mem: Add REP MOVSB/STOSB implementations

The assembly generated seems correct:
    https://rust.godbolt.org/z/9ERcGG

Signed-off-by: Joe Richey <[email protected]>

Author: josephlr

src/lib.rs

@@ -1,5 +1,6 @@
 #![cfg_attr(feature = "compiler-builtins", compiler_builtins)]
 #![feature(abi_unadjusted)]
+#![feature(asm)]
 #![feature(llvm_asm)]
 #![feature(global_asm)]
 #![feature(cfg_target_has_atomic)]

src/mem/mod.rs

@@ -6,7 +6,8 @@ type c_int = i16;
 type c_int = i32;
 
 // memcpy and friends have optimized implementations on some architectures
-#[path = "fallback.rs"]
+#[cfg_attr(target_arch = "x86_64", path = "x86_64.rs")]
+#[cfg_attr(not(target_arch = "x86_64"), path = "fallback.rs")]
 mod memcpy;
 pub use self::memcpy::*;
 

src/mem/x86_64.rs

@@ -0,0 +1,81 @@
+use super::c_int;
+
+#[inline(always)]
+unsafe fn copy_forward(dest: *mut u8, src: *const u8, count: usize) {
+    asm!(
+        "rep movsb [rdi], [rsi]",
+        inout("rdi") dest => _,
+        inout("rsi") src => _,
+        inout("rcx") count => _,
+        options(nostack, preserves_flags)
+    );
+}
+
+#[cfg_attr(all(feature = "mem", not(feature = "mangled-names")), no_mangle)]
+pub unsafe extern "C" fn memcpy(dest: *mut u8, src: *const u8, count: usize) -> *mut u8 {
+    copy_forward(dest, src, count);
+    dest
+}
+
+#[cfg_attr(all(feature = "mem", not(feature = "mangled-names")), no_mangle)]
+pub unsafe extern "C" fn memmove(dest: *mut u8, src: *const u8, count: usize) -> *mut u8 {
+    if src < dest as *const u8 {
+        // copy backwards
+        asm!(
+            "std
+            rep movsb [rdi], [rsi]
+            cld",
+            inout("rdi") dest.add(count).sub(1) => _,
+            inout("rsi") src.add(count).sub(1) => _,
+            inout("rcx") count => _,
+            options(nostack, preserves_flags)
+        );
+    } else {
+        copy_forward(dest, src, count);
+    }
+    dest
+}
+
+#[cfg_attr(all(feature = "mem", not(feature = "mangled-names")), no_mangle)]
+pub unsafe extern "C" fn memset(dest: *mut u8, c: c_int, count: usize) -> *mut u8 {
+    asm!(
+        "rep stosb [rdi], al",
+        inout("rdi") dest => _,
+        in("al") c as u8,
+        inout("rcx") count => _,
+        options(nostack, preserves_flags)
+    );
+    dest
+}
+
+#[cfg_attr(all(feature = "mem", not(feature = "mangled-names")), no_mangle)]
+pub unsafe extern "C" fn memcmp(s1: *const u8, s2: *const u8, count: usize) -> i32 {
+    let res: i32;
+    asm!(
+        "repe cmpsb [rsi], [rdi]
+        setb {0:l}
+        cmova {0:e}, {1:e}",
+        inout(reg_abcd) 0 => res,
+        in(reg) -1i32,
+        inout("rsi") s2 => _,
+        inout("rdi") s1 => _,
+        inout("rcx") count => _,
+        options(nostack, pure, readonly)
+    );
+    res
+}
+
+#[cfg_attr(all(feature = "mem", not(feature = "mangled-names")), no_mangle)]
+pub unsafe extern "C" fn bcmp(s1: *const u8, s2: *const u8, count: usize) -> i32 {
+    let res: i32;
+    asm!(
+        "repe cmpsb [rsi], [rdi]
+        setne {0:l}",
+        inout(reg_abcd) 0 => res,
+        inout("rsi") s2 => _,
+        inout("rdi") s1 => _,
+        inout("rcx") count => _,
+        options(nostack, pure, readonly)
+    );
+    res
+}