@device_code_agx errors when Metal Shader Validation is enabled

[christiangnrd]

MWE

julia> using Metal

julia> Metal.versioninfo()
macOS 15.0.1, Darwin 24.0.0

Toolchain:
- Julia: 1.11.1
- LLVM: 16.0.6

Julia packages: 
- Metal.jl: 1.4.0
- GPUArrays: 10.3.1
- GPUCompiler: 1.0.0
- KernelAbstractions: 0.9.28
- ObjectiveC: 3.1.0
- LLVM: 9.1.2
- LLVMDowngrader_jll: 0.3.0+1

Environment:
- MTL_SHADER_VALIDATION: 1
- MTL_DEBUG_LAYER: 1

2024-10-17 16:40:59.867 julia[17424:9358094] Metal API Validation Enabled
2024-10-17 16:40:59.867 julia[17424:9358094] Metal GPU Validation Enabled
1 device:
- Apple M2 Max (192.000 KiB allocated)

julia> dummy() = return
dummy (generic function with 1 method)

julia> @device_code_agx @metal dummy()
; GPUCompiler.CompilerJob{GPUCompiler.MetalCompilerTarget, Metal.MetalCompilerParams}(MethodInstance for dummy(), CompilerConfig for GPUCompiler.MetalCompilerTarget, 0x0000000000006845)

ERROR: Expected a universal binary, got a Vector{ObjectFile.MachO.MachOHandle{IOStream}}
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:35
  [2] extract_gpu_code(f::Metal.var"#188#191"{String, Base.TTY}, binary::String)
    @ Metal ~/.julia/dev/Metal/src/compiler/reflection.jl:96
  [3] #187
    @ ~/.julia/dev/Metal/src/compiler/reflection.jl:69 [inlined]
  [4] mktempdir(fn::Metal.var"#187#190"{Metal.MTL.MTLBinaryArchiveInstance, Base.TTY}, parent::String; prefix::String)
    @ Base.Filesystem ./file.jl:819
  [5] mktempdir (repeats 2 times)
    @ ./file.jl:815 [inlined]
  [6] macro expansion
    @ ~/.julia/dev/Metal/src/compiler/reflection.jl:61 [inlined]
  [7] (::Metal.var"#186#189"{Base.TTY, GPUCompiler.CompilerJob{GPUCompiler.MetalCompilerTarget, Metal.MetalCompilerParams}})()
    @ Metal ~/.julia/packages/ObjectiveC/C7BVt/src/foundation.jl:637
  [8] macro expansion
    @ ~/.julia/packages/ObjectiveC/C7BVt/src/foundation.jl:565 [inlined]
  [9] macro expansion
    @ ./lock.jl:273 [inlined]
 [10] ObjectiveC.Foundation.NSAutoreleasePool(f::Metal.var"#186#189"{Base.TTY, GPUCompiler.CompilerJob{…}})
    @ ObjectiveC.Foundation ~/.julia/packages/ObjectiveC/C7BVt/src/foundation.jl:557
 [11] code_agx
    @ ~/.julia/packages/ObjectiveC/C7BVt/src/foundation.jl:636 [inlined]
 [12] (::Metal.var"#hook#203"{…})(job::GPUCompiler.CompilerJob{…}; io::Base.TTY, kwargs::@Kwargs{})
    @ Metal ~/.julia/dev/Metal/src/compiler/reflection.jl:216
 [13] (::Metal.var"#hook#203"{…})(job::GPUCompiler.CompilerJob{…})
    @ Metal ~/.julia/dev/Metal/src/compiler/reflection.jl:213
 [14] var"#4#outer_hook"(job::GPUCompiler.CompilerJob{GPUCompiler.MetalCompilerTarget, Metal.MetalCompilerParams})
    @ Main ~/.julia/packages/GPUCompiler/QCNA0/src/reflection.jl:174
 [15] compile(target::Symbol, job::GPUCompiler.CompilerJob; kwargs::@Kwargs{})
    @ GPUCompiler ~/.julia/packages/GPUCompiler/QCNA0/src/driver.jl:76
 [16] compile
    @ ~/.julia/packages/GPUCompiler/QCNA0/src/driver.jl:74 [inlined]
 [17] (::Metal.var"#155#163"{GPUCompiler.CompilerJob{…}})(ctx::LLVM.Context)
    @ Metal ~/.julia/dev/Metal/src/compiler/compilation.jl:108
 [18] JuliaContext(f::Metal.var"#155#163"{GPUCompiler.CompilerJob{…}}; kwargs::@Kwargs{})
    @ GPUCompiler ~/.julia/packages/GPUCompiler/QCNA0/src/driver.jl:34
 [19] JuliaContext(f::Function)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/QCNA0/src/driver.jl:25
 [20] macro expansion
    @ ~/.julia/dev/Metal/src/compiler/compilation.jl:107 [inlined]
 [21] macro expansion
    @ ~/.julia/packages/ObjectiveC/C7BVt/src/os.jl:264 [inlined]
 [22] compile(job::GPUCompiler.CompilerJob)
    @ Metal ~/.julia/dev/Metal/src/compiler/compilation.jl:105
 [23] actual_compilation(cache::Dict{…}, src::Core.MethodInstance, world::UInt64, cfg::GPUCompiler.CompilerConfig{…}, compiler::typeof(Metal.compile), linker::typeof(Metal.link))
    @ GPUCompiler ~/.julia/packages/GPUCompiler/QCNA0/src/execution.jl:237
 [24] cached_compilation(cache::Dict{…}, src::Core.MethodInstance, cfg::GPUCompiler.CompilerConfig{…}, compiler::Function, linker::Function)
    @ GPUCompiler ~/.julia/packages/GPUCompiler/QCNA0/src/execution.jl:151
 [25] macro expansion
    @ ~/.julia/dev/Metal/src/compiler/execution.jl:189 [inlined]
 [26] macro expansion
    @ ./lock.jl:273 [inlined]
 [27] mtlfunction(f::typeof(dummy), tt::Type{Tuple{}}; name::Nothing, kwargs::@Kwargs{})
    @ Metal ~/.julia/dev/Metal/src/compiler/execution.jl:184
 [28] mtlfunction(f::typeof(dummy), tt::Type{Tuple{}})
    @ Metal ~/.julia/dev/Metal/src/compiler/execution.jl:182
 [29] macro expansion
    @ ~/.julia/dev/Metal/src/compiler/execution.jl:85 [inlined]
 [30] top-level scope
    @ ~/.julia/packages/GPUCompiler/QCNA0/src/reflection.jl:185
 [31] top-level scope
    @ ~/.julia/dev/Metal/src/initialization.jl:72
Some type information was truncated. Use `show(err)` to see complete types.

[maleadt]

Hah, interesting. I've considered removing this functionality, though, as the disassembler doesn't work on my M3, isn't actively maintained, and drags along several annoying dependencies (like Python_jll).