test: allow qwen3 tilelet cases on a3

Author: 牢大

.github/workflows/ci.yml

@@ -293,7 +293,7 @@ jobs:
           # suite (RUN_ONLY_CASES is empty), skip the non-matching variant based
           # on SOC_VERSION to keep the remote validation portable.
           A3_ONLY_CASES="partition5d,partition5d_dynamic,mrgsort,tmatmulk_autosync"
-          A5_ONLY_CASES="partition5d_a5,partition5d_dynamic_a5,mrgsort_a5,tmatmulk_autosync_a5,qwen3_decode_layer_incore_1,qwen3_decode_layer_incore_2,qwen3_decode_layer_incore_10,qwen3_decode_layer_incore_13,qwen3_decode_layer_incore_14"
+          A5_ONLY_CASES="partition5d_a5,partition5d_dynamic_a5,mrgsort_a5,tmatmulk_autosync_a5"
 
           sv_lc="$(printf '%s' "${SOC_VERSION}" | tr '[:upper:]' '[:lower:]')"
           is_a5=0

test/samples/Qwen3Tilelet/README.md

@@ -2,7 +2,7 @@ Qwen3 tilelet PTO kernels generated from `pypto-lib/examples/models/qwen3/qwen3_
 
 Scope:
 - compile-regression inputs for `ptoas`
-- A5-only kernels; `runop.sh` injects `--pto-arch a5 --pto-level=level3` for this directory unless the caller already overrides `PTOAS_FLAGS`
+- tilelet kernels that default to `--pto-arch a5 --pto-level=level3` in `runop.sh`, but can also be compiled on A3 when the caller overrides `PTOAS_FLAGS`
 
 Notes:
 - The source PyPTO program lowers to 20 `qwen3_decode_layer_incore_*.pto` fragments; this directory vendors the full emitted `.pto` set regenerated from the tilelet source with `BATCH_TILE=16`.

test/samples/runop.sh

@@ -154,8 +154,8 @@ process_one_dir() {
     use_ptobc_roundtrip=1
   fi
   # Qwen3 tilelet kernels currently serve as direct ptoas compile-regression
-  # coverage. They require A5/level3 lowering, but are not expected to
-  # roundtrip through ptobc yet.
+  # coverage. Default them to A5/level3 lowering when the caller does not
+  # provide an explicit arch, but let A3/A5 callers override PTOAS_FLAGS.
   if [[ "$A" == "Qwen3Tilelet" ]]; then
     use_ptobc_roundtrip=0
   fi
@@ -942,10 +942,6 @@ PY
       if [[ "$A" == "Qwen3Tilelet" ]]; then
         cpp="${out_subdir}/${base}-pto.cpp"
       fi
-      if [[ "$A" == "Qwen3Tilelet" && "$(printf '%s' "$target_arch" | tr '[:upper:]' '[:lower:]')" != "a5" ]]; then
-        echo -e "${A}(${base}.pto)\tSKIP\trequires --pto-arch=a5"
-        continue
-      fi
       local sample_use_ptobc_roundtrip="$use_ptobc_roundtrip"
 
       # TODO(ptobc): decode of this regression currently fails with