Apple's open-source XNU kernel on Android phones: its own loader and drivers, Metal on the Adreno. Runs from RAM; a reboot gives the phone back to Android.
C
docs/air: how far the rounding patch reaches, counted
vcvkkd04e96b6e2db
diff --git a/docs/air.md b/docs/air.mdindex ea1f82e..76b2bc2 100644--- a/docs/air.md+++ b/docs/air.md@@ -159,6 +159,19 @@ functions and not entry points at all.declares nothing and this frontend declares nothing, which is exactly why two independent pathsreturned the same wrong number. **The hardware can round; ir3 only asks when someone else hasasked first, and supplies no default where IEEE-754 and GLSL both require round-to-nearest.**+ **How much this reaches is a number and not an adjective.** Of the 456 modules of the corpus, 20+ contain a conversion the patch can change — a `u2f`, an `i2f` or an `f2f16`, the three that+ round; none contains only the exact kinds (`f2f32`, `b2f`), where setting the mode cannot alter a+ value. So on the Metal side the blast radius is 20 modules, not all 313 that translate, and+ whether each of those 20 actually changes depends on the values it meets at run time.++ That bounds this frontend's half and says nothing about the other. The port's OpenGL ES reaches+ ir3 through Mesa's own GLSL compiler, which this check does not look at, so **"it changes+ everything ir3 compiles, the window server included" is reasoning and not measurement.** A peer+ session is putting that to a visual test — an ordinary boot to the home screen with the patch,+ compared against one without it — because the claim is the part of this that would matter to+ someone deciding whether to carry the patch, and it is the part with no runs behind it.+The fix is one line, and unlike the `ir3_cmdline` bug this one is in a file upstream stillcarries, with reproduction down two independent paths — so it is the first thing in this workthat can actually be reported. The change is now carried here as `tools/mesa/ir3-round-default.patch`, measured on the