Auto merge of #76830 - Artoria2e5:tune, r=nagisa

Pass tune-cpu to LLVM

I think this is how it should work...

See https://internals.rust-lang.org/t/expose-tune-cpu-from-llvm/13088 for the background. Or the documentation diff.

2 files changed, 39 insertions, 2 deletions

diff --git a/src/doc/rustc/src/codegen-options/index.md b/src/doc/rustc/src/codegen-options/index.md
index bed10ca16d3..f6493e49c3c 100644
--- a/src/doc/rustc/src/codegen-options/index.md
+++ b/src/doc/rustc/src/codegen-options/index.md
@@ -497,8 +497,10 @@ point instructions in software. It takes one of the following values:
 This instructs `rustc` to generate code specifically for a particular processor.
 
 You can run `rustc --print target-cpus` to see the valid options to pass
-here. Additionally, `native` can be passed to use the processor of the host
-machine. Each target has a default base CPU.
+here. Each target has a default base CPU. Special values include:
+
+* `native` can be passed to use the processor of the host machine. 
+* `generic` refers to an LLVM target with minimal features but modern tuning.
 
 ## target-feature
 
@@ -530,6 +532,20 @@ This also supports the feature `+crt-static` and `-crt-static` to control
 Each target and [`target-cpu`](#target-cpu) has a default set of enabled
 features.
 
+## tune-cpu
+
+This instructs `rustc` to schedule code specifically for a particular
+processor. This does not affect the compatibility (instruction sets or ABI),
+but should make your code slightly more efficient on the selected CPU.
+
+The valid options are the same as those for [`target-cpu`](#target-cpu).
+The default is `None`, which LLVM translates as the `target-cpu`.
+
+This is an unstable option. Use `-Z tune-cpu=machine` to specify a value.
+
+Due to limitations in LLVM (12.0.0-git9218f92), this option is currently
+effective only for x86 targets.
+
 [option-emit]: ../command-line-arguments.md#option-emit
 [option-o-optimize]: ../command-line-arguments.md#option-o-optimize
 [profile-guided optimization]: ../profile-guided-optimization.md
diff --git a/src/test/codegen/tune-cpu-on-functions.rs b/src/test/codegen/tune-cpu-on-functions.rs
new file mode 100644
index 00000000000..9121799cdbf
--- /dev/null
+++ b/src/test/codegen/tune-cpu-on-functions.rs
@@ -0,0 +1,21 @@
+// This test makes sure that functions get annotated with the proper
+// "tune-cpu" attribute in LLVM.
+
+// no-prefer-dynamic
+// ignore-tidy-linelength
+// compile-flags: -C no-prepopulate-passes -C panic=abort -C linker-plugin-lto -Cpasses=name-anon-globals -Z tune-cpu=generic
+
+#![crate_type = "staticlib"]
+
+// CHECK-LABEL: define {{.*}} @exported() {{.*}} #0
+#[no_mangle]
+pub extern fn exported() {
+    not_exported();
+}
+
+// CHECK-LABEL: ; tune_cpu_on_functions::not_exported
+// CHECK-NEXT: ; Function Attrs:
+// CHECK-NEXT: define {{.*}}() {{.*}} #0
+fn not_exported() {}
+
+// CHECK: attributes #0 = {{.*}} "tune-cpu"="{{.*}}"