[oneandonlydean]

What

With no lane active, v_readfirstlane_b32 reads lane 0 on hardware (the RDNA ISA gives the lane as the first set bit of EXEC, or lane 0 when EXEC is zero). EmitReadFirstLane shuffled from the ballot's first lane without that case:

• one host lane per invocation: OpGroupNonUniformBallotFindLSB of an empty ballot is undefined in SPIR-V (on my RTX 4090 it gave lane 31);
• a wave64 program over two 32-lane halves: FirstLane takes FindILsb(high) + 32 when the low word is empty, and with FindILsb(0) = -1 that is lane 31.

The emitter now selects lane 0 when both ballot words are zero, and the first active lane otherwise.

The v_writelane half of this area went in with #254. #322 (v_readlane lane wrap) and #411 (v_writelane and EXEC) edit the functions right below this one; this branch merges cleanly with each of them, alone and both together.

Tested

• Linux (CachyOS, x86-64, RTX 4090, subgroup size 32): full build with BUILD_TESTING=ON and ctest pass.
• New execution test agc_driver_read_first_lane_tests (tests/execution/ReadFirstLane.cpp, exits 77 without Vulkan): a wave32 and a wave64 compute shader each do v_readfirstlane_b32 of 0x100 + lane with EXEC empty (expects lane 0) and with only lanes 4 and 5 (wave32) or 40 and 41 (wave64) active, and store both results from every lane. On this GPU the wave64 shader takes the two-lane path. Against main it fails with "read first lane wave32: lane 0 result 0 is 287, expected 256" (lane 31).
• PPSA21564: on my integration branch since 2026-09-30, through repeated passes of my visual regression check (checkpoint frames, flicker, shimmer, video tearing, FPS). I have no title symptom tied to this; it is from the ISA.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed in TechnicalDebt
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• AI-assisted: yes