From af3bfcccc484ace662c78f053f633bc883c73152 Mon Sep 17 00:00:00 2001
From: Tavian Barnes <tavianator@tavianator.com>
Date: Fri, 21 Feb 2025 13:38:16 -0500
Subject: Micro-optimize word-at-a-time loop tails

Compilers apparently aren't smart enough to infer that the odd tail
length fixup "loops" run at most once, and could be converted to ifs, so
do that manually.
---
 src/bfstd.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

(limited to 'src/bfstd.c')

diff --git a/src/bfstd.c b/src/bfstd.c
index f2938ad..219b8d0 100644
--- a/src/bfstd.c
+++ b/src/bfstd.c
@@ -786,7 +786,7 @@ size_t asciinlen(const char *str, size_t n) {
 	// Word-at-a-time isascii()
 #define CHUNK(n) CHUNK_(uint##n##_t, load8_leu##n)
 #define CHUNK_(type, load8) \
-	while (n - i >= sizeof(type)) { \
+	(n - i >= sizeof(type)) { \
 		type word = load8(ustr + i); \
 		type mask = (((type)-1) / 0xFF) << 7; /* 0x808080.. */ \
 		word &= mask; \
@@ -797,11 +797,13 @@ size_t asciinlen(const char *str, size_t n) {
 	}
 
 #if SIZE_WIDTH >= 64
-	CHUNK(64);
+	while CHUNK(64);
+	if CHUNK(32);
+#else
+	while CHUNK(32);
 #endif
-	CHUNK(32);
-	CHUNK(16);
-	CHUNK(8);
+	if CHUNK(16);
+	if CHUNK(8);
 
 #undef CHUNK_
 #undef CHUNK
-- 
cgit v1.2.3