[SOL] Optimize memory operations (anza-xyz#31)

Author: LucasSte

compiler-builtins/src/mem/mod.rs

@@ -67,7 +67,6 @@ intrinsics! {
 // number of store operations is greater than 8 the memory operation
 // is performed in the run-time system instead, by calling the
 // corresponding "C" function.
-
 #[cfg(all(target_os = "solana", not(target_feature = "static-syscalls")))]
 mod syscalls {
     extern "C" {
@@ -108,36 +107,14 @@ mod syscalls {
 #[cfg(target_os = "solana")]
 use self::syscalls::*;
 
-#[cfg(target_os = "solana")]
-const NSTORE_THRESHOLD: usize = 15;
-
 #[cfg(target_os = "solana")]
 #[cfg_attr(
     all(feature = "mem-unaligned", not(feature = "mangled-names")),
     no_mangle
 )]
 #[inline]
 pub unsafe extern "C" fn memcpy(dest: *mut u8, src: *const u8, n: usize) -> *mut u8 {
-    let chunks = (n / 8) as isize;
-    let nstore = n - (7 * chunks) as usize;
-    if nstore > NSTORE_THRESHOLD {
-        sol_memcpy_(dest, src, n as u64);
-        return dest;
-    }
-    let mut i: isize = 0;
-    if chunks != 0 {
-        let dest_64 = dest as *mut _ as *mut u64;
-        let src_64 = src as *const _ as *const u64;
-        while i < chunks {
-            *dest_64.offset(i) = *src_64.offset(i);
-            i += 1;
-        }
-        i *= 8;
-    }
-    while i < n as isize {
-        *dest.offset(i) = *src.offset(i);
-        i += 1;
-    }
+    sol_memcpy_(dest, src, n as u64);
     dest
 }
 
@@ -148,45 +125,7 @@ pub unsafe extern "C" fn memcpy(dest: *mut u8, src: *const u8, n: usize) -> *mut
 )]
 #[inline]
 pub unsafe extern "C" fn memmove(dest: *mut u8, src: *const u8, n: usize) -> *mut u8 {
-    let chunks = (n / 8) as isize;
-    let nstore = n - (7 * chunks) as usize;
-    if nstore > NSTORE_THRESHOLD {
-        sol_memmove_(dest, src, n as u64);
-        return dest;
-    }
-    if src < dest as *const u8 {
-        // copy from end
-        let mut i = n as isize;
-        while i > chunks * 8 {
-            i -= 1;
-            *dest.offset(i) = *src.offset(i);
-        }
-        i = chunks;
-        if i > 0 {
-            let dest_64 = dest as *mut _ as *mut u64;
-            let src_64 = src as *const _ as *const u64;
-            while i > 0 {
-                i -= 1;
-                *dest_64.offset(i) = *src_64.offset(i);
-            }
-        }
-    } else {
-        // copy from beginning
-        let mut i: isize = 0;
-        if chunks != 0 {
-            let dest_64 = dest as *mut _ as *mut u64;
-            let src_64 = src as *const _ as *const u64;
-            while i < chunks {
-                *dest_64.offset(i) = *src_64.offset(i);
-                i += 1;
-            }
-            i *= 8;
-        }
-        while i < n as isize {
-            *dest.offset(i) = *src.offset(i);
-            i += 1;
-        }
-    }
+    sol_memmove_(dest, src, n as u64);
     dest
 }
 
@@ -197,29 +136,7 @@ pub unsafe extern "C" fn memmove(dest: *mut u8, src: *const u8, n: usize) -> *mu
 )]
 #[inline]
 pub unsafe extern "C" fn memset(s: *mut u8, c: c_int, n: usize) -> *mut u8 {
-    let chunks = (n / 8) as isize;
-    let nstore = n - (7 * chunks) as usize;
-    if nstore > NSTORE_THRESHOLD {
-        sol_memset_(s, c as u8, n as u64);
-        return s;
-    }
-    let mut i: isize = 0;
-    if chunks != 0 {
-        let mut c_64 = c as u64 & 0xFF as u64;
-        c_64 |= c_64 << 8;
-        c_64 |= c_64 << 16;
-        c_64 |= c_64 << 32;
-        let s_64 = s as *mut _ as *mut u64;
-        while i < chunks {
-            *s_64.offset(i) = c_64;
-            i += 1;
-        }
-        i *= 8;
-    }
-    while i < n as isize {
-        *s.offset(i) = c as u8;
-        i += 1;
-    }
+    sol_memset_(s, c as u8, n as u64);
     s
 }
 
@@ -230,34 +147,7 @@ pub unsafe extern "C" fn memset(s: *mut u8, c: c_int, n: usize) -> *mut u8 {
 )]
 #[inline]
 pub unsafe extern "C" fn memcmp(s1: *const u8, s2: *const u8, n: usize) -> i32 {
-    let chunks = (n / 8) as isize;
-    let nstore = n - (7 * chunks) as usize;
-    if nstore > NSTORE_THRESHOLD {
-        let mut result = 0;
-        sol_memcmp_(s1, s2, n as u64, &mut result as *mut i32);
-        return result;
-    }
-    let mut i: isize = 0;
-    if chunks != 0 {
-        let s1_64 = s1 as *const _ as *const u64;
-        let s2_64 = s2 as *const _ as *const u64;
-        while i < chunks {
-            let a = *s1_64.offset(i);
-            let b = *s2_64.offset(i);
-            if a != b {
-                break;
-            }
-            i += 1;
-        }
-        i *= 8;
-    }
-    while i < n as isize {
-        let a = *s1.offset(i);
-        let b = *s2.offset(i);
-        if a != b {
-            return a as i32 - b as i32;
-        }
-        i += 1;
-    }
-    0
-}
+    let mut result = 0;
+    sol_memcmp_(s1, s2, n as u64, &mut result as *mut i32);
+    result
+}