*: deinline SWAP_xE64 on 32-bit CPUs. Wins !90 bytes both on 32 and 64 bits

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>

3 files changed, 22 insertions, 15 deletions

diff --git a/libbb/Kbuild.src b/libbb/Kbuild.src
index 0fd7302..0955d73 100644
--- a/libbb/Kbuild.src
+++ b/libbb/Kbuild.src
@@ -14,6 +14,7 @@ lib-y += appletlib.o
 lib-y += ask_confirmation.o
 lib-y += bb_askpass.o
 lib-y += bb_basename.o
+lib-y += bb_bswap_64.o
 lib-y += bb_do_delay.o
 lib-y += bb_pwd.o
 lib-y += bb_qsort.o
diff --git a/libbb/bb_bswap_64.c b/libbb/bb_bswap_64.c
new file mode 100644
index 0000000..ce9d53b
--- /dev/null
+++ b/libbb/bb_bswap_64.c
@@ -0,0 +1,16 @@
+/*
+ * Utility routines.
+ *
+ * Copyright (C) 2010 Denys Vlasenko
+ *
+ * Licensed under GPLv2, see file LICENSE in this source tree.
+ */
+
+#include "libbb.h"
+
+#if !(ULONG_MAX > 0xffffffff)
+uint64_t FAST_FUNC bb_bswap_64(uint64_t x)
+{
+	return bswap_64(x);
+}
+#endif
diff --git a/libbb/hash_sha.c b/libbb/hash_sha.c
index e7199d3..72d5092 100644
--- a/libbb/hash_sha.c
+++ b/libbb/hash_sha.c
@@ -51,16 +51,6 @@ static ALWAYS_INLINE uint64_t rotr64(uint64_t x, unsigned n)
 {
 	return (x >> n) | (x << (64 - n));
 }
-#if BB_LITTLE_ENDIAN
-/* ALWAYS_INLINE below sometimes hurts code size, using plain inline: */
-static inline uint64_t hton64(uint64_t v)
-{
-	return SWAP_BE64(v);
-}
-#else
-#define hton64(v) (v)
-#endif
-#define ntoh64(v) hton64(v)
 
 
 /* Some arch headers have conflicting defines */
@@ -274,7 +264,7 @@ static void FAST_FUNC sha512_process_block128(sha512_ctx_t *ctx)
 
 	/* Compute the message schedule according to FIPS 180-2:6.3.2 step 2.  */
 	for (t = 0; t < 16; ++t)
-		W[t] = ntoh64(words[t]);
+		W[t] = SWAP_BE64(words[t]);
 	for (/*t = 16*/; t < 80; ++t)
 		W[t] = R1(W[t - 2]) + W[t - 7] + R0(W[t - 15]) + W[t - 16];
 
@@ -475,7 +465,7 @@ void FAST_FUNC sha1_end(sha1_ctx_t *ctx, void *resbuf)
 		if (remaining >= 8) {
 			/* Store the 64-bit counter of bits in the buffer in BE format */
 			uint64_t t = ctx->total64 << 3;
-			t = hton64(t);
+			t = SWAP_BE64(t);
 			/* wbuffer is suitably aligned for this */
 			*(uint64_t *) (&ctx->wbuffer[64 - 8]) = t;
 		}
@@ -509,10 +499,10 @@ void FAST_FUNC sha512_end(sha512_ctx_t *ctx, void *resbuf)
 			/* Store the 128-bit counter of bits in the buffer in BE format */
 			uint64_t t;
 			t = ctx->total64[0] << 3;
-			t = hton64(t);
+			t = SWAP_BE64(t);
 			*(uint64_t *) (&ctx->wbuffer[128 - 8]) = t;
 			t = (ctx->total64[1] << 3) | (ctx->total64[0] >> 61);
-			t = hton64(t);
+			t = SWAP_BE64(t);
 			*(uint64_t *) (&ctx->wbuffer[128 - 16]) = t;
 		}
 		sha512_process_block128(ctx);
@@ -524,7 +514,7 @@ void FAST_FUNC sha512_end(sha512_ctx_t *ctx, void *resbuf)
 	if (BB_LITTLE_ENDIAN) {
 		unsigned i;
 		for (i = 0; i < ARRAY_SIZE(ctx->hash); ++i)
-			ctx->hash[i] = hton64(ctx->hash[i]);
+			ctx->hash[i] = SWAP_BE64(ctx->hash[i]);
 	}
 	memcpy(resbuf, ctx->hash, sizeof(ctx->hash));
 }