AMDGPU ci

[wsmoses]

No description provided.

[wsmoses]

@avik-pal this attr look familiar?

┌ Error: Compilation failed, MLIR module written to /tmp/reactant_Qiv1U1/module_000_reactant_mysoftmax!_post_xla_compile.mlir
--
  | └ @ Reactant.MLIR.IR /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/mlir/IR/Pass.jl:119
  | Basic softmax: Error During Test at /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/test/basic.jl:82
  | Got exception outside of a @test
  | INTERNAL: ttg.total-num-warps attribute not found.
  |  
  | Stacktrace:
  | [1] reactant_err(msg::Cstring)
  | @ Reactant.XLA /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/xla/Utils.jl:12
  | [2] (::Reactant.XLA.IFRT.var"#36#37"{Bool, Vector{Int64}, Int64, Int64, Bool, Reactant.XLA.IFRT.Client, Reactant.MLIR.IR.Module, Int64})()
  | @ Reactant.XLA.IFRT /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/xla/IFRT/LoadedExecutable.jl:89
  | [3] try_compile_dump_mlir(f::Reactant.XLA.IFRT.var"#36#37"{Bool, Vector{Int64}, Int64, Int64, Bool, Reactant.XLA.IFRT.Client, Reactant.MLIR.IR.Module, Int64}, mod::Reactant.MLIR.IR.Module, pm::Nothing)
  | @ Reactant.MLIR.IR /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/mlir/IR/Pass.jl:134
  | [4] try_compile_dump_mlir
  | @ /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/mlir/IR/Pass.jl:129 [inlined]
  | [5] #compile#35
  | @ /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/xla/IFRT/LoadedExecutable.jl:88 [inlined]
  | [6] compile_xla(f::Function, args::Tuple{ConcreteIFRTArray{Float64, 2, Nothing}}; before_xla_optimizations::Bool, client::Nothing, serializable::Bool, kwargs::@Kwargs{compile_options::CompileOptions, fn_kwargs::@NamedTuple{}})
  | @ Reactant.Compiler /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/Compiler.jl:3574
  | [7] compile_xla
  | @ /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/Compiler.jl:3502 [inlined]
  | [8] compile(f::Function, args::Tuple{ConcreteIFRTArray{Float64, 2, Nothing}}; kwargs::@Kwargs{fn_kwargs::@NamedTuple{}, client::Nothing, reshape_propagate::Symbol, raise_first::Bool, assert_nonallocating::Bool, legalize_chlo_to_stablehlo::Bool, transpose_propagate::Symbol, donated_args::Symbol, optimize_then_pad::Bool, cudnn_hlo_optimize::Bool, compile_options::Missing, sync::Bool, no_nan::Bool, raise::Bool, shardy_passes::Symbol, optimize::Bool, optimize_communications::Bool})
  | @ Reactant.Compiler /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/Compiler.jl:3606
  | [9] macro expansion
  | @ /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/Compiler.jl:2675 [inlined]
  | [10] macro expansion
  | @ /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/test/basic.jl:88 [inlined]
  | [11] macro expansion
  | @ ~/.cache/julia-buildkite-plugin/julia_installs/bin/linux/x64/1.10/julia-1.10-latest-linux-x86_64/share/julia/stdlib/v1.10/Test/src/Test.jl:1577 [inlined]
  | [12] top-level scope
  | @ /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/test/basic.jl:83
  | [13] include(mod::Module, _path::String)
  | @ Base ./Base.jl:495
  | [14] include(x::String)
  | @ Main.var"##Basic#228" ~/.cache/julia-buildkite-plugin/depots/0190ad31-dfb1-4a4c-ac5b-02d05277ba6c/packages/SafeTestsets/raUNr/src/SafeTestsets.jl:28
  | [15] macro expansion
  | @ ~/.cache/julia-buildkite-plugin/depots/0190ad31-dfb1-4a4c-ac5b-02d05277ba6c/packages/SafeTestsets/raUNr/src/SafeTestsets.jl:24 [inlined]
  | [16] macro expansion
  | @ ~/.cache/julia-buildkite-plugin/julia_installs/bin/linux/x64/1.10/julia-1.10-latest-linux-x86_64/share/julia/stdlib/v1.10/Test/src/Test.jl:1577 [inlined]
  | [17] top-level scope
  | @ ~/.cache/julia-buildkite-plugin/depots/0190ad31-dfb1-4a4c-ac5b-02d05277ba6c/packages/SafeTestsets/raUNr/src/SafeTestsets.jl:24
  | [18] eval(m::Module, e::Any)
  | @ Core ./boot.jl:385
  | [19] macro expansion
  | @ ~/.cache/julia-buildkite-plugin/depots/0190ad31-dfb1-4a4c-ac5b-02d05277ba6c/packages/SafeTestsets/raUNr/src/SafeTestsets.jl:28 [inlined]
  | [20] macro expansion
  | @ /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/test/runtests.jl:17 [inlined]
  | [21] macro expansion
  | @ ~/.cache/julia-buildkite-plugin/julia_installs/bin/linux/x64/1.10/julia-1.10-latest-linux-x86_64/share/julia/stdlib/v1.10/Test/src/Test.jl:1577 [inlined]
  | [22] top-level scope
  | @ /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/test/runtests.jl:10
  | [23] include(fname::String)
  | @ Base.MainInclude ./client.jl:494
  | [24] top-level scope
  | @ none:6
  | [25] eval
  | @ ./boot.jl:385 [inlined]
  | [26] exec_options(opts::Base.JLOptions)
  | @ Base ./client.jl:296
  | [27] _start()
  | @ Base ./client.jl:557
  | ┌ Error: Compilation failed, MLIR module written to /tmp/reactant_Qiv1U1/module_001_reactant_*_post_xla_compile.mlir
  | └ @ Reactant.MLIR.IR /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/src/mlir/IR/Pass.jl:119
  | matmul: Error During Test at /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/test/basic.jl:140
  | Test threw exception
  | Expression: #= /var/lib/buildkite-agent/builds/amdgpu1-luraess-com/julialang/reactant-dot-jl/test/basic.jl:140 =# @jit(c * d) ≈ ones(50, 70) * ones(70, 30)
  | INTERNAL: Autotuning failed for HLO: %custom-call.1 = (f64[30,50]{1,0}, s8[44800]{0}) custom-call(%arg2__path___args__2__.1, %arg1__path___args__1__.1), custom_call_target="__cublas$gemm", backend_config={"operation_queue_id":"0","wait_on_operation_queues":[],"gemm_backend_config":{"alpha_real":1,"beta":0,"dot_dimension_numbers":{"lhs_contracting_dimensions":["1"],"rhs_contracting_dimensions":["0"],"lhs_batch_dimensions":[],"rhs_batch_dimensions":[]},"alpha_imag":0,"precision_config":{"operand_precision":["DEFAULT","DEFAULT"],"algorithm":"ALG_UNSET"},"epilogue":"DEFAULT","lhs_stride":"2100","rhs_stride":"3500","grad_x":false,"grad_y":false,"damax_output":false},"force_earliest_schedule":false,"reification_cost":[],"device_type":"DEVICE_TYPE_INVALID"} with error: NOT_FOUND: No valid config found!

also some cublas dgemm nonsense

[avik-pal]

ttg.total-num-warps is from triton lowering.

[avik-pal]

https://github.com/openxla/xla/blob/802f4306b18ee544f4bb0fa127b55c61054fe069/xla/backends/gpu/codegen/triton/emitter_helpers.cc#L585-L638

[avik-pal]

https://github.com/openxla/xla/blob/802f4306b18ee544f4bb0fa127b55c61054fe069/xla/backends/gpu/codegen/triton/emitter_helpers.cc#L603-L615 is surely a cuda not rocm path?