nv2a: Implement texture LOD bias #2435

abaire · 2025-09-18T18:14:09Z

Fixes #767
Fixes #973
Fixes #1110
Fixes #1204
Fixes #2377

Tests: https://github.com/abaire/nxdk_pgraph_tests/blob/2aa0cb89543f3853a3830ba15a03c9f7dad58d69/src/tests/texture_lod_bias_tests.h#L1

HW results: https://abaire.github.io/nxdk_pgraph_tests_golden_results/results/Texture_LOD_Bias/index.html

PR results: https://github.com/abaire/xemu-dev_pgraph_test_results/blob/fix/implements_lod_bias/results/xemu-0.8.101-1-g3ece4e2a17-fix_implements_lod_bias-3ece4e2a170f38aef3f220ab007b090bbb29233b/Darwin_arm64/gl_Apple_Apple_M3_Max/gslv_4.10/Texture_LOD_Bias/LODBias.png

Fixes xemu-project#767 Tests: https://github.com/abaire/nxdk_pgraph_tests/blob/2aa0cb89543f3853a3830ba15a03c9f7dad58d69/src/tests/texture_lod_bias_tests.h#L1 HW results: https://abaire.github.io/nxdk_pgraph_tests_golden_results/results/Texture_LOD_Bias/index.html

Triticum0 · 2025-09-18T20:02:48Z

Seems to improve ground texture quality of Quantum Redshift. Though rather subtle
Masters

PR

abaire · 2025-09-18T20:13:57Z

@Triticum0 excellent, thanks for testing!

Could you also check Vulkan with the newest build? I don't have access to a VK capable machine at the moment and put in a prospective fix.

Triticum0 · 2025-09-18T20:34:28Z

Seems to fix Burnout-3-Takedown ground texture quality too.
Masters

PR

Tested vulkan matches opengl didn't crash and looked identical to opengl.

Triticum0 · 2025-09-18T20:35:24Z

Also tested crash and burn as well Wreckless The Yakuza Missions Both seem to be unrelated.

abaire · 2025-09-18T20:53:59Z

Thanks! I'll take a look at the traces you added to #2385, we might want to split them into a separate issue if they're due to a novel issue.

Triticum0 · 2025-09-18T21:37:49Z

Opps actually fixes Crash 'N' Burn floor rendering and tree rendering. Look at outline around far floor textures and outline around 2d tree texture. #1110

Masters

PR

~~Doesn't fix railing above the track. only realise other issues where fixed when double checking to get a capture.~~

~~Hardware screenshot from loading screen~~

Ingame

~~See theirs 5 parts in-game rails but on screenshot there at least double. don't know if its related send capture soon.~~ Found a hardware capture actually occurs on hardware https://www.youtube.com/watch?v=KIHMt-TKgYk

Triticum0 · 2025-09-18T21:43:13Z

Renderdoc are on this issue if your interested have add new one for the railing issue as I missed it before

Triticum0 · 2025-09-18T21:54:27Z

Doesn't fix Racing Evoluzione/Apex ground texture issue.

Triticum0 · 2025-09-18T22:03:03Z

Fixes #2377
Masters

PR

Triticum0 · 2025-09-18T22:18:26Z

Fixed metal mesh overhead sign next to Xicat logo rendering incorrectly. on Motor Trend Presents Lotus Challenge.
Masters

PR

Hopefully why car blue or green could be looked at only happens with no shader cache.

Other Issues in #1228 are unaffected

abaire · 2025-09-18T23:09:40Z

@Triticum0 just to confirm, does this fully resolve #1110 or are there still remaining issues? I don't want to accidentally close out that issue if there is still work to be done.

Triticum0 · 2025-09-18T23:18:56Z

Create new issue which supersede old issue with remaining issues so you can close it. other issue unrelated so I will move it

Triticum0 · 2025-09-18T23:21:59Z

Improves the behavior of burnout 2 grass texture. #1161
Master

PR

No 100% sure if it matches hardware seem behavior present on hardware but not if it the same. need hardware capture to verify. here video where you can see line on hardware https://youtu.be/HGxRSVldAHk?t=333

@abaire Leave it to you judgment where you want to close the issue. the other issues on the pr either occur on hardware or are already fixed so can be closed if you thing the fix sufficient for bug above.

Triticum0 · 2025-09-18T23:33:42Z

#2437 supersedes #1110 you can close it when it merged to masters.

abaire · 2025-09-18T23:55:25Z

Improves the behavior of burnout 2 grass texture. #1161
@abaire Leave it to you judgment where you want to close the issue. the other issues on the pr either occur on hardware or are already fixed so can be closed if you thing the fix sufficient for bug above.

Does the Z-fighting mentioned not occur anymore? I assumed that bug would need coldhex's #2240 to fix some parts of it. We might need to break that bug up into separate issues, maybe we can just do that after this change is merged and leave off whatever is fixed by LOD biasing.

#2437 supersedes #1110 you can close it when it merged to masters.
Perfect, marked #1110 as fixed by this; thanks!

Triticum0 · 2025-09-19T00:20:38Z

It hard to see but It also occurs on hardware https://youtu.be/HGxRSVldAHk?t=502 so you can close that too. ask cold-hex about similar issue and he mention it occurred on hardware so in most of my old z-fighting I mentioned weren't valid.

Triticum0 · 2025-09-19T11:18:43Z

Fixes rendering on distance building looking blurry on Driver: Parallel Lines. you can actually see windows on distance side buildings.
Master

PR

mborgerson · 2025-09-19T21:32:16Z

hw/xbox/nv2a/pgraph/vk/texture.c

        .minLod = mipmap_en ? MIN(state.min_mipmap_level, state.levels - 1) : 0.0,
        .maxLod = mipmap_en ? MIN(state.max_mipmap_level, state.levels - 1) : 0.0,
-        .mipLodBias = 0.0,
+        .mipLodBias = convert_lod_bias(


Absolute value needs to be clamped to device limits (see VkPhysicalDeviceLimits::maxSamplerLodBias)

Done, I think; will need to see if my change builds on CI since I can't build VK locally right now.

mborgerson · 2025-09-19T21:33:07Z

hw/xbox/nv2a/pgraph/gl/texture.c

    return possibly_dirty;
 }

+static inline float convert_lod_bias(uint32_t lod_bias)


If implementation is identical between renderers, prefer to have this in common code

mborgerson · 2025-09-19T21:34:43Z

hw/xbox/nv2a/pgraph/gl/texture.c

    unsigned int addru = GET_MASK(address, NV_PGRAPH_TEXADDRESS0_ADDRU);
    unsigned int addrv = GET_MASK(address, NV_PGRAPH_TEXADDRESS0_ADDRV);
    unsigned int addrp = GET_MASK(address, NV_PGRAPH_TEXADDRESS0_ADDRP);
+    unsigned int lod_bias =


nit: Prefer to group filter params

mborgerson · 2025-09-19T21:39:44Z

hw/xbox/nv2a/pgraph/gl/texture.c

                        pgraph_texture_mag_filter_gl_map[mag_filter]);
        binding->mag_filter = mag_filter;
    }
+    if (lod_bias != binding->lod_bias) {


binding->lod_bias not initialized?

Good catch, done.

mborgerson · 2025-09-19T21:41:48Z

hw/xbox/nv2a/pgraph/gl/renderer.h

    bool border_color_set;
    GLenum gl_target;
    GLuint gl_texture;
+    uint32_t lod_bias;


nit: Prefer to group filter params

mborgerson · 2025-09-19T22:59:22Z

hw/xbox/nv2a/pgraph/vk/texture.c

+    float lod_bias = pgraph_convert_lod_bias_to_float(
+        GET_MASK(filter, NV_PGRAPH_TEXFILTER0_MIPMAP_LOD_BIAS));
+    if (lod_bias > r->device_props.limits.maxSamplerLodBias) {
+        lod_bias = r->device_props.limits.maxSamplerLodBias;


Need to also handle negative bias clamp (absolute value must not exceed maxSamplerLodBias)

Oops, misread your comment, sorry!

That said, is it actually necessary to manually clamp these at the application level?

The docs lead me to believe that the driver would handle the clamping, since it ends up being a summation of the bias and the sampling so simply clamping the bias may be insufficient to remain in bounds anyway:

maxSamplerLodBias is the maximum absolute sampler LOD bias. The sum of the mipLodBias member of the [VkSamplerCreateInfo](https://registry.khronos.org/vulkan/specs/latest/man/html/VkSamplerCreateInfo.html) structure and the Bias operand of image sampling operations in shader modules (or 0 if no Bias operand is provided to an image sampling operation) are clamped to the range [-maxSamplerLodBias,+maxSamplerLodBias]. See https://registry.khronos.org/vulkan/specs/latest/html/vkspec.html#samplers-mipLodBias.

is it actually necessary

Maybe, maybe not. According to spec: The absolute value of mipLodBias must be less than or equal to VkPhysicalDeviceLimits::maxSamplerLodBias

the driver would handle the clamping

In general, Vulkan drivers are a lot less forgiving. Even when operating fully within spec, they can be problematic. If we are lucky, a case like this would trigger a validation warning and have no other impact, if we are unlucky it results in some random overflow and a crash, so it's best to follow spec guidance when populating struct params.

PatrickvL · 2025-09-20T07:10:02Z

hw/xbox/nv2a/pgraph/texture.h

+static inline float pgraph_convert_lod_bias_to_float(uint32_t lod_bias)
+{
+    int sign_extended_bias = lod_bias;
+    if (lod_bias & (1 << 12)) {


Does this 12th bit have any significant name or is it merely tested to "or in" the subsequent 0x00001FFF const? And whats the significance of that? Fixed point ceiling, because of the final division by 256.0f ? Xemu is not strong in code comments but this looks like a hard to maintain bit of logic of anyone ever needs to revisit this again.

Never mind, it is actually doing sign extension on the 12th bit onwards, assuming the trust contains a 12 bit signed fixed point number (4 integer + 8 fractional bits, in twos complement, a.k.a. Q3.8). Apparently there's no helper function for that yet.

Just 1 thought: the output range is now -8.0 to +7.99609375 which is not exactly 8.0
Would that be any problem?

I don't think there are many places where 5.8 fixed point numbers are used and it doesn't seem clearly useful to bother with a more complicated generic sign extended fixed point -> float conversion function.

I also doubt it'd 7.99609375 versus 8 would result in noticeable differences if it actually ends up being materially different from how the nv2a HW handles the 5.8 value. I'd guess there's likely more variance across HW/drivers than that fractional delta in any case.

Triticum0 · 2025-09-20T20:49:42Z

Fixes #973
Before

After

Triticum0 · 2025-09-20T22:50:21Z

Fixes #1204
Before

After

Triticum0 · 2025-09-21T00:08:47Z

Should Fix #1930 but haven't tested it

mborgerson · 2025-09-23T19:56:26Z

Thanks!

Triticum0 · 2025-09-24T17:54:44Z

Did fix #1930

abaire force-pushed the fix/implements_lod_bias branch 2 times, most recently from 3ece4e2 to a895ea2 Compare September 18, 2025 19:51

abaire force-pushed the fix/implements_lod_bias branch from a895ea2 to b8ada79 Compare September 18, 2025 19:52

abaire marked this pull request as ready for review September 18, 2025 20:52

abaire mentioned this pull request Sep 18, 2025

Mipmapping Issues (missing emulation of LOD bias and anisotropic filtering) #2385

Closed

mborgerson reviewed Sep 19, 2025

View reviewed changes

abaire added 2 commits September 19, 2025 15:45

SQUASHME: CR updates

9c9f43c

SQUASHME: Renamed convert_lod_bias for namespacing purposes and clarity

ebbb520

mborgerson reviewed Sep 19, 2025

View reviewed changes

abaire mentioned this pull request Sep 19, 2025

nv2a: Handle anisotropic filter setting #2450

Merged

SQUASHME: Clamp negative LOD bias as well

ec5ed5e

PatrickvL reviewed Sep 20, 2025

View reviewed changes

abaire mentioned this pull request Sep 21, 2025

NBA Inside Drive 2004 crowd rendering bug Opengl #2396

Open

mborgerson merged commit 2d26f18 into xemu-project:master Sep 23, 2025
21 checks passed

Uh oh!

nv2a: Implement texture LOD bias #2435

nv2a: Implement texture LOD bias #2435

Uh oh!

Conversation

abaire commented Sep 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Triticum0 commented Sep 18, 2025

Uh oh!

abaire commented Sep 18, 2025

Uh oh!

Triticum0 commented Sep 18, 2025

Uh oh!

Triticum0 commented Sep 18, 2025

Uh oh!

abaire commented Sep 18, 2025

Uh oh!

Triticum0 commented Sep 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Triticum0 commented Sep 18, 2025

Uh oh!

Triticum0 commented Sep 18, 2025

Uh oh!

Triticum0 commented Sep 18, 2025

Uh oh!

Triticum0 commented Sep 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

abaire commented Sep 18, 2025

Uh oh!

Triticum0 commented Sep 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Triticum0 commented Sep 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Triticum0 commented Sep 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

abaire commented Sep 18, 2025

Uh oh!

Triticum0 commented Sep 19, 2025

Uh oh!

Triticum0 commented Sep 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

PatrickvL Sep 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

abaire commented Sep 18, 2025 •

edited

Loading

Triticum0 commented Sep 18, 2025 •

edited

Loading

Triticum0 commented Sep 18, 2025 •

edited

Loading

Triticum0 commented Sep 18, 2025 •

edited

Loading

Triticum0 commented Sep 18, 2025 •

edited

Loading

Triticum0 commented Sep 18, 2025 •

edited

Loading

Triticum0 commented Sep 19, 2025 •

edited

Loading

PatrickvL Sep 20, 2025 •

edited

Loading