diff options
| author | Nicolas Boichat <drinkcat@google.com> | 2017-03-03 10:12:07 -0800 |
|---|---|---|
| committer | chrome-bot <chrome-bot@chromium.org> | 2018-05-28 22:46:28 -0700 |
| commit | 49ff62bf0be5cfc88e092f987d77c18a0b21390a (patch) | |
| tree | 4f7fc10f29d02422bcdba81d9964de1bf6d91eb8 /core/cortex-m0 | |
| parent | 6e364d59c1ae658876b8a0e58a38edd81233655d (diff) | |
| download | chrome-ec-49ff62bf0be5cfc88e092f987d77c18a0b21390a.tar.gz | |
rsa: Optimization of multiplications for Cortex-M0
We multiply 2 32-bit numbers (and not 64-bit numbers), and then add
another 32-bit number, which makes it possible to optimize the
assembly and save a few instructions.
With -O3, 3072-bit exponent, lower verification time from 122 ms to
104 ms on STM32F072 @48Mhz.
Optimized mac function from Dmitry Grinberg <dmitrygr@google.com>.
BRANCH=poppy
BUG=b:35647963
BUG=b:77608104
TEST=On staff, flash, verification successful
TEST=make test-rsa, make test-rsa3
TEST=Flash test-utils and test-rsa to hammer => pass
Change-Id: I584c54c631a3f59f691849a279b308e8d4b4b22d
Signed-off-by: Nicolas Boichat <drinkcat@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/449024
Reviewed-by: Vincent Palatin <vpalatin@chromium.org>
Diffstat (limited to 'core/cortex-m0')
| -rw-r--r-- | core/cortex-m0/build.mk | 2 | ||||
| -rw-r--r-- | core/cortex-m0/config_core.h | 2 | ||||
| -rw-r--r-- | core/cortex-m0/mula.S | 46 |
3 files changed, 49 insertions, 1 deletions
diff --git a/core/cortex-m0/build.mk b/core/cortex-m0/build.mk index 6a0f44b367..2023bd9607 100644 --- a/core/cortex-m0/build.mk +++ b/core/cortex-m0/build.mk @@ -18,7 +18,7 @@ CFLAGS_CPU+=-flto LDFLAGS_EXTRA+=-flto endif -core-y=cpu.o init.o thumb_case.o div.o lmul.o ldivmod.o uldivmod.o +core-y=cpu.o init.o thumb_case.o div.o lmul.o ldivmod.o mula.o uldivmod.o core-$(CONFIG_COMMON_PANIC_OUTPUT)+=panic.o core-$(CONFIG_COMMON_RUNTIME)+=switch.o task.o diff --git a/core/cortex-m0/config_core.h b/core/cortex-m0/config_core.h index 389ed8edec..19a22bc333 100644 --- a/core/cortex-m0/config_core.h +++ b/core/cortex-m0/config_core.h @@ -15,4 +15,6 @@ #define CONFIG_SOFTWARE_CTZ #define CONFIG_SOFTWARE_PANIC +#define CONFIG_ASSEMBLY_MULA32 + #endif /* __CROS_EC_CONFIG_CORE_H */ diff --git a/core/cortex-m0/mula.S b/core/cortex-m0/mula.S new file mode 100644 index 0000000000..fc0a6f3ee0 --- /dev/null +++ b/core/cortex-m0/mula.S @@ -0,0 +1,46 @@ +/* Copyright 2018 The Chromium OS Authors. All rights reserved. + * Use of this source code is governed by a BSD-style license that can be + * found in the LICENSE file. + * + * Cortex-M0 multiply-accumulate functions + */ + + .syntax unified + .text + .thumb + +@ uint64_t mula32(uint32_t a, uint32_t b, uint32_t c) +@ +@ Multiply a (r0) and b (r1), add c (r2) and return the product in r1:r0 +@ + .thumb_func + .section .text.mula32 + .global mula32 +mula32: + + push {r4, r5} + uxth r4, r0 /* r4 = a.lo16 */ + uxth r5, r1 /* r5 = b.lo16 */ + uxth r3, r2 /* r3 = c.lo16 */ + muls r4, r5 /* r4 = a.lo16 * b.lo16 */ + adds r4, r3 /* r4 = a.lo16 * b.lo16 + c.lo16 == r.lo32 */ + lsrs r3, r0, 16 /* r3 = a.hi16 */ + lsrs r2, r2, 16 /* r2 = c.hi16 */ + muls r5, r3 /* r5 = a.hi16 * b.lo16 */ + adds r5, r2 /* r5 = a.hi16 * b.lo16 + c.hi16 == r.mid32.1 */ + uxth r2, r0 /* r2 = a.lo16 */ + lsrs r1, r1, 16 /* r1 = b.hi16 */ + muls r2, r1 /* r2 = a.lo16 * b.hi16 == r.mid32.2 */ + muls r1, r3 /* r1 = b.hi16 * a.hi16 == r.hi32 */ + movs r3, 0 /* r3 = 0 */ + adds r5, r2 /* r5 = (r.mid32.1 + r.mid32.2).lo32 == r.mid.lo32 */ + adcs r3, r3 /* r3 = (r.mid32.1 + r.mid32.2).hi32 == r.mid.hi32 */ + lsls r0, r5, 16 /* r0 = r.mid.lo32.lo16 << 16 == r.mid.inpos.lo32 */ + lsrs r5, r5, 16 /* r5 = r.mid.lo32.hi16 >> 16 */ + lsls r3, r3, 16 /* r3 = r.mid.hi.lo16 << 16 */ + adds r3, r5 /* r3 = r5 + r3 = r.mid.inpos.hi32 */ + adds r0, r4 /* r0 = r.lo32 */ + adcs r1, r3 /* r1 = r.hi32 */ + pop {r4, r5} + bx lr + |