Add a comment to explain the #[inline(never)] annotation

and regenerate the platform intrinsics source files.

Author: ranma42

src/etc/platform-intrinsics/generator.py

@@ -737,6 +737,8 @@ def open(self, platform):
 use IntrinsicDef::Named;
 use rustc::middle::ty;
 
+// The default inlining settings trigger a pathological behaviour in
+// LLVM, which causes makes compilation very slow. See #28273.
 #[inline(never)]
 pub fn find<'tcx>(_tcx: &ty::ctxt<'tcx>, name: &str) -> Option<Intrinsic> {{
     if !name.starts_with("{0}") {{ return None }}

src/librustc_platform_intrinsics/aarch64.rs

@@ -17,6 +17,8 @@ use {Intrinsic, i, i_, u, u_, f, v, v_, agg, p, void};
 use IntrinsicDef::Named;
 use rustc::middle::ty;
 
+// The default inlining settings trigger a pathological behaviour in
+// LLVM, which causes makes compilation very slow. See #28273.
 #[inline(never)]
 pub fn find<'tcx>(_tcx: &ty::ctxt<'tcx>, name: &str) -> Option<Intrinsic> {
     if !name.starts_with("aarch64_v") { return None }

src/librustc_platform_intrinsics/arm.rs

@@ -13,10 +13,12 @@
 
 #![allow(unused_imports)]
 
-use {Intrinsic, i, i_, u, u_, f, v, agg, p};
+use {Intrinsic, i, i_, u, u_, f, v, v_, agg, p, void};
 use IntrinsicDef::Named;
 use rustc::middle::ty;
 
+// The default inlining settings trigger a pathological behaviour in
+// LLVM, which causes makes compilation very slow. See #28273.
 #[inline(never)]
 pub fn find<'tcx>(_tcx: &ty::ctxt<'tcx>, name: &str) -> Option<Intrinsic> {
     if !name.starts_with("arm_v") { return None }

src/librustc_platform_intrinsics/x86.rs

@@ -17,6 +17,8 @@ use {Intrinsic, i, i_, u, u_, f, v, v_, agg, p, void};
 use IntrinsicDef::Named;
 use rustc::middle::ty;
 
+// The default inlining settings trigger a pathological behaviour in
+// LLVM, which causes makes compilation very slow. See #28273.
 #[inline(never)]
 pub fn find<'tcx>(_tcx: &ty::ctxt<'tcx>, name: &str) -> Option<Intrinsic> {
     if !name.starts_with("x86_mm") { return None }