Do not add chess tests to CI unless strictly necessary

[newling]

The chess tests take 1 minute to compile each, if we're not careful we're going to quickly get 1hr+ CI runs which will slow development down quite badly. See Time spent in compilation: 59091.0 [ms] in following from CI:

Found HIP installation: /usr, version 5.7.31921
 "/home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/llvm-aie/bin/ld.lld" --nmagic /home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/test_aie_vs_cpu/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess/input.o /home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/test_aie_vs_cpu/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess/mm_npu1.o --gc-sections --orphan-handling=error -T /home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/test_aie_vs_cpu/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess/core_1_5.elf.ld -L/home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/llvm-aie/bin/../lib/aie2-none-unknown-elf -L/home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/llvm-aie/lib/clang/19/lib/aie2-none-unknown-elf /home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/llvm-aie/lib/clang/19/lib/aie2-none-unknown-elf/libclang_rt.builtins.a -lc -lm /home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/llvm-aie/bin/../lib/aie2-none-unknown-elf/crt0.o /home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/llvm-aie/bin/../lib/aie2-none-unknown-elf/crt1.o -o /home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/test_aie_vs_cpu/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess/core_1_5.elf
Time spent in compilation: 59091.0 [ms]
Run #1 of 2 for /home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/test_aie_vs_cpu/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess.mlir
Running the following command:
/home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/iree-install/bin/iree-run-module --module=/home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/test_aie_vs_cpu/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess_aie.vmfb --input=512x4096xbf16=@/home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/test_aie_vs_cpu/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess_input1.bin --input=4096x512xbf16=@/home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/test_aie_vs_cpu/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess_input2.bin --device=xrt-lite --output=@/home/nod/actions-runner/_work/iree-amd-aie/iree-amd-aie/test_aie_vs_cpu/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess/matmul_512_512_4096_bf16_f32_O3_npu1_4col_outline_ukernel_chess_aie.bin --xrt_lite_n_core_rows=4 --xrt_lite_n_core_cols=4
Standard output from script:
EXEC @matmul

Maybe we should split the CI tests in chess and non-chess tests too, so that we can see that all peano tests pass quickly?

Update: this includes all ukernel tests that use chess. Maybe the solution will be to change ukernel tests to use peano for assembling (currently this doesn't work out-of-the-box). @jtuyls thoughts?

[jtuyls]

Update: this includes all ukernel tests that use chess. Maybe the solution will be to change ukernel tests to use peano for assembling (currently this doesn't work out-of-the-box). @jtuyls thoughts?

Yeah, that's one of the reasons I have been working on getting in compilation of ukernels through peano. Except for a couple of (performance) tests, I think we should move the remaining tests to peano indeed.

[jtuyls]

#1114 reduced the E2E Strix CI time from 40+ min to 12min.