
[vulkan] Clean up make_id() Passing the control struct and a flag for style/nostyle makes the calls much easier to read.
2023-12-17 04:14:43 +00:00
.id_data = make_id(r, style_disable),
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
};
}
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
QFV_PacketScatterBuffer (packet, lframe->render_buffer,
1, &render_scatter, bb);
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
}
if (developer & SYS_lighting) {
Vulkan_Draw_String (vid.width - 32, 8,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
va (ctx->va_ctx, "%3d", light_count),
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
ctx);
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
if (light_count) {
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
for (int i = 1; i < ST_COUNT; i++) {
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
queue[i].start = queue[i - 1].start + queue[i - 1].count;
}
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
auto lids = (uint32_t *) packet_data;
qfv_scatter_t lid_scatter = {
.srcOffset = packet_data - packet_start,
.dstOffset = 0,
.length = sizeof (uint32_t[light_count]),
};
packet_data += lid_scatter.length;
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
for (int i = 0; i < ST_COUNT; i++) {
[vulkan] Implement mouse-picking for light entities Currently, only light entities get drawn to the entid buffer, and the ids are simply displayed in a window for now (not very useful yet).
2023-08-07 08:47:49 +00:00
memcpy (lids + queue[i].start, light_ids[i],
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
sizeof (uint32_t[queue[i].count]));
}
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
QFV_PacketScatterBuffer (packet, lframe->id_buffer,
1, &lid_scatter,
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
&bufferBarriers[qfv_BB_TransferWrite_to_IndexRead]);
[vulkan] Implement mouse-picking for light entities Currently, only light entities get drawn to the entid buffer, and the ids are simply displayed in a window for now (not very useful yet).
2023-08-07 08:47:49 +00:00
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
auto lradii = (float *) packet_data;
qfv_scatter_t lradius_scatter = {
.srcOffset = packet_data - packet_start,
.dstOffset = 0,
.length = sizeof (float[light_count]),
};
packet_data += lradius_scatter.length;
for (int i = 0; i < ST_COUNT; i++) {
memcpy (lradii + queue[i].start, light_radii[i],
sizeof (float[queue[i].count]));
}
QFV_PacketScatterBuffer (packet, lframe->radius_buffer,
1, &lradius_scatter,
&bufferBarriers[qfv_BB_TransferWrite_to_IndexRead]);
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
auto eids = (uint32_t *) packet_data;
qfv_scatter_t eid_scatter = {
.srcOffset = packet_data - packet_start,
.dstOffset = 0,
.length = sizeof (uint32_t[light_count]),
};
packet_data += eid_scatter.length;
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
for (int i = 0; i < ST_COUNT; i++) {
[vulkan] Implement mouse-picking for light entities Currently, only light entities get drawn to the entid buffer, and the ids are simply displayed in a window for now (not very useful yet).
2023-08-07 08:47:49 +00:00
memcpy (eids + queue[i].start, entids[i],
sizeof (uint32_t[queue[i].count]));
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
auto ir_barrier = &bufferBarriers[qfv_BB_TransferWrite_to_IndexRead];
QFV_PacketScatterBuffer (packet, lframe->entid_buffer, 1, &eid_scatter,
ir_barrier);
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
[vulkan] Handle spotlights pointing -X The problem didn't show up until the switch to float depth buffers (because they can store the resulting nan).
2023-12-19 08:56:13 +00:00
memset (lframe->id_radius, -1, MaxLights * sizeof (light_idrad_t));
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
for (int i = 0; i < ST_COUNT; i++) {
auto q = queue[i];
auto idr = &lframe->id_radius[q.start];
auto pos = &lframe->positions[q.start];
for (uint32_t j = 0; j < q.count; j++) {
idr[j] = (light_idrad_t) {
.id = light_ids[i][j],
.radius = light_radii[i][j],
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
.leafnum = light_leafs[i][j],
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
};
pos[j] = light_positions[i][j];
}
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
}
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
QFV_PacketSubmit (packet);
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
}
[vulkan] Register task functions for the pipelines They're currently just stubs, but this gets the render info loading working without any errors. The next step is to connect up pipelines and create the image resources, then implementing the task functions will have meaning.
2023-02-13 15:55:22 +00:00
static void
[vulkan] Add debug lines for light splats Disabled, but all that's needed is to uncomment the debug pipeline in the compose subpass.
2023-06-28 02:47:26 +00:00
lighting_update_descriptors (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
[vulkan] Register task functions for the pipelines They're currently just stubs, but this gets the render info loading working without any errors. The next step is to connect up pipelines and create the image resources, then implementing the task functions will have meaning.
2023-02-13 15:55:22 +00:00
{
[build] Add support for building with Tracy Tracy is a frame profiler: https://github.com/wolfpld/tracy This uses Tracy's C API to instrument the code (already added in several places). It turns out there is something very weird with the fence behavior between the staging buffers and render commands as the inter-frame delay occurs in a very strangle place (in the draw code's packet acquisition rather than the fence waiter that's there for that purpose). I suspect some tangled dependencies.
2023-11-28 06:54:55 +00:00
qfZoneNamed (zone, true);
[vulkan] Get lighting mostly working Lights seem to be good in renderdoc, but still need to get the compose subpass working.
2023-06-22 12:36:14 +00:00
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto device = ctx->device;
auto dfunc = device->funcs;
auto lctx = ctx->lighting_context;
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
auto lframe = &lctx->frames.a[ctx->curFrame];
[vulkan] Get lighting mostly working Lights seem to be good in renderdoc, but still need to get the compose subpass working.
2023-06-22 12:36:14 +00:00
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto job = ctx->render_context->job;
auto step = QFV_GetStep (params[0], job);
auto render = step->render;
auto fb = &render->active->framebuffer;
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
VkDescriptorImageInfo attachInfo[] = {
{ .imageView = fb->views[QFV_attachColor],
.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL },
{ .imageView = fb->views[QFV_attachEmission],
.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL },
{ .imageView = fb->views[QFV_attachNormal],
.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL },
{ .imageView = fb->views[QFV_attachPosition],
.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL },
};
VkWriteDescriptorSet attachWrite[] = {
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lframe->attach_set,
.dstBinding = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
.pImageInfo = &attachInfo[0], },
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lframe->attach_set,
.dstBinding = 1,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
.pImageInfo = &attachInfo[1], },
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lframe->attach_set,
.dstBinding = 2,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
.pImageInfo = &attachInfo[2], },
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lframe->attach_set,
.dstBinding = 3,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
.pImageInfo = &attachInfo[3], },
};
[vulkan] Get lighting mostly working Lights seem to be good in renderdoc, but still need to get the compose subpass working.
2023-06-22 12:36:14 +00:00
dfunc->vkUpdateDescriptorSets (device->dev,
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
LIGHTING_ATTACH_INFOS, attachWrite,
0, 0);
[vulkan] Add debug lines for light splats Disabled, but all that's needed is to uncomment the debug pipeline in the compose subpass.
2023-06-28 02:47:26 +00:00
}
static void
lighting_bind_descriptors (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
{
[build] Add support for building with Tracy Tracy is a frame profiler: https://github.com/wolfpld/tracy This uses Tracy's C API to instrument the code (already added in several places). It turns out there is something very weird with the fence behavior between the staging buffers and render commands as the inter-frame delay occurs in a very strangle place (in the draw code's packet acquisition rather than the fence waiter that's there for that purpose). I suspect some tangled dependencies.
2023-11-28 06:54:55 +00:00
qfZoneNamed (zone, true);
[vulkan] Add debug lines for light splats Disabled, but all that's needed is to uncomment the debug pipeline in the compose subpass.
2023-06-28 02:47:26 +00:00
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto device = ctx->device;
auto dfunc = device->funcs;
auto lctx = ctx->lighting_context;
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
if (!lctx->num_maps) {
return;
}
[vulkan] Add debug lines for light splats Disabled, but all that's needed is to uncomment the debug pipeline in the compose subpass.
2023-06-28 02:47:26 +00:00
auto cmd = taskctx->cmd;
auto layout = taskctx->pipeline->layout;
auto lframe = &lctx->frames.a[ctx->curFrame];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
auto shadow_type = *(int *) params[0]->value;
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
auto stage = *(int *) params[1]->value;
[vulkan] Use separate tracy GPU context for light culling My efforts (especially the collect zone (what was I thinking)) got tracy's knickers in a twist resulting in vanishing zones in the server. It looks like there are some synchronisation issues between cpu and gpu, but I'm not *too* worried about it at this stage.
2023-12-16 05:00:01 +00:00
if (stage == lighting_hull) {
[vulkan] Use OIT to visualize the light hulls Using the translucency pass made it easy to have depth-tested translucent "solid" light volumes instead of always visible lines (which are still an option as that's useful too). Most importantly, being able to see the surfaces helped no end in figuring out that my hulls were created with counter-clockwise windings instead of quake's usual clockwise windings and thus my hulls were being rendered inside-out in the occlusion pass.
2023-12-17 01:56:59 +00:00
bool planes = shadow_type == ST_PLANE;
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
VkDescriptorSet sets[] = {
Vulkan_Matrix_Descriptors (ctx, ctx->curFrame),
lframe->lights_set,
[vulkan] Use OIT to visualize the light hulls Using the translucency pass made it easy to have depth-tested translucent "solid" light volumes instead of always visible lines (which are still an option as that's useful too). Most importantly, being able to see the surfaces helped no end in figuring out that my hulls were created with counter-clockwise windings instead of quake's usual clockwise windings and thus my hulls were being rendered inside-out in the occlusion pass.
2023-12-17 01:56:59 +00:00
planes ? Vulkan_Translucent_Descriptors (ctx, ctx->curFrame) : 0,
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
};
dfunc->vkCmdBindDescriptorSets (cmd, VK_PIPELINE_BIND_POINT_GRAPHICS,
[vulkan] Use OIT to visualize the light hulls Using the translucency pass made it easy to have depth-tested translucent "solid" light volumes instead of always visible lines (which are still an option as that's useful too). Most importantly, being able to see the surfaces helped no end in figuring out that my hulls were created with counter-clockwise windings instead of quake's usual clockwise windings and thus my hulls were being rendered inside-out in the occlusion pass.
2023-12-17 01:56:59 +00:00
layout, 0, 2 + planes, sets, 0, 0);
[vulkan] Get lighting mostly working Lights seem to be good in renderdoc, but still need to get the compose subpass working.
2023-06-22 12:36:14 +00:00
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
VkBuffer buffers[] = {
lframe->id_buffer,
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
lframe->radius_buffer,
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
lctx->splat_verts,
};
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
VkDeviceSize offsets[] = { 0, 0, 0 };
dfunc->vkCmdBindVertexBuffers (cmd, 0, 3, buffers, offsets);
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
dfunc->vkCmdBindIndexBuffer (cmd, lctx->splat_inds, 0,
VK_INDEX_TYPE_UINT32);
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
} else {
VkDescriptorSet sets[] = {
lframe->shadowmat_set,
lframe->lights_set,
lframe->attach_set,
(VkDescriptorSet[]) {
lctx->shadow_2d_set,
lctx->shadow_2d_set,
lctx->shadow_2d_set,
lctx->shadow_cube_set
}[shadow_type],
};
dfunc->vkCmdBindDescriptorSets (cmd, VK_PIPELINE_BIND_POINT_GRAPHICS,
layout, 0, 4, sets, 0, 0);
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
}
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
}
static void
lighting_draw_splats (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
{
[build] Add support for building with Tracy Tracy is a frame profiler: https://github.com/wolfpld/tracy This uses Tracy's C API to instrument the code (already added in several places). It turns out there is something very weird with the fence behavior between the staging buffers and render commands as the inter-frame delay occurs in a very strangle place (in the draw code's packet acquisition rather than the fence waiter that's there for that purpose). I suspect some tangled dependencies.
2023-11-28 06:54:55 +00:00
qfZoneNamed (zone, true);
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto device = ctx->device;
auto dfunc = device->funcs;
auto lctx = ctx->lighting_context;
auto cmd = taskctx->cmd;
auto lframe = &lctx->frames.a[ctx->curFrame];
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
if (lframe->light_queue[ST_CUBE].count) {
auto q = lframe->light_queue[ST_CUBE];
dfunc->vkCmdDrawIndexed (cmd, num_ico_inds, q.count, 0, 0, q.start);
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
}
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
if (lframe->light_queue[ST_PLANE].count) {
auto q = lframe->light_queue[ST_PLANE];
dfunc->vkCmdDrawIndexed (cmd, num_cone_inds, q.count,
num_ico_inds, 12, q.start);
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
}
[vulkan] Register task functions for the pipelines They're currently just stubs, but this gets the render info loading working without any errors. The next step is to connect up pipelines and create the image resources, then implementing the task functions will have meaning.
2023-02-13 15:55:22 +00:00
}
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
static void
lighting_rewrite_ids (lightingframe_t *lframe, vulkan_ctx_t *ctx)
{
uint32_t count = 0;
[vulkan] Implement shadow map culling The rendering of the shadow maps now takes the culling information into account resulting in a drastic reduction of work. There's still more work to be done, but demo1 peaks at over 1000fps at 640x480, gmsp3v2 now gets 14fps (1920x1080) near the front gate (used to be 3, then 6), ad_tears is up to 3fps, but marcher is still unhappy, but it has infinite radius lights, so needs more culling work (clipped light volumes will help, I think). Also, culling lights for which nothing has moved within their volumes will help somewhat (though not as much for most id maps, I suspect).
2023-12-17 07:03:41 +00:00
auto lctx = ctx->lighting_context;
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
for (int i = 0; i < ST_COUNT; i++) {
auto q = &lframe->light_queue[i];
count += q->count;
}
uint32_t light_ids[count];
float light_radii[count];
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
uint32_t light_leafs[count];
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
uint32_t light_count = 0;
light_queue_t queue[ST_COUNT] = {};
for (int i = 0; i < ST_COUNT; i++) {
auto q = &lframe->light_queue[i];
for (uint32_t j = 0; j < q[0].count; j++) {
uint32_t id = lframe->id_radius[q[0].start + j].id;
float radius = lframe->id_radius[q[0].start + j].radius;
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
uint32_t leaf = lframe->id_radius[q[0].start + j].leafnum;
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
if (id != ~0u) {
light_ids[queue[i].start + queue[i].count] = id;
light_radii[queue[i].start + queue[i].count] = radius;
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
light_leafs[queue[i].start + queue[i].count] = leaf;
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
queue[i].count++;
}
}
if (i < ST_COUNT - 1) {
queue[i + 1].start = queue[i].start + queue[i].count;
} else {
light_count = queue[i].start + queue[i].count;
}
}
for (int i = 0; i < ST_COUNT; i++) {
lframe->light_queue[i] = queue[i];
}
[vulkan] Implement shadow map culling The rendering of the shadow maps now takes the culling information into account resulting in a drastic reduction of work. There's still more work to be done, but demo1 peaks at over 1000fps at 640x480, gmsp3v2 now gets 14fps (1920x1080) near the front gate (used to be 3, then 6), ad_tears is up to 3fps, but marcher is still unhappy, but it has infinite radius lights, so needs more culling work (clipped light volumes will help, I think). Also, culling lights for which nothing has moved within their volumes will help somewhat (though not as much for most id maps, I suspect).
2023-12-17 07:03:41 +00:00
for (int i = 0; i < LIGHTING_STAGES; i++) {
lframe->stage_queue[i].count = 0;
}
int matrix_id_count = 0;
for (uint32_t i = 0; i < light_count; i++) {
auto r = &lctx->light_control.a[light_ids[i]];
if (r->light_id != light_ids[i]) {
Sys_Error ("%d != %d", r->light_id, light_ids[i]);
}
lframe->stage_queue[r->stage_index].count += r->numLayers;
matrix_id_count += r->numLayers;
[vulkan] Apply light cull info to light pvs This makes a possible improvement to e1m3, only barely affects ad_tears, but makes about 30% difference to gmsp3v2 (21fps to 27, and from 3300 leafs to 2700).
2023-12-18 12:07:57 +00:00
lframe->id_radius[i] = (light_idrad_t) {
.id = light_ids[i],
.radius = light_radii[i],
.leafnum = light_leafs[i],
};
[vulkan] Implement shadow map culling The rendering of the shadow maps now takes the culling information into account resulting in a drastic reduction of work. There's still more work to be done, but demo1 peaks at over 1000fps at 640x480, gmsp3v2 now gets 14fps (1920x1080) near the front gate (used to be 3, then 6), ad_tears is up to 3fps, but marcher is still unhappy, but it has infinite radius lights, so needs more culling work (clipped light volumes will help, I think). Also, culling lights for which nothing has moved within their volumes will help somewhat (though not as much for most id maps, I suspect).
2023-12-17 07:03:41 +00:00
}
lframe->stage_queue[0].start = 0;
for (int i = 1; i < LIGHTING_STAGES; i++) {
auto q = &lframe->stage_queue[i];
q[0].start = q[-1].start + q[-1].count;
q[-1].count = 0;
}
lframe->stage_queue[LIGHTING_STAGES - 1].count = 0;
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
size_t packet_size = 0;
packet_size += sizeof (uint32_t[light_count]);
packet_size += sizeof (float[light_count]);
[vulkan] Implement shadow map culling The rendering of the shadow maps now takes the culling information into account resulting in a drastic reduction of work. There's still more work to be done, but demo1 peaks at over 1000fps at 640x480, gmsp3v2 now gets 14fps (1920x1080) near the front gate (used to be 3, then 6), ad_tears is up to 3fps, but marcher is still unhappy, but it has infinite radius lights, so needs more culling work (clipped light volumes will help, I think). Also, culling lights for which nothing has moved within their volumes will help somewhat (though not as much for most id maps, I suspect).
2023-12-17 07:03:41 +00:00
packet_size += sizeof (uint32_t[matrix_id_count]);
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
[vulkan] Bail if all lights have been culled When all lights have been culled, the updates wind up being 0 bytes and vulkan validation doesn't particularly like that.
2023-12-17 11:05:17 +00:00
if (!packet_size) {
return;
}
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
auto bb = &bufferBarriers[qfv_BB_TransferWrite_to_UniformRead];
auto packet = QFV_PacketAcquire (ctx->staging);
byte *packet_start = QFV_PacketExtend (packet, packet_size);
byte *packet_data = packet_start;
qfv_scatter_t id_scatter = {
.srcOffset = packet_data - packet_start,
.dstOffset = 0,
.length = sizeof (uint32_t[light_count]),
};
auto id_data = (uint32_t *) packet_data;
packet_data += id_scatter.length;
qfv_scatter_t radius_scatter = {
.srcOffset = packet_data - packet_start,
.dstOffset = 0,
.length = sizeof (uint32_t[light_count]),
};
auto radius_data = (uint32_t *) packet_data;
packet_data += radius_scatter.length;
[vulkan] Implement shadow map culling The rendering of the shadow maps now takes the culling information into account resulting in a drastic reduction of work. There's still more work to be done, but demo1 peaks at over 1000fps at 640x480, gmsp3v2 now gets 14fps (1920x1080) near the front gate (used to be 3, then 6), ad_tears is up to 3fps, but marcher is still unhappy, but it has infinite radius lights, so needs more culling work (clipped light volumes will help, I think). Also, culling lights for which nothing has moved within their volumes will help somewhat (though not as much for most id maps, I suspect).
2023-12-17 07:03:41 +00:00
qfv_scatter_t matrix_id_scater = {
.srcOffset = packet_data - packet_start,
.dstOffset = 0,
.length = sizeof (uint32_t[matrix_id_count]),
};
auto matrix_ids = (uint32_t *) packet_data;
packet_data += matrix_id_scater.length;
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
memcpy (id_data, light_ids, packet_size);
memcpy (radius_data, light_radii, packet_size);
[vulkan] Implement shadow map culling The rendering of the shadow maps now takes the culling information into account resulting in a drastic reduction of work. There's still more work to be done, but demo1 peaks at over 1000fps at 640x480, gmsp3v2 now gets 14fps (1920x1080) near the front gate (used to be 3, then 6), ad_tears is up to 3fps, but marcher is still unhappy, but it has infinite radius lights, so needs more culling work (clipped light volumes will help, I think). Also, culling lights for which nothing has moved within their volumes will help somewhat (though not as much for most id maps, I suspect).
2023-12-17 07:03:41 +00:00
for (uint32_t i = 0; i < light_count; i++) {
auto r = &lctx->light_control.a[light_ids[i]];
enqueue_map (matrix_ids, lframe, r);
}
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
QFV_PacketScatterBuffer (packet, lframe->id_buffer, 1, &id_scatter, bb);
QFV_PacketScatterBuffer (packet, lframe->radius_buffer,
1, &radius_scatter, bb);
[vulkan] Implement shadow map culling The rendering of the shadow maps now takes the culling information into account resulting in a drastic reduction of work. There's still more work to be done, but demo1 peaks at over 1000fps at 640x480, gmsp3v2 now gets 14fps (1920x1080) near the front gate (used to be 3, then 6), ad_tears is up to 3fps, but marcher is still unhappy, but it has infinite radius lights, so needs more culling work (clipped light volumes will help, I think). Also, culling lights for which nothing has moved within their volumes will help somewhat (though not as much for most id maps, I suspect).
2023-12-17 07:03:41 +00:00
QFV_PacketScatterBuffer (packet, lframe->shadowmat_id_buffer,
1, &matrix_id_scater, bb);
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
QFV_PacketSubmit (packet);
[vulkan] Implement shadow map culling The rendering of the shadow maps now takes the culling information into account resulting in a drastic reduction of work. There's still more work to be done, but demo1 peaks at over 1000fps at 640x480, gmsp3v2 now gets 14fps (1920x1080) near the front gate (used to be 3, then 6), ad_tears is up to 3fps, but marcher is still unhappy, but it has infinite radius lights, so needs more culling work (clipped light volumes will help, I think). Also, culling lights for which nothing has moved within their volumes will help somewhat (though not as much for most id maps, I suspect).
2023-12-17 07:03:41 +00:00
transition_shadow_targets (lframe, ctx);
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
if (developer & SYS_lighting) {
Vulkan_Draw_String (vid.width - 32, 16,
va (ctx->va_ctx, "%3d", light_count),
ctx);
}
}
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
static void
lighting_cull_select_renderpass (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
{
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto light_cull = QFV_GetStep (params[0], ctx->render_context->job);
auto render = light_cull->render;
if (scr_fisheye) {
render->active = &render->renderpasses[1];
} else {
render->active = &render->renderpasses[0];
}
}
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
static void
lighting_cull_lights (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
{
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto device = ctx->device;
auto dfunc = device->funcs;
auto lctx = ctx->lighting_context;
auto lframe = &lctx->frames.a[ctx->curFrame];
auto queue = lframe->light_queue;
uint32_t count = queue[ST_CUBE].count + queue[ST_PLANE].count;
if (!count) {
return;
}
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
if (scr_fisheye) {
count *= 6;
}
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
auto light_cull = QFV_GetStep (params[0], ctx->render_context->job);
auto render = light_cull->render;
auto cmd = QFV_GetCmdBuffer (ctx, false);
dfunc->vkBeginCommandBuffer (cmd, &(VkCommandBufferBeginInfo) {
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
});
[vulkan] Use separate tracy GPU context for light culling My efforts (especially the collect zone (what was I thinking)) got tracy's knickers in a twist resulting in vanishing zones in the server. It looks like there are some synchronisation issues between cpu and gpu, but I'm not *too* worried about it at this stage.
2023-12-16 05:00:01 +00:00
qftCVkCollect (lframe->qftVkCtx, cmd);
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
dfunc->vkCmdResetQueryPool (cmd, lframe->query, 0, MaxLights * 6);
[vulkan] Use separate tracy GPU context for light culling My efforts (especially the collect zone (what was I thinking)) got tracy's knickers in a twist resulting in vanishing zones in the server. It looks like there are some synchronisation issues between cpu and gpu, but I'm not *too* worried about it at this stage.
2023-12-16 05:00:01 +00:00
auto qftVkCtx = taskctx->frame->qftVkCtx;
taskctx->frame->qftVkCtx = lframe->qftVkCtx;
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
QFV_RunRenderPassCmd (cmd, ctx, render->active, 0);
[vulkan] Use separate tracy GPU context for light culling My efforts (especially the collect zone (what was I thinking)) got tracy's knickers in a twist resulting in vanishing zones in the server. It looks like there are some synchronisation issues between cpu and gpu, but I'm not *too* worried about it at this stage.
2023-12-16 05:00:01 +00:00
taskctx->frame->qftVkCtx = qftVkCtx;
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
dfunc->vkEndCommandBuffer (cmd);
qfMessageL ("submit");
auto dev_queue = &device->queue;
VkSubmitInfo submitInfo = {
.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
.commandBufferCount = 1,
.pCommandBuffers = &cmd,
};
dfunc->vkResetFences (device->dev, 1, &lframe->fence);
dfunc->vkQueueSubmit (dev_queue->queue, 1, &submitInfo, lframe->fence);
dfunc->vkWaitForFences (device->dev, 1, &lframe->fence, VK_TRUE, 200000000);
uint32_t frag_counts[count];
VkDeviceSize size = sizeof (frag_counts);
dfunc->vkGetQueryPoolResults (device->dev, lframe->query, 0, count,
size, frag_counts, sizeof (uint32_t),
VK_QUERY_RESULT_WAIT_BIT);
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
if (scr_fisheye) {
uint32_t *p = frag_counts;
for (uint32_t i = 0; i < count; i += 6) {
uint32_t sum = 0;
for (int j = 0; j < 6; j++) {
// care only about non-zero, not actual count
sum |= frag_counts[i + j];
}
*p++ = sum;
}
}
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
uint32_t c = 0;
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
uint32_t ci = 0;
vec4f_t cam = r_refdef.camera[3];
uint32_t id = 0;
if (lframe->light_queue[ST_CUBE].count) {
auto q = lframe->light_queue[ST_CUBE];
for (uint32_t i = 0; i < q.count; i++) {
uint32_t fc = frag_counts[id++];
c += fc != 0;
if (!fc) {
uint32_t hull = q.start + i;
vec4f_t dist = cam - lframe->positions[hull];
dist[3] = 0;
float rad = lframe->id_radius[hull].radius;
constexpr float s = 1.5835921350012616f;
bool inside = dotf(dist, dist)[0] < rad * rad * s;
ci += inside;
if (!inside) {
lframe->id_radius[hull].id = ~0u;
}
}
}
}
if (lframe->light_queue[ST_PLANE].count) {
auto q = lframe->light_queue[ST_PLANE];
for (uint32_t i = 0; i < q.count; i++) {
uint32_t fc = frag_counts[id++];
c += fc != 0;
if (!fc) {
uint32_t hull = q.start + i;
vec4f_t dist = cam - lframe->positions[hull];
dist[3] = 0;
float rad = lframe->id_radius[hull].radius;
bool inside = dotf(dist, dist)[0] < rad * rad;
ci += inside;
if (!inside) {
lframe->id_radius[hull].id = ~0u;
}
}
}
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
}
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
lighting_rewrite_ids (lframe, ctx);
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
}
static void
draw_hull (uint32_t indexCount, uint32_t firstIndex, int32_t vertOffset,
uint32_t hull, uint32_t id, VkCommandBuffer cmd, VkQueryPool query,
qfv_devfuncs_t *dfunc, qftVkCtx_t *vk)
{
qfZoneNamed (zone, true);
[vulkan] Implement shadow map culling The rendering of the shadow maps now takes the culling information into account resulting in a drastic reduction of work. There's still more work to be done, but demo1 peaks at over 1000fps at 640x480, gmsp3v2 now gets 14fps (1920x1080) near the front gate (used to be 3, then 6), ad_tears is up to 3fps, but marcher is still unhappy, but it has infinite radius lights, so needs more culling work (clipped light volumes will help, I think). Also, culling lights for which nothing has moved within their volumes will help somewhat (though not as much for most id maps, I suspect).
2023-12-17 07:03:41 +00:00
// qftVkScopedZoneC (vk, cmd, "draw_hull", 0xc0a000);
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
dfunc->vkCmdBeginQuery (cmd, query, id, 0);
dfunc->vkCmdDrawIndexed (cmd, indexCount, 1, firstIndex, vertOffset, hull);
dfunc->vkCmdEndQuery (cmd, query, id);
}
static void
lighting_draw_hulls (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
{
qfZoneNamed (zone, true);
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto device = ctx->device;
auto dfunc = device->funcs;
auto lctx = ctx->lighting_context;
auto cmd = taskctx->cmd;
auto lframe = &lctx->frames.a[ctx->curFrame];
uint32_t id = 0;
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
uint32_t id_step = scr_fisheye ? 6 : 1;
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
if (lframe->light_queue[ST_CUBE].count) {
auto q = lframe->light_queue[ST_CUBE];
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
for (uint32_t i = 0; i < q.count; i++, id += id_step) {
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
uint32_t hull = q.start + i;
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
draw_hull (num_ico_inds, 0, 0, hull, id,
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
cmd, lframe->query, dfunc, taskctx->frame->qftVkCtx);
}
}
if (lframe->light_queue[ST_PLANE].count) {
auto q = lframe->light_queue[ST_PLANE];
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
for (uint32_t i = 0; i < q.count; i++, id += id_step) {
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
uint32_t hull = q.start + i;
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
draw_hull (num_cone_inds, num_ico_inds, 12, hull, id,
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
cmd, lframe->query, dfunc, taskctx->frame->qftVkCtx);
}
}
}
[vulkan] Implement deferred fog It's not perfect (double fog on translucent surfaces, the scatter/absorption isn't right, and no local lighting on the fog itself), but it at least seems to look ok.
2024-01-23 05:32:30 +00:00
typedef struct {
vec4f_t fog;
vec4f_t CascadeDepths;
uint32_t queue;
} light_push_constants_t;
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
static void
lighting_draw_lights (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
{
[build] Add support for building with Tracy Tracy is a frame profiler: https://github.com/wolfpld/tracy This uses Tracy's C API to instrument the code (already added in several places). It turns out there is something very weird with the fence behavior between the staging buffers and render commands as the inter-frame delay occurs in a very strangle place (in the draw code's packet acquisition rather than the fence waiter that's there for that purpose). I suspect some tangled dependencies.
2023-11-28 06:54:55 +00:00
qfZoneNamed (zone, true);
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto device = ctx->device;
auto dfunc = device->funcs;
auto lctx = ctx->lighting_context;
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
auto layout = taskctx->pipeline->layout;
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
auto cmd = taskctx->cmd;
auto lframe = &lctx->frames.a[ctx->curFrame];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
auto shadow_type = *(int *) params[0]->value;
auto queue = lframe->light_queue[shadow_type];
if (!queue.count) {
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
return;
}
[vulkan] Implement deferred fog It's not perfect (double fog on translucent surfaces, the scatter/absorption isn't right, and no local lighting on the fog itself), but it at least seems to look ok.
2024-01-23 05:32:30 +00:00
vec4f_t fog = Fog_Get ();
// convert scatter to transmission (FIXME ignoring absorbtion)
fog = (vec4f_t) { 1, 1, 1, 0 } - fog;
fog[3] = -fog[3];
[vulkan] Implement direction shadow lighting Finally, even suns cast shadows :)
2023-08-13 17:33:19 +00:00
//FIXME dup of z_range (sort of)
vec4f_t depths = {
r_nearclip / 32, r_nearclip / 256, r_nearclip / 1024, 0,
};
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
qfv_push_constants_t push_constants[] = {
[vulkan] Implement deferred fog It's not perfect (double fog on translucent surfaces, the scatter/absorption isn't right, and no local lighting on the fog itself), but it at least seems to look ok.
2024-01-23 05:32:30 +00:00
{ VK_SHADER_STAGE_FRAGMENT_BIT,
field_offset (light_push_constants_t, fog),
sizeof (fog), &fog },
{ VK_SHADER_STAGE_FRAGMENT_BIT,
field_offset (light_push_constants_t, CascadeDepths),
sizeof (depths), &depths },
{ VK_SHADER_STAGE_FRAGMENT_BIT,
field_offset (light_push_constants_t, queue),
sizeof (queue), &queue },
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
};
[vulkan] Implement deferred fog It's not perfect (double fog on translucent surfaces, the scatter/absorption isn't right, and no local lighting on the fog itself), but it at least seems to look ok.
2024-01-23 05:32:30 +00:00
QFV_PushConstants (device, cmd, layout, 3, push_constants);
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
dfunc->vkCmdDraw (cmd, 3, 1, 0, 0);
}
[vulkan] Add job tasks to be run on new scene I think has been one of the biggest roadblocks to breaking free of quake, so having dual render paths and thus the different new scene load sequence has proven to be unexpected helpful. There's a lot more to be done to make the render graph actually usable by anyone but me, but just making scene load configurable frees up a lot. I think there needs to be renderer startup/shutdown configuration too, but this seems to be enough for now.
2024-01-21 04:36:17 +00:00
static void
lighting_load_lights (const exprval_t **params, exprval_t *result,
exprctx_t *ectx)
{
qfZoneNamed (zone, true);
auto taskctx = (qfv_taskctx_t *) ectx;
auto ctx = taskctx->ctx;
auto scene = (scene_t *) taskctx->data;
Vulkan_LoadLights (scene, ctx);
}
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
static exprenum_t lighting_stage_enum;
static exprtype_t lighting_stage_type = {
.name = "lighting_stage",
.size = sizeof (int),
.get_string = cexpr_enum_get_string,
.data = &lighting_stage_enum,
};
static int lighting_stage_values[] = {
lighting_main,
lighting_shadow,
[vulkan] Use separate tracy GPU context for light culling My efforts (especially the collect zone (what was I thinking)) got tracy's knickers in a twist resulting in vanishing zones in the server. It looks like there are some synchronisation issues between cpu and gpu, but I'm not *too* worried about it at this stage.
2023-12-16 05:00:01 +00:00
lighting_hull,
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
};
static exprsym_t lighting_stage_symbols[] = {
{"main", &lighting_stage_type, lighting_stage_values + 0},
{"shadow", &lighting_stage_type, lighting_stage_values + 1},
[vulkan] Use separate tracy GPU context for light culling My efforts (especially the collect zone (what was I thinking)) got tracy's knickers in a twist resulting in vanishing zones in the server. It looks like there are some synchronisation issues between cpu and gpu, but I'm not *too* worried about it at this stage.
2023-12-16 05:00:01 +00:00
{"hull", &lighting_stage_type, lighting_stage_values + 2},
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
{}
};
static exprtab_t lighting_stage_symtab = { .symbols = lighting_stage_symbols };
static exprenum_t lighting_stage_enum = {
&lighting_stage_type,
&lighting_stage_symtab,
};
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
static exprenum_t shadow_type_enum;
static exprtype_t shadow_type_type = {
.name = "shadow_type",
.size = sizeof (int),
.get_string = cexpr_enum_get_string,
.data = &shadow_type_enum,
};
static int shadow_type_values[] = { ST_NONE, ST_PLANE, ST_CASCADE, ST_CUBE };
static exprsym_t shadow_type_symbols[] = {
{"none", &shadow_type_type, shadow_type_values + 0},
{"plane", &shadow_type_type, shadow_type_values + 1},
{"cascade", &shadow_type_type, shadow_type_values + 2},
{"cube", &shadow_type_type, shadow_type_values + 3},
{}
};
static exprtab_t shadow_type_symtab = { .symbols = shadow_type_symbols };
static exprenum_t shadow_type_enum = {
&shadow_type_type,
&shadow_type_symtab,
};
static exprtype_t *shadow_type_param[] = {
&shadow_type_type,
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
&lighting_stage_type,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
};
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
static exprtype_t *stepref_param[] = {
&cexpr_string,
};
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
static exprfunc_t lighting_update_lights_func[] = {
{ .func = lighting_update_lights },
{}
};
[vulkan] Add debug lines for light splats Disabled, but all that's needed is to uncomment the debug pipeline in the compose subpass.
2023-06-28 02:47:26 +00:00
static exprfunc_t lighting_update_descriptors_func[] = {
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
{ .func = lighting_update_descriptors, .num_params = 1,
.param_types = stepref_param },
[vulkan] Add debug lines for light splats Disabled, but all that's needed is to uncomment the debug pipeline in the compose subpass.
2023-06-28 02:47:26 +00:00
{}
};
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
static exprfunc_t lighting_bind_descriptors_func[] = {
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
{ .func = lighting_bind_descriptors, .num_params = 2,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.param_types = shadow_type_param },
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
{}
};
static exprfunc_t lighting_draw_splats_func[] = {
{ .func = lighting_draw_splats },
{}
};
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
static exprfunc_t lighting_cull_select_renderpass_func[] = {
{ .func = lighting_cull_select_renderpass, .num_params = 1,
.param_types = stepref_param },
{}
};
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
static exprfunc_t lighting_cull_lights_func[] = {
{ .func = lighting_cull_lights, .num_params = 1,
.param_types = stepref_param },
{}
};
static exprfunc_t lighting_draw_hulls_func[] = {
{ .func = lighting_draw_hulls },
{}
};
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
static exprfunc_t lighting_draw_lights_func[] = {
[vulkan] Get debug light objects working again And also get them working for infinite radius lights.
2023-08-05 15:20:25 +00:00
{ .func = lighting_draw_lights, .num_params = 2,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.param_types = shadow_type_param },
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
{}
};
[vulkan] Rename bsp aux pass to shadow Much better name.
2023-08-08 09:04:58 +00:00
static exprfunc_t lighting_setup_shadow_func[] = {
{ .func = lighting_setup_shadow },
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
{}
};
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
static exprfunc_t lighting_draw_shadow_maps_func[] = {
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
{ .func = lighting_draw_shadow_maps, .num_params = 1,
.param_types = stepref_param },
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
{}
};
[vulkan] Add job tasks to be run on new scene I think has been one of the biggest roadblocks to breaking free of quake, so having dual render paths and thus the different new scene load sequence has proven to be unexpected helpful. There's a lot more to be done to make the render graph actually usable by anyone but me, but just making scene load configurable frees up a lot. I think there needs to be renderer startup/shutdown configuration too, but this seems to be enough for now.
2024-01-21 04:36:17 +00:00
static exprfunc_t lighting_load_lights_func[] = {
{ .func = lighting_load_lights },
{}
};
[vulkan] Register task functions for the pipelines They're currently just stubs, but this gets the render info loading working without any errors. The next step is to connect up pipelines and create the image resources, then implementing the task functions will have meaning.
2023-02-13 15:55:22 +00:00
static exprsym_t lighting_task_syms[] = {
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
{ "lighting_update_lights", &cexpr_function, lighting_update_lights_func },
[vulkan] Add debug lines for light splats Disabled, but all that's needed is to uncomment the debug pipeline in the compose subpass.
2023-06-28 02:47:26 +00:00
{ "lighting_update_descriptors", &cexpr_function,
lighting_update_descriptors_func },
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
{ "lighting_bind_descriptors", &cexpr_function,
lighting_bind_descriptors_func },
{ "lighting_draw_splats", &cexpr_function, lighting_draw_splats_func },
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
{ "lighting_cull_select_renderpass", &cexpr_function,
lighting_cull_select_renderpass_func },
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
{ "lighting_cull_lights", &cexpr_function, lighting_cull_lights_func },
{ "lighting_draw_hulls", &cexpr_function, lighting_draw_hulls_func },
[vulkan] Abandon light splats They were an interesting idea and might be useful in the future, but they don't work as well as I had hoped for quake's maps due to the overlapping light volumes causing contention while doing the additive blends in the frame buffer. The cause was made obvious when testing in the marcher map: most of its over 400 lights have infinite radius thus require full screen passes: all those passes fighting for the frame buffer did very nasty things to performance. However, light splats might be useful for many small, non-overlapping light volumes, thus the code is being kept (and I like the cleanups that came with it).
2023-06-28 04:19:10 +00:00
{ "lighting_draw_lights", &cexpr_function, lighting_draw_lights_func },
[vulkan] Rename bsp aux pass to shadow Much better name.
2023-08-08 09:04:58 +00:00
{ "lighting_setup_shadow", &cexpr_function, lighting_setup_shadow_func },
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
{ "lighting_draw_shadow_maps", &cexpr_function,
lighting_draw_shadow_maps_func },
[vulkan] Add job tasks to be run on new scene I think has been one of the biggest roadblocks to breaking free of quake, so having dual render paths and thus the different new scene load sequence has proven to be unexpected helpful. There's a lot more to be done to make the render graph actually usable by anyone but me, but just making scene load configurable frees up a lot. I think there needs to be renderer startup/shutdown configuration too, but this seems to be enough for now.
2024-01-21 04:36:17 +00:00
{ "lighting_load_lights", &cexpr_function, lighting_load_lights_func },
[vulkan] Register task functions for the pipelines They're currently just stubs, but this gets the render info loading working without any errors. The next step is to connect up pipelines and create the image resources, then implementing the task functions will have meaning.
2023-02-13 15:55:22 +00:00
{}
};
[vulkan] Apply normal light rules to dynamic lights ie, enforce shadow quanta and max light size.
2023-12-14 03:43:37 +00:00
static int
round_light_size (int size)
{
size = ((size + shadow_quanta - 1) / shadow_quanta) * shadow_quanta;
return min (size, 1024);
}
static void
dynlight_size_listener (void *data, const cvar_t *cvar)
{
dynlight_size = round_light_size (dynlight_size);
}
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
void
Vulkan_Lighting_Init (vulkan_ctx_t *ctx)
{
Clean up about 12000 allocations Some of them were actual leaks, but tracking memory should be a lot easier now. However, there's a lot of room for optimization of allocations (eg, recylcling of hierarchies. There is now 1 active allocation (according to tracy) when nq exits: Qgetline's string buffer (I think an api change is in order).
2024-01-04 05:10:57 +00:00
qfZoneScoped (true);
[vulkan] Nuke the old render pass code from orbit Gotta be sure :) With the new system mostly up and running (just bsp rendering and descriptor sets/layout handling to go, and they're independent of the old render pass system), the old system can finally be cleared out.
2023-06-24 01:34:18 +00:00
lightingctx_t *lctx = calloc (1, sizeof (lightingctx_t));
ctx->lighting_context = lctx;
[vulkan] Apply normal light rules to dynamic lights ie, enforce shadow quanta and max light size.
2023-12-14 03:43:37 +00:00
Cvar_Register (&dynlight_size_cvar, dynlight_size_listener, 0);
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
[vulkan] Register task functions for the pipelines They're currently just stubs, but this gets the render info loading working without any errors. The next step is to connect up pipelines and create the image resources, then implementing the task functions will have meaning.
2023-02-13 15:55:22 +00:00
QFV_Render_AddTasks (ctx, lighting_task_syms);
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
lctx->shadow_info = (qfv_attachmentinfo_t) {
.name = "$shadow",
[vulkan] Switch to float depth buffers This takes full advantage of the reversed (infinite far plane) depth buffers, and seems to have no performance cost.
2023-12-18 14:15:21 +00:00
.format = VK_FORMAT_D32_SFLOAT,
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
.samples = 1,
.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR,
.storeOp = VK_ATTACHMENT_STORE_OP_STORE,
.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE,
.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE,
.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
.finalLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,//FIXME plist
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
};
qfv_attachmentinfo_t *attachments[] = {
&lctx->shadow_info,
};
QFV_Render_AddAttachments (ctx, 1, attachments);
[vulkan] Use the descriptor set manager This cleans up a lot of ugly code that I always thought was rather dumb.
2023-06-24 12:42:46 +00:00
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
static void
make_default_map (int size, VkImage default_map, vulkan_ctx_t *ctx)
{
Clean up about 12000 allocations Some of them were actual leaks, but tracking memory should be a lot easier now. However, there's a lot of room for optimization of allocations (eg, recylcling of hierarchies. There is now 1 active allocation (according to tracy) when nq exits: Qgetline's string buffer (I think an api change is in order).
2024-01-04 05:10:57 +00:00
qfZoneScoped (true);
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
auto device = ctx->device;
auto dfunc = device->funcs;
auto packet = QFV_PacketAcquire (ctx->staging);
size_t imgsize = size * size * sizeof (uint32_t);
uint32_t *img = QFV_PacketExtend (packet, imgsize);
for (int i = 0; i < 64; i++) {
for (int j = 0; j < 64; j++) {
img[i * 64 + j] = ((j ^ i) & 1) ? 0x00ffffff : 0;
}
}
auto ib = imageBarriers[qfv_LT_Undefined_to_TransferDst];
ib.barrier.image = default_map;
ib.barrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
ib.barrier.subresourceRange.levelCount = VK_REMAINING_MIP_LEVELS;
ib.barrier.subresourceRange.layerCount = VK_REMAINING_ARRAY_LAYERS;
dfunc->vkCmdPipelineBarrier (packet->cmd, ib.srcStages, ib.dstStages,
0, 0, 0, 0, 0, 1, &ib.barrier);
VkBufferImageCopy copy_region[6];
for (int i = 0; i < 6; i++) {
copy_region[i] = (VkBufferImageCopy) {
.bufferOffset = packet->offset,
.bufferRowLength = 0,
.bufferImageHeight = 0,
.imageSubresource = {VK_IMAGE_ASPECT_DEPTH_BIT, 0, i, 1},
{0, 0, 0}, {size, size, 1},
};
}
dfunc->vkCmdCopyBufferToImage (packet->cmd, packet->stage->buffer,
default_map,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
6, copy_region);
ib = imageBarriers[qfv_LT_TransferDst_to_ShaderReadOnly];
ib.barrier.image = default_map;
ib.barrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
ib.barrier.subresourceRange.levelCount = VK_REMAINING_MIP_LEVELS;
ib.barrier.subresourceRange.layerCount = VK_REMAINING_ARRAY_LAYERS;
dfunc->vkCmdPipelineBarrier (packet->cmd, ib.srcStages, ib.dstStages,
0, 0, 0, 0, 0, 1, &ib.barrier);
QFV_PacketSubmit (packet);
}
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
static void
make_ico (qfv_packet_t *packet)
{
vec3_t *verts = QFV_PacketExtend (packet, sizeof (vec3_t[ico_verts]));
float p = (sqrt(5) + 1) / 2;
float a = sqrt (3) / p;
float b = a / p;
for (int i = 0; i < 3; i++) {
for (int j = 0; j < 4; j++) {
float my = j & 1 ? a : -a;
float mz = j & 2 ? b : -b;
int vind = i * 4 + j;
int ix = i;
int iy = (i + 1) % 3;
int iz = (i + 2) % 3;
verts[vind][ix] = 0;
verts[vind][iy] = my;
verts[vind][iz] = mz;
}
}
}
static void
make_cone (qfv_packet_t *packet)
{
vec3_t *verts = QFV_PacketExtend (packet, sizeof (vec3_t[cone_verts]));
float a = 2 / sqrt (3);
float b = 1 / sqrt (3);
VectorSet ( 0, 0, 0, verts[0]);
VectorSet ( a, 0, -1, verts[1]);
VectorSet ( b, 1, -1, verts[2]);
VectorSet (-b, 1, -1, verts[3]);
VectorSet (-a, 0, -1, verts[4]);
VectorSet (-b, -1, -1, verts[5]);
VectorSet ( b, -1, -1, verts[6]);
}
static void
write_inds (qfv_packet_t *packet)
{
uint32_t *inds = QFV_PacketExtend (packet, sizeof (ico_inds)
+ sizeof (cone_inds));
memcpy (inds, ico_inds, sizeof (ico_inds));
inds += num_ico_inds;
memcpy (inds, cone_inds, sizeof (cone_inds));
}
[vulkan] Use the descriptor set manager This cleans up a lot of ugly code that I always thought was rather dumb.
2023-06-24 12:42:46 +00:00
void
Vulkan_Lighting_Setup (vulkan_ctx_t *ctx)
{
Clean up about 12000 allocations Some of them were actual leaks, but tracking memory should be a lot easier now. However, there's a lot of room for optimization of allocations (eg, recylcling of hierarchies. There is now 1 active allocation (according to tracy) when nq exits: Qgetline's string buffer (I think an api change is in order).
2024-01-04 05:10:57 +00:00
qfZoneScoped (true);
[vulkan] Use the descriptor set manager This cleans up a lot of ugly code that I always thought was rather dumb.
2023-06-24 12:42:46 +00:00
qfvPushDebug (ctx, "lighting init");
[vulkan] Add some more debug stack contexts
2021-12-02 13:48:50 +00:00
[vulkan] Use the descriptor set manager This cleans up a lot of ugly code that I always thought was rather dumb.
2023-06-24 12:42:46 +00:00
auto device = ctx->device;
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
auto dfunc = device->funcs;
[vulkan] Use the descriptor set manager This cleans up a lot of ugly code that I always thought was rather dumb.
2023-06-24 12:42:46 +00:00
auto lctx = ctx->lighting_context;
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
[vulkan] Clean up the sampler config loading Samplers have no direct relation to render passes or pipelines, so should not necessarily be in the same config file. This makes all the old config files obsolete, and quite a bit of support code in vkparse.c.
2023-06-25 14:33:47 +00:00
lctx->sampler = QFV_Render_Sampler (ctx, "shadow_sampler");
[vulkan] Make script support objects more private This will help keep changes to the scripting system localized once things are further cleaned up.
2022-11-24 14:44:07 +00:00
Vulkan_Script_SetOutput (ctx,
[vulkan] Switch to float depth buffers This takes full advantage of the reversed (infinite far plane) depth buffers, and seems to have no performance cost.
2023-12-18 14:15:21 +00:00
&(qfv_output_t) { .format = VK_FORMAT_D32_SFLOAT });
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
DARRAY_INIT (&lctx->light_mats, 16);
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
DARRAY_INIT (&lctx->light_control, 16);
[vulkan] Load the map's lights into an array It's not used yet as work needs to be done to better support generic entities, but this is the next step to real-time lighting (though, to be honest, I expect it will be too slow to be usable).
2021-03-03 09:14:16 +00:00
[vulkan] Move frames from vulkan_ctx to renderctx Part of the command buffer fix, but also a step towards cleaning up vulkan_ctx.
2023-06-22 07:44:05 +00:00
auto rctx = ctx->render_context;
size_t frames = rctx->frames.size;
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
DARRAY_INIT (&lctx->frames, frames);
DARRAY_RESIZE (&lctx->frames, frames);
lctx->frames.grow = 0;
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
lctx->light_resources = malloc (sizeof (qfv_resource_t)
// splat vertices
+ sizeof (qfv_resobj_t)
// splat indices
+ sizeof (qfv_resobj_t)
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
// default shadow map and views
+ 3 * sizeof (qfv_resobj_t)
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
// light matrices
[vulkan] Implement mouse-picking for light entities Currently, only light entities get drawn to the entid buffer, and the ids are simply displayed in a window for now (not very useful yet).
2023-08-07 08:47:49 +00:00
+ sizeof (qfv_resobj_t[frames])
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
// light matrix ids
+ sizeof (qfv_resobj_t[frames])
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
// light ids
+ sizeof (qfv_resobj_t[frames])
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
// light radii
+ sizeof (qfv_resobj_t[frames])
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
// light data
+ sizeof (qfv_resobj_t[frames])
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
// light render
+ sizeof (qfv_resobj_t[frames])
// light styles
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
+ sizeof (qfv_resobj_t[frames])
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
// light entids
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
+ sizeof (qfv_resobj_t[frames]));
lctx->light_resources[0] = (qfv_resource_t) {
.name = "lights",
.va_ctx = ctx->va_ctx,
.memory_properties = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
.num_objects = 2 + 3 + 8 * frames,
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
.objects = (qfv_resobj_t *) &lctx->light_resources[1],
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
};
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
auto splat_verts = lctx->light_resources->objects;
auto splat_inds = &splat_verts[1];
auto default_map = &splat_inds[1];
auto default_view_cube = &default_map[1];
auto default_view_2d = &default_view_cube[1];
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
auto light_mats = &default_view_2d[1];
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
auto light_mat_ids = &light_mats[frames];
auto light_ids = &light_mat_ids[frames];
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
auto light_radii = &light_ids[frames];
auto light_data = &light_radii[frames];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
auto light_render = &light_data[frames];
auto light_styles = &light_render[frames];
[vulkan] Merge lighting updates into one packet While QFV_PacketScatterBuffer works on only one destination buffer, it turns out it's still useful for scattering to multiple buffers, just with multiple calls. This makes it pretty easy to combine multiple buffer updates into a single staging buffer packet, resulting in reducing lighting's packet use from up to 7 to just one, drastically reducing the pressure on the stating buffer packet pool, and thus reducing the chances of QFV_PacketAcquire stalling.
2023-12-05 15:15:42 +00:00
auto light_entids = &light_styles[frames];
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
splat_verts[0] = (qfv_resobj_t) {
.name = "splat:vertices",
.type = qfv_res_buffer,
.buffer = {
.size = (20 + 7) * sizeof (vec3_t),
.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT
| VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
},
};
splat_inds[0] = (qfv_resobj_t) {
.name = "splat:indices",
.type = qfv_res_buffer,
.buffer = {
.size = sizeof (ico_inds) + sizeof (cone_inds),
.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT
| VK_BUFFER_USAGE_INDEX_BUFFER_BIT,
},
};
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
default_map[0] = (qfv_resobj_t) {
.name = "default_map",
.type = qfv_res_image,
.image = {
.flags = VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT,
.type = VK_IMAGE_TYPE_2D,
[vulkan] Switch to float depth buffers This takes full advantage of the reversed (infinite far plane) depth buffers, and seems to have no performance cost.
2023-12-18 14:15:21 +00:00
.format = VK_FORMAT_D32_SFLOAT,
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
.extent = { 64, 64, 1 },
.num_mipmaps = 1,
.num_layers = 6,
.samples = VK_SAMPLE_COUNT_1_BIT,
.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT
| VK_IMAGE_USAGE_SAMPLED_BIT,
},
};
default_view_cube[0] = (qfv_resobj_t) {
.name = "default_map:view_cube",
.type = qfv_res_image_view,
.image_view = {
.image = default_map - lctx->light_resources->objects,
.type = VK_IMAGE_VIEW_TYPE_CUBE_ARRAY,
[vulkan] Switch to float depth buffers This takes full advantage of the reversed (infinite far plane) depth buffers, and seems to have no performance cost.
2023-12-18 14:15:21 +00:00
.format = VK_FORMAT_D32_SFLOAT,
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
.subresourceRange = {
.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
.levelCount = VK_REMAINING_MIP_LEVELS,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
},
};
default_view_2d[0] = (qfv_resobj_t) {
.name = "default_map:view_2d",
.type = qfv_res_image_view,
.image_view = {
.image = default_map - lctx->light_resources->objects,
.type = VK_IMAGE_VIEW_TYPE_2D_ARRAY,
[vulkan] Switch to float depth buffers This takes full advantage of the reversed (infinite far plane) depth buffers, and seems to have no performance cost.
2023-12-18 14:15:21 +00:00
.format = VK_FORMAT_D32_SFLOAT,
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
.subresourceRange = {
.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
.levelCount = VK_REMAINING_MIP_LEVELS,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
},
};
[vulkan] Implement deferred lighting for dlights Static lights are yet to come (so the screen is black most of the time), but dynamic lights work very nicely (and look very good) despite the falloff being incorrect.
2021-02-25 06:51:54 +00:00
for (size_t i = 0; i < frames; i++) {
[vulkan] Implement mouse-picking for light entities Currently, only light entities get drawn to the entid buffer, and the ids are simply displayed in a window for now (not very useful yet).
2023-08-07 08:47:49 +00:00
light_entids[i] = (qfv_resobj_t) {
.name = va (ctx->va_ctx, "entids:%zd", i),
.type = qfv_res_buffer,
.buffer = {
.size = MaxLights * sizeof (uint32_t),
.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT
| VK_BUFFER_USAGE_STORAGE_BUFFER_BIT
| VK_BUFFER_USAGE_TRANSFER_DST_BIT,
},
};
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
light_ids[i] = (qfv_resobj_t) {
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
.name = va (ctx->va_ctx, "ids:%zd", i),
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.type = qfv_res_buffer,
.buffer = {
[vulkan] Implement mouse-picking for light entities Currently, only light entities get drawn to the entid buffer, and the ids are simply displayed in a window for now (not very useful yet).
2023-08-07 08:47:49 +00:00
.size = MaxLights * sizeof (uint32_t),
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT
| VK_BUFFER_USAGE_STORAGE_BUFFER_BIT
| VK_BUFFER_USAGE_TRANSFER_DST_BIT,
},
};
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
light_radii[i] = (qfv_resobj_t) {
.name = va (ctx->va_ctx, "radii:%zd", i),
.type = qfv_res_buffer,
.buffer = {
.size = MaxLights * sizeof (float),
.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT
| VK_BUFFER_USAGE_TRANSFER_DST_BIT,
},
};
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
light_data[i] = (qfv_resobj_t) {
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
.name = va (ctx->va_ctx, "lights:%zd", i),
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
.type = qfv_res_buffer,
.buffer = {
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.size = sizeof (light_t[MaxLights]),
.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
| VK_BUFFER_USAGE_TRANSFER_DST_BIT,
},
};
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
light_render[i] = (qfv_resobj_t) {
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
.name = va (ctx->va_ctx, "render:%zd", i),
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.type = qfv_res_buffer,
.buffer = {
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.size = sizeof (qfv_light_render_t[MaxLights]),
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT
| VK_BUFFER_USAGE_TRANSFER_DST_BIT,
},
};
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
light_styles[i] = (qfv_resobj_t) {
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
.name = va (ctx->va_ctx, "styles:%zd", i),
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.type = qfv_res_buffer,
.buffer = {
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.size = sizeof (vec4f_t[NumStyles]),
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT
| VK_BUFFER_USAGE_TRANSFER_DST_BIT,
},
};
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
light_mats[i] = (qfv_resobj_t) {
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
.name = va (ctx->va_ctx, "matrices:%zd", i),
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
.type = qfv_res_buffer,
.buffer = {
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
// never need more than 6 matrices per light
.size = sizeof (mat4f_t[MaxLights * 6]),
.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
| VK_BUFFER_USAGE_TRANSFER_DST_BIT,
},
};
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
light_mat_ids[i] = (qfv_resobj_t) {
.name = va (ctx->va_ctx, "matrix ids:%zd", i),
.type = qfv_res_buffer,
.buffer = {
// never need more than 6 matrices per light
.size = sizeof (uint32_t[MaxLights * 6]),
.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT
| VK_BUFFER_USAGE_TRANSFER_DST_BIT,
},
};
[vulkan] Implement deferred lighting for dlights Static lights are yet to come (so the screen is black most of the time), but dynamic lights work very nicely (and look very good) despite the falloff being incorrect.
2021-02-25 06:51:54 +00:00
}
Fix a couple of windows build issues There's still the problem with unused variables when building for windows because of vulkan debug stuff, but this fixes the important errors. It actually still works (at least under wine).
2023-07-05 17:04:47 +00:00
QFV_CreateResource (device, lctx->light_resources);
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
lctx->splat_verts = splat_verts[0].buffer.buffer;
lctx->splat_inds = splat_inds[0].buffer.buffer;
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
lctx->default_map = default_map[0].image.image;
lctx->default_view_cube = default_view_cube[0].image_view.view;
lctx->default_view_2d = default_view_2d[0].image_view.view;
[vulkan] Implement deferred lighting for dlights Static lights are yet to come (so the screen is black most of the time), but dynamic lights work very nicely (and look very good) despite the falloff being incorrect.
2021-02-25 06:51:54 +00:00
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto shadow_mgr = QFV_Render_DSManager (ctx, "lighting_shadow");
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
lctx->shadow_cube_set = QFV_DSManager_AllocSet (shadow_mgr);
lctx->shadow_2d_set = QFV_DSManager_AllocSet (shadow_mgr);
QFV_duSetObjectName (device, VK_OBJECT_TYPE_DESCRIPTOR_SET,
lctx->shadow_cube_set, "lighting:shadow_cube_set");
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
QFV_duSetObjectName (device, VK_OBJECT_TYPE_DESCRIPTOR_SET,
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
lctx->shadow_2d_set, "lighting:shadow_2d_set");
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
lctx->shadow_sampler = QFV_Render_Sampler (ctx, "shadow_sampler");
[vulkan] Use the descriptor set manager This cleans up a lot of ugly code that I always thought was rather dumb.
2023-06-24 12:42:46 +00:00
auto attach_mgr = QFV_Render_DSManager (ctx, "lighting_attach");
auto lights_mgr = QFV_Render_DSManager (ctx, "lighting_lights");
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
auto shadowmat_mgr = QFV_Render_DSManager (ctx, "shadowmat_set");
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
for (size_t i = 0; i < frames; i++) {
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
auto lframe = &lctx->frames.a[i];
*lframe = (lightingframe_t) {
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.shadowmat_set = QFV_DSManager_AllocSet (shadowmat_mgr),
.lights_set = QFV_DSManager_AllocSet (lights_mgr),
.attach_set = QFV_DSManager_AllocSet (attach_mgr),
.shadowmat_buffer = light_mats[i].buffer.buffer,
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
.shadowmat_id_buffer = light_mat_ids[i].buffer.buffer,
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.light_buffer = light_data[i].buffer.buffer,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.render_buffer = light_render[i].buffer.buffer,
.style_buffer = light_styles[i].buffer.buffer,
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
.id_buffer = light_ids[i].buffer.buffer,
[vulkan] Move light radius calcs to the CPU No point in calculating them for every vertex, especially when I forgot to update the calculations for the entid vertex shader.
2023-12-16 06:44:44 +00:00
.radius_buffer = light_radii[i].buffer.buffer,
[vulkan] Implement mouse-picking for light entities Currently, only light entities get drawn to the entid buffer, and the ids are simply displayed in a window for now (not very useful yet).
2023-08-07 08:47:49 +00:00
.entid_buffer = light_entids[i].buffer.buffer,
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
};
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
QFV_duSetObjectName (device, VK_OBJECT_TYPE_DESCRIPTOR_SET,
lframe->attach_set,
[vulkan] Start work on shadows This gets the shaders needed for creating shadow maps, and the changes to the lighting pipeline for binding the shadow maps, but no generation or reading is done yet. It feels like parts of various systems are getting a little big for their britches and I need to do an audit of various things.
2021-04-24 01:40:39 +00:00
va (ctx->va_ctx, "lighting:attach_set:%zd", i));
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
QFV_duSetObjectName (device, VK_OBJECT_TYPE_DESCRIPTOR_SET,
lframe->lights_set,
[vulkan] Start work on shadows This gets the shaders needed for creating shadow maps, and the changes to the lighting pipeline for binding the shadow maps, but no generation or reading is done yet. It feels like parts of various systems are getting a little big for their britches and I need to do an audit of various things.
2021-04-24 01:40:39 +00:00
va (ctx->va_ctx, "lighting:lights_set:%zd", i));
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
QFV_duSetObjectName (device, VK_OBJECT_TYPE_DESCRIPTOR_SET,
lframe->shadowmat_set,
va (ctx->va_ctx, "lighting:shadowmat_set:%zd", i));
[vulkan] Run the shadow render passes The shadow maps are rendered incorrectly because the matrices aren't set correctly yet, but the dual-pass bsp rendering is working.
2023-07-28 16:17:53 +00:00
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
VkDescriptorBufferInfo bufferInfo[] = {
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
{ .buffer = lframe->shadowmat_buffer,
.offset = 0, .range = VK_WHOLE_SIZE, },
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
{ .buffer = lframe->shadowmat_id_buffer,
.offset = 0, .range = VK_WHOLE_SIZE, },
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
{ .buffer = lframe->id_buffer,
.offset = 0, .range = VK_WHOLE_SIZE, },
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
{ .buffer = lframe->light_buffer,
.offset = 0, .range = VK_WHOLE_SIZE, },
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
{ .buffer = lframe->render_buffer,
.offset = 0, .range = VK_WHOLE_SIZE, },
{ .buffer = lframe->style_buffer,
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.offset = 0, .range = VK_WHOLE_SIZE, },
[vulkan] Implement mouse-picking for light entities Currently, only light entities get drawn to the entid buffer, and the ids are simply displayed in a window for now (not very useful yet).
2023-08-07 08:47:49 +00:00
{ .buffer = lframe->entid_buffer,
.offset = 0, .range = VK_WHOLE_SIZE, },
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
};
VkWriteDescriptorSet bufferWrite[] = {
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.dstSet = lframe->shadowmat_set,
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.dstBinding = 0,
.descriptorCount = 1,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.pBufferInfo = &bufferInfo[0], },
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lframe->shadowmat_set,
.dstBinding = 1,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.pBufferInfo = &bufferInfo[1], },
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.dstSet = lframe->lights_set,
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
.dstBinding = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
.pBufferInfo = &bufferInfo[2], },
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lframe->lights_set,
.dstBinding = 1,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
.pBufferInfo = &bufferInfo[3], },
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lframe->lights_set,
.dstBinding = 2,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
.pBufferInfo = &bufferInfo[4], },
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lframe->lights_set,
.dstBinding = 3,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
.pBufferInfo = &bufferInfo[5], },
[vulkan] Implement mouse-picking for light entities Currently, only light entities get drawn to the entid buffer, and the ids are simply displayed in a window for now (not very useful yet).
2023-08-07 08:47:49 +00:00
{ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lframe->lights_set,
.dstBinding = 4,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
.pBufferInfo = &bufferInfo[6], },
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
};
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
dfunc->vkUpdateDescriptorSets (device->dev, 7, bufferWrite, 0, 0);
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
dfunc->vkCreateQueryPool (device->dev, &(VkQueryPoolCreateInfo) {
.sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO,
.queryType = VK_QUERY_TYPE_OCCLUSION,
[vulkan] Support fisheye for light culling Nicely, the people who created multiview thought of queries and how they need to interact.
2023-12-17 18:07:29 +00:00
.queryCount = MaxLights * 6, // 6 for cube maps
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
}, 0, &lframe->query);
[vulkan] Destroy the tracy GPU contexts I had forgotten about doing so because I was doing my profiling without validation layers.
2023-12-22 04:44:31 +00:00
QFV_duSetObjectName (device, VK_OBJECT_TYPE_QUERY_POOL, lframe->query,
va (ctx->va_ctx, "light_cull:%zd", i));
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
lframe->fence = QFV_CreateFence (device, 1);
[vulkan] Use separate tracy GPU context for light culling My efforts (especially the collect zone (what was I thinking)) got tracy's knickers in a twist resulting in vanishing zones in the server. It looks like there are some synchronisation issues between cpu and gpu, but I'm not *too* worried about it at this stage.
2023-12-16 05:00:01 +00:00
#ifdef TRACY_ENABLE
auto instance = ctx->instance->instance;
auto physdev = ctx->device->physDev->dev;
auto gipa = ctx->vkGetInstanceProcAddr;
auto gdpa = ctx->instance->funcs->vkGetDeviceProcAddr;
lframe->qftVkCtx = qftCVkContextHostCalibrated (instance, physdev,
device->dev,
gipa, gdpa);
#endif
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
size_t target_count = MaxLights * 6;
size_t target_size = frames * sizeof (uint16_t[target_count]);
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
size_t idr_size = frames * sizeof (light_idrad_t[MaxLights]);
size_t position_size = frames * sizeof (vec4f_t[MaxLights]);
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
lctx->frames.a[0].stage_targets = malloc (target_size);
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
lctx->frames.a[0].id_radius = malloc (idr_size);
lctx->frames.a[0].positions = malloc (position_size);
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
for (size_t i = 1; i < frames; i++) {
auto lframe = &lctx->frames.a[i];
lframe[0].stage_targets = lframe[-1].stage_targets + target_count;
[vulkan] Apply the light culling information The results of the occlusion queries give the lights that don't have a visible hull, but unfortunately that includes any lights which the camera is inside, but simple distance checks sort that out (with a fudge-factor for the icosahedron vertices (1.583 (3(2+p)/(2+3p), p is golden ratio)).
2023-12-17 01:45:59 +00:00
lframe[0].id_radius = lframe[-1].id_radius + MaxLights;
lframe[0].positions = lframe[-1].positions + MaxLights;
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
}
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
make_default_map (64, lctx->default_map, ctx);
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
auto packet = QFV_PacketAcquire (ctx->staging);
make_ico (packet);
make_cone (packet);
QFV_PacketCopyBuffer (packet, splat_verts[0].buffer.buffer, 0,
&bufferBarriers[qfv_BB_TransferWrite_to_UniformRead]);
QFV_PacketSubmit (packet);
packet = QFV_PacketAcquire (ctx->staging);
write_inds (packet);
QFV_PacketCopyBuffer (packet, splat_inds[0].buffer.buffer, 0,
&bufferBarriers[qfv_BB_TransferWrite_to_IndexRead]);
QFV_PacketSubmit (packet);
[vulkan] Add some more debug stack contexts
2021-12-02 13:48:50 +00:00
qfvPopDebug (ctx);
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
}
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
static void
clear_shadows (vulkan_ctx_t *ctx)
{
[vulkan] Add gpu profiling zones This relies on my fork of tracy: https://github.com/taniwha/tracy on the wip-c-vulkan branch. Everything is still rather flaky though. This necessitated the jump to vulkan 1.2 as a requirement.
2023-12-04 12:38:43 +00:00
qfZoneScoped (true);
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
qfv_device_t *device = ctx->device;
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
auto dfunc = device->funcs;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
lightingctx_t *lctx = ctx->lighting_context;
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
if (lctx->shadow_resources) {
QFV_DestroyResource (device, lctx->shadow_resources);
[vulkan] Free shadow map resource block I forgot I had to free the memory too. Oops.
2022-10-27 02:23:49 +00:00
free (lctx->shadow_resources);
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
lctx->shadow_resources = 0;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
for (int i = 0; i < LIGHTING_STAGES; i++) {
for (int j = 0; j < 32; j++) {
auto framebuffer = lctx->stage_framebuffers[j][i];
if (framebuffer) {
dfunc->vkDestroyFramebuffer (device->dev, framebuffer, 0);
}
lctx->stage_framebuffers[j][i] = 0;
}
// images and views freed via shadow_resources
lctx->stage_images[i] = 0;
lctx->stage_views[i] = 0;
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
free (lctx->map_images);
free (lctx->map_views);
free (lctx->map_cube);
lctx->map_images = 0;
lctx->map_views = 0;
lctx->map_cube = 0;
lctx->num_maps = 0;
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
lctx->light_control.size = 0;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
void
Vulkan_Lighting_Shutdown (vulkan_ctx_t *ctx)
{
[vulkan] Add tracy zones to most shutdown functions This helped find the 5s hang on exit.
2024-01-16 04:46:37 +00:00
qfZoneScoped (true);
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
auto device = ctx->device;
auto dfunc = device->funcs;
auto lctx = ctx->lighting_context;
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
clear_shadows (ctx);
[vulkan] Rework lighting to use splats It's currently slower, and the cone splats are buggy, but the lighting code itself got some nice cleanups.
2023-06-27 16:01:56 +00:00
QFV_DestroyResource (device, lctx->light_resources);
free (lctx->light_resources);
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
for (size_t i = 0; i < lctx->frames.size; i++) {
auto lframe = &lctx->frames.a[i];
dfunc->vkDestroyQueryPool (device->dev, lframe->query, 0);
dfunc->vkDestroyFence (device->dev, lframe->fence, 0);
[vulkan] Destroy the tracy GPU contexts I had forgotten about doing so because I was doing my profiling without validation layers.
2023-12-22 04:44:31 +00:00
qftCVkContextDestroy (lframe->qftVkCtx);
[vulkan] Use occlusion queries for culling lights The info isn't used yet, but this shows that vulkan's occlusion queries are at least somewhat useful. However, the technique isn't perfect: infinite radius lights (1/r and 1/r^2) are difficult to cull, and all lights can poke through thin enough walls, and then lights containing the camera get culled incorrectly (will need a separate test). Still, it looks like it will help once everything is tied together.
2023-12-15 14:56:41 +00:00
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
free (lctx->frames.a[0].stage_targets);
Clean up about 12000 allocations Some of them were actual leaks, but tracking memory should be a lot easier now. However, there's a lot of room for optimization of allocations (eg, recylcling of hierarchies. There is now 1 active allocation (according to tracy) when nq exits: Qgetline's string buffer (I think an api change is in order).
2024-01-04 05:10:57 +00:00
free (lctx->frames.a[0].id_radius);
free (lctx->frames.a[0].positions);
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
DARRAY_CLEAR (&lctx->light_mats);
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
DARRAY_CLEAR (&lctx->light_control);
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
free (lctx->map_images);
free (lctx->map_views);
free (lctx->map_cube);
[vulkan] Implement lighting and compose passes Lighting doesn't actually do lights yet, but it's producing pixels. Translucent seems to be working (2d draw uses it), and compose seems to be working.
2021-02-24 10:58:31 +00:00
free (lctx->frames.a);
free (lctx);
}
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
static void
create_light_matrices (lightingctx_t *lctx)
{
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
auto reg = lctx->scene->reg;
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
auto light_pool = &reg->comp_pools[lctx->scene->base + scene_light];
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
auto light_data = (light_t *) light_pool->data;
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
uint16_t mat_count = 0;
for (uint32_t i = 0; i < light_pool->count; i++) {
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
entity_t ent = { .reg = reg, .id = light_pool->dense[i],
.base = lctx->scene->base };
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
uint32_t id = get_lightid (ent);
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto r = &lctx->light_control.a[id];
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
r->matrix_id = mat_count;
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
mat_count += r->numLayers;
}
DARRAY_RESIZE (&lctx->light_mats, mat_count);
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
lctx->dynamic_matrix_base = mat_count;
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
for (uint32_t i = 0; i < dynlight_max; i++) {
auto r = &lctx->light_control.a[lctx->dynamic_base + i];
r->matrix_id = lctx->dynamic_matrix_base + i * 6;
}
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
for (uint32_t i = 0; i < light_pool->count; i++) {
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
light_t *light = &light_data[i];
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
entity_t ent = { .reg = reg, .id = light_pool->dense[i],
.base = lctx->scene->base };
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
uint32_t id = get_lightid (ent);
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto r = &lctx->light_control.a[id];
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
auto lm = &lctx->light_mats.a[r->matrix_id];
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
mat4f_t view;
mat4f_t proj;
Implement clang support Fixing a load of issues related to autoconf and some small source-level issues to re-add clang support. autoconf feature detection probably needs some addressing - partially as -Werror is applied late.
2023-08-10 22:11:52 +00:00
vec4f_t dir;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
switch (r->mode) {
[vulkan] Fix an uninitialized variable warning semi-bogus, but I prefer false positives to false negatives.
2021-06-01 10:09:03 +00:00
default:
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
case ST_NONE:
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
continue;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
case ST_CUBE:
mat4fidentity (view);
break;
case ST_CASCADE:
case ST_PLANE:
Implement clang support Fixing a load of issues related to autoconf and some small source-level issues to re-add clang support. autoconf feature detection probably needs some addressing - partially as -Werror is applied late.
2023-08-10 22:11:52 +00:00
dir = light->direction;
[vulkan] Rework lighting model to be more algebraic This leaves only the one conditional in the shader code, that being the distance check. It doesn't seem to make any noticeable difference to performance, but other than explosion sprites being blue, lighting quality seems to have improved. However, I really need to get shadows working: marcher is just silly-bright without them, and light levels changing as I move around is a bit disconcerting (but reasonable as those lights' leaf nodes go in and out of visibility).
2022-05-04 23:40:02 +00:00
dir[3] = 0;
[vulkan] Handle spotlights pointing -X The problem didn't show up until the switch to float depth buffers (because they can store the resulting nan).
2023-12-19 08:56:13 +00:00
vec4f_t q = dir[0] == -1 ? (vec4f_t) { 0, 0, 1, 0 }
: qrotf (dir, ref_direction);
mat4fquat (view, q);
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
break;
}
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
vec4f_t pos = -light->position;
pos[3] = 1;
view[3] = mvmulf (view, pos);
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
switch (r->mode) {
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
case ST_NONE:
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
continue;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
case ST_CUBE:
[vulkan] Make near and far clip explicit parameters This improves the projection API in that near clip is a parameter rather than being taken directly from the cvar, and a far clip (ie, finite far plane) version is available (necessary for cascaded shadow maps as it's rather hard to fit a box to an infinite frustum). Also, the orthographic projection matrix is now reversed as per the perspective matrix (and the code tidied up a little), and a version that takes min and max vectors is available.
2023-08-13 08:25:17 +00:00
QFV_PerspectiveTan (proj, 1, 1, lnearclip);
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
for (int j = 0; j < 6; j++) {
mat4f_t side_view;
mat4f_t rotinv;
mat4ftranspose (rotinv, qfv_box_rotations[j]);
mmulf (side_view, rotinv, view);
mmulf (side_view, qfv_z_up, side_view);
mmulf (lm[j], proj, side_view);
}
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
break;
case ST_CASCADE:
// dependent on view fustrum and cascade level
mat4fidentity (proj);
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
mmulf (view, qfv_z_up, view);
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
for (int j = 0; j < num_cascade; j++) {
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
mmulf (lm[j], proj, view);
}
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
break;
case ST_PLANE:
[vulkan] Make near and far clip explicit parameters This improves the projection API in that near clip is a parameter rather than being taken directly from the cvar, and a far clip (ie, finite far plane) version is available (necessary for cascaded shadow maps as it's rather hard to fit a box to an infinite frustum). Also, the orthographic projection matrix is now reversed as per the perspective matrix (and the code tidied up a little), and a version that takes min and max vectors is available.
2023-08-13 08:25:17 +00:00
QFV_PerspectiveCos (proj, -light->direction[3], lnearclip);
[vulkan] Compute correct matrices for spots and points Directional lights don't get correct matrices yet as I need to study the math involved for cascaded shadow maps (and id maps don't have directional lights). Getting spotlights working correctly was insanely frustrating: I just couldn't get the entities into the view of the spotlight with any sensible combination of inverses and the z_up matrix. It turned out it was all due to an incorrect reference vector: it was +Z instead of +X.
2023-07-30 02:59:47 +00:00
mmulf (view, qfv_z_up, view);
mmulf (lm[0], proj, view);
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
break;
}
}
}
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
static void
upload_light_matrices (lightingctx_t *lctx, vulkan_ctx_t *ctx)
{
auto packet = QFV_PacketAcquire (ctx->staging);
size_t mat_size = sizeof (mat4f_t[lctx->light_mats.size]);
void *mat_data = QFV_PacketExtend (packet, mat_size);
memcpy (mat_data, lctx->light_mats.a, mat_size);
auto bb = &bufferBarriers[qfv_BB_TransferWrite_to_UniformRead];
for (size_t i = 0; i < lctx->frames.size; i++) {
auto lframe = &lctx->frames.a[i];
QFV_PacketCopyBuffer (packet, lframe->shadowmat_buffer, 0, bb);
}
QFV_PacketSubmit (packet);
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
packet = QFV_PacketAcquire (ctx->staging);
size_t id_size = sizeof (uint32_t[MaxLights * 6]);
uint32_t *id_data = QFV_PacketExtend (packet, id_size);
[vulkan] Handle spotlights pointing -X The problem didn't show up until the switch to float depth buffers (because they can store the resulting nan).
2023-12-19 08:56:13 +00:00
memset (id_data, -1, id_size);
[vulkan] Add a level of indirection to shadow matrices Batching shadow map rendering needs be able to reference matrices for multiple lights in a single batch, but the only input is the view index, so use that to look up the matrix index rather than using it to index the matrices directly (modulo the base index that's still there).
2023-12-14 01:01:27 +00:00
for (size_t i = 0; i < lctx->frames.size; i++) {
auto lframe = &lctx->frames.a[i];
QFV_PacketCopyBuffer (packet, lframe->shadowmat_id_buffer, 0, bb);
}
QFV_PacketSubmit (packet);
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
}
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
static void
upload_light_data (lightingctx_t *lctx, vulkan_ctx_t *ctx)
{
auto reg = lctx->scene->reg;
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
auto light_pool = &reg->comp_pools[lctx->scene->base + scene_light];
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto lights = (light_t *) light_pool->data;
uint32_t count = light_pool->count;
auto packet = QFV_PacketAcquire (ctx->staging);
auto light_data = QFV_PacketExtend (packet, sizeof (light_t[count]));
memcpy (light_data, lights, sizeof (light_t[count]));
auto bb = &bufferBarriers[qfv_BB_TransferWrite_to_UniformRead];
for (size_t i = 0; i < lctx->frames.size; i++) {
auto lframe = &lctx->frames.a[i];
QFV_PacketCopyBuffer (packet, lframe->light_buffer, 0, bb);
}
QFV_PacketSubmit (packet);
packet = QFV_PacketAcquire (ctx->staging);
uint32_t r_size = sizeof (qfv_light_render_t[count]);
qfv_light_render_t *render = QFV_PacketExtend (packet, r_size);
for (uint32_t i = 0; i < count; i++) {
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
entity_t ent = { .reg = reg, .id = light_pool->dense[i],
.base = lctx->scene->base };
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
uint32_t id = get_lightid (ent);
[vulkan] Avoid drawing too many dynamic lights Trying to draw too many dynamic lights results in various crashes due to the resources being preallocated.
2023-08-04 16:25:26 +00:00
if (id >= lctx->light_control.size) {
continue;
}
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto r = &lctx->light_control.a[id];
render[i] = (qfv_light_render_t) {
[vulkan] Clean up make_id() Passing the control struct and a flag for style/nostyle makes the calls much easier to read.
2023-12-17 04:14:43 +00:00
.id_data = make_id(r, style_enable),
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.style = get_lightstyle (ent),
};
}
for (size_t i = 0; i < lctx->frames.size; i++) {
auto lframe = &lctx->frames.a[i];
QFV_PacketCopyBuffer (packet, lframe->render_buffer, 0, bb);
}
QFV_PacketSubmit (packet);
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
static int
light_shadow_type (const light_t *light)
{
if (!light->position[3]) {
if (!VectorIsZero (light->direction)) {
return ST_CASCADE;
}
} else {
if (light->direction[3] > -0.5) {
return ST_CUBE;
} else {
return ST_PLANE;
}
}
return ST_NONE;
}
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
static int
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
light_compare (const void *_li2, const void *_li1, void *_lights)
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
{
[vulkan] Resurrect the shadow map resource creation However, this time it doesn't modify the light array when it sorts the lights by size since the lights are now located before the renderer gets to see them, and having the fix up the light leafs array would be too painful (and probably the completely wrong thing to do anyway: the light array should be treated as constant by the renderer). 1.6GB of memory for gmsp3v2's lights (a little better than marcher: more smaller lights?). For reference: gmsp3v2: shadow maps: 8330 layers in 29 images: 1647706112 marcher: shadow maps: 2440 layers in 11 images: 2358575104
2022-05-10 02:15:52 +00:00
const int *li1 = _li1;
const int *li2 = _li2;
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
const light_t *lights = _lights;
const light_t *l1 = &lights[*li1];
const light_t *l2 = &lights[*li2];
[vulkan] Use absolute light intensity/radius for size Many modders use negative lights for interesting effects, but vulkan doesn't like the result of a negative int treated as unsigned when it comes to texture sizes.
2022-05-11 04:29:17 +00:00
int s1 = abs ((int) l1->color[3]);
int s2 = abs ((int) l2->color[3]);
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
[vulkan] Use absolute light intensity/radius for size Many modders use negative lights for interesting effects, but vulkan doesn't like the result of a negative int treated as unsigned when it comes to texture sizes.
2022-05-11 04:29:17 +00:00
if (s1 == s2) {
[vulkan] Add some comments to lighting code
2023-12-12 12:13:14 +00:00
// same size
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
if (l1->position[3] == l2->position[3]) {
[vulkan] Add some comments to lighting code
2023-12-12 12:13:14 +00:00
// same "type" (point/spot vs directional)
// sort by spot size (1 for point/directional)
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
return (l2->direction[3] > -0.5) - (l1->direction[3] > -0.5);
}
[vulkan] Add some comments to lighting code
2023-12-12 12:13:14 +00:00
// sort by "type" (point/spot vs directional)
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
return l2->position[3] - l1->position[3];
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
[vulkan] Add some comments to lighting code
2023-12-12 12:13:14 +00:00
// sort by size
[vulkan] Use absolute light intensity/radius for size Many modders use negative lights for interesting effects, but vulkan doesn't like the result of a negative int treated as unsigned when it comes to texture sizes.
2022-05-11 04:29:17 +00:00
return s1 - s2;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
typedef struct {
int size;
int layers;
int cube;
} mapdesc_t;
typedef struct {
mapdesc_t *maps;
int numMaps;
int numLights;
const light_t *lights;
int *imageMap;
const int *lightMap;
light_control_t *control;
int maxLayers;
} mapctx_t;
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
static int
allocate_map (mapctx_t *mctx, int type, int (*getsize) (const light_t *light))
{
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
int size = -1;
int numLayers = 0;
int totalLayers = 0;
[vulkan] Implement CSM rendering This covers only the rendering of the shadow maps (actual use still needs to be implemented). Working with orthographic projection matrices is surprisingly difficult, partly because creating one includes the translations needed to get the scene into the view (and depth range), which means care needs to be taken with the view (camera) matrix in order to avoid double-translating depending on just how the orthographic matrix is set up (if it's set up to focus on the origin, then the camera matrix will need translation, otherwise the camera matrix needs to avoid translation).
2023-08-13 08:36:32 +00:00
int layers = ((int[ST_COUNT]) { 0, 1, num_cascade, 6 })[type];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
int cube = type == ST_CUBE;
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
for (int i = 0; i < mctx->numLights; i++) {
auto li = mctx->lightMap[i];
auto lr = &mctx->control[li];
[vulkan] Rework lighting model to be more algebraic This leaves only the one conditional in the shader code, that being the distance check. It doesn't seem to make any noticeable difference to performance, but other than explosion sprites being blue, lighting quality seems to have improved. However, I really need to get shadows working: marcher is just silly-bright without them, and light levels changing as I move around is a bit disconcerting (but reasonable as those lights' leaf nodes go in and out of visibility).
2022-05-04 23:40:02 +00:00
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
if (lr->mode != type) {
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
continue;
}
[vulkan] Round shadow map sizes correctly I'd added some developer output to see how the layers were distributed between images and found the image widths to be... odd. It turns out I was double-adding the shadow_quanta. Oops. Results in ~164MB less memory used by marcher (for 32 pixel quanta).
2023-12-12 15:03:25 +00:00
int light_size = getsize (&mctx->lights[li]);
[vulkan] Apply normal light rules to dynamic lights ie, enforce shadow quanta and max light size.
2023-12-14 03:43:37 +00:00
light_size = round_light_size (light_size);
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
if (size != light_size || numLayers + layers > mctx->maxLayers) {
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
if (numLayers) {
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
mctx->maps[mctx->numMaps++] = (mapdesc_t) {
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
.size = size,
.layers = numLayers,
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
.cube = cube,
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
};
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
numLayers = 0;
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
size = light_size;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
mctx->imageMap[li] = mctx->numMaps;
[vulkan] Hook up the shadow map render pass There are some issues with the light renderers getting mangled, and only the first light is even touched (just begin and end of render pass), but this gets a lot of the framework into place.
2022-05-31 03:43:04 +00:00
lr->size = size;
lr->layer = numLayers;
lr->numLayers = layers;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
numLayers += layers;
totalLayers += layers;
}
if (numLayers) {
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
mctx->maps[mctx->numMaps++] = (mapdesc_t) {
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
.size = size,
.layers = numLayers,
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
.cube = cube,
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
};
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
return totalLayers;
}
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
static int
allocate_dynlight_map (mapctx_t *mctx)
{
int size = -1;
int numLayers = 0;
int totalLayers = 0;
int layers = 6;
int cube = 1;
for (int i = 0; i < dynlight_max; i++) {
if (size != dynlight_size || numLayers + layers > mctx->maxLayers) {
if (numLayers) {
mctx->maps[mctx->numMaps++] = (mapdesc_t) {
.size = size,
.layers = numLayers,
.cube = cube,
};
numLayers = 0;
}
size = dynlight_size;
}
auto li = mctx->numLights + i;
auto lr = &mctx->control[li];
*lr = (light_control_t) {
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
.stage_index = (size / shadow_quanta) - 1,
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
.map_index = mctx->numMaps,
.size = size,
.layer = numLayers,
.numLayers = layers,
.mode = ST_CUBE,
.light_id = li,
};
numLayers += layers;
totalLayers += layers;
}
if (numLayers) {
mctx->maps[mctx->numMaps++] = (mapdesc_t) {
.size = size,
.layers = numLayers,
.cube = cube,
};
}
return totalLayers;
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
static int
get_point_size (const light_t *light)
{
return abs ((int) light->color[3]);
}
static int
get_spot_size (const light_t *light)
{
float c = light->direction[3];
float s = sqrt (1 - c * c);
return abs ((int) (s * light->color[3]));
}
static int
get_direct_size (const light_t *light)
{
return 1024;
}
static void
build_shadow_maps (lightingctx_t *lctx, vulkan_ctx_t *ctx)
{
qfv_device_t *device = ctx->device;
qfv_physdev_t *physDev = device->physDev;
[vulkan] Switch to vulkan 1.1/1.2 prop/feat structs Since switching to the 1.2 api as a requirement, might as well use the relevant structs instead of extension struct (for multiview). Came up when double-checking the max views property due to running into what appears to be an nvidia bug where > 29 views (any bit pattern) cause a segfault when creating the pipeline.
2023-12-13 11:25:23 +00:00
int maxLayers = physDev->p.properties.limits.maxImageArrayLayers;
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
if (maxLayers > 2048) {
maxLayers = 2048;
}
auto reg = lctx->scene->reg;
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
auto light_pool = &reg->comp_pools[lctx->scene->base + scene_light];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
auto lights = (light_t *) light_pool->data;
int numLights = light_pool->count;
int totalLayers = 0;
int imageMap[numLights];
int lightMap[numLights];
[vulkan] Allocate map spaces for dynamic lights I'm actually surprised anything worked, though I guess it was just the one entry getting corrupted (and not 32, but I figured allocate slots for all of the dynamic lights just in case). Or none, really, since larger scenes (ie, those with multiple lights that fit in the same image size) would result in not all the maps getting used and thus one spare for dynamic lights.
2023-08-16 01:52:31 +00:00
mapdesc_t maps[numLights + dynlight_max];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
for (int i = 0; i < numLights; i++) {
lightMap[i] = i;
}
[vulkan] Add some comments to lighting code
2023-12-12 12:13:14 +00:00
// sort lights by size, type, spot-size
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
heapsort_r (lightMap, numLights, sizeof (int), light_compare, lights);
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
DARRAY_RESIZE (&lctx->light_control, numLights + dynlight_max);
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
for (int i = 0; i < numLights; i++) {
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
auto li = lightMap[i];
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto lr = &lctx->light_control.a[li];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
*lr = (light_control_t) {
.mode = light_shadow_type (&lights[li]),
.light_id = li,
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
};
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
entity_t ent = { .reg = reg, .id = light_pool->dense[li],
.base = lctx->scene->base };
set_lightid (ent, li);
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
// assume all lights have no shadows
imageMap[li] = -1;
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
mapctx_t mctx = {
.maps = maps,
.numLights = numLights,
.lights = lights,
.imageMap = imageMap,
.lightMap = lightMap,
.control = lctx->light_control.a,
.maxLayers = maxLayers,
};
totalLayers += allocate_map (&mctx, ST_CUBE, get_point_size);
totalLayers += allocate_map (&mctx, ST_PLANE, get_spot_size);
totalLayers += allocate_map (&mctx, ST_CASCADE, get_direct_size);
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
totalLayers += allocate_dynlight_map (&mctx);
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
lctx->num_maps = mctx.numMaps;
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
int stage_layers[LIGHTING_STAGES] = {};
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
if (mctx.numMaps) {
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
for (int i = 0; i < mctx.numMaps; i++) {
int ind = (maps[i].size / shadow_quanta) - 1;
stage_layers[ind] += maps[i].layers;
}
int stage_maps = 0;
for (int i = 0; i < LIGHTING_STAGES; i++) {
stage_layers[i] = min (stage_layers[i], max_views);
if (stage_layers[i]) {
stage_maps++;
}
}
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
qfv_resource_t *shad = calloc (1, sizeof (qfv_resource_t)
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
+ sizeof (qfv_resobj_t[mctx.numMaps])
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
+ sizeof (qfv_resobj_t[mctx.numMaps])
+ sizeof (qfv_resobj_t[stage_maps])
+ sizeof (qfv_resobj_t[stage_maps]));
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
lctx->shadow_resources = shad;
*shad = (qfv_resource_t) {
.name = "shadow",
.va_ctx = ctx->va_ctx,
.memory_properties = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
.num_objects = 2 * (mctx.numMaps + stage_maps),
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
.objects = (qfv_resobj_t *) &shad[1],
};
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
lctx->map_images = malloc (sizeof (VkImage[mctx.numMaps]));
lctx->map_views = malloc (sizeof (VkImageView[mctx.numMaps]));
lctx->map_cube = malloc (sizeof (bool[mctx.numMaps]));
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto images = shad->objects;
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
auto stage_images = &images[mctx.numMaps];
auto views = &stage_images[stage_maps];
auto stage_views = &views[mctx.numMaps];
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
for (int i = 0; i < mctx.numMaps; i++) {
int cube = maps[i].cube;
lctx->map_cube[i] = cube;
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
images[i] = (qfv_resobj_t) {
[vulkan] Up max lights to 2048 and quantize sizes This seems excessive, but gmsp3v2 map has 1399 lights. Worse, it has a lot of different light sizes that go up by small increments (generally around 10) resulting in 33 shadow map images (1 too many). Quantizing the sizes to 32 drops this nicely to 20, and reduces memory consumption slightly too (image buffer overhead, I guess).
2023-08-15 05:34:23 +00:00
.name = va (ctx->va_ctx, "map:image:%02d:%d", i, maps[i].size),
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
.type = qfv_res_image,
.image = {
[vulkan] Complete resource image and image view I don't remember why I kept the abbreviated configs for images and image views, but it because such that I need to be able to specify them completely. In addition, image views support external images. The rest was just cleaning up after the changes to qfv_resobj_t.
2023-02-19 03:25:13 +00:00
.flags = cube ? VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT : 0,
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
.type = VK_IMAGE_TYPE_2D,
[vulkan] Switch to float depth buffers This takes full advantage of the reversed (infinite far plane) depth buffers, and seems to have no performance cost.
2023-12-18 14:15:21 +00:00
.format = VK_FORMAT_D32_SFLOAT,
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
.extent = { maps[i].size, maps[i].size, 1 },
.num_mipmaps = 1,
.num_layers = maps[i].layers,
.samples = VK_SAMPLE_COUNT_1_BIT,
.usage = VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
| VK_IMAGE_USAGE_TRANSFER_DST_BIT
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
| VK_IMAGE_USAGE_SAMPLED_BIT,
},
};
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
views[i] = (qfv_resobj_t) {
[vulkan] Up max lights to 2048 and quantize sizes This seems excessive, but gmsp3v2 map has 1399 lights. Worse, it has a lot of different light sizes that go up by small increments (generally around 10) resulting in 33 shadow map images (1 too many). Quantizing the sizes to 32 drops this nicely to 20, and reduces memory consumption slightly too (image buffer overhead, I guess).
2023-08-15 05:34:23 +00:00
.name = va (ctx->va_ctx, "map:view:%02d:%d", i, maps[i].size),
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.type = qfv_res_image_view,
.image_view = {
.image = i,
.type = cube ? VK_IMAGE_VIEW_TYPE_CUBE_ARRAY
: VK_IMAGE_VIEW_TYPE_2D_ARRAY,
[vulkan] Switch to float depth buffers This takes full advantage of the reversed (infinite far plane) depth buffers, and seems to have no performance cost.
2023-12-18 14:15:21 +00:00
.format = VK_FORMAT_D32_SFLOAT,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.subresourceRange = {
.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
.levelCount = VK_REMAINING_MIP_LEVELS,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
},
};
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
for (int i = 0, ind = 0; i < LIGHTING_STAGES; i++) {
if (!stage_layers[i]) {
continue;
}
int size = (i + 1) * shadow_quanta;
stage_images[ind] = (qfv_resobj_t) {
.name = va (ctx->va_ctx, "stage_map:image:%02d:%d", ind, size),
.type = qfv_res_image,
.image = {
.type = VK_IMAGE_TYPE_2D,
[vulkan] Switch to float depth buffers This takes full advantage of the reversed (infinite far plane) depth buffers, and seems to have no performance cost.
2023-12-18 14:15:21 +00:00
.format = VK_FORMAT_D32_SFLOAT,
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
.extent = { size, size, 1 },
.num_mipmaps = 1,
.num_layers = stage_layers[i],
.samples = VK_SAMPLE_COUNT_1_BIT,
.usage = VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT
| VK_IMAGE_USAGE_TRANSFER_SRC_BIT,
},
};
stage_views[ind] = (qfv_resobj_t) {
.name = va (ctx->va_ctx, "stage_map:view:%02d:%d", ind, size),
.type = qfv_res_image_view,
.image_view = {
.image = mctx.numMaps + ind,
.type = VK_IMAGE_VIEW_TYPE_2D_ARRAY,
[vulkan] Switch to float depth buffers This takes full advantage of the reversed (infinite far plane) depth buffers, and seems to have no performance cost.
2023-12-18 14:15:21 +00:00
.format = VK_FORMAT_D32_SFLOAT,
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
.subresourceRange = {
.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT,
.levelCount = VK_REMAINING_MIP_LEVELS,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
},
};
ind++;
}
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
QFV_CreateResource (device, shad);
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
for (int i = 0; i < mctx.numMaps; i++) {
lctx->map_images[i] = images[i].image.image;
lctx->map_views[i] = views[i].image_view.view;
[vulkan] Use the resource functions for shadow maps Another step towards moving all resource creation into the one place. The motivation for doing the change was getting my test scene to work with only ambient lights or no lights at all.
2022-10-27 02:02:45 +00:00
}
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
for (int i = 0, ind = 0; i < LIGHTING_STAGES; i++) {
if (!stage_layers[i]) {
continue;
}
lctx->stage_images[i] = stage_images[ind].image.image;
lctx->stage_views[i] = stage_views[ind].image_view.view;
ind++;
}
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
for (int i = 0; i < numLights; i++) {
[vulkan] Hook up a shadow render job step It doesn't do much yet, but did help in getting light ids working.
2023-07-24 01:23:05 +00:00
int li = lightMap[i];
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
auto lr = &lctx->light_control.a[li];
[vulkan] Resurrect the shadow map resource creation However, this time it doesn't modify the light array when it sorts the lights by size since the lights are now located before the renderer gets to see them, and having the fix up the light leafs array would be too painful (and probably the completely wrong thing to do anyway: the light array should be treated as constant by the renderer). 1.6GB of memory for gmsp3v2's lights (a little better than marcher: more smaller lights?). For reference: gmsp3v2: shadow maps: 8330 layers in 29 images: 1647706112 marcher: shadow maps: 2440 layers in 11 images: 2358575104
2022-05-10 02:15:52 +00:00
if (imageMap[li] == -1) {
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
continue;
}
[vulkan] Rework lighting model to be more algebraic This leaves only the one conditional in the shader code, that being the distance check. It doesn't seem to make any noticeable difference to performance, but other than explosion sprites being blue, lighting quality seems to have improved. However, I really need to get shadows working: marcher is just silly-bright without them, and light levels changing as I move around is a bit disconcerting (but reasonable as those lights' leaf nodes go in and out of visibility).
2022-05-04 23:40:02 +00:00
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
lr->stage_index = -1;
if (lr->numLayers) {
lr->stage_index = (lr->size / shadow_quanta) - 1;
[vulkan] Rework lighting model to be more algebraic This leaves only the one conditional in the shader code, that being the distance check. It doesn't seem to make any noticeable difference to performance, but other than explosion sprites being blue, lighting quality seems to have improved. However, I really need to get shadows working: marcher is just silly-bright without them, and light levels changing as I move around is a bit disconcerting (but reasonable as those lights' leaf nodes go in and out of visibility).
2022-05-04 23:40:02 +00:00
}
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
lr->map_index = imageMap[li];
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
Sys_MaskPrintf (SYS_lighting,
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
"shadow maps: %d layers in %d images: %"PRId64"\n",
totalLayers, lctx->num_maps,
lctx->shadow_resources ? lctx->shadow_resources->size
: (VkDeviceSize) 0);
[vulkan] Round shadow map sizes correctly I'd added some developer output to see how the layers were distributed between images and found the image widths to be... odd. It turns out I was double-adding the shadow_quanta. Oops. Results in ~164MB less memory used by marcher (for 32 pixel quanta).
2023-12-12 15:03:25 +00:00
if (developer & SYS_lighting) {
auto images = lctx->shadow_resources->objects;
for (int i = 0; i < lctx->num_maps; i++) {
[vulkan] Implement staged shadow maps This doesn't make much of a difference on the GPU, but it drastically cuts down CPU usage, especially for ad_tears: shadow map drawing is down from 16.3ms to 3.7ms thanks to no having to run the alias model queues as often.
2023-12-14 12:40:41 +00:00
Sys_Printf ("map id:%d width:%d layers:%d\n", i,
images[i].image.extent.width,
images[i].image.num_layers);
}
for (int i = 0; i < LIGHTING_STAGES; i++) {
Sys_Printf ("stage:%d width:%d layers:%d\n", i,
(i + 1) * shadow_quanta, stage_layers[i]);
[vulkan] Round shadow map sizes correctly I'd added some developer output to see how the layers were distributed between images and found the image widths to be... odd. It turns out I was double-adding the shadow_quanta. Oops. Results in ~164MB less memory used by marcher (for 32 pixel quanta).
2023-12-12 15:03:25 +00:00
}
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
}
static void
transition_shadow_maps (lightingctx_t *lctx, vulkan_ctx_t *ctx)
{
auto device = ctx->device;
auto dfunc = device->funcs;
VkCommandBufferAllocateInfo aInfo = {
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO,
.commandPool = ctx->cmdpool,
.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY,
.commandBufferCount = 1,
};
VkCommandBuffer cmd;
dfunc->vkAllocateCommandBuffers (device->dev, &aInfo, &cmd);
VkCommandBufferBeginInfo bInfo = {
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
};
dfunc->vkBeginCommandBuffer (cmd, &bInfo);
auto ib = imageBarriers[qfv_LT_Undefined_to_ShaderReadOnly];
ib.barrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
ib.barrier.subresourceRange.levelCount = VK_REMAINING_MIP_LEVELS;
ib.barrier.subresourceRange.layerCount = VK_REMAINING_ARRAY_LAYERS;
VkImageMemoryBarrier barriers[lctx->num_maps];
for (int i = 0; i < lctx->num_maps; i++) {
barriers[i] = ib.barrier;
barriers[i].image = lctx->map_images[i];
}
dfunc->vkCmdPipelineBarrier (cmd, ib.srcStages, ib.dstStages,
0, 0, 0, 0, 0, lctx->num_maps, barriers);
dfunc->vkEndCommandBuffer (cmd);
VkSubmitInfo submitInfo = {
.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
.commandBufferCount = 1,
.pCommandBuffers = &cmd,
};
dfunc->vkQueueSubmit (device->queue.queue, 1, &submitInfo, 0);
[vulkan] Create shadow maps and basic data I will definitely need to look into sparse images later (2.3GB for marcher) but the shadow map images, views and matrices are created.
2021-04-29 11:54:38 +00:00
}
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
static void
update_shadow_descriptors (lightingctx_t *lctx, vulkan_ctx_t *ctx)
{
auto device = ctx->device;
auto dfunc = device->funcs;
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
VkDescriptorImageInfo imageInfoCube[32];
VkDescriptorImageInfo imageInfo2d[32];
for (int i = 0; i < 32; i++) {
VkImageView viewCube = lctx->default_view_cube;
VkImageView view2d = lctx->default_view_2d;
if (i < lctx->num_maps) {
if (lctx->map_cube[i]) {
viewCube = lctx->map_views[i];
} else {
view2d = lctx->map_views[i];
}
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
imageInfoCube[i] = (VkDescriptorImageInfo) {
.sampler = lctx->shadow_sampler,
.imageView = viewCube,
.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL,
};
imageInfo2d[i] = (VkDescriptorImageInfo) {
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.sampler = lctx->shadow_sampler,
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
.imageView = view2d,
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL,
};
}
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
VkWriteDescriptorSet imageWrite[2] = {
{
.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lctx->shadow_cube_set,
.dstBinding = 0,
.descriptorCount = 32,
.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
.pImageInfo = imageInfoCube,
},
{
.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.dstSet = lctx->shadow_2d_set,
.dstBinding = 0,
.descriptorCount = 32,
.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
.pImageInfo = imageInfo2d,
},
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
};
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
dfunc->vkUpdateDescriptorSets (device->dev, 2, imageWrite, 0, 0);
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
}
[vulkan] Add more light debug info Lines are drawn for a light's leaf, the leafs visible to it, or those in its efrags chain. Still no idea why lights are drawing when they shouldn't. Deek suggest holes in the map, but I think if that was the case, there'd be something visible. My suspicion is I'm doing something wrong in with efrags.
2023-08-10 00:38:42 +00:00
static void
mark_leaves (bsp_pass_t *pass, set_t *pvs)
{
visstate_t visstate = {
.node_visframes = pass->node_frames,
.leaf_visframes = pass->leaf_frames,
.face_visframes = pass->face_frames,
.visframecount = pass->vis_frame,
.brush = pass->brush,
};
R_MarkLeavesPVS (&visstate, pvs);
pass->vis_frame = visstate.visframecount;
}
[vulkan] Show surfaces for selected light leaf This has resulted in some rather interesting information: it seems the surfaces (and thus, presumably bounding boxes) for leafs have little to do with the actual leaf node's volume.
2023-08-08 16:23:13 +00:00
static void
show_leaves (vulkan_ctx_t *ctx, uint32_t leafnum, efrag_t *efrags)
{
auto pass = Vulkan_Bsp_GetPass (ctx, QFV_bspDebug);
auto brush = pass->brush;
set_t pvs = SET_STATIC_INIT (brush->visleafs, alloca);
set_empty (&pvs);
if (leafnum) {
set_add (&pvs, leafnum - 1);
[vulkan] Add more light debug info Lines are drawn for a light's leaf, the leafs visible to it, or those in its efrags chain. Still no idea why lights are drawing when they shouldn't. Deek suggest holes in the map, but I think if that was the case, there'd be something visible. My suspicion is I'm doing something wrong in with efrags.
2023-08-10 00:38:42 +00:00
} else {
for (auto e = efrags; e; e = e->entnext) {
set_add (&pvs, e->leaf - brush->leafs - 1);
}
[vulkan] Show surfaces for selected light leaf This has resulted in some rather interesting information: it seems the surfaces (and thus, presumably bounding boxes) for leafs have little to do with the actual leaf node's volume.
2023-08-08 16:23:13 +00:00
}
[vulkan] Add more light debug info Lines are drawn for a light's leaf, the leafs visible to it, or those in its efrags chain. Still no idea why lights are drawing when they shouldn't. Deek suggest holes in the map, but I think if that was the case, there'd be something visible. My suspicion is I'm doing something wrong in with efrags.
2023-08-10 00:38:42 +00:00
mark_leaves (pass, &pvs);
[vulkan] Show surfaces for selected light leaf This has resulted in some rather interesting information: it seems the surfaces (and thus, presumably bounding boxes) for leafs have little to do with the actual leaf node's volume.
2023-08-08 16:23:13 +00:00
}
Fix most of the hacks for clang gcc didn't like a couple of the changes (rightly so: one was actually incorrect), and the fix for qfcc I didn't think to suggest while working with Emily. The general CFLAGS etc fixes mostly required just getting the order of operations right: check for attributes after setting the warnings flags, though those needed some care for gcc as it began warning about main wanting the const attribute. Fixing the imui link errors required moving the ui functions and setup to vulkan_lighting.c, which is really the only place they're used.
2023-08-11 09:05:45 +00:00
static void
light_dyn_light_ui (void *comp, imui_ctx_t *imui_ctx,
ecs_registry_t *reg, uint32_t ent, void *data)
{
dlight_t *dlight = comp;
UI_Horizontal {
UI_Labelf ("Origin: ");
UI_FlexibleSpace ();
UI_Labelf ("%6.1f %6.1f %6.1f %6g", VEC4_EXP (dlight->origin));
}
UI_Horizontal {
UI_Label ("Color: ");
UI_FlexibleSpace ();
UI_Labelf ("%5.3f %5.3f %5.3f %5.3f", VEC4_EXP (dlight->color));
}
UI_Horizontal {
UI_Labelf ("Radius: ");
UI_FlexibleSpace ();
UI_Labelf ("%6g", dlight->radius);
}
UI_Horizontal {
UI_Labelf ("Die: ");
UI_FlexibleSpace ();
UI_Labelf ("%6.2f", dlight->die);
}
UI_Horizontal {
UI_Labelf ("Decay: ");
UI_FlexibleSpace ();
UI_Labelf ("%6.2f", dlight->decay);
}
UI_Horizontal {
UI_Labelf ("Min Light: ");
UI_FlexibleSpace ();
UI_Labelf ("%6g", dlight->minlight);
}
}
static void
light_light_ui (void *comp, imui_ctx_t *imui_ctx,
ecs_registry_t *reg, uint32_t ent, void *data)
{
light_t *light = comp;
UI_Horizontal {
UI_Label ("Color: ");
UI_FlexibleSpace ();
UI_Labelf ("%5.3f %5.3f %5.3f %3g", VEC4_EXP (light->color));
}
UI_Horizontal {
UI_Labelf ("Position: ");
UI_FlexibleSpace ();
UI_Labelf ("%6.1f %6.1f %6.1f %6g", VEC4_EXP (light->position));
}
UI_Horizontal {
UI_Labelf ("Direction: ");
UI_FlexibleSpace ();
UI_Labelf ("%6.3f %6.3f %6.3f %6.3g", VEC4_EXP (light->direction));
}
UI_Horizontal {
UI_Labelf ("Attenuation: ");
UI_FlexibleSpace ();
UI_Labelf ("%g %g %g %g", VEC4_EXP (light->attenuation));
}
}
[vulkan] Show surfaces for selected light leaf This has resulted in some rather interesting information: it seems the surfaces (and thus, presumably bounding boxes) for leafs have little to do with the actual leaf node's volume.
2023-08-08 16:23:13 +00:00
static void
scene_efrags_ui (void *comp, imui_ctx_t *imui_ctx,
ecs_registry_t *reg, uint32_t ent, void *data)
{
[vulkan] Add more light debug info Lines are drawn for a light's leaf, the leafs visible to it, or those in its efrags chain. Still no idea why lights are drawing when they shouldn't. Deek suggest holes in the map, but I think if that was the case, there'd be something visible. My suspicion is I'm doing something wrong in with efrags.
2023-08-10 00:38:42 +00:00
vulkan_ctx_t *ctx = data;
[vulkan] Show surfaces for selected light leaf This has resulted in some rather interesting information: it seems the surfaces (and thus, presumably bounding boxes) for leafs have little to do with the actual leaf node's volume.
2023-08-08 16:23:13 +00:00
auto efrags = *(efrag_t **) comp;
uint32_t len = 0;
bool valid = true;
for (auto e = efrags; e; e = e->entnext, len++) {
valid &= e->entity.id == ent;
}
UI_Horizontal {
[vulkan] Add more light debug info Lines are drawn for a light's leaf, the leafs visible to it, or those in its efrags chain. Still no idea why lights are drawing when they shouldn't. Deek suggest holes in the map, but I think if that was the case, there'd be something visible. My suspicion is I'm doing something wrong in with efrags.
2023-08-10 00:38:42 +00:00
if (UI_Button (va (ctx->va_ctx, "Show##lightefrags_ui.%08x", ent))) {
show_leaves (ctx, 0, efrags);
}
[vulkan] Show surfaces for selected light leaf This has resulted in some rather interesting information: it seems the surfaces (and thus, presumably bounding boxes) for leafs have little to do with the actual leaf node's volume.
2023-08-08 16:23:13 +00:00
UI_FlexibleSpace ();
UI_Labelf ("%4s %5u", valid ? "good" : "bad", len);
}
}
static void
scene_lightleaf_ui (void *comp, imui_ctx_t *imui_ctx,
ecs_registry_t *reg, uint32_t ent, void *data)
{
vulkan_ctx_t *ctx = data;
auto leaf = *(uint32_t *) comp;
UI_Horizontal {
if (UI_Button (va (ctx->va_ctx, "Show##lightleaf_ui.%08x", ent))) {
show_leaves (ctx, leaf, 0);
}
UI_FlexibleSpace ();
UI_Labelf ("%5u", leaf);
[vulkan] Add more light debug info Lines are drawn for a light's leaf, the leafs visible to it, or those in its efrags chain. Still no idea why lights are drawing when they shouldn't. Deek suggest holes in the map, but I think if that was the case, there'd be something visible. My suspicion is I'm doing something wrong in with efrags.
2023-08-10 00:38:42 +00:00
auto pass = Vulkan_Bsp_GetPass (ctx, QFV_bspDebug);
auto brush = pass->brush;
set_t pvs = SET_STATIC_INIT (brush->visleafs, alloca);
Mod_LeafPVS_set (brush->leafs + leaf, brush, 0, &pvs);
UI_FlexibleSpace ();
if (UI_Button (va (ctx->va_ctx, "Vis##lightleaf_ui.%08x", ent))) {
mark_leaves (pass, &pvs);
}
UI_FlexibleSpace ();
UI_Labelf ("%5u", set_count (&pvs));
[vulkan] Show surfaces for selected light leaf This has resulted in some rather interesting information: it seems the surfaces (and thus, presumably bounding boxes) for leafs have little to do with the actual leaf node's volume.
2023-08-08 16:23:13 +00:00
}
}
[vulkan] Show light style and id in the debug UI Style includes the resulting scalar for the light value.
2023-12-17 02:25:25 +00:00
static void
scene_lightstyle_ui (void *comp, imui_ctx_t *imui_ctx,
ecs_registry_t *reg, uint32_t ent, void *data)
{
auto style = *(uint32_t *) comp;
UI_Horizontal {
UI_Labelf ("%5u", style);
UI_FlexibleSpace ();
auto val = d_lightstylevalue[style];
UI_Labelf ("%3f", val / 65536.0);
}
}
static void
scene_lightid_ui (void *comp, imui_ctx_t *imui_ctx,
ecs_registry_t *reg, uint32_t ent, void *data)
{
auto id = *(uint32_t *) comp;
UI_Horizontal {
UI_Labelf ("%5u", id);
UI_FlexibleSpace ();
}
}
[vulkan] Load the map's lights into an array It's not used yet as work needs to be done to better support generic entities, but this is the next step to real-time lighting (though, to be honest, I expect it will be too slow to be usable).
2021-03-03 09:14:16 +00:00
void
[vulkan] Move non-specific lighting code out of Vulkan The parsing of light data from maps is now in the client library, and basic light management is in scene. Putting the light loading code into the Vulkan renderer was a mistake I've wanted to correct for a while. The client code still needs a bit of cleanup, but the basics are working nicely.
2022-05-05 05:58:47 +00:00
Vulkan_LoadLights (scene_t *scene, vulkan_ctx_t *ctx)
[vulkan] Load the map's lights into an array It's not used yet as work needs to be done to better support generic entities, but this is the next step to real-time lighting (though, to be honest, I expect it will be too slow to be usable).
2021-03-03 09:14:16 +00:00
{
[vulkan] Add gpu profiling zones This relies on my fork of tracy: https://github.com/taniwha/tracy on the wip-c-vulkan branch. Everything is still rather flaky though. This necessitated the jump to vulkan 1.2 as a requirement.
2023-12-04 12:38:43 +00:00
qfZoneScoped (true);
[vulkan] Load the map's lights into an array It's not used yet as work needs to be done to better support generic entities, but this is the next step to real-time lighting (though, to be honest, I expect it will be too slow to be usable).
2021-03-03 09:14:16 +00:00
lightingctx_t *lctx = ctx->lighting_context;
[vulkan] Move non-specific lighting code out of Vulkan The parsing of light data from maps is now in the client library, and basic light management is in scene. Putting the light loading code into the Vulkan renderer was a mistake I've wanted to correct for a while. The client code still needs a bit of cleanup, but the basics are working nicely.
2022-05-05 05:58:47 +00:00
lctx->scene = scene;
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
[vulkan] Resurrect the shadow map resource creation However, this time it doesn't modify the light array when it sorts the lights by size since the lights are now located before the renderer gets to see them, and having the fix up the light leafs array would be too painful (and probably the completely wrong thing to do anyway: the light array should be treated as constant by the renderer). 1.6GB of memory for gmsp3v2's lights (a little better than marcher: more smaller lights?). For reference: gmsp3v2: shadow maps: 8330 layers in 29 images: 1647706112 marcher: shadow maps: 2440 layers in 11 images: 2358575104
2022-05-10 02:15:52 +00:00
clear_shadows (ctx);
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
lctx->ldata = 0;
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
if (lctx->scene) {
auto reg = lctx->scene->reg;
[ecs] Implement hierarchies as components The main goal of this change was to make it easier to tell when a hierarchy has been deleted, but as a side benefit, it got rid of the use of PR_RESMAP. Also, it's easy to track the number of hierarchies. Unfortunately, it showed how brittle the component side of the ECS is (scene and canvas registries assumed their components were the first (no long the case), thus the sweeping changes). Centerprint doesn't work (but it hasn't for a while).
2024-01-01 11:21:07 +00:00
reg->components.a[lctx->scene->base + scene_dynlight].ui = light_dyn_light_ui;
reg->components.a[lctx->scene->base + scene_light].ui = light_light_ui;
reg->components.a[lctx->scene->base + scene_efrags].ui = scene_efrags_ui;
reg->components.a[lctx->scene->base + scene_lightstyle].ui = scene_lightstyle_ui;
reg->components.a[lctx->scene->base + scene_lightleaf].ui = scene_lightleaf_ui;
reg->components.a[lctx->scene->base + scene_lightid].ui = scene_lightid_ui;
auto light_pool = &reg->comp_pools[lctx->scene->base + scene_light];
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
if (light_pool->count) {
[vulkan] Implement shadows for dynamic lights This also fixes the segfault in the previous commit. Dynamic light shadow sizes are fixed, but can be controlled via the dynlight_size cvar (defaults to 250).
2023-08-04 06:44:07 +00:00
lctx->dynamic_base = light_pool->count;
lctx->dynamic_count = 0;
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
build_shadow_maps (lctx, ctx);
[vulkan] Split the lighting pass into per-type passes This takes care of the type punning issue by each pass using the correct sampler type with the correct view types bound. Also, point light and spot light shadow maps are now guaranteed to be separated (it was just luck that they were before) and spot light maps may be significantly smaller as their cone angle is taken into account. Lighting is quite borked, but at least the engine is running again.
2023-08-02 08:52:57 +00:00
transition_shadow_maps (lctx, ctx);
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
update_shadow_descriptors (lctx, ctx);
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
create_light_matrices (lctx);
[vulkan] Upload light matrices for shadow maps This involved rewriting the descriptor update code too, but that now feels cleaner. The matrices are loaded into a storage buffer as it can get quite big at 6 matrices per light (and the current max lights is 768).
2023-07-30 02:28:42 +00:00
upload_light_matrices (lctx, ctx);
[vulkan] Hook up all the shadow resources This gets everything but the actual shadow map bindings working: the validation layers don't like my type punning (which may well be the right thing) and specialization constants don't help (yet, anyway) but I want to get things into git.
2023-08-01 14:34:08 +00:00
upload_light_data (lctx, ctx);
[vulkan] Get shadow map image creation working again The switch to using the ECS for lights temporarily broke the generation of the images, but getting things going again was fairly easy.
2023-07-22 13:41:05 +00:00
}
[vulkan] Connect the shadow pass to the renderers It's only partially complete, but R_VisitWorldNodes is now run for both the main pass and the shadow auxiliary pass.
2023-07-27 07:09:33 +00:00
lctx->ldata = scene->lights;
[vulkan] Resurrect the shadow map resource creation However, this time it doesn't modify the light array when it sorts the lights by size since the lights are now located before the renderer gets to see them, and having the fix up the light leafs array would be too painful (and probably the completely wrong thing to do anyway: the light array should be treated as constant by the renderer). 1.6GB of memory for gmsp3v2's lights (a little better than marcher: more smaller lights?). For reference: gmsp3v2: shadow maps: 8330 layers in 29 images: 1647706112 marcher: shadow maps: 2440 layers in 11 images: 2358575104
2022-05-10 02:15:52 +00:00
}
[vulkan] Load the map's lights into an array It's not used yet as work needs to be done to better support generic entities, but this is the next step to real-time lighting (though, to be honest, I expect it will be too slow to be usable).
2021-03-03 09:14:16 +00:00
}
[vulkan] Use the shadow matrices when rendering maps It turns out bsp faces are still back-face culled despite the null point being on the front of every possible plane... or really, because it's on the front of every possible plane: sometimes the back face is the front face, and this breaks the face selection code (a separate traversal function will be needed for non-culling rendering). Despite that, other than having to deal with different pipelines, getting the model renderers working went better than expected.
2023-07-30 02:52:13 +00:00
VkDescriptorSet
Vulkan_Lighting_Descriptors (vulkan_ctx_t *ctx, int frame)
{
auto lctx = ctx->lighting_context;
return lctx->frames.a[frame].shadowmat_set;
}
Reference in a new issue Copy permalink