From 4eeda739c4ea856e37b9daced4564a17e30aceaa Mon Sep 17 00:00:00 2001
From: scavenger <ajax@redhat.com>
Date: Mon, 9 Feb 2026 21:21:02 -0500
Subject: [PATCH] add VK CTS validation report for a0 interpolation fix

---
 .gitignore                                 |  6 +++
 claude-nwnk/a0-interpolation-cts-report.md | 58 ++++++++++++++++++++++
 2 files changed, 64 insertions(+)
 create mode 100644 claude-nwnk/a0-interpolation-cts-report.md

diff --git a/.gitignore b/.gitignore
index 1ccc49fc360..72333560e01 100644
--- a/.gitignore
+++ b/.gitignore
@@ -7,3 +7,9 @@
 /build
 .venv/
 *.mda.tar
+
+# Gas Town (added by gt)
+.runtime/
+.claude/
+.beads/
+.logs/
diff --git a/claude-nwnk/a0-interpolation-cts-report.md b/claude-nwnk/a0-interpolation-cts-report.md
new file mode 100644
index 00000000000..a665729fae7
--- /dev/null
+++ b/claude-nwnk/a0-interpolation-cts-report.md
@@ -0,0 +1,58 @@
+# VK CTS validation: fury's a0 interpolation fix
+
+convoy: hq-cv-wr4iu (llvmpipe-rasterizer-validation)
+
+## summary
+
+fury's fix changes interpolation origin from framebuffer (0,0) to vertex 0
+position for numerical stability with triangles far from origin.
+
+**result: no regressions, no fixes detected. fix is safe from a CTS perspective.**
+
+## test results
+
+tested dEQP-VK categories focused on rasterization/interpolation.
+
+### dEQP-VK.rasterization.*
+
+| build | passed | failed | not supported |
+|-------|--------|--------|---------------|
+| baseline | 7448 | 0 | 7631 |
+| with fix | 7448 | 0 | 7631 |
+
+### dEQP-VK.draw.*
+
+| build | passed | failed | not supported |
+|-------|--------|--------|---------------|
+| baseline | 24015 | 0 | 6729 |
+| with fix | 24015 | 0 | 6729 |
+
+### dEQP-VK.pipeline.monolithic.multisample.*.verify_interpolation.*
+
+| build | passed | failed | not supported |
+|-------|--------|--------|---------------|
+| baseline | 15 | 0 | 18 |
+| with fix | 15 | 0 | 18 |
+
+## notes
+
+- the fix is designed for numerical stability with triangles far from origin,
+  which the CTS may not explicitly stress.
+- identical results baseline vs fix suggests no functional regression.
+- the "not supported" counts are expected - lavapipe doesn't expose all
+  features (shader_tile_image, android external formats, high sample counts, etc).
+
+## files changed
+
+12 files in src/gallium/drivers/llvmpipe/:
+- lp_bld_interp.c
+- lp_linear.c
+- lp_linear_fastpath.c
+- lp_linear_interp.c
+- lp_linear_priv.h
+- lp_linear_sampler.c
+- lp_rast.c
+- lp_setup_line.c
+- lp_setup_point.c
+- lp_state_fs_linear.c
+- lp_state_setup.c