| 1 |
/* |
| 2 |
* (C) 2020 Jack Lloyd |
| 3 |
* |
| 4 |
* Botan is released under the Simplified BSD License (see license.txt) |
| 5 |
*/ |
| 6 |
|
| 7 |
#include <botan/internal/shacal2.h> |
| 8 |
|
| 9 |
#include <botan/internal/isa_extn.h> |
| 10 |
#include <arm_neon.h> |
| 11 |
|
| 12 |
namespace Botan { |
| 13 |
|
| 14 |
/* |
| 15 |
Only encryption is supported since the inverse round function would |
| 16 |
require a different instruction |
| 17 |
*/ |
| 18 |
BOTAN_FN_ISA_SHA2 |
| 19 |
void SHACAL2::armv8_encrypt_blocks(const uint8_t in[], uint8_t out[], size_t blocks) const { |
| 20 |
const uint32_t* input32 = reinterpret_cast<const uint32_t*>(in); |
| 21 |
uint32_t* output32 = reinterpret_cast<uint32_t*>(out); |
| 22 |
|
| 23 |
while(blocks >= 2) { |
| 24 |
uint32x4_t B0_0 = vld1q_u32(input32 + 0); |
| 25 |
uint32x4_t B0_1 = vld1q_u32(input32 + 4); |
| 26 |
uint32x4_t B1_0 = vld1q_u32(input32 + 8); |
| 27 |
uint32x4_t B1_1 = vld1q_u32(input32 + 12); |
| 28 |
|
| 29 |
B0_0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0_0))); |
| 30 |
B0_1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0_1))); |
| 31 |
B1_0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1_0))); |
| 32 |
B1_1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1_1))); |
| 33 |
|
| 34 |
for(size_t i = 0; i != 8; ++i) { |
| 35 |
const auto RK0 = vld1q_u32(&m_RK[8 * i]); |
| 36 |
const auto RK1 = vld1q_u32(&m_RK[8 * i + 4]); |
| 37 |
|
| 38 |
const auto T0_0 = vsha256hq_u32(B0_0, B0_1, RK0); |
| 39 |
const auto T0_1 = vsha256h2q_u32(B0_1, B0_0, RK0); |
| 40 |
const auto T1_0 = vsha256hq_u32(B1_0, B1_1, RK0); |
| 41 |
const auto T1_1 = vsha256h2q_u32(B1_1, B1_0, RK0); |
| 42 |
|
| 43 |
B0_0 = vsha256hq_u32(T0_0, T0_1, RK1); |
| 44 |
B0_1 = vsha256h2q_u32(T0_1, T0_0, RK1); |
| 45 |
B1_0 = vsha256hq_u32(T1_0, T1_1, RK1); |
| 46 |
B1_1 = vsha256h2q_u32(T1_1, T1_0, RK1); |
| 47 |
} |
| 48 |
|
| 49 |
B0_0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0_0))); |
| 50 |
B0_1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0_1))); |
| 51 |
B1_0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1_0))); |
| 52 |
B1_1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1_1))); |
| 53 |
|
| 54 |
vst1q_u32(&output32[0], B0_0); |
| 55 |
vst1q_u32(&output32[4], B0_1); |
| 56 |
vst1q_u32(&output32[8], B1_0); |
| 57 |
vst1q_u32(&output32[12], B1_1); |
| 58 |
|
| 59 |
blocks -= 2; |
| 60 |
input32 += 16; |
| 61 |
output32 += 16; |
| 62 |
} |
| 63 |
|
| 64 |
while(blocks > 0) { |
| 65 |
uint32x4_t B0 = vld1q_u32(input32 + 0); |
| 66 |
uint32x4_t B1 = vld1q_u32(input32 + 4); |
| 67 |
|
| 68 |
B0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0))); |
| 69 |
B1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1))); |
| 70 |
|
| 71 |
for(size_t i = 0; i != 8; ++i) { |
| 72 |
const auto RK0 = vld1q_u32(&m_RK[8 * i]); |
| 73 |
const auto RK1 = vld1q_u32(&m_RK[8 * i + 4]); |
| 74 |
|
| 75 |
const auto T0 = vsha256hq_u32(B0, B1, RK0); |
| 76 |
const auto T1 = vsha256h2q_u32(B1, B0, RK0); |
| 77 |
|
| 78 |
B0 = vsha256hq_u32(T0, T1, RK1); |
| 79 |
B1 = vsha256h2q_u32(T1, T0, RK1); |
| 80 |
} |
| 81 |
|
| 82 |
B0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B0))); |
| 83 |
B1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(B1))); |
| 84 |
|
| 85 |
vst1q_u32(&output32[0], B0); |
| 86 |
vst1q_u32(&output32[4], B1); |
| 87 |
|
| 88 |
blocks--; |
| 89 |
input32 += 8; |
| 90 |
output32 += 8; |
| 91 |
} |
| 92 |
} |
| 93 |
|
| 94 |
} // namespace Botan |