Merge d6c3e89 into a7fc463

Author: scottmcm

library/core/src/cmp.rs

@@ -29,6 +29,7 @@ mod bytewise;
 pub(crate) use bytewise::BytewiseEq;
 
 use self::Ordering::*;
+use crate::ops::ControlFlow::{self, Break, Continue};
 
 /// Trait for comparisons using the equality operator.
 ///
@@ -1446,6 +1447,54 @@ pub macro PartialOrd($item:item) {
     /* compiler built-in */
 }
 
+/// Helpers for chaining together field PartialOrds into the full type's ordering.
+///
+/// If the two values are equal, returns `ControlFlow::Continue`.
+/// If the two values are not equal, returns `ControlFlow::Break(self OP other)`.
+///
+/// This allows simple types like `i32` and `f64` to just emit two comparisons
+/// directly, instead of needing to optimize the 3-way comparison.
+///
+/// Currently this is done using specialization, but it doesn't need that:
+/// it could be provided methods on `PartialOrd` instead and work fine.
+pub(crate) trait SpecChainingPartialOrd<Rhs>: PartialOrd<Rhs> {
+    fn spec_chain_lt(&self, other: &Rhs) -> ControlFlow<bool>;
+    fn spec_chain_le(&self, other: &Rhs) -> ControlFlow<bool>;
+    fn spec_chain_gt(&self, other: &Rhs) -> ControlFlow<bool>;
+    fn spec_chain_ge(&self, other: &Rhs) -> ControlFlow<bool>;
+}
+
+impl<T: PartialOrd<U>, U> SpecChainingPartialOrd<U> for T {
+    #[inline]
+    default fn spec_chain_lt(&self, other: &U) -> ControlFlow<bool> {
+        match PartialOrd::partial_cmp(self, other) {
+            Some(Equal) => Continue(()),
+            c => Break(c.is_some_and(Ordering::is_lt)),
+        }
+    }
+    #[inline]
+    default fn spec_chain_le(&self, other: &U) -> ControlFlow<bool> {
+        match PartialOrd::partial_cmp(self, other) {
+            Some(Equal) => Continue(()),
+            c => Break(c.is_some_and(Ordering::is_le)),
+        }
+    }
+    #[inline]
+    default fn spec_chain_gt(&self, other: &U) -> ControlFlow<bool> {
+        match PartialOrd::partial_cmp(self, other) {
+            Some(Equal) => Continue(()),
+            c => Break(c.is_some_and(Ordering::is_gt)),
+        }
+    }
+    #[inline]
+    default fn spec_chain_ge(&self, other: &U) -> ControlFlow<bool> {
+        match PartialOrd::partial_cmp(self, other) {
+            Some(Equal) => Continue(()),
+            c => Break(c.is_some_and(Ordering::is_ge)),
+        }
+    }
+}
+
 /// Compares and returns the minimum of two values.
 ///
 /// Returns the first argument if the comparison determines them to be equal.
@@ -1741,6 +1790,7 @@ where
 mod impls {
     use crate::cmp::Ordering::{self, Equal, Greater, Less};
     use crate::hint::unreachable_unchecked;
+    use crate::ops::ControlFlow::{self, Break, Continue};
 
     macro_rules! partial_eq_impl {
         ($($t:ty)*) => ($(
@@ -1779,6 +1829,36 @@ mod impls {
 
     eq_impl! { () bool char usize u8 u16 u32 u64 u128 isize i8 i16 i32 i64 i128 }
 
+    macro_rules! chaining_impl {
+        ($t:ty) => {
+            // These implementations are the same for `Ord` or `PartialOrd` types
+            // because if either is NAN the `==` test will fail so we end up in
+            // the `Break` case and the comparison will correctly return `false`.
+            impl super::SpecChainingPartialOrd<$t> for $t {
+                #[inline]
+                fn spec_chain_lt(&self, other: &Self) -> ControlFlow<bool> {
+                    let (lhs, rhs) = (*self, *other);
+                    if lhs == rhs { Continue(()) } else { Break(lhs < rhs) }
+                }
+                #[inline]
+                fn spec_chain_le(&self, other: &Self) -> ControlFlow<bool> {
+                    let (lhs, rhs) = (*self, *other);
+                    if lhs == rhs { Continue(()) } else { Break(lhs <= rhs) }
+                }
+                #[inline]
+                fn spec_chain_gt(&self, other: &Self) -> ControlFlow<bool> {
+                    let (lhs, rhs) = (*self, *other);
+                    if lhs == rhs { Continue(()) } else { Break(lhs > rhs) }
+                }
+                #[inline]
+                fn spec_chain_ge(&self, other: &Self) -> ControlFlow<bool> {
+                    let (lhs, rhs) = (*self, *other);
+                    if lhs == rhs { Continue(()) } else { Break(lhs >= rhs) }
+                }
+            }
+        };
+    }
+
     macro_rules! partial_ord_impl {
         ($($t:ty)*) => ($(
             #[stable(feature = "rust1", since = "1.0.0")]
@@ -1801,6 +1881,8 @@ mod impls {
                 #[inline(always)]
                 fn gt(&self, other: &$t) -> bool { (*self) > (*other) }
             }
+
+            chaining_impl!($t);
         )*)
     }
 
@@ -1840,6 +1922,8 @@ mod impls {
                 fn gt(&self, other: &$t) -> bool { (*self) > (*other) }
             }
 
+            chaining_impl!($t);
+
             #[stable(feature = "rust1", since = "1.0.0")]
             impl Ord for $t {
                 #[inline]

library/core/src/tuple.rs

@@ -1,7 +1,9 @@
 // See core/src/primitive_docs.rs for documentation.
 
 use crate::cmp::Ordering::{self, *};
+use crate::cmp::SpecChainingPartialOrd;
 use crate::marker::{ConstParamTy_, StructuralPartialEq, UnsizedConstParamTy};
+use crate::ops::ControlFlow::{Break, Continue};
 
 // Recursive macro for implementing n-ary tuple functions and operations
 //
@@ -80,19 +82,19 @@ macro_rules! tuple_impls {
                 }
                 #[inline]
                 fn lt(&self, other: &($($T,)+)) -> bool {
-                    lexical_ord!(lt, Less, $( ${ignore($T)} self.${index()}, other.${index()} ),+)
+                    lexical_ord!(lt, spec_chain_lt, $( ${ignore($T)} self.${index()}, other.${index()} ),+)
                 }
                 #[inline]
                 fn le(&self, other: &($($T,)+)) -> bool {
-                    lexical_ord!(le, Less, $( ${ignore($T)} self.${index()}, other.${index()} ),+)
+                    lexical_ord!(le, spec_chain_le, $( ${ignore($T)} self.${index()}, other.${index()} ),+)
                 }
                 #[inline]
                 fn ge(&self, other: &($($T,)+)) -> bool {
-                    lexical_ord!(ge, Greater, $( ${ignore($T)} self.${index()}, other.${index()} ),+)
+                    lexical_ord!(ge, spec_chain_ge, $( ${ignore($T)} self.${index()}, other.${index()} ),+)
                 }
                 #[inline]
                 fn gt(&self, other: &($($T,)+)) -> bool {
-                    lexical_ord!(gt, Greater, $( ${ignore($T)} self.${index()}, other.${index()} ),+)
+                    lexical_ord!(gt, spec_chain_gt, $( ${ignore($T)} self.${index()}, other.${index()} ),+)
                 }
             }
         }
@@ -171,15 +173,16 @@ macro_rules! maybe_tuple_doc {
 // `(a1, a2, a3) < (b1, b2, b3)` would be `lexical_ord!(lt, opt_is_lt, a1, b1,
 // a2, b2, a3, b3)` (and similarly for `lexical_cmp`)
 //
-// `$ne_rel` is only used to determine the result after checking that they're
-// not equal, so `lt` and `le` can both just use `Less`.
+// `$chain_rel` is the method from `SpecChainingPartialOrd` to use for all but the
+// final value, to produce better results for simple primitives.
 macro_rules! lexical_ord {
-    ($rel: ident, $ne_rel: ident, $a:expr, $b:expr, $($rest_a:expr, $rest_b:expr),+) => {{
-        let c = PartialOrd::partial_cmp(&$a, &$b);
-        if c != Some(Equal) { c == Some($ne_rel) }
-        else { lexical_ord!($rel, $ne_rel, $($rest_a, $rest_b),+) }
+    ($rel: ident, $chain_rel: ident, $a:expr, $b:expr, $($rest_a:expr, $rest_b:expr),+) => {{
+        match SpecChainingPartialOrd::$chain_rel(&$a, &$b) {
+            Break(val) => val,
+            Continue(()) => lexical_ord!($rel, $chain_rel, $($rest_a, $rest_b),+),
+        }
     }};
-    ($rel: ident, $ne_rel: ident, $a:expr, $b:expr) => {
+    ($rel: ident, $chain_rel: ident, $a:expr, $b:expr) => {
         // Use the specific method for the last element
         PartialOrd::$rel(&$a, &$b)
     };

tests/mir-opt/pre-codegen/tuple_ord.demo_ge_partial.PreCodegen.after.mir

@@ -0,0 +1,70 @@
+// MIR for `demo_ge_partial` after PreCodegen
+
+fn demo_ge_partial(_1: &(f32, f32), _2: &(f32, f32)) -> bool {
+    debug a => _1;
+    debug b => _2;
+    let mut _0: bool;
+    scope 1 (inlined std::cmp::impls::<impl PartialOrd for &(f32, f32)>::le) {
+        scope 2 (inlined core::tuple::<impl PartialOrd for (f32, f32)>::le) {
+            let mut _7: std::ops::ControlFlow<bool>;
+            let _8: bool;
+            scope 3 {
+            }
+            scope 4 (inlined std::cmp::impls::<impl std::cmp::SpecChainingPartialOrd<f32> for f32>::spec_chain_le) {
+                let mut _3: f32;
+                let mut _4: f32;
+                let mut _5: bool;
+                let mut _6: bool;
+                scope 5 {
+                }
+            }
+            scope 6 (inlined std::cmp::impls::<impl PartialOrd for f32>::le) {
+                let mut _9: f32;
+                let mut _10: f32;
+            }
+        }
+    }
+
+    bb0: {
+        StorageLive(_7);
+        StorageLive(_3);
+        StorageLive(_4);
+        _3 = copy ((*_1).0: f32);
+        _4 = copy ((*_2).0: f32);
+        StorageLive(_5);
+        _5 = Eq(copy _3, copy _4);
+        switchInt(move _5) -> [0: bb1, otherwise: bb2];
+    }
+
+    bb1: {
+        StorageLive(_6);
+        _6 = Le(copy _3, copy _4);
+        _7 = ControlFlow::<bool>::Break(move _6);
+        StorageDead(_6);
+        StorageDead(_5);
+        StorageDead(_4);
+        StorageDead(_3);
+        _8 = copy ((_7 as Break).0: bool);
+        _0 = copy _8;
+        goto -> bb3;
+    }
+
+    bb2: {
+        StorageDead(_5);
+        StorageDead(_4);
+        StorageDead(_3);
+        StorageLive(_9);
+        _9 = copy ((*_1).1: f32);
+        StorageLive(_10);
+        _10 = copy ((*_2).1: f32);
+        _0 = Le(move _9, move _10);
+        StorageDead(_10);
+        StorageDead(_9);
+        goto -> bb3;
+    }
+
+    bb3: {
+        StorageDead(_7);
+        return;
+    }
+}

tests/mir-opt/pre-codegen/tuple_ord.demo_le_total.PreCodegen.after.mir

@@ -0,0 +1,70 @@
+// MIR for `demo_le_total` after PreCodegen
+
+fn demo_le_total(_1: &(u16, i16), _2: &(u16, i16)) -> bool {
+    debug a => _1;
+    debug b => _2;
+    let mut _0: bool;
+    scope 1 (inlined std::cmp::impls::<impl PartialOrd for &(u16, i16)>::le) {
+        scope 2 (inlined core::tuple::<impl PartialOrd for (u16, i16)>::le) {
+            let mut _7: std::ops::ControlFlow<bool>;
+            let _8: bool;
+            scope 3 {
+            }
+            scope 4 (inlined std::cmp::impls::<impl std::cmp::SpecChainingPartialOrd<u16> for u16>::spec_chain_le) {
+                let mut _3: u16;
+                let mut _4: u16;
+                let mut _5: bool;
+                let mut _6: bool;
+                scope 5 {
+                }
+            }
+            scope 6 (inlined std::cmp::impls::<impl PartialOrd for i16>::le) {
+                let mut _9: i16;
+                let mut _10: i16;
+            }
+        }
+    }
+
+    bb0: {
+        StorageLive(_7);
+        StorageLive(_3);
+        StorageLive(_4);
+        _3 = copy ((*_1).0: u16);
+        _4 = copy ((*_2).0: u16);
+        StorageLive(_5);
+        _5 = Eq(copy _3, copy _4);
+        switchInt(move _5) -> [0: bb1, otherwise: bb2];
+    }
+
+    bb1: {
+        StorageLive(_6);
+        _6 = Le(copy _3, copy _4);
+        _7 = ControlFlow::<bool>::Break(move _6);
+        StorageDead(_6);
+        StorageDead(_5);
+        StorageDead(_4);
+        StorageDead(_3);
+        _8 = copy ((_7 as Break).0: bool);
+        _0 = copy _8;
+        goto -> bb3;
+    }
+
+    bb2: {
+        StorageDead(_5);
+        StorageDead(_4);
+        StorageDead(_3);
+        StorageLive(_9);
+        _9 = copy ((*_1).1: i16);
+        StorageLive(_10);
+        _10 = copy ((*_2).1: i16);
+        _0 = Le(move _9, move _10);
+        StorageDead(_10);
+        StorageDead(_9);
+        goto -> bb3;
+    }
+
+    bb3: {
+        StorageDead(_7);
+        return;
+    }
+}

tests/mir-opt/pre-codegen/tuple_ord.rs

@@ -0,0 +1,16 @@
+//@ compile-flags: -O -Zmir-opt-level=2 -Cdebuginfo=0
+//@ needs-unwind
+
+#![crate_type = "lib"]
+
+// EMIT_MIR tuple_ord.demo_le_total.PreCodegen.after.mir
+pub fn demo_le_total(a: &(u16, i16), b: &(u16, i16)) -> bool {
+    // CHECK-LABEL: demo_le_total
+    a <= b
+}
+
+// EMIT_MIR tuple_ord.demo_ge_partial.PreCodegen.after.mir
+pub fn demo_ge_partial(a: &(f32, f32), b: &(f32, f32)) -> bool {
+    // CHECK-LABEL: demo_ge_partial
+    a <= b
+}