Auto merge of rust-lang#119911 - NCGThompson:is-statically-known, r=oli-obk

Replacement of rust-lang#114390: Add new intrinsic `is_var_statically_known` and optimize pow for powers of two

This adds a new intrinsic `is_val_statically_known` that lowers to [``@llvm.is.constant.*`](https://llvm.org/docs/LangRef.html#llvm-is-constant-intrinsic).` It also applies the intrinsic in the int_pow methods to recognize and optimize the idiom `2isize.pow(x)`. See rust-lang#114390 for more discussion.

While I have extended the scope of the power of two optimization from rust-lang#114390, I haven't added any new uses for the intrinsic. That can be done in later pull requests.

Note: When testing or using the library, be sure to use `--stage 1` or higher. Otherwise, the intrinsic will be a noop and the doctests will be skipped. If you are trying out edits, you may be interested in [`--keep-stage 0`](https://rustc-dev-guide.rust-lang.org/building/suggested.html#faster-builds-with---keep-stage).

Fixes rust-lang#47234
Resolves rust-lang#114390
`@Centri3`

Author: bors

compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs

@@ -443,6 +443,12 @@ fn codegen_regular_intrinsic_call<'tcx>(
 
             ret.write_cvalue(fx, a);
         }
+        sym::is_val_statically_known => {
+            intrinsic_args!(fx, args => (_a); intrinsic);
+
+            let res = fx.bcx.ins().iconst(types::I8, 0);
+            ret.write_cvalue(fx, CValue::by_val(res, ret.layout()));
+        }
         sym::breakpoint => {
             intrinsic_args!(fx, args => (); intrinsic);
 

compiler/rustc_codegen_gcc/src/context.rs

@@ -196,15 +196,16 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
 
         let mut functions = FxHashMap::default();
         let builtins = [
-            "__builtin_unreachable", "abort", "__builtin_expect", "__builtin_add_overflow", "__builtin_mul_overflow",
-            "__builtin_saddll_overflow", /*"__builtin_sadd_overflow",*/ "__builtin_smulll_overflow", /*"__builtin_smul_overflow",*/
+            "__builtin_unreachable", "abort", "__builtin_expect", /*"__builtin_expect_with_probability",*/
+            "__builtin_constant_p", "__builtin_add_overflow", "__builtin_mul_overflow", "__builtin_saddll_overflow",
+            /*"__builtin_sadd_overflow",*/ "__builtin_smulll_overflow", /*"__builtin_smul_overflow",*/
             "__builtin_ssubll_overflow", /*"__builtin_ssub_overflow",*/ "__builtin_sub_overflow", "__builtin_uaddll_overflow",
             "__builtin_uadd_overflow", "__builtin_umulll_overflow", "__builtin_umul_overflow", "__builtin_usubll_overflow",
             "__builtin_usub_overflow", "sqrtf", "sqrt", "__builtin_powif", "__builtin_powi", "sinf", "sin", "cosf", "cos",
             "powf", "pow", "expf", "exp", "exp2f", "exp2", "logf", "log", "log10f", "log10", "log2f", "log2", "fmaf",
             "fma", "fabsf", "fabs", "fminf", "fmin", "fmaxf", "fmax", "copysignf", "copysign", "floorf", "floor", "ceilf",
             "ceil", "truncf", "trunc", "rintf", "rint", "nearbyintf", "nearbyint", "roundf", "round",
-            "__builtin_expect_with_probability",
+           
         ];
 
         for builtin in builtins.iter() {

compiler/rustc_codegen_gcc/src/intrinsic/mod.rs

@@ -123,6 +123,12 @@ impl<'a, 'gcc, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
                 sym::unlikely => {
                     self.expect(args[0].immediate(), false)
                 }
+                sym::is_val_statically_known => {
+                    let a = args[0].immediate();
+                    let builtin = self.context.get_builtin_function("__builtin_constant_p");
+                    let res = self.context.new_call(None, builtin, &[a]);
+                    self.icmp(IntPredicate::IntEQ, res, self.const_i32(0))
+                }
                 kw::Try => {
                     try_intrinsic(
                         self,

compiler/rustc_codegen_llvm/src/context.rs

@@ -916,6 +916,20 @@ impl<'ll> CodegenCx<'ll, '_> {
         ifn!("llvm.lifetime.start.p0i8", fn(t_i64, ptr) -> void);
         ifn!("llvm.lifetime.end.p0i8", fn(t_i64, ptr) -> void);
 
+        // FIXME: This is an infinitesimally small portion of the types you can
+        // pass to this intrinsic, if we can ever lazily register intrinsics we
+        // should register these when they're used, that way any type can be
+        // passed.
+        ifn!("llvm.is.constant.i1", fn(i1) -> i1);
+        ifn!("llvm.is.constant.i8", fn(t_i8) -> i1);
+        ifn!("llvm.is.constant.i16", fn(t_i16) -> i1);
+        ifn!("llvm.is.constant.i32", fn(t_i32) -> i1);
+        ifn!("llvm.is.constant.i64", fn(t_i64) -> i1);
+        ifn!("llvm.is.constant.i128", fn(t_i128) -> i1);
+        ifn!("llvm.is.constant.isize", fn(t_isize) -> i1);
+        ifn!("llvm.is.constant.f32", fn(t_f32) -> i1);
+        ifn!("llvm.is.constant.f64", fn(t_f64) -> i1);
+
         ifn!("llvm.expect.i1", fn(i1, i1) -> i1);
         ifn!("llvm.eh.typeid.for", fn(ptr) -> t_i32);
         ifn!("llvm.localescape", fn(...) -> void);

compiler/rustc_codegen_llvm/src/intrinsic.rs

@@ -119,6 +119,10 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> {
             sym::likely => {
                 self.call_intrinsic("llvm.expect.i1", &[args[0].immediate(), self.const_bool(true)])
             }
+            sym::is_val_statically_known => self.call_intrinsic(
+                &format!("llvm.is.constant.{:?}", args[0].layout.immediate_llvm_type(self.cx)),
+                &[args[0].immediate()],
+            ),
             sym::unlikely => self
                 .call_intrinsic("llvm.expect.i1", &[args[0].immediate(), self.const_bool(false)]),
             kw::Try => {

compiler/rustc_const_eval/src/const_eval/machine.rs

@@ -531,6 +531,11 @@ impl<'mir, 'tcx> interpret::Machine<'mir, 'tcx> for CompileTimeInterpreter<'mir,
                     )?;
                 }
             }
+            // The intrinsic represents whether the value is known to the optimizer (LLVM).
+            // We're not doing any optimizations here, so there is no optimizer that could know the value.
+            // (We know the value here in the machine of course, but this is the runtime of that code,
+            // not the optimization stage.)
+            sym::is_val_statically_known => ecx.write_scalar(Scalar::from_bool(false), dest)?,
             _ => {
                 throw_unsup_format!(
                     "intrinsic `{intrinsic_name}` is not supported at compile-time"

compiler/rustc_hir_analysis/src/check/intrinsic.rs

@@ -453,6 +453,8 @@ pub fn check_intrinsic_type(tcx: TyCtxt<'_>, it: &hir::ForeignItem<'_>) {
 
             sym::black_box => (1, vec![param(0)], param(0)),
 
+            sym::is_val_statically_known => (1, vec![param(0)], tcx.types.bool),
+
             sym::const_eval_select => (4, vec![param(0), param(1), param(2)], param(3)),
 
             sym::vtable_size | sym::vtable_align => {

compiler/rustc_span/src/symbol.rs

@@ -910,6 +910,7 @@ symbols! {
         io_stderr,
         io_stdout,
         irrefutable_let_patterns,
+        is_val_statically_known,
         isa_attribute,
         isize,
         issue,

library/core/src/intrinsics.rs

@@ -2517,6 +2517,66 @@ extern "rust-intrinsic" {
     where
         G: FnOnce<ARG, Output = RET>,
         F: FnOnce<ARG, Output = RET>;
+
+    /// Returns whether the argument's value is statically known at
+    /// compile-time.
+    ///
+    /// This is useful when there is a way of writing the code that will
+    /// be *faster* when some variables have known values, but *slower*
+    /// in the general case: an `if is_val_statically_known(var)` can be used
+    /// to select between these two variants. The `if` will be optimized away
+    /// and only the desired branch remains.
+    ///
+    /// Formally speaking, this function non-deterministically returns `true`
+    /// or `false`, and the caller has to ensure sound behavior for both cases.
+    /// In other words, the following code has *Undefined Behavior*:
+    ///
+    /// ```
+    /// #![feature(is_val_statically_known)]
+    /// #![feature(core_intrinsics)]
+    /// # #![allow(internal_features)]
+    /// use std::hint::unreachable_unchecked;
+    /// use std::intrinsics::is_val_statically_known;
+    ///
+    /// unsafe {
+    ///    if !is_val_statically_known(0) { unreachable_unchecked(); }
+    /// }
+    /// ```
+    ///
+    /// This also means that the following code's behavior is unspecified; it
+    /// may panic, or it may not:
+    ///
+    /// ```no_run
+    /// #![feature(is_val_statically_known)]
+    /// #![feature(core_intrinsics)]
+    /// # #![allow(internal_features)]
+    /// use std::intrinsics::is_val_statically_known;
+    ///
+    /// unsafe {
+    ///     assert_eq!(is_val_statically_known(0), is_val_statically_known(0));
+    /// }
+    /// ```
+    ///
+    /// Unsafe code may not rely on `is_val_statically_known` returning any
+    /// particular value, ever. However, the compiler will generally make it
+    /// return `true` only if the value of the argument is actually known.
+    ///
+    /// When calling this in a `const fn`, both paths must be semantically
+    /// equivalent, that is, the result of the `true` branch and the `false`
+    /// branch must return the same value and have the same side-effects *no
+    /// matter what*.
+    #[rustc_const_unstable(feature = "is_val_statically_known", issue = "none")]
+    #[rustc_nounwind]
+    #[cfg(not(bootstrap))]
+    pub fn is_val_statically_known<T: Copy>(arg: T) -> bool;
+}
+
+// FIXME: Seems using `unstable` here completely ignores `rustc_allow_const_fn_unstable`
+// and thus compiling stage0 core doesn't work.
+#[rustc_const_stable(feature = "is_val_statically_known", since = "0.0.0")]
+#[cfg(bootstrap)]
+pub const unsafe fn is_val_statically_known<T: Copy>(_arg: T) -> bool {
+    false
 }
 
 // Some functions are defined here because they accidentally got made

library/core/src/lib.rs

@@ -200,6 +200,7 @@
 //
 // Language features:
 // tidy-alphabetical-start
+#![cfg_attr(not(bootstrap), feature(is_val_statically_known))]
 #![feature(abi_unadjusted)]
 #![feature(adt_const_params)]
 #![feature(allow_internal_unsafe)]