diff --git a/src/compiler.jl b/src/compiler.jl index 8405ccbdec..f62f0cdeb7 100644 --- a/src/compiler.jl +++ b/src/compiler.jl @@ -4366,6 +4366,9 @@ function GPUCompiler.codegen(output::Symbol, job::CompilerJob{<:EnzymeTarget}; mod, meta = GPUCompiler.codegen(:llvm, primal_job; optimize=false, toplevel=toplevel, cleanup=false, validate=false, parent_job=parent_job) prepare_llvm(mod, primal_job, meta) + for f in functions(mod) + permit_inlining!(f) + end LLVM.ModulePassManager() do pm API.AddPreserveNVVMPass!(pm, #=Begin=#true)