Rollup merge of rust-lang#147457 - the8472:slice_fill_memset2, r=RalfJung,joboet

specialize slice::fill to use memset when possible

It helps const eval performance rust-lang/miri#4616, debug builds and the gcc backend.

Previously attempted in rust-lang#83245 but reverted due to unsoundness rust-lang#87891 around potentially-uninitialized types. This PR only handles primitives where the problem does not arise.

split off from rust-lang#147294

Author: Zalathar

library/core/src/slice/specialize.rs

@@ -15,9 +15,54 @@ impl<T: Clone> SpecFill<T> for [T] {
 }
 
 impl<T: Copy> SpecFill<T> for [T] {
-    fn spec_fill(&mut self, value: T) {
+    default fn spec_fill(&mut self, value: T) {
         for item in self.iter_mut() {
             *item = value;
         }
     }
 }
+
+impl SpecFill<u8> for [u8] {
+    fn spec_fill(&mut self, value: u8) {
+        // SAFETY: The pointer is derived from a reference, so it's writable.
+        unsafe {
+            crate::intrinsics::write_bytes(self.as_mut_ptr(), value, self.len());
+        }
+    }
+}
+
+impl SpecFill<i8> for [i8] {
+    fn spec_fill(&mut self, value: i8) {
+        // SAFETY: The pointer is derived from a reference, so it's writable.
+        unsafe {
+            crate::intrinsics::write_bytes(self.as_mut_ptr(), value.cast_unsigned(), self.len());
+        }
+    }
+}
+
+macro spec_fill_int {
+    ($($type:ty)*) => {$(
+        impl SpecFill<$type> for [$type] {
+            #[inline]
+            fn spec_fill(&mut self, value: $type) {
+                // We always take this fastpath in Miri for long slices as the manual `for`
+                // loop can be prohibitively slow.
+                if (cfg!(miri) && self.len() > 32) || crate::intrinsics::is_val_statically_known(value) {
+                    let bytes = value.to_ne_bytes();
+                    if value == <$type>::from_ne_bytes([bytes[0]; size_of::<$type>()]) {
+                        // SAFETY: The pointer is derived from a reference, so it's writable.
+                        unsafe {
+                            crate::intrinsics::write_bytes(self.as_mut_ptr(), bytes[0], self.len());
+                        }
+                        return;
+                    }
+                }
+                for item in self.iter_mut() {
+                    *item = value;
+                }
+            }
+        }
+    )*}
+}
+
+spec_fill_int! { u16 i16 u32 i32 u64 i64 u128 i128 usize isize }

tests/codegen-llvm/lib-optimizations/slice_fill.rs

@@ -0,0 +1,28 @@
+//@ compile-flags: -Copt-level=3
+#![crate_type = "lib"]
+
+use std::mem::MaybeUninit;
+
+// CHECK-LABEL: @slice_fill_pass_undef
+#[no_mangle]
+pub fn slice_fill_pass_undef(s: &mut [MaybeUninit<u8>], v: MaybeUninit<u8>) {
+    // CHECK: tail call void @llvm.memset.{{.*}}(ptr nonnull align 1 %s.0, i8 %v, {{.*}} %s.1, i1 false)
+    // CHECK: ret
+    s.fill(v);
+}
+
+// CHECK-LABEL: @slice_fill_uninit
+#[no_mangle]
+pub fn slice_fill_uninit(s: &mut [MaybeUninit<u8>]) {
+    // CHECK-NOT: call
+    // CHECK: ret void
+    s.fill(MaybeUninit::uninit());
+}
+
+// CHECK-LABEL: @slice_wide_memset
+#[no_mangle]
+pub fn slice_wide_memset(s: &mut [u16]) {
+    // CHECK: tail call void @llvm.memset.{{.*}}(ptr nonnull align 2 %s.0, i8 -1
+    // CHECK: ret
+    s.fill(0xFFFF);
+}