Auto merge of #121223 - RalfJung:simd-intrinsics, r=Amanieu

intrinsics::simd: add missing functions, avoid UB-triggering fast-math

Turns out stdarch declares a bunch more SIMD intrinsics that are still missing from libcore.
I hope I got the docs and in particular the safety requirements right for these "unordered" and "nanless" intrinsics.

Many of these are unused even in stdarch, but they are implemented in the codegen backend, so we may as well list them here.

r? `@Amanieu`
Cc `@calebzulawski` `@workingjubilee`

1 files changed, 14 insertions, 0 deletions

diff --git a/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp b/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp
index 7326f2e8e2a..af2353fbb19 100644
--- a/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp
+++ b/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp
@@ -450,6 +450,20 @@ extern "C" void LLVMRustSetAlgebraicMath(LLVMValueRef V) {
   }
 }
 
+// Enable the reassoc fast-math flag, allowing transformations that pretend
+// floating-point addition and multiplication are associative.
+//
+// Note that this does NOT enable any flags which can cause a floating-point operation on
+// well-defined inputs to return poison, and therefore this function can be used to build
+// safe Rust intrinsics (such as fadd_algebraic).
+//
+// https://llvm.org/docs/LangRef.html#fast-math-flags
+extern "C" void LLVMRustSetAllowReassoc(LLVMValueRef V) {
+  if (auto I = dyn_cast<Instruction>(unwrap<Value>(V))) {
+    I->setHasAllowReassoc(true);
+  }
+}
+
 extern "C" LLVMValueRef
 LLVMRustBuildAtomicLoad(LLVMBuilderRef B, LLVMTypeRef Ty, LLVMValueRef Source,
                         const char *Name, LLVMAtomicOrdering Order) {