test: add A5-safe incore RMSNorm/decode projection PTO cases

[TaoTao-real]

Summary

• Add two Sync sample PTO testcases for the reported A5 failure cases:
  • test/samples/Sync/rmsnorm_incore_0.pto
  • test/samples/Sync/decode_projection_incore_0.pto
• Make both cases A5-safe by forcing partition row offsets to %c0 instead of %arg3 to avoid out-of-range row slicing under multi-core launch.

Motivation

• These two cases run on A3 but fail/hang on A5 in field testing.
• Root cause is row partition offsets using %arg3 while tensor view first dimension is fixed to 16; non-zero %arg3 can generate invalid accesses on A5 dispatch.

Testing

• ptoas compile check (A5 level3) passed:
  • /Users/lishengtao/Documents/PTO/PTOAS_main_upstream_b07733b/build/tools/ptoas/ptoas --pto-arch=a5 --pto-level=level3 test/samples/Sync/rmsnorm_incore_0.pto -o /tmp/rmsnorm_incore_0.cpp
  • /Users/lishengtao/Documents/PTO/PTOAS_main_upstream_b07733b/build/tools/ptoas/ptoas --pto-arch=a5 --pto-level=level3 test/samples/Sync/decode_projection_incore_0.pto -o /tmp/decode_projection_incore_0.cpp

Risk / Rollback

• Risk is low: only adds two testcase files and constrains row offsets to single-tile semantics.
• Rollback: revert this PR.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[TaoTao-real]

/run A5 decode_projection_incore_0 --pto-level=level3

[reedhecre]

A5 板测失败

• 触发方式：manual
• 源码提交：3d62d76daa91
• 结果汇总：OK 0 / FAIL 0 / SKIP 0
• 日志：/root/ptoas-board-monitor-a5/logs/20260403_114805_manual_pr437.log
• 手动指令：/run a5 decode_projection_incore_0 --pto-level=level3
• 触发人：TaoTao-real
• 指定用例：decode_projection_incore_0
• PTOAS 参数：--pto-level=level3
• 触发评论：test: add A5-safe incore RMSNorm/decode projection PTO cases #437 (comment)
• 失败阶段：sample-build-and-test / exit=1

日志尾部

ples/planmemory/plan_memory_peak_exact_capacity.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_peak_8_overlapping.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_no_reuse_overlap.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_nested_loops.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_loop_no_reuse_outer_live.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_loop_in_if.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_if_yield.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_if_in_loop.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_fragmentation_two_holes.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_fragmentation_hole_fit.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_for_iter_args_yield.py [not in RUN_ONLY_CASES], test/samples/planmemory/plan_memory_bind_tile_alias_liveness.py [not in RUN_ONLY_CASES], test/samples/Xors/xors_golden.py [not in RUN_ONLY_CASES], test/samples/Xors/xors_compare.py [not in RUN_ONLY_CASES], test/samples/Xors/xors.py [not in RUN_ONLY_CASES], test/samples/Xor/xor_golden.py [not in RUN_ONLY_CASES], test/samples/Xor/xor_compare.py [not in RUN_ONLY_CASES], test/samples/Xor/xor.py [not in RUN_ONLY_CASES], ... (+410 more)

===== STAGE sample-build-and-test @ 2026-04-03 11:49:41 =====
bash test/samples/runop.sh --enablebc all
PTOAS_OUT_DIR=/tmp/ptoas-board-monitor-a5/runs/20260403_114805_manual_pr437/payload/test/samples
========== SUMMARY ==========
Sync(decode_projection_incore_0.pto) FAIL ptobc encode failed: decode_projection_incore_0.pto
-----------------------------
OK=0  FAIL=1  SKIP=0
=============================
===== END STAGE sample-build-and-test rc=1 @ 2026-04-03 11:49:42 =====