Rollup merge of rust-lang#87916 - nbdd0121:black_box, r=nagisa

Implement `black_box` using intrinsic

Introduce `black_box` intrinsic, as suggested in rust-lang#87590 (comment).

This is still codegenned as empty inline assembly for LLVM. For MIR interpretation and cranelift it's treated as identity.

cc `@Amanieu` as this is related to inline assembly
cc `@bjorn3` for rustc_codegen_cranelift changes
cc `@RalfJung` as this affects MIRI

r? `@nagisa` I suppose

compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs

@@ -1136,6 +1136,11 @@ pub(crate) fn codegen_intrinsic_call<'tcx>(
                 };
             ret.write_cvalue(fx, CValue::by_val(is_eq_value, ret.layout()));
         };
+
+        black_box, (c a) {
+            // FIXME implement black_box semantics
+            ret.write_cvalue(fx, a);
+        };
     }
 
     if let Some((_, dest)) = destination {

compiler/rustc_codegen_llvm/src/asm.rs

@@ -425,7 +425,7 @@ impl AsmMethods for CodegenCx<'ll, 'tcx> {
     }
 }
 
-fn inline_asm_call(
+pub(crate) fn inline_asm_call(
     bx: &mut Builder<'a, 'll, 'tcx>,
     asm: &str,
     cons: &str,

compiler/rustc_codegen_llvm/src/intrinsic.rs

@@ -7,6 +7,7 @@ use crate::type_of::LayoutLlvmExt;
 use crate::va_arg::emit_va_arg;
 use crate::value::Value;
 
+use rustc_ast as ast;
 use rustc_codegen_ssa::base::{compare_simd_types, wants_msvc_seh};
 use rustc_codegen_ssa::common::span_invalid_monomorphization_error;
 use rustc_codegen_ssa::common::{IntPredicate, TypeKind};
@@ -327,6 +328,31 @@ impl IntrinsicCallMethods<'tcx> for Builder<'a, 'll, 'tcx> {
                 }
             }
 
+            sym::black_box => {
+                args[0].val.store(self, result);
+
+                // We need to "use" the argument in some way LLVM can't introspect, and on
+                // targets that support it we can typically leverage inline assembly to do
+                // this. LLVM's interpretation of inline assembly is that it's, well, a black
+                // box. This isn't the greatest implementation since it probably deoptimizes
+                // more than we want, but it's so far good enough.
+                crate::asm::inline_asm_call(
+                    self,
+                    "",
+                    "r,~{memory}",
+                    &[result.llval],
+                    self.type_void(),
+                    true,
+                    false,
+                    ast::LlvmAsmDialect::Att,
+                    &[span],
+                )
+                .unwrap_or_else(|| bug!("failed to generate inline asm call for `black_box`"));
+
+                // We have copied the value to `result` already.
+                return;
+            }
+
             _ if name_str.starts_with("simd_") => {
                 match generic_simd_intrinsic(self, name, callee_ty, args, ret_ty, llret_ty, span) {
                     Ok(llval) => llval,

compiler/rustc_mir/src/interpret/intrinsics.rs

@@ -465,7 +465,7 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
                 );
                 self.copy_op(&self.operand_index(&args[0], index)?, dest)?;
             }
-            sym::likely | sym::unlikely => {
+            sym::likely | sym::unlikely | sym::black_box => {
                 // These just return their argument
                 self.copy_op(&args[0], dest)?;
             }

compiler/rustc_span/src/symbol.rs

@@ -335,6 +335,7 @@ symbols! {
         bitreverse,
         bitxor,
         bitxor_assign,
+        black_box,
         block,
         bool,
         borrowck_graphviz_format,

compiler/rustc_typeck/src/check/intrinsic.rs

@@ -102,6 +102,7 @@ pub fn intrinsic_operation_unsafety(intrinsic: Symbol) -> hir::Unsafety {
         | sym::maxnumf64
         | sym::type_name
         | sym::forget
+        | sym::black_box
         | sym::variant_count => hir::Unsafety::Normal,
         _ => hir::Unsafety::Unsafe,
     }
@@ -387,6 +388,8 @@ pub fn check_intrinsic_type(tcx: TyCtxt<'_>, it: &hir::ForeignItem<'_>) {
                 (1, vec![param_ty; 2], tcx.types.bool)
             }
 
+            sym::black_box => (1, vec![param(0)], param(0)),
+
             other => {
                 tcx.sess.emit_err(UnrecognizedIntrinsicFunction { span: it.span, name: other });
                 return;

library/core/src/hint.rs

@@ -152,23 +152,19 @@ pub fn spin_loop() {
 /// backend used. Programs cannot rely on `black_box` for *correctness* in any way.
 ///
 /// [`std::convert::identity`]: crate::convert::identity
-#[cfg_attr(not(miri), inline)]
-#[cfg_attr(miri, inline(never))]
+#[inline]
 #[unstable(feature = "bench_black_box", issue = "64102")]
-#[cfg_attr(miri, allow(unused_mut))]
+#[cfg_attr(not(bootstrap), allow(unused_mut))]
 pub fn black_box<T>(mut dummy: T) -> T {
-    // We need to "use" the argument in some way LLVM can't introspect, and on
-    // targets that support it we can typically leverage inline assembly to do
-    // this. LLVM's interpretation of inline assembly is that it's, well, a black
-    // box. This isn't the greatest implementation since it probably deoptimizes
-    // more than we want, but it's so far good enough.
-
-    #[cfg(not(miri))] // This is just a hint, so it is fine to skip in Miri.
+    #[cfg(bootstrap)]
     // SAFETY: the inline assembly is a no-op.
     unsafe {
-        // FIXME: Cannot use `asm!` because it doesn't support MIPS and other architectures.
         llvm_asm!("" : : "r"(&mut dummy) : "memory" : "volatile");
+        dummy
     }
 
-    dummy
+    #[cfg(not(bootstrap))]
+    {
+        crate::intrinsics::black_box(dummy)
+    }
 }

library/core/src/intrinsics.rs

@@ -1933,6 +1933,12 @@ extern "rust-intrinsic" {
     /// which is UB if any of their inputs are `undef`.)
     #[rustc_const_unstable(feature = "const_intrinsic_raw_eq", issue = "none")]
     pub fn raw_eq<T>(a: &T, b: &T) -> bool;
+
+    /// See documentation of [`std::hint::black_box`] for details.
+    ///
+    /// [`std::hint::black_box`]: crate::hint::black_box
+    #[cfg(not(bootstrap))]
+    pub fn black_box<T>(dummy: T) -> T;
 }
 
 // Some functions are defined here because they accidentally got made