f09d088fd

Denys Vlasenko

2021-12-31

SUCCESS Duration: 154s

Subject

libbb/sha1: shrink and speed up fully unrolled version

No binary (build failed)

Architecture: x86_64-linux-gnu Compiler: gcc (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0

—

(log yok)

libbb/Config.src+1 -1

@@ -59,7 +59,7 @@ config SHA1_SMALL
	Trade binary size versus speed for the sha1 algorithm.
	throughput MB/s size of sha1_process_block64
	value 486 x86-64 486 x86-64
-	0 339 374 4149 4167
+	0 360 374 3950 4167
	1 224 229 654 732
	2,3 200 195 358 380

libbb/hash_md5_sha.c+22 -0

@@ -509,6 +509,27 @@ static void FAST_FUNC sha1_process_block64(sha1_ctx_t *ctx)
	d = ctx->hash[3];
	e = ctx->hash[4];

+	/* From kernel source comments:
+	* """
+	* If you have 32 registers or more, the compiler can (and should)
+	* try to change the array[] accesses into registers. However, on
+	* machines with less than ~25 registers, that won't really work,
+	* and at least gcc will make an unholy mess of it.
+	*
+	* So to avoid that mess which just slows things down, we force
+	* the stores to memory to actually happen (we might be better off
+	* with a 'W(t)=(val);asm("":"+m" (W(t))' there instead, as
+	* suggested by Artur Skawina - that will also make gcc unable to
+	* try to do the silly "optimize away loads" part because it won't
+	* see what the value will be).
+	* """
+	*/
+	#if defined(__i386__)
+	# define DO_NOT_TRY_PROPAGATING(m) asm("":"+m"(m))
+	#else
+	# define DO_NOT_TRY_PROPAGATING(m) ((void)0)
+	#endif
+
	#undef OP
	#define OP(A,B,C,D,E, n) \
	do { \
@@ -517,6 +538,7 @@ static void FAST_FUNC sha1_process_block64(sha1_ctx_t *ctx)
	work += W[n & 15] = SWAP_BE32(((uint32_t*)ctx->wbuffer)[n]); \
	if (n >= 16) \
	work += W[n & 15] = rotl32(W[(n+13) & 15] ^ W[(n+8) & 15] ^ W[(n+2) & 15] ^ W[n & 15], 1); \
+	DO_NOT_TRY_PROPAGATING(W[n & 15]); \
	E += work + rotl32(A, 5) + rconsts[n / 20]; \
	B = rotl32(B, 30); \
	} while (0)