nv2a: Fix DstAlpha blending for const-alpha surface modes #2478

abaire · 2025-10-08T06:24:30Z

This remaps DstAlpha to alternate source/destination factors when performing blending on target surfaces with forced alpha bits (e.g., NV097_SET_SURFACE_FORMAT_COLOR_LE_X8R8G8B8_Z8R8G8B8).

Fixes #2473

Tests: https://github.com/abaire/nxdk_pgraph_tests/blob/de548172ac3166f03f3a75999829ba17aade73aa/src/tests/blend_surface_tests.cpp#L249
HW results: https://abaire.github.io/nxdk_pgraph_tests_golden_results/results/Blend_surface/index.html

PR results: https://abaire.github.io/xemu-dev_pgraph_test_results/fix_2473_prevent_alpha_updates_for_forced_value_surfaces/index.html

abaire · 2025-10-08T06:26:03Z

Needs to be tested with Vulkan and verified that it fixes #2473 (may need to be tested along with #2472, I'm not sure if the bug manifests with broken env mapping)

antoniodesousa · 2025-10-08T13:09:28Z

It fixes the issue on Vulkan and OpenGL. But the water is missing on OpenGL. It probably needs #2472 to render properly.

Vulkan

OpenGL

abaire · 2025-10-08T14:20:30Z

Thanks for confirming!

The missing env mapping on GL is indeed expected without #2472 (though I didn't realize it always worked on Vulkan)

Triticum0 · 2025-10-08T15:24:12Z

On NVIDIA Water rendering missing on both renders using vulkan makes no differences on my 3060ti. Issue is resolved on my end as well smoke effect all seem to be fixed haven't got to fog area yet but from what you written seem unrelated.

antoniodesousa · 2025-10-08T15:30:20Z

Interesting, so AMD has better handling of 2D textures in 3D modes than Nvidia on Vulkan? 😄
Anyways, the fog is still broken but it's unrelated to this ticket.

mborgerson · 2026-01-15T05:22:38Z

/gemini review

gemini-code-assist

Code Review

This pull request correctly addresses an issue with DstAlpha blending on surfaces with forced alpha bits by introducing a fixup_blend_factor_for_surface function. This new function remaps the blend factors for both OpenGL and Vulkan renderers, which should fix the incorrect blending behavior. The changes are well-structured and the addition of a new surface.c file to contain this logic is a good approach. My review includes a few suggestions to improve code readability and maintainability by caching a register value to avoid redundant lookups and refactoring a series of if statements into a switch.

gemini-code-assist · 2026-01-15T05:23:46Z

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

+        uint32_t sfactor = fixup_blend_factor_for_surface(
+            GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_BLEND),
+                     NV_PGRAPH_BLEND_SFACTOR),
+            &pg->surface_shape);
+        uint32_t dfactor = fixup_blend_factor_for_surface(
+            GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_BLEND),
+                     NV_PGRAPH_BLEND_DFACTOR),
+            &pg->surface_shape);


For improved readability and to avoid multiple lookups of the same register, you could cache the value of pgraph_reg_r(pg, NV_PGRAPH_BLEND) in a local variable.

const uint32_t blend_reg = pgraph_reg_r(pg, NV_PGRAPH_BLEND); uint32_t sfactor = fixup_blend_factor_for_surface( GET_MASK(blend_reg, NV_PGRAPH_BLEND_SFACTOR), &pg->surface_shape); uint32_t dfactor = fixup_blend_factor_for_surface( GET_MASK(blend_reg, NV_PGRAPH_BLEND_DFACTOR), &pg->surface_shape);

gemini-code-assist · 2026-01-15T05:23:46Z

hw/xbox/nv2a/pgraph/surface.c

+        if (blend_factor == NV_PGRAPH_BLEND_SFACTOR_DST_ALPHA) {
+            return NV_PGRAPH_BLEND_SFACTOR_ONE;
+        }
+        if (blend_factor ==
+            NV_PGRAPH_BLEND_SFACTOR_ONE_MINUS_DST_ALPHA) {
+            return NV_PGRAPH_BLEND_SFACTOR_ZERO;
+        }
+        if (blend_factor == NV_PGRAPH_BLEND_DFACTOR_DST_ALPHA) {
+            return NV_PGRAPH_BLEND_DFACTOR_ONE;
+        }
+        if (blend_factor ==
+            NV_PGRAPH_BLEND_DFACTOR_ONE_MINUS_DST_ALPHA) {
+            return NV_PGRAPH_BLEND_DFACTOR_ZERO;
+        }


To improve readability and make the logic clearer, consider refactoring the series of if statements into a switch statement on blend_factor.

Suggested change

if (blend_factor == NV_PGRAPH_BLEND_SFACTOR_DST_ALPHA) {

return NV_PGRAPH_BLEND_SFACTOR_ONE;

}

if (blend_factor ==

NV_PGRAPH_BLEND_SFACTOR_ONE_MINUS_DST_ALPHA) {

return NV_PGRAPH_BLEND_SFACTOR_ZERO;

}

if (blend_factor == NV_PGRAPH_BLEND_DFACTOR_DST_ALPHA) {

return NV_PGRAPH_BLEND_DFACTOR_ONE;

}

if (blend_factor ==

NV_PGRAPH_BLEND_DFACTOR_ONE_MINUS_DST_ALPHA) {

return NV_PGRAPH_BLEND_DFACTOR_ZERO;

}

switch (blend_factor) {

case NV_PGRAPH_BLEND_SFACTOR_DST_ALPHA:

return NV_PGRAPH_BLEND_SFACTOR_ONE;

case NV_PGRAPH_BLEND_SFACTOR_ONE_MINUS_DST_ALPHA:

return NV_PGRAPH_BLEND_SFACTOR_ZERO;

case NV_PGRAPH_BLEND_DFACTOR_DST_ALPHA:

return NV_PGRAPH_BLEND_DFACTOR_ONE;

case NV_PGRAPH_BLEND_DFACTOR_ONE_MINUS_DST_ALPHA:

return NV_PGRAPH_BLEND_DFACTOR_ZERO;

}

gemini-code-assist · 2026-01-15T05:23:46Z

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

+        uint32_t sfactor = fixup_blend_factor_for_surface(
+            GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_BLEND),
+                     NV_PGRAPH_BLEND_SFACTOR),
+            &pg->surface_shape);
+        uint32_t dfactor = fixup_blend_factor_for_surface(
+            GET_MASK(pgraph_reg_r(pg, NV_PGRAPH_BLEND),
+                     NV_PGRAPH_BLEND_DFACTOR),
+            &pg->surface_shape);


Similar to the change in gl/draw.c, you can cache the value of pgraph_reg_r(pg, NV_PGRAPH_BLEND) to improve readability and avoid redundant register reads.

const uint32_t blend_reg = pgraph_reg_r(pg, NV_PGRAPH_BLEND); uint32_t sfactor = fixup_blend_factor_for_surface( GET_MASK(blend_reg, NV_PGRAPH_BLEND_SFACTOR), &pg->surface_shape); uint32_t dfactor = fixup_blend_factor_for_surface( GET_MASK(blend_reg, NV_PGRAPH_BLEND_DFACTOR), &pg->surface_shape);

Copilot

Pull request overview

This PR fixes an issue with destination alpha (DstAlpha) blending on surface formats that have forced alpha bits (e.g., X8R8G8B8 formats where "X" represents constant/unused alpha bits). The fix remaps DstAlpha blend factors to appropriate alternatives when the target surface format has forced alpha values.

Changes:

Adds a new fixup_blend_factor_for_surface() function to handle blend factor remapping for special surface formats
Updates both GL and Vulkan rendering backends to apply the blend factor fixup before using the factors
Minor whitespace cleanup in GL draw code

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
hw/xbox/nv2a/pgraph/surface.c	New file implementing the blend factor fixup logic for surfaces with forced alpha bits
hw/xbox/nv2a/pgraph/surface.h	Adds function declaration and includes for the new fixup function
hw/xbox/nv2a/pgraph/vk/draw.c	Applies blend factor fixup before creating Vulkan pipeline
hw/xbox/nv2a/pgraph/gl/draw.c	Applies blend factor fixup before setting OpenGL blend function, includes whitespace fix
hw/xbox/nv2a/pgraph/meson.build	Adds surface.c to the build

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-15T05:25:30Z

hw/xbox/nv2a/pgraph/surface.c

+        if (blend_factor == NV_PGRAPH_BLEND_SFACTOR_DST_ALPHA) {
+            return NV_PGRAPH_BLEND_SFACTOR_ONE;
+        }
+        if (blend_factor ==
+            NV_PGRAPH_BLEND_SFACTOR_ONE_MINUS_DST_ALPHA) {
+            return NV_PGRAPH_BLEND_SFACTOR_ZERO;
+        }
+        if (blend_factor == NV_PGRAPH_BLEND_DFACTOR_DST_ALPHA) {
+            return NV_PGRAPH_BLEND_DFACTOR_ONE;
+        }
+        if (blend_factor ==
+            NV_PGRAPH_BLEND_DFACTOR_ONE_MINUS_DST_ALPHA) {
+            return NV_PGRAPH_BLEND_DFACTOR_ZERO;
+        }


The function checks both SFACTOR and DFACTOR constants for the same blend_factor parameter. While this works because the numeric values are identical (DST_ALPHA=6, ONE_MINUS_DST_ALPHA=7 for both), it's confusing and could be error-prone if the constants ever diverge. Consider simplifying by checking numeric values or documenting this assumption clearly. For example, you could add a comment explaining that SFACTOR and DFACTOR constants share the same numeric values.

This remaps DstAlpha to alternate source/destination factors when performing blending on target surfaces with forced alpha bits (e.g., NV097_SET_SURFACE_FORMAT_COLOR_LE_X8R8G8B8_Z8R8G8B8). Fixes xemu-project#2473 Tests: https://github.com/abaire/nxdk_pgraph_tests/blob/de548172ac3166f03f3a75999829ba17aade73aa/src/tests/blend_surface_tests.cpp#L249 HW results: https://abaire.github.io/nxdk_pgraph_tests_golden_results/results/Blend_surface/index.html

abaire marked this pull request as ready for review October 8, 2025 14:17

abaire mentioned this pull request Oct 8, 2025

Chase: Hollywood Stunt Driver Black water. - Passing cubemap to shader expecting sampler2D #2384

Closed

mborgerson requested a review from Copilot January 15, 2026 05:22

Copilot started reviewing on behalf of mborgerson January 15, 2026 05:23 View session

gemini-code-assist bot reviewed Jan 15, 2026

View reviewed changes

Copilot AI reviewed Jan 15, 2026

View reviewed changes

abaire force-pushed the fix_2473_prevent_alpha_updates_for_forced_value_surfaces branch 3 times, most recently from 9fbb60a to e211bf4 Compare January 19, 2026 14:45

abaire force-pushed the fix_2473_prevent_alpha_updates_for_forced_value_surfaces branch from e211bf4 to 77eec63 Compare January 19, 2026 14:46

Uh oh!

nv2a: Fix DstAlpha blending for const-alpha surface modes #2478

Are you sure you want to change the base?

nv2a: Fix DstAlpha blending for const-alpha surface modes #2478

Conversation

abaire commented Oct 8, 2025

Uh oh!

abaire commented Oct 8, 2025

Uh oh!

antoniodesousa commented Oct 8, 2025

Vulkan

OpenGL

Uh oh!

abaire commented Oct 8, 2025

Uh oh!

Triticum0 commented Oct 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

antoniodesousa commented Oct 8, 2025

Uh oh!

mborgerson commented Jan 15, 2026

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

abaire Jan 17, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

abaire Jan 17, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

abaire Jan 17, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 15, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Triticum0 commented Oct 8, 2025 •

edited

Loading