From 99977dbf006d851df53ae081551e1cb0bf1fd909 Mon Sep 17 00:00:00 2001
From: Alfred Klomp <git@alfredklomp.com>
Date: Fri, 22 Nov 2019 00:09:45 +0100
Subject: [PATCH] Generic64: enc: factor encoding loop into inline function

---
 lib/arch/generic/64/enc_loop.c | 69 ++++++++++++++++++++--------------
 lib/arch/generic/codec.c       |  6 ++-
 lib/arch/neon64/codec.c        |  3 +-
 3 files changed, 48 insertions(+), 30 deletions(-)

diff --git a/lib/arch/generic/64/enc_loop.c b/lib/arch/generic/64/enc_loop.c
index 74a7add1..05236f64 100644
--- a/lib/arch/generic/64/enc_loop.c
+++ b/lib/arch/generic/64/enc_loop.c
@@ -1,30 +1,43 @@
-// If we have 64-bit ints, pick off 6 bytes at a time for as long as we can,
-// but ensure that there are at least 8 bytes available to avoid segfaulting:
-while (srclen >= 8)
+static inline void
+enc_loop_generic_64 (const uint8_t **s, size_t *slen, uint8_t **o, size_t *olen)
 {
-	uint64_t str;
-
-	// Load string:
-	memcpy(&str, c, sizeof (str));
-
-	// Reorder to 64-bit big-endian, if not already in that format. The
-	// workset must be in big-endian, otherwise the shifted bits do not
-	// carry over properly among adjacent bytes:
-	str = BASE64_HTOBE64(str);
-
-	// Shift input by 6 bytes each round and mask in only the lower 6 bits;
-	// look up the character in the Base64 encoding table and write it to
-	// the output location:
-	*o++ = base64_table_enc[(str >> 58) & 0x3F];
-	*o++ = base64_table_enc[(str >> 52) & 0x3F];
-	*o++ = base64_table_enc[(str >> 46) & 0x3F];
-	*o++ = base64_table_enc[(str >> 40) & 0x3F];
-	*o++ = base64_table_enc[(str >> 34) & 0x3F];
-	*o++ = base64_table_enc[(str >> 28) & 0x3F];
-	*o++ = base64_table_enc[(str >> 22) & 0x3F];
-	*o++ = base64_table_enc[(str >> 16) & 0x3F];
-
-	c += 6;		// 6 bytes of input
-	outl += 8;	// 8 bytes of output
-	srclen -= 6;
+	if (*slen < 8) {
+		return;
+	}
+
+	// Process blocks of 6 bytes at a time. Because blocks are loaded 8
+	// bytes at a time, ensure that there will be at least 2 remaining
+	// bytes after the last round, so that the final read will not pass
+	// beyond the bounds of the input buffer:
+	size_t rounds = (*slen - 2) / 6;
+
+	*slen -= rounds * 6;	// 6 bytes consumed per round
+	*olen += rounds * 8;	// 8 bytes produced per round
+
+	do {
+		uint64_t src;
+
+		// Load input:
+		memcpy(&src, *s, sizeof (src));
+
+		// Reorder to 64-bit big-endian, if not already in that format.
+		// The workset must be in big-endian, otherwise the shifted
+		// bits do not carry over properly among adjacent bytes:
+		src = BASE64_HTOBE64(src);
+
+		// Shift input by 6 bytes each round and mask in only the lower
+		// 6 bits; look up the character in the Base64 encoding table
+		// and write it to the output location:
+		*(*o)++ = base64_table_enc[(src >> 58) & 0x3F];
+		*(*o)++ = base64_table_enc[(src >> 52) & 0x3F];
+		*(*o)++ = base64_table_enc[(src >> 46) & 0x3F];
+		*(*o)++ = base64_table_enc[(src >> 40) & 0x3F];
+		*(*o)++ = base64_table_enc[(src >> 34) & 0x3F];
+		*(*o)++ = base64_table_enc[(src >> 28) & 0x3F];
+		*(*o)++ = base64_table_enc[(src >> 22) & 0x3F];
+		*(*o)++ = base64_table_enc[(src >> 16) & 0x3F];
+
+		*s += 6;
+
+	} while (--rounds > 0);
 }
diff --git a/lib/arch/generic/codec.c b/lib/arch/generic/codec.c
index 69cec3b3..89eda760 100644
--- a/lib/arch/generic/codec.c
+++ b/lib/arch/generic/codec.c
@@ -5,13 +5,17 @@
 #include "../../../include/libbase64.h"
 #include "../../codecs.h"
 
+#if BASE64_WORDSIZE == 64
+#  include "64/enc_loop.c"
+#endif
+
 BASE64_ENC_FUNCTION(plain)
 {
 	#include "enc_head.c"
 #if BASE64_WORDSIZE == 32
 	#include "32/enc_loop.c"
 #elif BASE64_WORDSIZE == 64
-	#include "64/enc_loop.c"
+	enc_loop_generic_64(&c, &srclen, &o, &outl);
 #endif
 	#include "enc_tail.c"
 }
diff --git a/lib/arch/neon64/codec.c b/lib/arch/neon64/codec.c
index 7c67c35e..49e65d70 100644
--- a/lib/arch/neon64/codec.c
+++ b/lib/arch/neon64/codec.c
@@ -77,6 +77,7 @@ static const uint8_t base64_dec_lut2[] =
 // Therefore, valid characters will be mapped to the valid [0..63] range and all invalid characters will be mapped
 // to values greater than 63.
 
+#include "../generic/64/enc_loop.c"
 #include "enc_loop.c"
 
 #endif	// BASE64_USE_NEON64
@@ -90,7 +91,7 @@ BASE64_ENC_FUNCTION(neon64)
 #ifdef BASE64_USE_NEON64
 	#include "../generic/enc_head.c"
 	enc_loop_neon64(&c, &srclen, &o, &outl);
-	#include "../generic/64/enc_loop.c"
+	enc_loop_generic_64(&c, &srclen, &o, &outl);
 	#include "../generic/enc_tail.c"
 #else
 	BASE64_ENC_STUB