bump llvm

[fifield]

update llvm to 1cde124
account for expand_shape changes

[jsetoain]

Yes, I was going to say. Between the version I originally used to replicate the other issues and the latest one, another change broke another test 😬

[jsetoain]

Yes, I was going to say. Between the version I originally used to replicate the other issues and the latest one, another change broke another test 😬

@makslevental It looks like we're not running the integration tests any more (the ones activated with -DAIE_INCLUDE_INTEGRATION_TESTS=ON), that's why CI didn't catch the issues.

[fifield]

Yes, I was going to say. Between the version I originally used to replicate the other issues and the latest one, another change broke another test 😬

@makslevental It looks like we're not running the integration tests any more (the ones activated with -DAIE_INCLUDE_INTEGRATION_TESTS=ON), that's why CI didn't catch the issues.

It looks like I wasn't running those tests either. I get a few failures after turning them on.

Failed Tests (7):
  AIE :: Integration/Dialect/TOSA/bf16_ceil_v16/bf16_ceil.mlir
  AIE :: Integration/Dialect/TOSA/bf16_ceil_v32/bf16_ceil.mlir
  AIE :: Integration/Dialect/TOSA/bf16_floor_v16/bf16_floor.mlir
  AIE :: Integration/Dialect/TOSA/bf16_floor_v32/bf16_floor.mlir
  AIE :: Integration/Dialect/TOSA/bf16_sigmoid_v16bf16/bf16_sigmoid.mlir
  AIE :: Integration/Dialect/TOSA/bf16_sigmoid_v32bf16/bf16_sigmoid.mlir
  AIE :: Integration/julia_by_lines/aie.mlir

[jsetoain]

Yes, I was going to say. Between the version I originally used to replicate the other issues and the latest one, another change broke another test 😬

@makslevental It looks like we're not running the integration tests any more (the ones activated with -DAIE_INCLUDE_INTEGRATION_TESTS=ON), that's why CI didn't catch the issues.

It looks like I wasn't running those tests either. I get a few failures after turning them on.

The same happened to me due to recent dev. env shenanigans. Anyway, the patch I sneaked in right before this one fixes those issues, but it looks like after rebase a test failed. I suspect it was one of those spurious failures we're getting later because the previous run finished successfully, might be worth looking into it.

I forgot, this is the failed attempt: https://github.com/Xilinx/mlir-aie/actions/runs/8942799480/attempts/1?pr=1455