oops

core/lib/string.c

@@ -1,8 +1,8 @@
 #include <zinnia/compiler.h>
+#include <zinnia/string.h>
 #include <zinnia/utils.h>
 #include <stddef.h>
 #include <stdint.h>
-#include <zinnia/string.h>
 
 [[__hot]]
 size_t strlen(const char* str) {
@@ -29,45 +29,37 @@ int memcmp(const void* s1, const void* s2, size_t size) {
 
     for (size_t i = 0; i < size; i++) {
         if (s1ptr[i] != s2ptr[i])
-            return s1ptr[i] - s2ptr[i];
+            return (int)s1ptr[i] - (int)s2ptr[i];
     }
     return 0;
 }
 
+typedef size_t word_t [[__may_alias]]
+;
+
 [[__hot]]
 void* memcpy(void* restrict dest, const void* restrict src, size_t n) {
-    if (__unlikely(n == 0))
-        return dest;
+    uint8_t* d = dest;
+    const uint8_t* s = src;
 
-    size_t d = (size_t)dest;
-    size_t s = (size_t)src;
-
-    if (d % sizeof(size_t) != 0 || s % sizeof(size_t) != 0) {
-        while (n && (d % sizeof(size_t) != 0) && (s % sizeof(size_t) != 0)) {
-            *((uint8_t*)d) = *((uint8_t*)s);
-            d++;
-            s++;
+    // Word copies are only possible if both pointers can be aligned at the same time.
+    if (((uintptr_t)d - (uintptr_t)s) % sizeof(word_t) == 0) {
+        while (n && (uintptr_t)d % sizeof(word_t) != 0) {
+            *d++ = *s++;
             n--;
         }
+
+        word_t* word_dest = (word_t*)d;
+        const word_t* word_src = (const word_t*)s;
+        for (; n >= sizeof(word_t); n -= sizeof(word_t))
+            *word_dest++ = *word_src++;
+
+        d = (uint8_t*)word_dest;
+        s = (const uint8_t*)word_src;
     }
 
-    size_t* qword_dest = (size_t*)d;
-    const size_t* qword_src = (const size_t*)s;
-    const size_t word_count = n / sizeof(size_t);
-
-    for (size_t i = 0; i < word_count; i++) {
-        qword_dest[i] = qword_src[i];
-    }
-
-    size_t remaining_bytes = n % sizeof(size_t);
-    d = (size_t)(qword_dest + word_count);
-    s = (size_t)(qword_src + word_count);
-
-    while (remaining_bytes--) {
-        *((uint8_t*)d) = *((uint8_t*)s);
-        d++;
-        s++;
-    }
+    while (n--)
+        *d++ = *s++;
 
     return dest;
 }
@@ -76,7 +68,8 @@ void* memcpy(void* restrict dest, const void* restrict src, size_t n) {
 void* memmove(void* dstptr, const void* srcptr, size_t size) {
     uint8_t* dst = (uint8_t*)dstptr;
     const uint8_t* src = (const uint8_t*)srcptr;
-    if (dst < src) {
+
+    if ((uintptr_t)dst < (uintptr_t)src) {
         for (size_t i = 0; i < size; i++)
             dst[i] = src[i];
     } else {
@@ -88,38 +81,22 @@ void* memmove(void* dstptr, const void* srcptr, size_t size) {
 
 [[__hot]]
 void* memset(void* dest, int value, size_t n) {
-    if (__unlikely(n == 0))
-        return dest;
+    uint8_t* d = dest;
+    const uint8_t byte = (uint8_t)value;
 
-    size_t d = (size_t)dest;
-
-    if (d % sizeof(size_t) != 0) {
-        while (n && (d % sizeof(size_t) != 0)) {
-            *((uint8_t*)d) = value;
-            d++;
-            n--;
-        }
+    while (n && (uintptr_t)d % sizeof(word_t) != 0) {
+        *d++ = byte;
+        n--;
     }
 
-    size_t* qword_dest = (size_t*)d;
-    size_t qword_value = (uint8_t)value;
-    qword_value |= (qword_value << 8);
-    qword_value |= (qword_value << 16);
-    qword_value |= (qword_value << 32);
+    const word_t word_value = (SIZE_MAX / 0xFF) * byte;
+    word_t* word_dest = (word_t*)d;
+    for (; n >= sizeof(word_t); n -= sizeof(word_t))
+        *word_dest++ = word_value;
 
-    const size_t word_count = n / sizeof(size_t);
-
-    for (size_t i = 0; i < word_count; i++) {
-        qword_dest[i] = qword_value;
-    }
-
-    size_t remaining_bytes = n % sizeof(size_t);
-    d = (size_t)(qword_dest + word_count);
-
-    while (remaining_bytes--) {
-        *((uint8_t*)d) = value;
-        d++;
-    }
+    d = (uint8_t*)word_dest;
+    while (n--)
+        *d++ = byte;
 
     return dest;
 }