Rollup merge of #139700 - EnzymeAD:autodiff-flags, r=oli-obk

Autodiff flags

Interestingly, it seems that some other projects have conflicts with exactly the same LLVM optimization passes as autodiff.
At least `LLVMRustOptimize` has exactly the flags that we need to disable problematic opt passes.

This PR enables us to compile code where users differentiate two identical functions in the same module. This has been especially common in test cases, but it's not impossible to encounter in the wild.

It also enables two new flags for testing/debugging. I consider writing an MCP to upgrade PrintPasses to be a standalone -Z flag, since it is *not* the same as `-Z print-llvm-passes`, which IMHO gives less useful output. A discussion can be found here: [#t-compiler/llvm > Print llvm passes. @ 💬](https://rust-lang.zulipchat.com/#narrow/channel/187780-t-compiler.2Fllvm/topic/Print.20llvm.20passes.2E/near/511533038)

Finally, it improves `PrintModBefore` and `PrintModAfter`. They used to work reliable, but now we just schedule enzyme as part of an existing ModulePassManager (MPM). Since Enzyme is last in the MPM scheduling, PrintModBefore became very inaccurate. It used to print the input module, which we gave to the Enzyme and was great to create llvm-ir reproducer. However, lately the MPM would run the whole `default<O3>` pipeline, which heavily modifies the llvm module, before we pass it to Enzyme. That made it impossible to use the flag to create llvm-ir reproducers for Enzyme bugs. We now schedule a PrintModule pass just before Enzyme, solving this problem.

Based on the PrintPass output, it also _seems_ like changing `registerEnzymeAndPassPipeline(PB, true);` to `registerEnzymeAndPassPipeline(PB, false);` has no effect. In theory, the bool should tell Enzyme to schedule some helpful passes in the PassBuilder. However, since it doesn't do anything and I'm not 100% sure anymore on whether we really need it, I'll just disable it for now and postpone investigations.

r? ``@oli-obk``

closes #139471

Tracking:

- https://github.com/rust-lang/rust/issues/124509

1 files changed, 45 insertions, 0 deletions

diff --git a/tests/codegen/autodiff/identical_fnc.rs b/tests/codegen/autodiff/identical_fnc.rs
new file mode 100644
index 00000000000..1c3277f52b4
--- /dev/null
+++ b/tests/codegen/autodiff/identical_fnc.rs
@@ -0,0 +1,45 @@
+//@ compile-flags: -Zautodiff=Enable -C opt-level=3  -Clto=fat
+//@ no-prefer-dynamic
+//@ needs-enzyme
+//
+// Each autodiff invocation creates a new placeholder function, which we will replace on llvm-ir
+// level. If a user tries to differentiate two identical functions within the same compilation unit,
+// then LLVM might merge them in release mode before AD. In that case we can't rewrite one of the
+// merged placeholder function anymore, and compilation would fail. We prevent this by disabling
+// LLVM's merge_function pass before AD. Here we implicetely test that our solution keeps working.
+// We also explicetly test that we keep running merge_function after AD, by checking for two
+// identical function calls in the LLVM-IR, while having two different calls in the Rust code.
+#![feature(autodiff)]
+
+use std::autodiff::autodiff;
+
+#[autodiff(d_square, Reverse, Duplicated, Active)]
+fn square(x: &f64) -> f64 {
+    x * x
+}
+
+#[autodiff(d_square2, Reverse, Duplicated, Active)]
+fn square2(x: &f64) -> f64 {
+    x * x
+}
+
+// CHECK:; identical_fnc::main
+// CHECK-NEXT:; Function Attrs:
+// CHECK-NEXT:define internal void @_ZN13identical_fnc4main17hf4dbc69c8d2f9130E()
+// CHECK-NEXT:start:
+// CHECK-NOT:br
+// CHECK-NOT:ret
+// CHECK:; call identical_fnc::d_square
+// CHECK-NEXT:  call fastcc void @_ZN13identical_fnc8d_square17h4c364207a2f8e06dE(double %x.val, ptr noalias noundef nonnull align 8 dereferenceable(8) %dx1)
+// CHECK-NEXT:; call identical_fnc::d_square
+// CHECK-NEXT:  call fastcc void @_ZN13identical_fnc8d_square17h4c364207a2f8e06dE(double %x.val, ptr noalias noundef nonnull align 8 dereferenceable(8) %dx2)
+
+fn main() {
+    let x = std::hint::black_box(3.0);
+    let mut dx1 = std::hint::black_box(1.0);
+    let mut dx2 = std::hint::black_box(1.0);
+    let _ = d_square(&x, &mut dx1, 1.0);
+    let _ = d_square2(&x, &mut dx2, 1.0);
+    assert_eq!(dx1, 6.0);
+    assert_eq!(dx2, 6.0);
+}