liblzma: SHA-256: Optimize the way rotations are done.
This looks weird because the rotations become sequential,
but it helps quite a bit on both 32-bit and 64-bit x86:
- It requires fewer instructions on two-operand
instruction sets like x86.
- It requires one register less which matters especially
on 32-bit x86.
I hope this doesn't hurt other archs.
I didn't invent this idea myself, but I don't remember where
I saw it first.
diff --git a/src/liblzma/check/sha256.c b/src/liblzma/check/sha256.c
index 6e2f65f..e0e2f10 100644
--- a/src/liblzma/check/sha256.c
+++ b/src/liblzma/check/sha256.c
@@ -23,8 +23,13 @@
#include "check.h"
-// At least on x86, GCC is able to optimize this to a rotate instruction.
-#define rotr_32(num, amount) ((num) >> (amount) | (num) << (32 - (amount)))
+// Rotate a uint32_t. GCC can optimize this to a rotate instruction
+// at least on x86.
+static inline uint32_t
+rotr_32(uint32_t num, unsigned amount)
+{
+ return (num >> amount) | (num << (32 - amount));
+}
#define blk0(i) (W[i] = conv32be(data[i]))
#define blk2(i) (W[i & 15] += s1(W[(i - 2) & 15]) + W[(i - 7) & 15] \
@@ -49,10 +54,10 @@
#define R0(i) R(i, 0, blk0(i))
#define R2(i) R(i, j, blk2(i))
-#define S0(x) (rotr_32(x, 2) ^ rotr_32(x, 13) ^ rotr_32(x, 22))
-#define S1(x) (rotr_32(x, 6) ^ rotr_32(x, 11) ^ rotr_32(x, 25))
-#define s0(x) (rotr_32(x, 7) ^ rotr_32(x, 18) ^ (x >> 3))
-#define s1(x) (rotr_32(x, 17) ^ rotr_32(x, 19) ^ (x >> 10))
+#define S0(x) rotr_32(x ^ rotr_32(x ^ rotr_32(x, 9), 11), 2)
+#define S1(x) rotr_32(x ^ rotr_32(x ^ rotr_32(x, 14), 5), 6)
+#define s0(x) (rotr_32(x ^ rotr_32(x, 11), 7) ^ (x >> 3))
+#define s1(x) (rotr_32(x ^ rotr_32(x, 2), 17) ^ (x >> 10))
static const uint32_t SHA256_K[64] = {
