WIP: fix(vk): float adds native on Panther Lake iGPUs, where the loop lost the exposure's adds #816

Draft

ertomas wants to merge 1 commit from ertomas/Longshot:fix/vk-ptl-native-fadd into main

pull from: ertomas/Longshot:fix/vk-ptl-native-fadd

merge into: MagicBeanDev:main

MagicBeanDev:main

Conversation 8 Commits 1 Files changed 5 +32 -15

ertomas commented 2026-11-15 15:09:25 +00:00

First-time contributor

• Copy link
• Copy Markdown

Type: fix · Risk: low · Tested on: Windows 11, Intel Arc B390 (Panther Lake iGPU), driver 101.9150 · Needs: a Linux build of the renderer (Test Farm)

Why

On Intel's Panther Lake laptops (the Arc B390 iGPU), Kokiri Forest is covered by a steady, milky near-white haze that washes out the trees, the houses, the ground and the characters. The game measures how bright the scene is to set its exposure, and on these GPUs that measurement lost part of its sums, so the game exposed the picture about 8 times too bright. An Arc B580 at the same spot shows only the forest's light mist.

The same sums are the program the GPU was lost in on this PC a few minutes into play (#606), and #444 made the same change on the Arc B-series for that program: with this change, the device was not lost in 40 minutes of play here.

Fixes #618
See #606

How to reproduce

Seen on: Windows 11 Insider 26220, Intel Arc B390 (Panther Lake iGPU, PCI 0xB080, no VRAM mirror), driver 32.0.101.9150, Core Ultra X7 358H, 32 GB
Settings and switches: defaults (Auto (this PC) preset); RX_VK_NO_INTEL_SERIAL=1 for the shader preparation only
Steps: load a save in Kokiri Forest, stand by Link's house next to Saria, or walk the path towards the Deku Tree
Expected: the forest with its light mist and sun shafts, full contrast and colour (as on an Arc B580)
Actual: the whole picture behind a near-white veil, steady, on every build of main tried (f647a072, fbd42024, 414f56e2, and #189, #443, #459)

What changed

The shaders' float adds to storage buffers are now the GPU's own atomic (OpAtomicFAddEXT) by default on Intel's Windows driver for Panther Lake iGPUs (PCI device 0xB0xx) from driver 101.9150, as they already are on Meteor Lake, Arrow Lake and Battlemage. Before, these GPUs ran the compare-exchange loop, which gives up after 1024 attempts and loses that add.

• vk: native_fadd_wanted() in gpu/vk/vk_main.cpp takes 0xB0xx from 101.9150; the log line vk: float atomic add on storage buffers: names Panther Lake.
• tests: tests/test_native_float_atomic_add.py's table has six Panther Lake cases: the B390 at 101.9150 and another 0xB0xx device native; an older driver, Meteor Lake at 9150, Mesa's driver and RX_VK_NATIVE_FADD=0 the loop. It fails on main (device 0xb080, version 101.9150: the loop) and passes here.
• docs: the RX_VK_NATIVE_FADD row in docs/development.md; the runs in docs/tested/2026-11-14-igpu-haze-ertomas-run01.md and docs/tested/2026-11-15-igpu-haze-ertomas-run02.md.

How it was found (frame dumps, RX_VK_DUMP, on this PC):

• Six runtime switches left the haze as it was: RX_VK_MIRROR=1, RX_VK_UB_SNAPSHOT=0 (=1 is this PC's default), RX_VK_SPLIT_UBFLUSH=1, RX_VK_READBACK_LATE=newest, RX_VK_CHUNK=0. So it is not stale or half-written memory without a VRAM mirror.
• In the hazy frame, everything after the tonemap (ab368194) passes the picture through faithfully: the TAAU stand-in's output equals the tonemap's, and the last copy's sRGB encode is the game's own (NVN_FORMAT_RGBA8_SRGB). The tonemap's linear output had a mean of 0.71, against about 0.19 in a B580 screenshot of the same spot.
• With RX_VK_NATIVE_FADD=1, every write to the scene target was about 8x darker from the first one on (the sky, 489438d7: mean 20.7 → 2.59; after the fog pass 1a7482c0: 0.78 → 0.11): the game's pre-exposure. The frame's float adds include 499dbfd0, one workgroup summing a 256x256 sky texture (#444's "sky's light sums"): one workgroup adding into a few words is the case where the loop runs out of attempts.

Measurements

Verdict: fixed. In four interleaved runs at Saria's spot (main, this PR, main, this PR), the screen with this PR is as bright as on an Arc B580 (linear mean 0.205 against 0.199), where main is about 4x brighter (0.778); the two runs of each side agree within 0.003. No device loss in about 40 minutes of play with native adds; one with the loop in about 20 minutes of short runs, at 499dbfd0.

Hardware: Windows 11 Insider 26220, Intel Arc B390 (Panther Lake iGPU, 0xB080, no VRAM mirror), driver 32.0.101.9150, Core Ultra X7 358H, 32 GB
Scene: Kokiri Forest, child, by Link's house next to Saria, daytime; Auto (this PC) preset, the game's upscaler (the TAAU stand-in on this card)
Method: screenshots (Win+Alt+PrtScn), mean of the scene area (5-70 % across, 5-75 % down, HUD left out) in linear light; the scene target's mean from the frame dump's log lines (RX_VK_DUMP); device losses from longshot_log.txt
How to reproduce: scripts/dev_packs.py runtimes: main 53f96d78 (its own renderer pack) and this PR over 53f96d78 (its renderer pack), no switch, run alternately; the frame dumps from runtimes over main 414f56e2, one with RX_VK_NATIVE_FADD=1

Interleaved A/B, screen linear mean (R, G, B), Saria's spot, in the order run:

Run	Before: main @ 53f96d78	After: this PR's change over 53f96d78
1	0.779 0.727 0.568	0.204 0.197 0.164
2	0.777 0.724 0.563	0.206 0.199 0.164
Median	0.778 0.726 0.566	0.205 0.198 0.164

Arc B580 at the same spot (screenshot from #459): 0.199 0.191 0.156.

Measured over main 53f96d78; the PR is now on bbe6fdf2. The numbers still hold: this PR's diff is the same, and none of main's commits since touches the float-add choice, the float-add translation or the passes that set the exposure (git diff 53f96d78 bbe6fdf2 has no change to native_fadd_wanted, OpAtomicFAddEXT or the compare-exchange loop).

Where the brightness comes from (frame dumps, runtimes over main 414f56e2):

	Before: main @ 414f56e2	After, RX_VK_NATIVE_FADD=1: main @ 414f56e2
Scene after the fog pass 1a7482c0, mean	0.780 0.590 0.354	0.111 0.097 0.077
Tonemap output ab368194, mean	0.712 0.646 0.455	0.154 0.139 0.099

Device lost: with the loop, once (499dbfd0) in about 20 minutes of short runs on main 414f56e2; with native adds, none in 37 minutes (this change over 414f56e2) plus the A/B runs.

main:   vk: float atomic add on storage buffers: compare-exchange loop, by default only with NVIDIA's driver and Intel's Windows driver on Meteor Lake and Arrow Lake (RX_VK_NATIVE_FADD=1: native where the GPU has it)
main:   vk: dump 249 pass c0 prog 1a7482c0ec9991ea #3485: ... mean 0.7797 0.5901 0.3535 1, ... max 33 24.25 14.5 1
main:   vk: dump 286 pass c0 prog ab3681946f93594a #4247: ... mean 0.7119 0.6461 0.4553 1, ... max 1 1 1 1
main:   vk: device lost: 2 checkpoints (1085666 markers set)
main:   vk:   stage 0x1: marker 1085480 (dispatch of program 499dbfd0721d2858, frame 2170)
switch: vk: float atomic add on storage buffers: native (VK_EXT_shader_atomic_float), RX_VK_NATIVE_FADD=1
switch: vk: dump 237 pass c0 prog 1a7482c0ec9991ea #200: ... mean 0.1107 0.09746 0.07739 1, ... max 3.219 2.438 1.438 1
switch: vk: dump ... pass c0 prog ab3681946f93594a #1077: ... mean 0.1542 0.1389 0.09899 1, ... max 0.9688 0.9531 0.9062 1
this PR: vk: float atomic add on storage buffers: native (VK_EXT_shader_atomic_float), Intel's Windows driver, Meteor Lake, Arrow Lake, Panther Lake or Battlemage

Screenshots of the same spot are in the first comment: the four A/B runs, the Deku Tree path, and the B580 reference.

Flags and settings

The default of RX_VK_NATIVE_FADD changes on Panther Lake iGPUs; no switch is added.

Flag	What it is	Before this PR	After this PR	Effect	Helps / doesn't help / untested	Recommendation
RX_VK_NATIVE_FADD (unset)	float adds to storage buffers: the GPU's own atomic, or the compare-exchange loop	the loop on Panther Lake	native on Panther Lake (0xB0xx) from driver 101.9150	the game's exposure right; no haze	helps on the Arc B390 (0xB080), 101.9150; other 0xB0xx devices untested	on for Panther Lake; RX_VK_NATIVE_FADD=0 gives the loop back

Testing

Tested:

• In the game, this change rebased on main 53f96d78 (the same diff as here; the PR is now on main bbe6fdf2), Windows 11 Insider 26220, Arc B390 (0xB080), driver 101.9150: four interleaved runs at Saria's spot against main 53f96d78 (main, this PR, main, this PR): the haze in both main runs, gone in both runs of this PR; the logs show the loop on main and native adds by the new default here; no device loss.
• Earlier, in the game, Windows 11 Insider 26220, Arc B390 (0xB080), driver 101.9150: Kokiri Forest at Saria's spot and on the Deku Tree path, with this change's renderer pack over main 414f56e2 and no switch: the haze is gone, the picture as on a B580; 37 minutes of play, no device loss, a clean close. The log line confirms native adds by the new default.
• RX_VK_NATIVE_FADD=1 on main 414f56e2: the same picture, so the native add alone is what changes it.
• tests/test_native_float_atomic_add.py run directly with clang (llvm-mingw) and the Vulkan SDK's spirv-val: fails on main, passes here (48 cases), and main's own version of the check passes on this rule. run_all skips it on Windows (it needs spirv-val).
• python -I -B tests/run_all.py on Windows, this PR @ f121f6bd over main bbe6fdf2: 552 passed, 18 failed, 36 skipped (Windows, 16 at once). None of the 18 reads this change, and each fails the same way on main bbe6fdf2 when run alone (test_tested_configurations is a CRLF/LF mismatch on Windows; test_vk_occlusion_queries doesn't find its C++ compiler on Windows): test_appimage_player_build, test_cpp_preflight, test_crash_dialog, test_env_on, test_lift_cleanup_error, test_linux_preflight, test_make_game_cache, test_make_game_prepare, test_missing_nup, test_oot_state_tap, test_options_budget_once, test_portable_builder_home, test_portable_builder_startup, test_readback_late, test_tested_configurations, test_vk_occlusion_queries, test_vk_pgo, test_walk_key.

Not tested:

• Panther Lake devices other than the Arc B390, and drivers newer than 101.9150: none here.
• Linux: Mesa's Intel driver keeps the loop (the check's table), so nothing changes there; the renderer was not built on Linux.

Must be tested:

• Linux, any GPU (Test Farm: @Beanstalk linux, build the renderer): the renderer builds; no change in play is expected, since the rule only changes Intel's Windows driver.

Risks

What could break: on a Panther Lake iGPU, a driver whose native float add misbehaves; it was run on 101.9150 only. 0xB0xx is taken as Panther Lake's range, as 0x7Dxx is for Meteor Lake and Arrow Lake.

Turning it off: RX_VK_NATIVE_FADD=0 in longshot_env.txt, or revert the commit.

Compatibility: on Panther Lake iGPUs, the first start after updating translates the shaders and prepares the pipelines again (the program cache's stamp holds the float-add choice): one long start, about 24 minutes here with compiles side by side. Nothing changes on other GPUs, saves or the options file; no vk ABI change.

Issues

Fixes #618
See #606

Checklist

Everyone

• One topic.
• The title is a Conventional Commit (type(scope): description) of the status line's type.
• Why gives the reason plainly and links the issues that fit (the tracker searched).
• Measurements has a verdict, or one line saying why there is nothing to measure.
• Testing names what was tested, what wasn't, and what must still be tested, where and how.
• Rebased on current main.
• python -I -B tests/run_all.py passes (any failure that also fails on main is named).
• No game files, dumps, build output or generated files.
• third_party/ untouched (a new folder with a PROVENANCE.md instead).
• The license is OK: GPL-3.0-or-later, nothing incompatible and nothing decompiled.

When it applies

• AI assistance recorded with Assisted-by: / Co-Authored-By: trailers.
• Numbers from the same PC, scene and settings, at least 2 interleaved runs each, the median reported.
• Switches documented (docs/development.md; the README for a menu row) and shipped at their recommended value.
• Screenshots of the same spot, captioned like the Measurements columns, for a change that shows on screen.
• A file in docs/tested/ for each new measured run (format in docs/tested-configurations.md).
• A game-free test added for the fix, failing on main and passing here.
• A check under tests/ that reads changed source updated in this PR, not weakened.
• Host code builds on both Windows (llvm-mingw) and Linux, with UTF-8 paths. (Built on Windows; Linux not built: Must be tested.)
• Docs updated.

🤖 Generated with Claude Code

**Type:** fix · **Risk:** low · **Tested on:** Windows 11, Intel Arc B390 (Panther Lake iGPU), driver 101.9150 · **Needs:** a Linux build of the renderer (Test Farm) ## Why On Intel's Panther Lake laptops (the Arc B390 iGPU), Kokiri Forest is covered by a steady, milky near-white haze that washes out the trees, the houses, the ground and the characters. The game measures how bright the scene is to set its exposure, and on these GPUs that measurement lost part of its sums, so the game exposed the picture about 8 times too bright. An Arc B580 at the same spot shows only the forest's light mist. The same sums are the program the GPU was lost in on this PC a few minutes into play (#606), and #444 made the same change on the Arc B-series for that program: with this change, the device was not lost in 40 minutes of play here. Fixes #618 See #606 ## How to reproduce **Seen on:** Windows 11 Insider 26220, Intel Arc B390 (Panther Lake iGPU, PCI 0xB080, no VRAM mirror), driver 32.0.101.9150, Core Ultra X7 358H, 32 GB **Settings and switches:** defaults (Auto (this PC) preset); `RX_VK_NO_INTEL_SERIAL=1` for the shader preparation only **Steps:** load a save in Kokiri Forest, stand by Link's house next to Saria, or walk the path towards the Deku Tree **Expected:** the forest with its light mist and sun shafts, full contrast and colour (as on an Arc B580) **Actual:** the whole picture behind a near-white veil, steady, on every build of main tried (f647a072, fbd42024, 414f56e2, and #189, #443, #459) ## What changed The shaders' float adds to storage buffers are now the GPU's own atomic (`OpAtomicFAddEXT`) by default on Intel's Windows driver for Panther Lake iGPUs (PCI device 0xB0xx) from driver 101.9150, as they already are on Meteor Lake, Arrow Lake and Battlemage. Before, these GPUs ran the compare-exchange loop, which gives up after 1024 attempts and loses that add. - vk: `native_fadd_wanted()` in `gpu/vk/vk_main.cpp` takes 0xB0xx from 101.9150; the log line `vk: float atomic add on storage buffers:` names Panther Lake. - tests: `tests/test_native_float_atomic_add.py`'s table has six Panther Lake cases: the B390 at 101.9150 and another 0xB0xx device native; an older driver, Meteor Lake at 9150, Mesa's driver and `RX_VK_NATIVE_FADD=0` the loop. It fails on main (`device 0xb080, version 101.9150: the loop`) and passes here. - docs: the `RX_VK_NATIVE_FADD` row in `docs/development.md`; the runs in `docs/tested/2026-11-14-igpu-haze-ertomas-run01.md` and `docs/tested/2026-11-15-igpu-haze-ertomas-run02.md`. How it was found (frame dumps, `RX_VK_DUMP`, on this PC): - Six runtime switches left the haze as it was: `RX_VK_MIRROR=1`, `RX_VK_UB_SNAPSHOT=0` (`=1` is this PC's default), `RX_VK_SPLIT_UBFLUSH=1`, `RX_VK_READBACK_LATE=newest`, `RX_VK_CHUNK=0`. So it is not stale or half-written memory without a VRAM mirror. - In the hazy frame, everything after the tonemap (`ab368194`) passes the picture through faithfully: the TAAU stand-in's output equals the tonemap's, and the last copy's sRGB encode is the game's own (`NVN_FORMAT_RGBA8_SRGB`). The tonemap's linear output had a mean of 0.71, against about 0.19 in a B580 screenshot of the same spot. - With `RX_VK_NATIVE_FADD=1`, every write to the scene target was about 8x darker from the first one on (the sky, `489438d7`: mean 20.7 → 2.59; after the fog pass `1a7482c0`: 0.78 → 0.11): the game's pre-exposure. The frame's float adds include `499dbfd0`, one workgroup summing a 256x256 sky texture (#444's "sky's light sums"): one workgroup adding into a few words is the case where the loop runs out of attempts. ## Measurements **Verdict:** fixed. In four interleaved runs at Saria's spot (main, this PR, main, this PR), the screen with this PR is as bright as on an Arc B580 (linear mean 0.205 against 0.199), where main is about 4x brighter (0.778); the two runs of each side agree within 0.003. No device loss in about 40 minutes of play with native adds; one with the loop in about 20 minutes of short runs, at `499dbfd0`. Hardware: Windows 11 Insider 26220, Intel Arc B390 (Panther Lake iGPU, 0xB080, no VRAM mirror), driver 32.0.101.9150, Core Ultra X7 358H, 32 GB Scene: Kokiri Forest, child, by Link's house next to Saria, daytime; Auto (this PC) preset, the game's upscaler (the TAAU stand-in on this card) Method: screenshots (Win+Alt+PrtScn), mean of the scene area (5-70 % across, 5-75 % down, HUD left out) in linear light; the scene target's mean from the frame dump's log lines (`RX_VK_DUMP`); device losses from `longshot_log.txt` How to reproduce: `scripts/dev_packs.py` runtimes: main 53f96d78 (its own renderer pack) and this PR over 53f96d78 (its renderer pack), no switch, run alternately; the frame dumps from runtimes over main 414f56e2, one with `RX_VK_NATIVE_FADD=1` **Interleaved A/B, screen linear mean (R, G, B)**, Saria's spot, in the order run: | Run | Before: main @ 53f96d78 | After: this PR's change over 53f96d78 | |---|---|---| | 1 | 0.779 0.727 0.568 | 0.204 0.197 0.164 | | 2 | 0.777 0.724 0.563 | 0.206 0.199 0.164 | | **Median** | **0.778 0.726 0.566** | **0.205 0.198 0.164** | Arc B580 at the same spot (screenshot from #459): 0.199 0.191 0.156. Measured over main 53f96d78; the PR is now on bbe6fdf2. The numbers still hold: this PR's diff is the same, and none of main's commits since touches the float-add choice, the float-add translation or the passes that set the exposure (`git diff 53f96d78 bbe6fdf2` has no change to `native_fadd_wanted`, `OpAtomicFAddEXT` or the compare-exchange loop). **Where the brightness comes from** (frame dumps, runtimes over main 414f56e2): | | Before: main @ 414f56e2 | After, `RX_VK_NATIVE_FADD=1`: main @ 414f56e2 | |---|---|---| | Scene after the fog pass `1a7482c0`, mean | 0.780 0.590 0.354 | 0.111 0.097 0.077 | | Tonemap output `ab368194`, mean | 0.712 0.646 0.455 | 0.154 0.139 0.099 | **Device lost:** with the loop, once (`499dbfd0`) in about 20 minutes of short runs on main 414f56e2; with native adds, none in 37 minutes (this change over 414f56e2) plus the A/B runs. ``` main: vk: float atomic add on storage buffers: compare-exchange loop, by default only with NVIDIA's driver and Intel's Windows driver on Meteor Lake and Arrow Lake (RX_VK_NATIVE_FADD=1: native where the GPU has it) main: vk: dump 249 pass c0 prog 1a7482c0ec9991ea #3485: ... mean 0.7797 0.5901 0.3535 1, ... max 33 24.25 14.5 1 main: vk: dump 286 pass c0 prog ab3681946f93594a #4247: ... mean 0.7119 0.6461 0.4553 1, ... max 1 1 1 1 main: vk: device lost: 2 checkpoints (1085666 markers set) main: vk: stage 0x1: marker 1085480 (dispatch of program 499dbfd0721d2858, frame 2170) switch: vk: float atomic add on storage buffers: native (VK_EXT_shader_atomic_float), RX_VK_NATIVE_FADD=1 switch: vk: dump 237 pass c0 prog 1a7482c0ec9991ea #200: ... mean 0.1107 0.09746 0.07739 1, ... max 3.219 2.438 1.438 1 switch: vk: dump ... pass c0 prog ab3681946f93594a #1077: ... mean 0.1542 0.1389 0.09899 1, ... max 0.9688 0.9531 0.9062 1 this PR: vk: float atomic add on storage buffers: native (VK_EXT_shader_atomic_float), Intel's Windows driver, Meteor Lake, Arrow Lake, Panther Lake or Battlemage ``` Screenshots of the same spot are in the first comment: the four A/B runs, the Deku Tree path, and the B580 reference. ### Flags and settings The default of `RX_VK_NATIVE_FADD` changes on Panther Lake iGPUs; no switch is added. | Flag | What it is | Before this PR | After this PR | Effect | Helps / doesn't help / untested | Recommendation | |---|---|---|---|---|---|---| | `RX_VK_NATIVE_FADD` (unset) | float adds to storage buffers: the GPU's own atomic, or the compare-exchange loop | the loop on Panther Lake | native on Panther Lake (0xB0xx) from driver 101.9150 | the game's exposure right; no haze | helps on the Arc B390 (0xB080), 101.9150; other 0xB0xx devices untested | on for Panther Lake; `RX_VK_NATIVE_FADD=0` gives the loop back | ## Testing **Tested:** - In the game, this change rebased on main 53f96d78 (the same diff as here; the PR is now on main bbe6fdf2), Windows 11 Insider 26220, Arc B390 (0xB080), driver 101.9150: four interleaved runs at Saria's spot against main 53f96d78 (main, this PR, main, this PR): the haze in both main runs, gone in both runs of this PR; the logs show the loop on main and native adds by the new default here; no device loss. - Earlier, in the game, Windows 11 Insider 26220, Arc B390 (0xB080), driver 101.9150: Kokiri Forest at Saria's spot and on the Deku Tree path, with this change's renderer pack over main 414f56e2 and no switch: the haze is gone, the picture as on a B580; 37 minutes of play, no device loss, a clean close. The log line confirms native adds by the new default. - `RX_VK_NATIVE_FADD=1` on main 414f56e2: the same picture, so the native add alone is what changes it. - `tests/test_native_float_atomic_add.py` run directly with clang (llvm-mingw) and the Vulkan SDK's `spirv-val`: fails on main, passes here (48 cases), and main's own version of the check passes on this rule. `run_all` skips it on Windows (it needs `spirv-val`). - `python -I -B tests/run_all.py` on Windows, this PR @ f121f6bd over main bbe6fdf2: 552 passed, 18 failed, 36 skipped (Windows, 16 at once). None of the 18 reads this change, and each fails the same way on main bbe6fdf2 when run alone (test_tested_configurations is a CRLF/LF mismatch on Windows; test_vk_occlusion_queries doesn't find its C++ compiler on Windows): test_appimage_player_build, test_cpp_preflight, test_crash_dialog, test_env_on, test_lift_cleanup_error, test_linux_preflight, test_make_game_cache, test_make_game_prepare, test_missing_nup, test_oot_state_tap, test_options_budget_once, test_portable_builder_home, test_portable_builder_startup, test_readback_late, test_tested_configurations, test_vk_occlusion_queries, test_vk_pgo, test_walk_key. **Not tested:** - Panther Lake devices other than the Arc B390, and drivers newer than 101.9150: none here. - Linux: Mesa's Intel driver keeps the loop (the check's table), so nothing changes there; the renderer was not built on Linux. **Must be tested:** - Linux, any GPU (Test Farm: `@Beanstalk linux, build the renderer`): the renderer builds; no change in play is expected, since the rule only changes Intel's Windows driver. ## Risks **What could break:** on a Panther Lake iGPU, a driver whose native float add misbehaves; it was run on 101.9150 only. `0xB0xx` is taken as Panther Lake's range, as `0x7Dxx` is for Meteor Lake and Arrow Lake. **Turning it off:** `RX_VK_NATIVE_FADD=0` in `longshot_env.txt`, or revert the commit. **Compatibility:** on Panther Lake iGPUs, the first start after updating translates the shaders and prepares the pipelines again (the program cache's stamp holds the float-add choice): one long start, about 24 minutes here with compiles side by side. Nothing changes on other GPUs, saves or the options file; no vk ABI change. ## Issues Fixes #618 See #606 ## Checklist ### Everyone - [x] One topic. - [x] The title is a Conventional Commit (`type(scope): description`) of the status line's type. - [x] Why gives the reason plainly and links the issues that fit (the tracker searched). - [x] Measurements has a verdict, or one line saying why there is nothing to measure. - [x] Testing names what was tested, what wasn't, and what must still be tested, where and how. - [x] Rebased on current `main`. - [x] `python -I -B tests/run_all.py` passes (any failure that also fails on `main` is named). - [x] No game files, dumps, build output or generated files. - [x] `third_party/` untouched (a new folder with a `PROVENANCE.md` instead). - [x] The license is OK: GPL-3.0-or-later, nothing incompatible and nothing decompiled. ### When it applies - [x] AI assistance recorded with `Assisted-by:` / `Co-Authored-By:` trailers. - [x] Numbers from the same PC, scene and settings, at least 2 interleaved runs each, the median reported. - [x] Switches documented (docs/development.md; the README for a menu row) and shipped at their recommended value. - [x] Screenshots of the same spot, captioned like the Measurements columns, for a change that shows on screen. - [x] A file in `docs/tested/` for each new measured run (format in `docs/tested-configurations.md`). - [x] A game-free test added for the fix, failing on `main` and passing here. - [x] A check under `tests/` that reads changed source updated in this PR, not weakened. - [ ] Host code builds on both Windows (llvm-mingw) and Linux, with UTF-8 paths. (Built on Windows; Linux not built: Must be tested.) - [x] Docs updated. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

ertomas added 1 commit 2026-11-15 15:09:25 +00:00

fix(vk): float adds native on Panther Lake iGPUs, where the loop lost the exposure's adds ... ee768482f1

On an Arc B390 (Panther Lake, 0xB080, driver 101.9150) the shaders' float adds to
storage buffers ran as the compare-exchange loop, and the game's exposure came out
about 8x too high: every write to the scene, from the sky on, was 8x too bright, and
a milky white haze covered Kokiri Forest (issue 618). With native adds the scene
matched an Arc B580's at the same spot, and the device was not lost in 40 minutes of
play, where the loop lost it at 499dbfd0, the sky's light sums (issue 606; #444 made
Battlemage native for the same program).

native_fadd_wanted now takes Intel's Windows driver on 0xB0xx from 101.9150 too,
beside Meteor Lake, Arrow Lake and Battlemage. The check's table has the Panther Lake
cases (failing on main, passing here); the log line and the switch's row in
docs/development.md name Panther Lake.

Fixes #618
See #606

Assisted-by: Claude Opus 5.5
Co-Authored-By: Claude Opus 5.5 <[email protected]>

ertomas commented 2026-11-15 15:09:32 +00:00

Author

First-time contributor

• Copy link
• Copy Markdown

Screenshots, Windows 11, Arc B390 (Panther Lake iGPU, 0xB080), driver 101.9150, Kokiri Forest, the Auto (this PC) preset.

By Link's house, next to Saria

Four interleaved runs, in the order played:

The path towards the Deku Tree (the same area; the camera is not at exactly the same place)

Screenshots, Windows 11, Arc B390 (Panther Lake iGPU, 0xB080), driver 101.9150, Kokiri Forest, the Auto (this PC) preset. **By Link's house, next to Saria** Four interleaved runs, in the order played:      **The path towards the Deku Tree** (the same area; the camera is not at exactly the same place)  

reference_from_forge.png

2.6 MiB

b390_main_414f56e2_deku_path.png

3.5 MiB

fix618_deku_path.png

7.8 MiB

ab_fix1.png

8.5 MiB

ab_main1.png

6.7 MiB

ab_main2.png

7.1 MiB

ab_fix2.png

8.6 MiB

ertomas referenced this pull request 2026-11-15 15:10:18 +00:00

Game: Arc B390 (Panther Lake iGPU, Windows): Kokiri Forest is washed out by a bright haze; an Arc B580 at the same spot is not #618

ertomas commented 2026-11-15 15:10:40 +00:00

Author

First-time contributor

• Copy link
• Copy Markdown

Rebased on main aa95d35f before opening; the description says why the A/B numbers from 53f96d78 still hold, and run_all was run again on the rebase (the same 18 Windows failures as on main).

Rebased on main aa95d35f before opening; the description says why the A/B numbers from 53f96d78 still hold, and run_all was run again on the rebase (the same 18 Windows failures as on main).

ertomas commented 2026-11-15 15:14:37 +00:00

Author

First-time contributor

• Copy link
• Copy Markdown

@Beanstalk linux, build the renderer: the build succeeds; no change in play expected (the rule only changes Intel's Windows driver)

@Beanstalk linux, build the renderer: the build succeeds; no change in play expected (the rule only changes Intel's Windows driver)

Beanstalk commented 2026-11-15 15:14:38 +00:00

Member

• Copy link
• Copy Markdown

@ertomas only members of BigBeans can start the farm.

One of them can approve this request by reacting 👍 to it. This request was for ee76848, and the PR has changed since (f121f6b), so a 👍 no longer approves it. Comment @Beanstalk again to ask for the new commit.

@ertomas only members of `BigBeans` can start the farm. ~~One of them can approve this request by reacting 👍 to it.~~ This request was for `ee76848`, and the PR has changed since (`f121f6b`), so a 👍 no longer approves it. Comment `@Beanstalk` again to ask for the new commit.

fireblast327 referenced this pull request 2026-11-15 17:05:54 +00:00

Game: random starts show a black picture all session (the exposure goes NaN in the first frames), GTX 1060 6GB #841

HumanBean force-pushed fix/vk-ptl-native-fadd from ee768482f1 to f121f6bd5b 2026-11-15 18:45:01 +00:00 Compare

HumanBean commented 2026-11-15 18:46:28 +00:00

Owner

• Copy link
• Copy Markdown

Rebased onto main @ bbe6fdf2: no conflicts. No tests were run here; the test boxes stay yours.

Rebased onto `main` @ bbe6fdf2: no conflicts. No tests were run here; the test boxes stay yours.

ertomas commented 2026-11-15 19:02:38 +00:00

Author

First-time contributor

• Copy link
• Copy Markdown

Description updated for HumanBean's rebase onto main bbe6fdf2 (f121f6bd): run_all run again there (552 passed; the 18 failures fail the same way on main bbe6fdf2 alone), the fix's own check passes, and none of main's commits since the A/B touches the float-add rule.

Description updated for HumanBean's rebase onto main bbe6fdf2 (f121f6bd): run_all run again there (552 passed; the 18 failures fail the same way on main bbe6fdf2 alone), the fix's own check passes, and none of main's commits since the A/B touches the float-add rule.

ertomas commented 2026-11-15 19:02:39 +00:00

Author

First-time contributor

• Copy link
• Copy Markdown

@Beanstalk linux, build the renderer: the build succeeds; no change in play expected (the rule only changes Intel's Windows driver)

@Beanstalk linux, build the renderer: the build succeeds; no change in play expected (the rule only changes Intel's Windows driver)

Beanstalk commented 2026-11-15 19:02:39 +00:00

Member

• Copy link
• Copy Markdown

@ertomas only members of BigBeans can start the farm. One of them can approve this request by reacting 👍 to it.

@ertomas only members of `BigBeans` can start the farm. One of them can approve this request by reacting 👍 to it.

Some checks failed

Hide checks

CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped

Details

CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped

Details

CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped

Details

Third-party parts / The same as upstream (pull_request) Has been skipped

Details

Website / Website on Cloudflare (pull_request) Has been skipped

Details

beanstalk/approval waiting for a BigBean's 👍 on the request

Details

PR description / Description follows the template (pull_request_target) Successful in 3s

Details

CI / Conventional Commit title (pull_request) Successful in 3s

Details

CI / Workflow lint (actionlint) (pull_request) Successful in 5s

Details

Guide screenshots / Guide images and local UI source review (pull_request) Failing after 4s

Details

Third-party parts / As their notes say (pull_request) Successful in 8s

Details

This pull request is marked as a work in progress.

This branch is out-of-date with the base branch

View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u 'fix/vk-ptl-native-fadd:ertomas-fix/vk-ptl-native-fadd'

git switch 'ertomas-fix/vk-ptl-native-fadd'

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

area/accessibility

Barriers for people with disabilities

  

area/audio

Sound and music

  

area/input

Controllers, keyboard, mouse

  

area/install

Installing, updating, building from your game

  

area/menu

In-game menu and settings

  

area/renderer

Vulkan renderer, shaders, effects

  

area/saves

Save files

  

area/upscaling

DLSS, FSR, frame generation

  

beanstalk: approved @ 1697603

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 2c5a62e

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 4fd82b7

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 64d6d09

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 6f91fae

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 70a78f9

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 8015c9a

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 8d361dc

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 9d8356e

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ abf2227

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ d82f202

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ e5f1e40

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ ee8ab5f

Set by Beanstalk for show; the farm's database is what counts

  

ci: approved

A maintainer approved CircleCI for this fork PR (sticky; removed after 14 quiet days)

  

good first issue

A good start for a new contributor

  

gpu/amd

AMD GPUs

  

gpu/igpu

Integrated GPUs

  

gpu/intel

Intel GPUs

  

gpu/nvidia

NVIDIA GPUs

  

help wanted

Extra hands or knowledge wanted

  

kind

bug

Something doesn't work as it should

  

kind

chore

Build, CI, dependencies, tidying

  

kind

docs

Documentation

  

kind

feature

A new feature or an improvement

  

kind

performance

Frame rate, frame time, loading, memory

  

kind

question

A question rather than a change

  

needs measurements

Waiting for before/after measurements

  

platform/linux

Affects Linux (a change for every system carries both Windows and Linux)

  

platform/macos

macOS (Apple silicon or Intel Macs)

  

platform/steam-deck

Steam Deck

  

platform/windows

Affects Windows (a change for every system carries both Windows and Linux)

  

preview

Maintainers: deploy a preview of this PR's website (read its changes first). Removed after the run.

  

priority

critical

Crashes, data loss, or blocks most players

  

priority

high

Important: next in line

  

priority

low

Nice to have

  

regression

Worked in an earlier build, broken in a newer one: highest priority, above other bugs

  

status

blocked

Waits on something else first

  

status

confirmed

Reproduced or agreed on

  

status

duplicate

Already reported or proposed

  

status

invalid

Not a problem in this project

  

status

needs-info

Waiting for more information from the reporter

  

status

wontfix

Will not be worked on

  

testing wanted

Wanted: a test on other hardware (posts a test request in #test-requests)

  

ci: approved

No labels

area/accessibility

area/audio

area/input

area/install

area/menu

area/renderer

area/saves

area/upscaling

beanstalk: approved @ 1697603

beanstalk: approved @ 2c5a62e

beanstalk: approved @ 4fd82b7

beanstalk: approved @ 64d6d09

beanstalk: approved @ 6f91fae

beanstalk: approved @ 70a78f9

beanstalk: approved @ 8015c9a

beanstalk: approved @ 8d361dc

beanstalk: approved @ 9d8356e

beanstalk: approved @ abf2227

beanstalk: approved @ d82f202

beanstalk: approved @ e5f1e40

beanstalk: approved @ ee8ab5f

ci: approved

good first issue

gpu/amd

gpu/igpu

gpu/intel

gpu/nvidia

help wanted

kind

bug

kind

chore

kind

docs

kind

feature

kind

performance

kind

question

needs measurements

platform/linux

platform/macos

platform/steam-deck

platform/windows

preview

priority

critical

priority

high

priority

low

regression

status

blocked

status

confirmed

status

duplicate

status

invalid

status

needs-info

status

wontfix

testing wanted

ci: approved

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

3 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

MagicBeanDev/Longshot!816

No description provided.