From 8fe321d5be965419fc2bfe652f35f5363a27a992 Mon Sep 17 00:00:00 2001
From: Juliusz Sosinowicz <juliusz@wolfssl.com>
Date: Thu, 23 Mar 2023 10:46:09 +0100
Subject: [PATCH] Try to XOR as many words as possible in xorbuf APIs

---
 wolfcrypt/src/misc.c     | 62 ++++++++++++++++++++++++----------------
 wolfssl/wolfcrypt/misc.h |  6 ++--
 2 files changed, 41 insertions(+), 27 deletions(-)

diff --git a/wolfcrypt/src/misc.c b/wolfcrypt/src/misc.c
index 3c47e517c..d0045a0e5 100644
--- a/wolfcrypt/src/misc.c
+++ b/wolfcrypt/src/misc.c
@@ -256,42 +256,50 @@ WC_MISC_STATIC WC_INLINE void ByteReverseWords64(word64* out, const word64* in,
 #ifndef WOLFSSL_NO_XOR_OPS
 /* This routine performs a bitwise XOR operation of <*r> and <*a> for <n> number
 of wolfssl_words, placing the result in <*r>. */
-WC_MISC_STATIC WC_INLINE void XorWordsOut(wolfssl_word* r,
-                         const wolfssl_word* a, const wolfssl_word* b, word32 n)
+WC_MISC_STATIC WC_INLINE void XorWordsOut(wolfssl_word** r,
+                       const wolfssl_word** a, const wolfssl_word** b, word32 n)
 {
     word32 i;
 
-    for (i = 0; i < n; i++) r[i] = a[i] ^ b[i];
+    for (i = 0; i < n; i++)
+        *(*r)++ = *(*a)++ ^ *(*b)++;
 }
 
 /* This routine performs a bitwise XOR operation of <*buf> and <*mask> of n
 counts, placing the result in <*buf>. */
 
-WC_MISC_STATIC WC_INLINE void xorbufout(void*out, const void* buf,
+WC_MISC_STATIC WC_INLINE void xorbufout(void* out, const void* buf,
                                         const void* mask, word32 count)
 {
-    if (((wc_ptr_t)out | (wc_ptr_t)buf | (wc_ptr_t)mask | count) %
-                                                         WOLFSSL_WORD_SIZE == 0)
-        XorWordsOut( (wolfssl_word*)out, (wolfssl_word*)buf,
-                     (const wolfssl_word*)mask, count / WOLFSSL_WORD_SIZE);
-    else {
-        word32 i;
-        byte*       o = (byte*)out;
-        byte*       b = (byte*)buf;
-        const byte* m = (const byte*)mask;
+    word32      i;
+    byte*       o;
+    byte*       b;
+    const byte* m;
 
-        for (i = 0; i < count; i++) o[i] = b[i] ^ m[i];
+    if (((wc_ptr_t)buf | (wc_ptr_t)mask) % WOLFSSL_WORD_SIZE == 0) {
+        /* Alignment checks out. Possible to XOR words. */
+        XorWordsOut( (wolfssl_word**)&out, (const wolfssl_word**)&buf,
+                     (const wolfssl_word**)&mask, count / WOLFSSL_WORD_SIZE);
+        count %= WOLFSSL_WORD_SIZE;
     }
+
+    o = (byte*)out;
+    b = (byte*)buf;
+    m = (const byte*)mask;
+
+    for (i = 0; i < count; i++)
+        o[i] = b[i] ^ m[i];
 }
 
 /* This routine performs a bitwise XOR operation of <*r> and <*a> for <n> number
 of wolfssl_words, placing the result in <*r>. */
-WC_MISC_STATIC WC_INLINE void XorWords(wolfssl_word* r, const wolfssl_word* a,
+WC_MISC_STATIC WC_INLINE void XorWords(wolfssl_word** r, const wolfssl_word** a,
                                        word32 n)
 {
     word32 i;
 
-    for (i = 0; i < n; i++) r[i] ^= a[i];
+    for (i = 0; i < n; i++)
+        *(*r)++ ^= *(*a)++;
 }
 
 /* This routine performs a bitwise XOR operation of <*buf> and <*mask> of n
@@ -299,16 +307,22 @@ counts, placing the result in <*buf>. */
 
 WC_MISC_STATIC WC_INLINE void xorbuf(void* buf, const void* mask, word32 count)
 {
-    if (((wc_ptr_t)buf | (wc_ptr_t)mask | count) % WOLFSSL_WORD_SIZE == 0)
-        XorWords( (wolfssl_word*)buf,
-                  (const wolfssl_word*)mask, count / WOLFSSL_WORD_SIZE);
-    else {
-        word32 i;
-        byte*       b = (byte*)buf;
-        const byte* m = (const byte*)mask;
+    word32      i;
+    byte*       b;
+    const byte* m;
 
-        for (i = 0; i < count; i++) b[i] ^= m[i];
+    if (((wc_ptr_t)buf | (wc_ptr_t)mask) % WOLFSSL_WORD_SIZE == 0) {
+        /* Alignment checks out. Possible to XOR words. */
+        XorWords( (wolfssl_word**)&buf,
+                  (const wolfssl_word**)&mask, count / WOLFSSL_WORD_SIZE);
+        count %= WOLFSSL_WORD_SIZE;
     }
+
+    b = (byte*)buf;
+    m = (const byte*)mask;
+
+    for (i = 0; i < count; i++)
+        b[i] ^= m[i];
 }
 #endif
 
diff --git a/wolfssl/wolfcrypt/misc.h b/wolfssl/wolfcrypt/misc.h
index 418246787..e433a1a3c 100644
--- a/wolfssl/wolfcrypt/misc.h
+++ b/wolfssl/wolfcrypt/misc.h
@@ -59,12 +59,12 @@ WOLFSSL_LOCAL
 void   ByteReverseWords(word32* out, const word32* in, word32 byteCount);
 
 WOLFSSL_LOCAL
-void XorWordsOut(wolfssl_word* r, const wolfssl_word* a, const wolfssl_word* b,
-                 word32 n);
+void XorWordsOut(wolfssl_word** r, const wolfssl_word** a,
+        const wolfssl_word** b, word32 n);
 WOLFSSL_LOCAL
 void xorbufout(void* out, const void* buf, const void* mask, word32 count);
 WOLFSSL_LOCAL
-void XorWords(wolfssl_word* r, const wolfssl_word* a, word32 n);
+void XorWords(wolfssl_word** r, const wolfssl_word** a, word32 n);
 WOLFSSL_LOCAL
 void xorbuf(void* buf, const void* mask, word32 count);