common/sha256: agressive SHA-256 unrolling as an option

Reduces "hash done" time from ~1.30 to ~1.15s on soraka.

BRANCH=none
BUG=chromium:702378
BUG=b:64196191
TEST=Boot soraka, looks at hash done time.
TEST=make run-sha256 run-sha256_unrolled passes.

Change-Id: Ia29ee27404d6e9aa615ff59755b59d3f26648e71
Signed-off-by: Nicolas Boichat <drinkcat@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/652327
Reviewed-by: Furquan Shaikh <furquan@chromium.org>
Reviewed-by: Vincent Palatin <vpalatin@chromium.org>

common/sha256.c

@@ -131,12 +131,37 @@ static void SHA256_transform(struct sha256_ctx *ctx, const uint8_t *message,
 		for (j = 0; j < 16; j++)
 			PACK32(&sub_block[j << 2], &w[j]);
 
+#ifdef CONFIG_SHA256_UNROLLED
+		for (j = 16; j < 64; j += 8) {
+			SHA256_SCR(j);
+			SHA256_SCR(j+1);
+			SHA256_SCR(j+2);
+			SHA256_SCR(j+3);
+			SHA256_SCR(j+4);
+			SHA256_SCR(j+5);
+			SHA256_SCR(j+6);
+			SHA256_SCR(j+7);
+		}
+#else
 		for (j = 16; j < 64; j++)
 			SHA256_SCR(j);
+#endif
 
 		for (j = 0; j < 8; j++)
 			wv[j] = ctx->h[j];
 
+#ifdef CONFIG_SHA256_UNROLLED
+		for (j = 0; j < 64; j += 8) {
+			SHA256_EXP(0, 1, 2, 3, 4, 5, 6, 7, j);
+			SHA256_EXP(7, 0, 1, 2, 3, 4, 5, 6, j+1);
+			SHA256_EXP(6, 7, 0, 1, 2, 3, 4, 5, j+2);
+			SHA256_EXP(5, 6, 7, 0, 1, 2, 3, 4, j+3);
+			SHA256_EXP(4, 5, 6, 7, 0, 1, 2, 3, j+4);
+			SHA256_EXP(3, 4, 5, 6, 7, 0, 1, 2, j+5);
+			SHA256_EXP(2, 3, 4, 5, 6, 7, 0, 1, j+6);
+			SHA256_EXP(1, 2, 3, 4, 5, 6, 7, 0, j+7);
+		}
+#else
 		for (j = 0; j < 64; j++) {
 			t1 = wv[7] + SHA256_F2(wv[4]) + CH(wv[4], wv[5], wv[6])
 				+ sha256_k[j] + w[j];
@@ -150,6 +175,7 @@ static void SHA256_transform(struct sha256_ctx *ctx, const uint8_t *message,
 			wv[1] = wv[0];
 			wv[0] = t1 + t2;
 		}
+#endif
 
 		for (j = 0; j < 8; j++)
 			ctx->h[j] += wv[j];

include/config.h

@@ -2109,6 +2109,9 @@
 /* Support computing of other hash sizes (without the VBOOT code) */
 #undef CONFIG_SHA256
 
+/* Unroll some loops in SHA256_transform for better performance. */
+#undef CONFIG_SHA256_UNROLLED
+
 /* Emulate the CLZ (Count Leading Zeros) in software for CPU lacking support */
 #undef CONFIG_SOFTWARE_CLZ
 

test/build.mk

@@ -71,6 +71,7 @@ test-list-host += rsa
 test-list-host += rsa3
 test-list-host += sbs_charging_v2
 test-list-host += sha256
+test-list-host += sha256_unrolled
 test-list-host += shmalloc
 test-list-host += system
 test-list-host += thermal
@@ -121,6 +122,7 @@ rsa3-y=rsa.o
 sbs_charging-y=sbs_charging.o
 sbs_charging_v2-y=sbs_charging_v2.o
 sha256-y=sha256.o
+sha256_unrolled-y=sha256.o
 shmalloc-y=shmalloc.o
 stress-y=stress.o
 system-y=system.o

test/sha256_unrolled.tasklist

@@ -0,0 +1 @@
+sha256.tasklist

test/test_config.h

@@ -85,6 +85,11 @@
 #define CONFIG_SHA256
 #endif
 
+#ifdef TEST_SHA256_UNROLLED
+#define CONFIG_SHA256
+#define CONFIG_SHA256_UNROLLED
+#endif
+
 #ifdef TEST_SHMALLOC
 #define CONFIG_MALLOC
 #endif