Remove invariant loads on primal computation

EnzymeAD · Mar 5, 2024 · 11608c9 · 11608c9
1 parent b71fc1c
commit 11608c9
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 0 deletions.
diff --git 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
@@ -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)