`Symbols not found: [ cblas_xerbla ]`

[vchuravy]

julia> function f_test_1(A, x)
           u = A*x[2:end] .+ x[1]
           return u
       end
f_test_1 (generic function with 1 method)

julia> J_f_1(A, x) = Enzyme.jacobian(Reverse, θ -> f_test_1(A, θ), x, Val(1))
J_f_1 (generic function with 1 method)

julia> x = ones(6)

julia> using LinearAlgebra

julia> A = Matrix{Float64}(LinearAlgebra.I, 5, 5)

From

https://discourse.julialang.org/t/mutating-array-workaround-for-reverse-ad/81009

[vchuravy]

julia> J_f_1(A, x)
warning: Linking two modules of different target triples: 'bcloader' is 'x86_64-unknown-linux-gnu' whereas 'text' is 'x86_64-pc-linux-gnu'

warning: Linking two modules of different target triples: 'bcloader' is 'x86_64-unknown-linux-gnu' whereas 'text' is 'x86_64-pc-linux-gnu'

┌ Warning: Using fallback BLAS replacements, performance may be degraded
└ @ Enzyme.Compiler ~/src/GPUCompiler/src/utils.jl:35
warning: didn't implement memmove, using memcpy as fallback which can result in errors
JIT session error: Symbols not found: [ cblas_xerbla ]
warning: Linking two modules of different target triples: 'bcloader' is 'x86_64-unknown-linux-gnu' whereas 'text' is 'x86_64-pc-linux-gnu'

warning: Linking two modules of different target triples: 'bcloader' is 'x86_64-unknown-linux-gnu' whereas 'text' is 'x86_64-pc-linux-gnu'

┌ Warning: Using fallback BLAS replacements, performance may be degraded
└ @ Enzyme.Compiler ~/src/GPUCompiler/src/utils.jl:35
warning: didn't implement memmove, using memcpy as fallback which can result in errors
JIT session error: Symbols not found: [ cblas_xerbla ]
warning: Linking two modules of different target triples: 'bcloader' is 'x86_64-unknown-linux-gnu' whereas 'text' is 'x86_64-pc-linux-gnu'

warning: Linking two modules of different target triples: 'bcloader' is 'x86_64-unknown-linux-gnu' whereas 'text' is 'x86_64-pc-linux-gnu'

┌ Warning: Using fallback BLAS replacements, performance may be degraded
└ @ Enzyme.Compiler ~/src/GPUCompiler/src/utils.jl:35
warning: didn't implement memmove, using memcpy as fallback which can result in errors
JIT session error: Symbols not found: [ cblas_xerbla ]
ERROR: LLVM error: Failed to materialize symbols: { (main, { augmented_julia__11_3790_inner_5wrap, diffejulia__11_3790_inner_5wrap }) }

@wsmoses would also be great if we can get rid of the merging warning

[wsmoses]

Module warning comes from LLVM proper and there's no way to disable. At best we force the triples to be the same to make the warning irrelevant...

[vchuravy]

We can change the triple to be the expected one before inlining

…

On Mon, May 23, 2022, 12:27 William Moses ***@***.***> wrote: Module warning comes from LLVM proper and there's no way to disable. At best we force the triples to be the same to make the warning irrelevant... — Reply to this email directly, view it on GitHub <#312 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABDO2QGT4EGD6PNQDK3D3DVLOWX7ANCNFSM5V3IXVVA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[vchuravy]

JIT session error: Symbols not found: [ cblas_xerbla ]
ERROR: LLVM error: Failed to materialize symbols: { (main, { diffejulia__5_3352_inner_5wrap, augmented_julia__5_3352_inner_5wrap }) }
Stacktrace:

from

using Enzyme, LinearAlgebra

 function f_test_1(A, x)
        u = A*x[2:end] .+ x[1]
        return u
    end

    J_f_1(A, x) = Enzyme.jacobian(Reverse, θ -> f_test_1(A, θ), x, Val(1))

    x = ones(6)
    A = Matrix{Float64}(LinearAlgebra.I, 5, 5)
    J_f_1(A, x)

[wsmoses]

Okay, this I may just need your help for to duck code something I did wrong. The example, I tested locally passed (when I didn't have the if function in module check). With the check (necessary for CI/elsewhere), the if body never runs.

[vchuravy]

I opened #319 to add tests for this