diff --git a/src/compiler.jl b/src/compiler.jl
index 771dc896db..52ccb6290f 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)
diff --git a/src/compiler/optimize.jl b/src/compiler/optimize.jl
index f430830dd4..70ee9b6912 100644
--- a/src/compiler/optimize.jl
+++ b/src/compiler/optimize.jl
@@ -1690,6 +1690,7 @@ function post_optimze!(mod, tm, machine=true)
         addOptimizationPasses!(pm)
         run!(pm, mod)
     end
+    println(string(mod))
     if machine
         # TODO enable validate_return_roots
         # validate_return_roots!(mod)