9ed5dcdc27
GPU: Use inversesqrt for light distance and half-vector calculations
...
Replace length()/normalize() with inversesqrt() pattern for better
performance on mobile GPUs. The inversesqrt instruction is a single
hardware operation, while length() requires sqrt(dot()) and normalize()
requires inversesqrt internally plus a multiply.
Changes:
1. Light distance calculation:
Before: distance = length(toLight); toLight /= distance;
After: distSq = dot(toLight, toLight);
invDist = inversesqrt(distSq);
distance = distSq * invDist;
toLight *= invDist;
2. Specular half-vector normalization:
Before: ldot = dot(normalize(toLight + vec3(0,0,1)), worldnormal);
After: halfVec = toLight + vec3(0,0,1);
halfInvLen = inversesqrt(dot(halfVec, halfVec));
ldot = dot(halfVec, worldnormal) * halfInvLen;
Also reuses distSq in attenuation calculation to avoid recomputing
distance*distance.
Offline shader compiler results (vertex shader with 4-light ubershader):
Mali (malioc):
G31 (Bifrost): 44.90 -> 40.50 cycles (-9.8%)
G52 (Bifrost): 14.97 -> 13.50 cycles (-9.8%)
G310 (Valhall): 13.88 -> 12.88 cycles (-7.2%)
G57 (Valhall): 7.60 -> 6.97 cycles (-8.3%)
PowerVR (Profiling Compilers):
Series 8 GE8300: 322 -> 302 ALU cycles (-6.2%)
Series 9 GM9445: 296 -> 280 ALU cycles (-5.4%)
Series B BXM: 560 -> 540 instructions (-3.6%)
Signed-off-by: Chris Healy <cphealy@gmail.com >
2026-03-08 19:13:41 -07:00
6cfea96e58
Remove D3D9 support, to make future changes easier
2025-06-10 15:07:16 +02:00
df6ed8cfc9
Do some cleanup of #includes in GPU
2024-12-18 13:57:26 +01:00
702e77454a
Hardware transform: Clamp the specular coefficient to 0.0 before taking the power.
2024-09-25 21:09:13 +02:00
3154a5fac1
OpenXR - VR camera features on any platform
2024-08-07 09:25:12 +02:00
fab034d866
Fix minor difference between ubershader and specialized vshaders
2024-07-24 20:50:33 +02:00
a9b5a4739e
Fix math of simple stereo
2024-06-25 16:06:03 +02:00
27152b1e47
OpenXR - Forgotten usages
2024-06-24 19:43:17 +02:00
dd9bc42b0a
OpenXR - Remove multiview from shader
2024-06-24 19:24:46 +02:00
7b089effa0
OpenXR - Use single VR projection matrix
2024-06-03 18:29:06 +02:00
e3177ac870
Make some global string pointers const, not just the strings.
...
Minor cleanup.
2023-12-29 14:09:45 +01:00
0f3dca5cea
Revert a bad "optimization" I made in #18587
2023-12-25 12:03:09 +01:00
2238fab601
Minor lighting optimization in vshader
2023-12-20 14:47:49 +01:00
0748ce610f
[GPU/Common/D3D11/Directx9/GLES/Vulkan] Using reserve if possible
2023-12-15 14:08:22 +03:00
b42670cf59
Manually revert the rest of the lmode optimization
2023-05-09 18:44:24 +02:00
0babac5c5f
Manually revert "Remove LM flag from vshader too" ( #16718 )
...
Reverts #16718
Though maybe we can keep this one? Let's see.
2023-05-09 18:44:18 +02:00
f42c682d34
Revert "Merge pull request #16628 from hrydgard/remove-fog-fshader-flag"
...
This reverts commit 10dee90c8334c11c8acf
2023-05-08 22:01:38 +02:00
9fcc1509e4
GPU: Correct depth clip/cull for zero scale.
2023-03-05 08:51:45 -08:00
0f3f2e361f
GLES: More uint const paranoia.
...
We don't use indexing on GLES now, but better to stay consistent.
2023-01-12 17:41:55 -08:00
86f34c52a4
Revert bad shader optimization
2023-01-12 00:30:41 +01:00
ca63bb1e5b
Even more optimization/cleanup. Don't need to check ldot that we already know is >= 0.0.
2023-01-11 19:34:58 +01:00
965007d65b
Tighten up the lighting shader code a bit, hopefully eliminating a branch or two.
2023-01-11 19:29:02 +01:00
4c2a41cc83
Break out the attenuation term, too
2023-01-11 19:29:02 +01:00
a508f7e345
Shave a few instructions from lighting shaders by prenormalizing the light dir used for spotlights
2023-01-11 19:29:02 +01:00
7b62b4626f
Comment fixes
2023-01-11 15:29:52 +01:00
3b46409350
Vertex shaders: On platforms with uniform buffers, use indexing and loop for real over the lights.
...
Strangely, greatly speeds up pipeline creation on PowerVR.
2023-01-11 15:24:56 +01:00
3d376b0ab7
Remove the rather redundant DoTexture flag from vshaders.
...
Slightly reduces the number of unique vertex shaders but doesn't do much
for the pipeline count, as the fragment shader has a tex flag. Still
worth doing for the simplification.
2023-01-11 14:16:46 +01:00
f09c09caa8
GLES: Avoid GLSL redefinition error.
2023-01-08 22:30:15 -08:00
f49f7c6371
D3D9: Force branch usage for fog compute.
...
When using `[flatten]`/default, it picks the wrong value, seemingly always
taking the else case (even if the condition is changed or reversed.)
2023-01-06 17:25:04 -08:00
81f4a76ac9
GPU: Don't bother with fog uniform in sw transform.
2023-01-06 17:24:44 -08:00
f3acdb9da8
GLES: Avoid a shader constant error.
...
Some Adreno drivers interpret 0u as an integer.
2023-01-05 17:56:58 -08:00
a58bf45540
Removes LM flag from vshader in the common case, if ubershader is enabled.
2023-01-05 13:09:49 +01:00
4e30c5c0c6
Unbreak fog, oops.
...
Fixes #16722
2023-01-04 13:01:00 +01:00
b84e24cb89
Invert the nesting of specularIsZero and lmode conditions in VS.
2023-01-04 11:04:24 +01:00
ae383147db
Minor simplification in VS
2023-01-04 11:04:24 +01:00
18d00b0718
Remove lmode flag bit from fragment and geometry shaders
2023-01-04 11:04:23 +01:00
5cae5615ea
Use more unusual values as "fog-disabled".
2023-01-04 10:14:12 +01:00
26c748f959
Make fog-enable driven by uniform instead of fragment shader flag bit
2023-01-04 10:14:11 +01:00
86d748fa0c
GLES: Use uint for uint shift amounts.
...
This seems to cause trouble for some Adreno drivers as well.
2023-01-02 14:37:38 -08:00
2b4a182b46
GLES: Use hex for uint constants.
...
Apparently some Adreno drivers have issues (fixed in 2014):
https://developer.qualcomm.com/forum/qdn-forums/maximize-hardware/mobile-gaming-graphics-adreno/27945
2023-01-02 13:13:48 -08:00
26884150d7
Remove the 0th descriptor set, move everything else back to 0
2022-12-16 13:05:40 +01:00
7950a00c14
Headless buildfix. Crashfix in shader generator test.
2022-12-01 22:49:00 +01:00
d466ae161a
OpenXR - HUD max scale fixed
2022-11-28 19:46:46 +01:00
7eee7f2573
Fix shader debug description issue.
2022-11-23 15:41:20 +01:00
8f103f3f47
Extract the Vulkan descriptor binding cleanup from #16345
2022-11-21 20:30:20 +01:00
eae85836c9
GPU: Add a small error-compensation to depth clip.
...
This was causing a depth that should've calculated as zero to be clipped,
because the multiply and division resulted in a value that was off by
about 2 / pow(2, 24) in the negative direction.
2022-11-12 20:01:46 -08:00
72d197fc47
Revert "OpenXR - Disable range culling properly"
...
This reverts commit d1dabd40ee
2022-11-10 18:44:11 +01:00
c6084c16c3
Fix GLSL compilation issues on really old Adreno drivers.
...
They don't seem to accept unsigned integers as switch cases.
Fixes some stuff in #16295 , not sure if all of it.
2022-11-08 22:59:43 +01:00
87bd7aaccc
Merge pull request #16356 from lvonasek/cleanup_camera_control
...
OpenXR - Camera adjust using any controller
2022-11-08 12:28:44 +01:00
4e8f97d64d
Output shader descriptions directly in the source code
2022-11-08 00:21:08 +01:00
Chris Healy