CryptoNight v8 ReverseWaltz (+asm code and all test vectors)

Co-Authored-By: Vitalius Parubochyi <eddragonwolf@users.noreply.github.com>
This commit is contained in:
SChernykh 2019-03-04 19:57:33 +01:00
parent 10eb579270
commit 154806ac74
18 changed files with 1353 additions and 29 deletions

View file

@ -484,6 +484,7 @@ void xmrig::Pool::rebuild()
m_algorithms.push_back(m_algorithm);
# ifndef XMRIG_PROXY_PROJECT
addVariant(VARIANT_RWZ);
addVariant(VARIANT_4);
addVariant(VARIANT_WOW);
addVariant(VARIANT_2);

View file

@ -66,6 +66,7 @@ static AlgoData const algorithms[] = {
{ "cryptonight/xtlv9", "cn/xtlv9", xmrig::CRYPTONIGHT, xmrig::VARIANT_HALF },
{ "cryptonight/wow", "cn/wow", xmrig::CRYPTONIGHT, xmrig::VARIANT_WOW },
{ "cryptonight/r", "cn/r", xmrig::CRYPTONIGHT, xmrig::VARIANT_4 },
{ "cryptonight/rwz", "cn/rwz", xmrig::CRYPTONIGHT, xmrig::VARIANT_RWZ },
# ifndef XMRIG_NO_AEON
{ "cryptonight-lite", "cn-lite", xmrig::CRYPTONIGHT_LITE, xmrig::VARIANT_AUTO },
@ -133,6 +134,7 @@ static const char *variants[] = {
"gpu",
"wow",
"r",
"rwz"
};

View file

@ -76,6 +76,7 @@ enum Variant {
VARIANT_GPU = 11, // CryptoNight-GPU (Ryo)
VARIANT_WOW = 12, // CryptoNightR (Wownero)
VARIANT_4 = 13, // CryptoNightR (Monero's variant 4)
VARIANT_RWZ = 14, // CryptoNight variant 2 with 3/4 iterations and reversed shuffle operation (Graft)
VARIANT_MAX
};

View file

@ -436,7 +436,7 @@ static inline void cryptonight_monero_tweak(const uint8_t* l, uint64_t idx, __m1
uint64_t* mem_out = (uint64_t*)&l[idx];
if (BASE == xmrig::VARIANT_2) {
VARIANT2_SHUFFLE(l, idx, ax0, bx0, bx1, cx);
VARIANT2_SHUFFLE(l, idx, ax0, bx0, bx1, cx, (VARIANT == xmrig::VARIANT_RWZ ? 1 : 0));
_mm_store_si128((__m128i *)mem_out, _mm_xor_si128(bx0, cx));
} else {
__m128i tmp = _mm_xor_si128(bx0, cx);
@ -530,9 +530,9 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
if (BASE == xmrig::VARIANT_2) {
if (VARIANT == xmrig::VARIANT_4) {
VARIANT2_SHUFFLE(l0, idx0 & MASK, ax0, bx0, bx1, cx);
VARIANT2_SHUFFLE(l0, idx0 & MASK, ax0, bx0, bx1, cx, 0);
} else {
VARIANT2_SHUFFLE2(l0, idx0 & MASK, ax0, bx0, bx1, hi, lo);
VARIANT2_SHUFFLE2(l0, idx0 & MASK, ax0, bx0, bx1, hi, lo, (VARIANT == xmrig::VARIANT_RWZ ? 1 : 0));
}
}
@ -709,9 +709,9 @@ inline void cryptonight_double_hash(const uint8_t *__restrict__ input, size_t si
if (BASE == xmrig::VARIANT_2) {
if (VARIANT == xmrig::VARIANT_4) {
VARIANT2_SHUFFLE(l0, idx0 & MASK, ax0, bx00, bx01, cx0);
VARIANT2_SHUFFLE(l0, idx0 & MASK, ax0, bx00, bx01, cx0, 0);
} else {
VARIANT2_SHUFFLE2(l0, idx0 & MASK, ax0, bx00, bx01, hi, lo);
VARIANT2_SHUFFLE2(l0, idx0 & MASK, ax0, bx00, bx01, hi, lo, (VARIANT == xmrig::VARIANT_RWZ ? 1 : 0));
}
}
@ -767,9 +767,9 @@ inline void cryptonight_double_hash(const uint8_t *__restrict__ input, size_t si
if (BASE == xmrig::VARIANT_2) {
if (VARIANT == xmrig::VARIANT_4) {
VARIANT2_SHUFFLE(l1, idx1 & MASK, ax1, bx10, bx11, cx1);
VARIANT2_SHUFFLE(l1, idx1 & MASK, ax1, bx10, bx11, cx1, 0);
} else {
VARIANT2_SHUFFLE2(l1, idx1 & MASK, ax1, bx10, bx11, hi, lo);
VARIANT2_SHUFFLE2(l1, idx1 & MASK, ax1, bx10, bx11, hi, lo, (VARIANT == xmrig::VARIANT_RWZ ? 1 : 0));
}
}

View file

@ -42,6 +42,7 @@ constexpr const uint32_t CRYPTONIGHT_MASK = 0x1FFFF0;
constexpr const uint32_t CRYPTONIGHT_ITER = 0x80000;
constexpr const uint32_t CRYPTONIGHT_HALF_ITER = 0x40000;
constexpr const uint32_t CRYPTONIGHT_XAO_ITER = 0x100000;
constexpr const uint32_t CRYPTONIGHT_WALTZ_ITER = 0x60000;
constexpr const uint32_t CRYPTONIGHT_GPU_ITER = 0xC000;
constexpr const uint32_t CRYPTONIGHT_GPU_MASK = 0x1FFFC0;
@ -134,6 +135,7 @@ template<> inline constexpr uint32_t cn_select_iter<CRYPTONIGHT, VARIANT_MSR>()
template<> inline constexpr uint32_t cn_select_iter<CRYPTONIGHT, VARIANT_XAO>() { return CRYPTONIGHT_XAO_ITER; }
template<> inline constexpr uint32_t cn_select_iter<CRYPTONIGHT, VARIANT_RTO>() { return CRYPTONIGHT_ITER; }
template<> inline constexpr uint32_t cn_select_iter<CRYPTONIGHT, VARIANT_GPU>() { return CRYPTONIGHT_GPU_ITER; }
template<> inline constexpr uint32_t cn_select_iter<CRYPTONIGHT, VARIANT_RWZ>() { return CRYPTONIGHT_WALTZ_ITER; }
template<> inline constexpr uint32_t cn_select_iter<CRYPTONIGHT_LITE, VARIANT_0>() { return CRYPTONIGHT_LITE_ITER; }
template<> inline constexpr uint32_t cn_select_iter<CRYPTONIGHT_LITE, VARIANT_1>() { return CRYPTONIGHT_LITE_ITER; }
template<> inline constexpr uint32_t cn_select_iter<CRYPTONIGHT_HEAVY, VARIANT_0>() { return CRYPTONIGHT_HEAVY_ITER; }
@ -158,6 +160,9 @@ inline uint32_t cn_select_iter(Algo algorithm, Variant variant)
case VARIANT_TRTL:
return CRYPTONIGHT_TRTL_ITER;
case VARIANT_RWZ:
return CRYPTONIGHT_WALTZ_ITER;
default:
break;
}
@ -199,6 +204,7 @@ template<> inline constexpr Variant cn_base_variant<VARIANT_TRTL>() { return VA
template<> inline constexpr Variant cn_base_variant<VARIANT_GPU>() { return VARIANT_GPU; }
template<> inline constexpr Variant cn_base_variant<VARIANT_WOW>() { return VARIANT_2; }
template<> inline constexpr Variant cn_base_variant<VARIANT_4>() { return VARIANT_2; }
template<> inline constexpr Variant cn_base_variant<VARIANT_RWZ>() { return VARIANT_2; }
template<Variant variant> inline constexpr bool cn_is_cryptonight_r() { return false; }

View file

@ -83,11 +83,11 @@
sqrt_result_xmm_##part = int_sqrt_v2(cx_0 + division_result); \
} while (0)
# define VARIANT2_SHUFFLE(base_ptr, offset, _a, _b, _b1, _c) \
# define VARIANT2_SHUFFLE(base_ptr, offset, _a, _b, _b1, _c, reverse) \
do { \
const __m128i chunk1 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10))); \
const __m128i chunk1 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ (reverse ? 0x30 : 0x10)))); \
const __m128i chunk2 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20))); \
const __m128i chunk3 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x30))); \
const __m128i chunk3 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ (reverse ? 0x10 : 0x30)))); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10)), _mm_add_epi64(chunk3, _b1)); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20)), _mm_add_epi64(chunk1, _b)); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x30)), _mm_add_epi64(chunk2, _a)); \
@ -96,15 +96,20 @@
} \
} while (0)
# define VARIANT2_SHUFFLE2(base_ptr, offset, _a, _b, _b1, hi, lo) \
# define VARIANT2_SHUFFLE2(base_ptr, offset, _a, _b, _b1, hi, lo, reverse) \
do { \
const __m128i chunk1 = _mm_xor_si128(_mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10))), _mm_set_epi64x(lo, hi)); \
const __m128i chunk2 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20))); \
hi ^= ((uint64_t*)((base_ptr) + ((offset) ^ 0x20)))[0]; \
lo ^= ((uint64_t*)((base_ptr) + ((offset) ^ 0x20)))[1]; \
const __m128i chunk3 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x30))); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10)), _mm_add_epi64(chunk3, _b1)); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20)), _mm_add_epi64(chunk1, _b)); \
if (reverse) { \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10)), _mm_add_epi64(chunk1, _b1)); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20)), _mm_add_epi64(chunk3, _b)); \
} else { \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10)), _mm_add_epi64(chunk3, _b1)); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20)), _mm_add_epi64(chunk1, _b)); \
} \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x30)), _mm_add_epi64(chunk2, _a)); \
} while (0)
@ -128,11 +133,11 @@
sqrt_result_##part += ((r2 + b > sqrt_input) ? -1 : 0) + ((r2 + (1ULL << 32) < sqrt_input - s) ? 1 : 0); \
} while (0)
# define VARIANT2_SHUFFLE(base_ptr, offset, _a, _b, _b1, _c) \
# define VARIANT2_SHUFFLE(base_ptr, offset, _a, _b, _b1, _c, reverse) \
do { \
const uint64x2_t chunk1 = vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x10))); \
const uint64x2_t chunk1 = vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ (reverse ? 0x30 : 0x10)))); \
const uint64x2_t chunk2 = vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x20))); \
const uint64x2_t chunk3 = vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x30))); \
const uint64x2_t chunk3 = vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ (reverse ? 0x10 : 0x30)))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x10)), vaddq_u64(chunk3, vreinterpretq_u64_u8(_b1))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x20)), vaddq_u64(chunk1, vreinterpretq_u64_u8(_b))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x30)), vaddq_u64(chunk2, vreinterpretq_u64_u8(_a))); \
@ -141,15 +146,20 @@
} \
} while (0)
# define VARIANT2_SHUFFLE2(base_ptr, offset, _a, _b, _b1, hi, lo) \
# define VARIANT2_SHUFFLE2(base_ptr, offset, _a, _b, _b1, hi, lo, reverse) \
do { \
const uint64x2_t chunk1 = veorq_u64(vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x10))), vcombine_u64(vcreate_u64(hi), vcreate_u64(lo))); \
const uint64x2_t chunk2 = vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x20))); \
hi ^= ((uint64_t*)((base_ptr) + ((offset) ^ 0x20)))[0]; \
lo ^= ((uint64_t*)((base_ptr) + ((offset) ^ 0x20)))[1]; \
const uint64x2_t chunk3 = vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x30))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x10)), vaddq_u64(chunk3, vreinterpretq_u64_u8(_b1))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x20)), vaddq_u64(chunk1, vreinterpretq_u64_u8(_b))); \
if (reverse) { \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x10)), vaddq_u64(chunk1, vreinterpretq_u64_u8(_b1))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x20)), vaddq_u64(chunk3, vreinterpretq_u64_u8(_b))); \
} else { \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x10)), vaddq_u64(chunk3, vreinterpretq_u64_u8(_b1))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x20)), vaddq_u64(chunk1, vreinterpretq_u64_u8(_b))); \
} \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x30)), vaddq_u64(chunk2, vreinterpretq_u64_u8(_a))); \
} while (0)
#endif

View file

@ -230,6 +230,18 @@ const static uint8_t test_output_rto[160] = {
0xE7, 0x81, 0x4E, 0x2A, 0xBD, 0x62, 0xC1, 0x1B, 0x7C, 0xB9, 0x33, 0x7B, 0xEE, 0x95, 0x80, 0xB3
};
const static uint8_t test_output_rwz[160] = {
0x5f, 0x56, 0xc6, 0xb0, 0x99, 0x6b, 0xa2, 0x3e, 0x0b, 0xba, 0x07, 0x29, 0xc9, 0x90, 0x74, 0x85,
0x5a, 0x10, 0xe3, 0x08, 0x7f, 0xdb, 0xfe, 0x94, 0x75, 0x33, 0x54, 0x73, 0x76, 0xf0, 0x75, 0xb8,
0x8b, 0x70, 0x43, 0x9a, 0xfc, 0xf5, 0xeb, 0x15, 0xbb, 0xf9, 0xad, 0x9d, 0x2a, 0xbd, 0x72, 0x52,
0x49, 0x54, 0x0b, 0x91, 0xea, 0x61, 0x7f, 0x98, 0x7d, 0x39, 0x17, 0xb7, 0xd7, 0x65, 0xff, 0x75,
0x13, 0x21, 0x1d, 0xce, 0x61, 0x5a, 0xdc, 0x5f, 0x8c, 0xcb, 0x1f, 0x6f, 0xbb, 0x92, 0x88, 0xc3,
0xe3, 0xe2, 0xfc, 0x4f, 0x62, 0xfb, 0xf0, 0x48, 0x02, 0x01, 0xd3, 0xbe, 0x77, 0x6a, 0x40, 0xca,
0x9a, 0xe9, 0xba, 0x0c, 0xc0, 0x2b, 0x11, 0xf6, 0x9b, 0xee, 0x24, 0x3a, 0xd8, 0x86, 0x18, 0xd0,
0xe8, 0xeb, 0xcb, 0x38, 0x2c, 0xf5, 0x99, 0x83, 0x14, 0x7b, 0x0c, 0x20, 0xbe, 0x50, 0xf4, 0x87,
0x83, 0x41, 0x75, 0xd8, 0xd1, 0xdd, 0x4b, 0x73, 0xb3, 0x92, 0x8f, 0xe6, 0x1c, 0x72, 0x70, 0xf5,
0x7c, 0xf6, 0x23, 0x3a, 0xb4, 0x5f, 0xdf, 0xde, 0xa6, 0x5a, 0x58, 0xec, 0x13, 0x5a, 0x23, 0x2f,
};
#ifndef XMRIG_NO_AEON
// "cn-lite/0"

View file

@ -531,7 +531,7 @@ template<xmrig::Variant VARIANT, xmrig::Variant BASE>
static inline void cryptonight_monero_tweak(uint64_t* mem_out, const uint8_t* l, uint64_t idx, __m128i ax0, __m128i bx0, __m128i bx1, __m128i& cx)
{
if (BASE == xmrig::VARIANT_2) {
VARIANT2_SHUFFLE(l, idx, ax0, bx0, bx1, cx);
VARIANT2_SHUFFLE(l, idx, ax0, bx0, bx1, cx, (VARIANT == xmrig::VARIANT_RWZ ? 1 : 0));
_mm_store_si128((__m128i *)mem_out, _mm_xor_si128(bx0, cx));
} else {
__m128i tmp = _mm_xor_si128(bx0, cx);
@ -653,9 +653,9 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
if (BASE == xmrig::VARIANT_2) {
if (VARIANT == xmrig::VARIANT_4) {
VARIANT2_SHUFFLE(l0, idx0 & MASK, ax0, bx0, bx1, cx);
VARIANT2_SHUFFLE(l0, idx0 & MASK, ax0, bx0, bx1, cx, 0);
} else {
VARIANT2_SHUFFLE2(l0, idx0 & MASK, ax0, bx0, bx1, hi, lo);
VARIANT2_SHUFFLE2(l0, idx0 & MASK, ax0, bx0, bx1, hi, lo, (VARIANT == xmrig::VARIANT_RWZ ? 1 : 0));
}
}
@ -754,6 +754,8 @@ extern "C" void cnv2_mainloop_ivybridge_asm(cryptonight_ctx *ctx);
extern "C" void cnv2_mainloop_ryzen_asm(cryptonight_ctx *ctx);
extern "C" void cnv2_mainloop_bulldozer_asm(cryptonight_ctx *ctx);
extern "C" void cnv2_double_mainloop_sandybridge_asm(cryptonight_ctx* ctx0, cryptonight_ctx* ctx1);
extern "C" void cnv2_rwz_mainloop_asm(cryptonight_ctx *ctx);
extern "C" void cnv2_rwz_double_mainloop_asm(cryptonight_ctx* ctx0, cryptonight_ctx* ctx1);
extern xmrig::CpuThread::cn_mainloop_fun cn_half_mainloop_ivybridge_asm;
extern xmrig::CpuThread::cn_mainloop_fun cn_half_mainloop_ryzen_asm;
@ -843,6 +845,9 @@ inline void cryptonight_single_hash_asm(const uint8_t *__restrict__ input, size_
cn_trtl_mainloop_bulldozer_asm(ctx[0]);
}
}
else if (VARIANT == xmrig::VARIANT_RWZ) {
cnv2_rwz_mainloop_asm(ctx[0]);
}
else if (xmrig::cn_is_cryptonight_r<VARIANT>()) {
ctx[0]->generated_code(ctx[0]);
}
@ -881,6 +886,9 @@ inline void cryptonight_double_hash_asm(const uint8_t *__restrict__ input, size_
else if (VARIANT == xmrig::VARIANT_TRTL) {
cn_trtl_double_mainloop_sandybridge_asm(ctx[0], ctx[1]);
}
else if (VARIANT == xmrig::VARIANT_RWZ) {
cnv2_rwz_double_mainloop_asm(ctx[0], ctx[1]);
}
else if (xmrig::cn_is_cryptonight_r<VARIANT>()) {
ctx[0]->generated_code_double(ctx[0], ctx[1]);
}
@ -995,9 +1003,9 @@ inline void cryptonight_double_hash(const uint8_t *__restrict__ input, size_t si
if (BASE == xmrig::VARIANT_2) {
if (VARIANT == xmrig::VARIANT_4) {
VARIANT2_SHUFFLE(l0, idx0 & MASK, ax0, bx00, bx01, cx0);
VARIANT2_SHUFFLE(l0, idx0 & MASK, ax0, bx00, bx01, cx0, 0);
} else {
VARIANT2_SHUFFLE2(l0, idx0 & MASK, ax0, bx00, bx01, hi, lo);
VARIANT2_SHUFFLE2(l0, idx0 & MASK, ax0, bx00, bx01, hi, lo, (VARIANT == xmrig::VARIANT_RWZ ? 1 : 0));
}
}
@ -1051,9 +1059,9 @@ inline void cryptonight_double_hash(const uint8_t *__restrict__ input, size_t si
if (BASE == xmrig::VARIANT_2) {
if (VARIANT == xmrig::VARIANT_4) {
VARIANT2_SHUFFLE(l1, idx1 & MASK, ax1, bx10, bx11, cx1);
VARIANT2_SHUFFLE(l1, idx1 & MASK, ax1, bx10, bx11, cx1, 0);
} else {
VARIANT2_SHUFFLE2(l1, idx1 & MASK, ax1, bx10, bx11, hi, lo);
VARIANT2_SHUFFLE2(l1, idx1 & MASK, ax1, bx10, bx11, hi, lo, (VARIANT == xmrig::VARIANT_RWZ ? 1 : 0));
}
}
@ -1155,9 +1163,9 @@ inline void cryptonight_double_hash(const uint8_t *__restrict__ input, size_t si
lo = __umul128(idx, cl##part, &hi); \
if (BASE == xmrig::VARIANT_2) { \
if (VARIANT == xmrig::VARIANT_4) { \
VARIANT2_SHUFFLE(l, idx & MASK, a, b0, b1, c); \
VARIANT2_SHUFFLE(l, idx & MASK, a, b0, b1, c, 0); \
} else { \
VARIANT2_SHUFFLE2(l, idx & MASK, a, b0, b1, hi, lo); \
VARIANT2_SHUFFLE2(l, idx & MASK, a, b0, b1, hi, lo, (VARIANT == xmrig::VARIANT_RWZ ? 1 : 0)); \
} \
} \
if (VARIANT == xmrig::VARIANT_4) { \

View file

@ -0,0 +1,410 @@
mov rax, rsp
push rbx
push rbp
push rsi
push rdi
push r12
push r13
push r14
push r15
sub rsp, 184
stmxcsr DWORD PTR [rsp+272]
mov DWORD PTR [rsp+276], 24448
ldmxcsr DWORD PTR [rsp+276]
mov r13, QWORD PTR [rcx+224]
mov r9, rdx
mov r10, QWORD PTR [rcx+32]
mov r8, rcx
xor r10, QWORD PTR [rcx]
mov r14d, 393216
mov r11, QWORD PTR [rcx+40]
xor r11, QWORD PTR [rcx+8]
mov rsi, QWORD PTR [rdx+224]
mov rdx, QWORD PTR [rcx+56]
xor rdx, QWORD PTR [rcx+24]
mov rdi, QWORD PTR [r9+32]
xor rdi, QWORD PTR [r9]
mov rbp, QWORD PTR [r9+40]
xor rbp, QWORD PTR [r9+8]
movq xmm0, rdx
movaps XMMWORD PTR [rax-88], xmm6
movaps XMMWORD PTR [rax-104], xmm7
movaps XMMWORD PTR [rax-120], xmm8
movaps XMMWORD PTR [rsp+112], xmm9
movaps XMMWORD PTR [rsp+96], xmm10
movaps XMMWORD PTR [rsp+80], xmm11
movaps XMMWORD PTR [rsp+64], xmm12
movaps XMMWORD PTR [rsp+48], xmm13
movaps XMMWORD PTR [rsp+32], xmm14
movaps XMMWORD PTR [rsp+16], xmm15
mov rdx, r10
movq xmm4, QWORD PTR [r8+96]
and edx, 2097136
mov rax, QWORD PTR [rcx+48]
xorps xmm13, xmm13
xor rax, QWORD PTR [rcx+16]
mov rcx, QWORD PTR [rcx+88]
xor rcx, QWORD PTR [r8+72]
movq xmm5, QWORD PTR [r8+104]
movq xmm7, rax
mov eax, 1
shl rax, 52
movq xmm14, rax
punpcklqdq xmm14, xmm14
mov eax, 1023
shl rax, 52
movq xmm12, rax
punpcklqdq xmm12, xmm12
mov rax, QWORD PTR [r8+80]
xor rax, QWORD PTR [r8+64]
punpcklqdq xmm7, xmm0
movq xmm0, rcx
mov rcx, QWORD PTR [r9+56]
xor rcx, QWORD PTR [r9+24]
movq xmm3, rax
mov rax, QWORD PTR [r9+48]
xor rax, QWORD PTR [r9+16]
punpcklqdq xmm3, xmm0
movq xmm0, rcx
mov QWORD PTR [rsp], r13
mov rcx, QWORD PTR [r9+88]
xor rcx, QWORD PTR [r9+72]
movq xmm6, rax
mov rax, QWORD PTR [r9+80]
xor rax, QWORD PTR [r9+64]
punpcklqdq xmm6, xmm0
movq xmm0, rcx
mov QWORD PTR [rsp+256], r10
mov rcx, rdi
mov QWORD PTR [rsp+264], r11
movq xmm8, rax
and ecx, 2097136
punpcklqdq xmm8, xmm0
movq xmm0, QWORD PTR [r9+96]
punpcklqdq xmm4, xmm0
movq xmm0, QWORD PTR [r9+104]
lea r8, QWORD PTR [rcx+rsi]
movdqu xmm11, XMMWORD PTR [r8]
punpcklqdq xmm5, xmm0
lea r9, QWORD PTR [rdx+r13]
movdqu xmm15, XMMWORD PTR [r9]
ALIGN(64)
rwz_main_loop_double:
movdqu xmm9, xmm15
mov eax, edx
mov ebx, edx
xor eax, 16
xor ebx, 32
xor edx, 48
movq xmm0, r11
movq xmm2, r10
punpcklqdq xmm2, xmm0
aesenc xmm9, xmm2
movdqu xmm0, XMMWORD PTR [rdx+r13]
movdqu xmm1, XMMWORD PTR [rbx+r13]
paddq xmm0, xmm7
paddq xmm1, xmm2
movdqu XMMWORD PTR [rbx+r13], xmm0
movdqu xmm0, XMMWORD PTR [rax+r13]
movdqu XMMWORD PTR [rdx+r13], xmm1
paddq xmm0, xmm3
movdqu XMMWORD PTR [rax+r13], xmm0
movq r11, xmm9
mov edx, r11d
and edx, 2097136
movdqa xmm0, xmm9
pxor xmm0, xmm7
movdqu XMMWORD PTR [r9], xmm0
lea rbx, QWORD PTR [rdx+r13]
mov r10, QWORD PTR [rdx+r13]
movdqu xmm10, xmm11
movq xmm0, rbp
movq xmm11, rdi
punpcklqdq xmm11, xmm0
aesenc xmm10, xmm11
mov eax, ecx
mov r12d, ecx
xor eax, 16
xor r12d, 32
xor ecx, 48
movdqu xmm0, XMMWORD PTR [rcx+rsi]
paddq xmm0, xmm6
movdqu xmm1, XMMWORD PTR [r12+rsi]
movdqu XMMWORD PTR [r12+rsi], xmm0
paddq xmm1, xmm11
movdqu xmm0, XMMWORD PTR [rax+rsi]
movdqu XMMWORD PTR [rcx+rsi], xmm1
paddq xmm0, xmm8
movdqu XMMWORD PTR [rax+rsi], xmm0
movq rcx, xmm10
and ecx, 2097136
movdqa xmm0, xmm10
pxor xmm0, xmm6
movdqu XMMWORD PTR [r8], xmm0
mov r12, QWORD PTR [rcx+rsi]
mov r9, QWORD PTR [rbx+8]
xor edx, 16
mov r8d, edx
mov r15d, edx
movq rdx, xmm5
shl rdx, 32
movq rax, xmm4
xor rdx, rax
xor r10, rdx
mov rax, r10
mul r11
mov r11d, r8d
xor r11d, 48
movq xmm0, rdx
xor rdx, [r11+r13]
movq xmm1, rax
xor rax, [r11+r13+8]
punpcklqdq xmm0, xmm1
pxor xmm0, XMMWORD PTR [r8+r13]
movdqu xmm1, XMMWORD PTR [r11+r13]
paddq xmm0, xmm3
paddq xmm1, xmm2
movdqu XMMWORD PTR [r8+r13], xmm0
xor r8d, 32
movdqu xmm0, XMMWORD PTR [r8+r13]
movdqu XMMWORD PTR [r8+r13], xmm1
paddq xmm0, xmm7
movdqu XMMWORD PTR [r11+r13], xmm0
mov r11, QWORD PTR [rsp+256]
add r11, rdx
mov rdx, QWORD PTR [rsp+264]
add rdx, rax
mov QWORD PTR [rbx], r11
xor r11, r10
mov QWORD PTR [rbx+8], rdx
xor rdx, r9
mov QWORD PTR [rsp+256], r11
and r11d, 2097136
mov QWORD PTR [rsp+264], rdx
mov QWORD PTR [rsp+8], r11
lea r15, QWORD PTR [r11+r13]
movdqu xmm15, XMMWORD PTR [r11+r13]
lea r13, QWORD PTR [rsi+rcx]
movdqa xmm0, xmm5
psrldq xmm0, 8
movaps xmm2, xmm13
movq r10, xmm0
psllq xmm5, 1
shl r10, 32
movdqa xmm0, xmm9
psrldq xmm0, 8
movdqa xmm1, xmm10
movq r11, xmm0
psrldq xmm1, 8
movq r8, xmm1
psrldq xmm4, 8
movaps xmm0, xmm13
movq rax, xmm4
xor r10, rax
movaps xmm1, xmm13
xor r10, r12
lea rax, QWORD PTR [r11+1]
shr rax, 1
movdqa xmm3, xmm9
punpcklqdq xmm3, xmm10
paddq xmm5, xmm3
movq rdx, xmm5
psrldq xmm5, 8
cvtsi2sd xmm2, rax
or edx, -2147483647
lea rax, QWORD PTR [r8+1]
shr rax, 1
movq r9, xmm5
cvtsi2sd xmm0, rax
or r9d, -2147483647
cvtsi2sd xmm1, rdx
unpcklpd xmm2, xmm0
movaps xmm0, xmm13
cvtsi2sd xmm0, r9
unpcklpd xmm1, xmm0
divpd xmm2, xmm1
paddq xmm2, xmm14
cvttsd2si rax, xmm2
psrldq xmm2, 8
mov rbx, rax
imul rax, rdx
sub r11, rax
js rwz_div_fix_1
rwz_div_fix_1_ret:
cvttsd2si rdx, xmm2
mov rax, rdx
imul rax, r9
movd xmm2, r11d
movd xmm4, ebx
sub r8, rax
js rwz_div_fix_2
rwz_div_fix_2_ret:
movd xmm1, r8d
movd xmm0, edx
punpckldq xmm2, xmm1
punpckldq xmm4, xmm0
punpckldq xmm4, xmm2
paddq xmm3, xmm4
movdqa xmm0, xmm3
psrlq xmm0, 12
paddq xmm0, xmm12
sqrtpd xmm1, xmm0
movq r9, xmm1
movdqa xmm5, xmm1
psrlq xmm5, 19
test r9, 524287
je rwz_sqrt_fix_1
rwz_sqrt_fix_1_ret:
movq r9, xmm10
psrldq xmm1, 8
movq r8, xmm1
test r8, 524287
je rwz_sqrt_fix_2
rwz_sqrt_fix_2_ret:
mov r12d, ecx
mov r8d, ecx
xor r12d, 16
xor r8d, 32
xor ecx, 48
mov rax, r10
mul r9
movq xmm0, rax
movq xmm3, rdx
punpcklqdq xmm3, xmm0
movdqu xmm0, XMMWORD PTR [r12+rsi]
pxor xmm0, xmm3
movdqu xmm1, XMMWORD PTR [r8+rsi]
xor rdx, [r8+rsi]
xor rax, [r8+rsi+8]
movdqu xmm3, XMMWORD PTR [rcx+rsi]
paddq xmm3, xmm6
paddq xmm1, xmm11
paddq xmm0, xmm8
movdqu XMMWORD PTR [r8+rsi], xmm3
movdqu XMMWORD PTR [rcx+rsi], xmm1
movdqu XMMWORD PTR [r12+rsi], xmm0
add rdi, rdx
mov QWORD PTR [r13], rdi
xor rdi, r10
mov ecx, edi
and ecx, 2097136
lea r8, QWORD PTR [rcx+rsi]
mov rdx, QWORD PTR [r13+8]
add rbp, rax
mov QWORD PTR [r13+8], rbp
movdqu xmm11, XMMWORD PTR [rcx+rsi]
xor rbp, rdx
mov r13, QWORD PTR [rsp]
movdqa xmm3, xmm7
mov rdx, QWORD PTR [rsp+8]
movdqa xmm8, xmm6
mov r10, QWORD PTR [rsp+256]
movdqa xmm7, xmm9
mov r11, QWORD PTR [rsp+264]
movdqa xmm6, xmm10
mov r9, r15
dec r14d
jne rwz_main_loop_double
ldmxcsr DWORD PTR [rsp+272]
movaps xmm13, XMMWORD PTR [rsp+48]
lea r11, QWORD PTR [rsp+184]
movaps xmm6, XMMWORD PTR [r11-24]
movaps xmm7, XMMWORD PTR [r11-40]
movaps xmm8, XMMWORD PTR [r11-56]
movaps xmm9, XMMWORD PTR [r11-72]
movaps xmm10, XMMWORD PTR [r11-88]
movaps xmm11, XMMWORD PTR [r11-104]
movaps xmm12, XMMWORD PTR [r11-120]
movaps xmm14, XMMWORD PTR [rsp+32]
movaps xmm15, XMMWORD PTR [rsp+16]
mov rsp, r11
pop r15
pop r14
pop r13
pop r12
pop rdi
pop rsi
pop rbp
pop rbx
jmp rwz_cnv2_double_mainloop_asm_endp
rwz_div_fix_1:
dec rbx
add r11, rdx
jmp rwz_div_fix_1_ret
rwz_div_fix_2:
dec rdx
add r8, r9
jmp rwz_div_fix_2_ret
rwz_sqrt_fix_1:
movq r8, xmm3
movdqa xmm0, xmm5
psrldq xmm0, 8
dec r9
mov r11d, -1022
shl r11, 32
mov rax, r9
shr r9, 19
shr rax, 20
mov rdx, r9
sub rdx, rax
lea rdx, [rdx+r11+1]
add rax, r11
imul rdx, rax
sub rdx, r8
adc r9, 0
movq xmm5, r9
punpcklqdq xmm5, xmm0
jmp rwz_sqrt_fix_1_ret
rwz_sqrt_fix_2:
psrldq xmm3, 8
movq r11, xmm3
dec r8
mov ebx, -1022
shl rbx, 32
mov rax, r8
shr r8, 19
shr rax, 20
mov rdx, r8
sub rdx, rax
lea rdx, [rdx+rbx+1]
add rax, rbx
imul rdx, rax
sub rdx, r11
adc r8, 0
movq xmm0, r8
punpcklqdq xmm5, xmm0
jmp rwz_sqrt_fix_2_ret
rwz_cnv2_double_mainloop_asm_endp:

View file

@ -0,0 +1,186 @@
mov QWORD PTR [rsp+24], rbx
push rbp
push rsi
push rdi
push r12
push r13
push r14
push r15
sub rsp, 80
stmxcsr DWORD PTR [rsp]
mov DWORD PTR [rsp+4], 24448
ldmxcsr DWORD PTR [rsp+4]
mov rax, QWORD PTR [rcx+48]
mov r9, rcx
xor rax, QWORD PTR [rcx+16]
mov esi, 393216
mov r8, QWORD PTR [rcx+32]
mov r13d, -2147483647
xor r8, QWORD PTR [rcx]
mov r11, QWORD PTR [rcx+40]
mov r10, r8
mov rdx, QWORD PTR [rcx+56]
movq xmm4, rax
xor rdx, QWORD PTR [rcx+24]
xor r11, QWORD PTR [rcx+8]
mov rbx, QWORD PTR [rcx+224]
mov rax, QWORD PTR [r9+80]
xor rax, QWORD PTR [r9+64]
movq xmm0, rdx
mov rcx, QWORD PTR [rcx+88]
xor rcx, QWORD PTR [r9+72]
movq xmm3, QWORD PTR [r9+104]
movaps XMMWORD PTR [rsp+64], xmm6
movaps XMMWORD PTR [rsp+48], xmm7
movaps XMMWORD PTR [rsp+32], xmm8
and r10d, 2097136
movq xmm5, rax
xor eax, eax
mov QWORD PTR [rsp+16], rax
mov ax, 1023
shl rax, 52
movq xmm8, rax
mov r15, QWORD PTR [r9+96]
punpcklqdq xmm4, xmm0
movq xmm0, rcx
punpcklqdq xmm5, xmm0
movdqu xmm6, XMMWORD PTR [r10+rbx]
ALIGN(64)
rwz_main_loop:
lea rdx, QWORD PTR [r10+rbx]
mov ecx, r10d
mov eax, r10d
mov rdi, r15
xor ecx, 16
xor eax, 32
xor r10d, 48
movq xmm0, r11
movq xmm7, r8
punpcklqdq xmm7, xmm0
aesenc xmm6, xmm7
movq rbp, xmm6
mov r9, rbp
and r9d, 2097136
movdqu xmm0, XMMWORD PTR [rcx+rbx]
movdqu xmm1, XMMWORD PTR [rax+rbx]
movdqu xmm2, XMMWORD PTR [r10+rbx]
paddq xmm0, xmm5
paddq xmm1, xmm7
paddq xmm2, xmm4
movdqu XMMWORD PTR [rcx+rbx], xmm0
movdqu XMMWORD PTR [rax+rbx], xmm2
movdqu XMMWORD PTR [r10+rbx], xmm1
mov r10, r9
xor r10d, 32
movq rcx, xmm3
mov rax, rcx
shl rax, 32
xor rdi, rax
movdqa xmm0, xmm6
pxor xmm0, xmm4
movdqu XMMWORD PTR [rdx], xmm0
xor rdi, QWORD PTR [r9+rbx]
lea r14, QWORD PTR [r9+rbx]
mov r12, QWORD PTR [r14+8]
xor edx, edx
lea r9d, DWORD PTR [ecx+ecx]
add r9d, ebp
movdqa xmm0, xmm6
psrldq xmm0, 8
or r9d, r13d
movq rax, xmm0
div r9
xorps xmm3, xmm3
mov eax, eax
shl rdx, 32
add rdx, rax
lea r9, QWORD PTR [rdx+rbp]
mov r15, rdx
mov rax, r9
shr rax, 12
movq xmm0, rax
paddq xmm0, xmm8
sqrtsd xmm3, xmm0
psubq xmm3, XMMWORD PTR [rsp+16]
movq rdx, xmm3
test edx, 524287
je rwz_sqrt_fixup
psrlq xmm3, 19
rwz_sqrt_fixup_ret:
mov ecx, r10d
mov rax, rdi
mul rbp
movq xmm2, rdx
xor rdx, [rcx+rbx]
add r8, rdx
mov QWORD PTR [r14], r8
xor r8, rdi
mov edi, r8d
and edi, 2097136
movq xmm0, rax
xor rax, [rcx+rbx+8]
add r11, rax
mov QWORD PTR [r14+8], r11
punpcklqdq xmm2, xmm0
mov r9d, r10d
xor r9d, 48
xor r10d, 16
pxor xmm2, XMMWORD PTR [r9+rbx]
movdqu xmm0, XMMWORD PTR [r10+rbx]
paddq xmm0, xmm4
movdqu xmm1, XMMWORD PTR [rcx+rbx]
paddq xmm2, xmm5
paddq xmm1, xmm7
movdqa xmm5, xmm4
movdqu XMMWORD PTR [r9+rbx], xmm2
movdqa xmm4, xmm6
movdqu XMMWORD PTR [rcx+rbx], xmm0
movdqu XMMWORD PTR [r10+rbx], xmm1
movdqu xmm6, [rdi+rbx]
mov r10d, edi
xor r11, r12
dec rsi
jne rwz_main_loop
ldmxcsr DWORD PTR [rsp]
mov rbx, QWORD PTR [rsp+160]
movaps xmm6, XMMWORD PTR [rsp+64]
movaps xmm7, XMMWORD PTR [rsp+48]
movaps xmm8, XMMWORD PTR [rsp+32]
add rsp, 80
pop r15
pop r14
pop r13
pop r12
pop rdi
pop rsi
pop rbp
jmp cnv2_rwz_main_loop_endp
rwz_sqrt_fixup:
dec rdx
mov r13d, -1022
shl r13, 32
mov rax, rdx
shr rdx, 19
shr rax, 20
mov rcx, rdx
sub rcx, rax
add rax, r13
not r13
sub rcx, r13
mov r13d, -2147483647
imul rcx, rax
sub rcx, r9
adc rdx, 0
movq xmm3, rdx
jmp rwz_sqrt_fixup_ret
cnv2_rwz_main_loop_endp:

View file

@ -15,6 +15,8 @@
.global FN_PREFIX(cnv2_mainloop_ryzen_asm)
.global FN_PREFIX(cnv2_mainloop_bulldozer_asm)
.global FN_PREFIX(cnv2_double_mainloop_sandybridge_asm)
.global FN_PREFIX(cnv2_rwz_mainloop_asm)
.global FN_PREFIX(cnv2_rwz_double_mainloop_asm)
ALIGN(64)
FN_PREFIX(cnv2_mainloop_ivybridge_asm):
@ -52,3 +54,21 @@ FN_PREFIX(cnv2_double_mainloop_sandybridge_asm):
add rsp, 48
ret 0
mov eax, 3735929054
ALIGN(64)
FN_PREFIX(cnv2_rwz_mainloop_asm):
sub rsp, 48
mov rcx, rdi
#include "cn2/cnv2_rwz_main_loop.inc"
add rsp, 48
ret 0
mov eax, 3735929054
ALIGN(64)
FN_PREFIX(cnv2_rwz_double_mainloop_asm):
sub rsp, 48
mov rcx, rdi
#include "cn2/cnv2_rwz_double_main_loop.inc"
add rsp, 48
ret 0
mov eax, 3735929054

View file

@ -3,6 +3,8 @@ PUBLIC cnv2_mainloop_ivybridge_asm
PUBLIC cnv2_mainloop_ryzen_asm
PUBLIC cnv2_mainloop_bulldozer_asm
PUBLIC cnv2_double_mainloop_sandybridge_asm
PUBLIC cnv2_rwz_mainloop_asm
PUBLIC cnv2_rwz_double_mainloop_asm
ALIGN(64)
cnv2_mainloop_ivybridge_asm PROC
@ -32,5 +34,19 @@ cnv2_double_mainloop_sandybridge_asm PROC
mov eax, 3735929054
cnv2_double_mainloop_sandybridge_asm ENDP
ALIGN(64)
cnv2_rwz_mainloop_asm PROC
INCLUDE cn2/cnv2_rwz_main_loop.inc
ret 0
mov eax, 3735929054
cnv2_rwz_mainloop_asm ENDP
ALIGN(64)
cnv2_rwz_double_mainloop_asm PROC
INCLUDE cn2/cnv2_rwz_double_main_loop.inc
ret 0
mov eax, 3735929054
cnv2_rwz_double_mainloop_asm ENDP
_TEXT_CNV2_MAINLOOP ENDS
END

View file

@ -0,0 +1,410 @@
mov rax, rsp
push rbx
push rbp
push rsi
push rdi
push r12
push r13
push r14
push r15
sub rsp, 184
stmxcsr DWORD PTR [rsp+272]
mov DWORD PTR [rsp+276], 24448
ldmxcsr DWORD PTR [rsp+276]
mov r13, QWORD PTR [rcx+224]
mov r9, rdx
mov r10, QWORD PTR [rcx+32]
mov r8, rcx
xor r10, QWORD PTR [rcx]
mov r14d, 393216
mov r11, QWORD PTR [rcx+40]
xor r11, QWORD PTR [rcx+8]
mov rsi, QWORD PTR [rdx+224]
mov rdx, QWORD PTR [rcx+56]
xor rdx, QWORD PTR [rcx+24]
mov rdi, QWORD PTR [r9+32]
xor rdi, QWORD PTR [r9]
mov rbp, QWORD PTR [r9+40]
xor rbp, QWORD PTR [r9+8]
movd xmm0, rdx
movaps XMMWORD PTR [rax-88], xmm6
movaps XMMWORD PTR [rax-104], xmm7
movaps XMMWORD PTR [rax-120], xmm8
movaps XMMWORD PTR [rsp+112], xmm9
movaps XMMWORD PTR [rsp+96], xmm10
movaps XMMWORD PTR [rsp+80], xmm11
movaps XMMWORD PTR [rsp+64], xmm12
movaps XMMWORD PTR [rsp+48], xmm13
movaps XMMWORD PTR [rsp+32], xmm14
movaps XMMWORD PTR [rsp+16], xmm15
mov rdx, r10
movd xmm4, QWORD PTR [r8+96]
and edx, 2097136
mov rax, QWORD PTR [rcx+48]
xorps xmm13, xmm13
xor rax, QWORD PTR [rcx+16]
mov rcx, QWORD PTR [rcx+88]
xor rcx, QWORD PTR [r8+72]
movd xmm5, QWORD PTR [r8+104]
movd xmm7, rax
mov eax, 1
shl rax, 52
movd xmm14, rax
punpcklqdq xmm14, xmm14
mov eax, 1023
shl rax, 52
movd xmm12, rax
punpcklqdq xmm12, xmm12
mov rax, QWORD PTR [r8+80]
xor rax, QWORD PTR [r8+64]
punpcklqdq xmm7, xmm0
movd xmm0, rcx
mov rcx, QWORD PTR [r9+56]
xor rcx, QWORD PTR [r9+24]
movd xmm3, rax
mov rax, QWORD PTR [r9+48]
xor rax, QWORD PTR [r9+16]
punpcklqdq xmm3, xmm0
movd xmm0, rcx
mov QWORD PTR [rsp], r13
mov rcx, QWORD PTR [r9+88]
xor rcx, QWORD PTR [r9+72]
movd xmm6, rax
mov rax, QWORD PTR [r9+80]
xor rax, QWORD PTR [r9+64]
punpcklqdq xmm6, xmm0
movd xmm0, rcx
mov QWORD PTR [rsp+256], r10
mov rcx, rdi
mov QWORD PTR [rsp+264], r11
movd xmm8, rax
and ecx, 2097136
punpcklqdq xmm8, xmm0
movd xmm0, QWORD PTR [r9+96]
punpcklqdq xmm4, xmm0
movd xmm0, QWORD PTR [r9+104]
lea r8, QWORD PTR [rcx+rsi]
movdqu xmm11, XMMWORD PTR [r8]
punpcklqdq xmm5, xmm0
lea r9, QWORD PTR [rdx+r13]
movdqu xmm15, XMMWORD PTR [r9]
ALIGN(64)
rwz_main_loop_double:
movdqu xmm9, xmm15
mov eax, edx
mov ebx, edx
xor eax, 16
xor ebx, 32
xor edx, 48
movd xmm0, r11
movd xmm2, r10
punpcklqdq xmm2, xmm0
aesenc xmm9, xmm2
movdqu xmm0, XMMWORD PTR [rdx+r13]
movdqu xmm1, XMMWORD PTR [rbx+r13]
paddq xmm0, xmm7
paddq xmm1, xmm2
movdqu XMMWORD PTR [rbx+r13], xmm0
movdqu xmm0, XMMWORD PTR [rax+r13]
movdqu XMMWORD PTR [rdx+r13], xmm1
paddq xmm0, xmm3
movdqu XMMWORD PTR [rax+r13], xmm0
movd r11, xmm9
mov edx, r11d
and edx, 2097136
movdqa xmm0, xmm9
pxor xmm0, xmm7
movdqu XMMWORD PTR [r9], xmm0
lea rbx, QWORD PTR [rdx+r13]
mov r10, QWORD PTR [rdx+r13]
movdqu xmm10, xmm11
movd xmm0, rbp
movd xmm11, rdi
punpcklqdq xmm11, xmm0
aesenc xmm10, xmm11
mov eax, ecx
mov r12d, ecx
xor eax, 16
xor r12d, 32
xor ecx, 48
movdqu xmm0, XMMWORD PTR [rcx+rsi]
paddq xmm0, xmm6
movdqu xmm1, XMMWORD PTR [r12+rsi]
movdqu XMMWORD PTR [r12+rsi], xmm0
paddq xmm1, xmm11
movdqu xmm0, XMMWORD PTR [rax+rsi]
movdqu XMMWORD PTR [rcx+rsi], xmm1
paddq xmm0, xmm8
movdqu XMMWORD PTR [rax+rsi], xmm0
movd rcx, xmm10
and ecx, 2097136
movdqa xmm0, xmm10
pxor xmm0, xmm6
movdqu XMMWORD PTR [r8], xmm0
mov r12, QWORD PTR [rcx+rsi]
mov r9, QWORD PTR [rbx+8]
xor edx, 16
mov r8d, edx
mov r15d, edx
movd rdx, xmm5
shl rdx, 32
movd rax, xmm4
xor rdx, rax
xor r10, rdx
mov rax, r10
mul r11
mov r11d, r8d
xor r11d, 48
movd xmm0, rdx
xor rdx, [r11+r13]
movd xmm1, rax
xor rax, [r11+r13+8]
punpcklqdq xmm0, xmm1
pxor xmm0, XMMWORD PTR [r8+r13]
movdqu xmm1, XMMWORD PTR [r11+r13]
paddq xmm0, xmm3
paddq xmm1, xmm2
movdqu XMMWORD PTR [r8+r13], xmm0
xor r8d, 32
movdqu xmm0, XMMWORD PTR [r8+r13]
movdqu XMMWORD PTR [r8+r13], xmm1
paddq xmm0, xmm7
movdqu XMMWORD PTR [r11+r13], xmm0
mov r11, QWORD PTR [rsp+256]
add r11, rdx
mov rdx, QWORD PTR [rsp+264]
add rdx, rax
mov QWORD PTR [rbx], r11
xor r11, r10
mov QWORD PTR [rbx+8], rdx
xor rdx, r9
mov QWORD PTR [rsp+256], r11
and r11d, 2097136
mov QWORD PTR [rsp+264], rdx
mov QWORD PTR [rsp+8], r11
lea r15, QWORD PTR [r11+r13]
movdqu xmm15, XMMWORD PTR [r11+r13]
lea r13, QWORD PTR [rsi+rcx]
movdqa xmm0, xmm5
psrldq xmm0, 8
movaps xmm2, xmm13
movd r10, xmm0
psllq xmm5, 1
shl r10, 32
movdqa xmm0, xmm9
psrldq xmm0, 8
movdqa xmm1, xmm10
movd r11, xmm0
psrldq xmm1, 8
movd r8, xmm1
psrldq xmm4, 8
movaps xmm0, xmm13
movd rax, xmm4
xor r10, rax
movaps xmm1, xmm13
xor r10, r12
lea rax, QWORD PTR [r11+1]
shr rax, 1
movdqa xmm3, xmm9
punpcklqdq xmm3, xmm10
paddq xmm5, xmm3
movd rdx, xmm5
psrldq xmm5, 8
cvtsi2sd xmm2, rax
or edx, -2147483647
lea rax, QWORD PTR [r8+1]
shr rax, 1
movd r9, xmm5
cvtsi2sd xmm0, rax
or r9d, -2147483647
cvtsi2sd xmm1, rdx
unpcklpd xmm2, xmm0
movaps xmm0, xmm13
cvtsi2sd xmm0, r9
unpcklpd xmm1, xmm0
divpd xmm2, xmm1
paddq xmm2, xmm14
cvttsd2si rax, xmm2
psrldq xmm2, 8
mov rbx, rax
imul rax, rdx
sub r11, rax
js rwz_div_fix_1
rwz_div_fix_1_ret:
cvttsd2si rdx, xmm2
mov rax, rdx
imul rax, r9
movd xmm2, r11d
movd xmm4, ebx
sub r8, rax
js rwz_div_fix_2
rwz_div_fix_2_ret:
movd xmm1, r8d
movd xmm0, edx
punpckldq xmm2, xmm1
punpckldq xmm4, xmm0
punpckldq xmm4, xmm2
paddq xmm3, xmm4
movdqa xmm0, xmm3
psrlq xmm0, 12
paddq xmm0, xmm12
sqrtpd xmm1, xmm0
movd r9, xmm1
movdqa xmm5, xmm1
psrlq xmm5, 19
test r9, 524287
je rwz_sqrt_fix_1
rwz_sqrt_fix_1_ret:
movd r9, xmm10
psrldq xmm1, 8
movd r8, xmm1
test r8, 524287
je rwz_sqrt_fix_2
rwz_sqrt_fix_2_ret:
mov r12d, ecx
mov r8d, ecx
xor r12d, 16
xor r8d, 32
xor ecx, 48
mov rax, r10
mul r9
movd xmm0, rax
movd xmm3, rdx
punpcklqdq xmm3, xmm0
movdqu xmm0, XMMWORD PTR [r12+rsi]
pxor xmm0, xmm3
movdqu xmm1, XMMWORD PTR [r8+rsi]
xor rdx, [r8+rsi]
xor rax, [r8+rsi+8]
movdqu xmm3, XMMWORD PTR [rcx+rsi]
paddq xmm3, xmm6
paddq xmm1, xmm11
paddq xmm0, xmm8
movdqu XMMWORD PTR [r8+rsi], xmm3
movdqu XMMWORD PTR [rcx+rsi], xmm1
movdqu XMMWORD PTR [r12+rsi], xmm0
add rdi, rdx
mov QWORD PTR [r13], rdi
xor rdi, r10
mov ecx, edi
and ecx, 2097136
lea r8, QWORD PTR [rcx+rsi]
mov rdx, QWORD PTR [r13+8]
add rbp, rax
mov QWORD PTR [r13+8], rbp
movdqu xmm11, XMMWORD PTR [rcx+rsi]
xor rbp, rdx
mov r13, QWORD PTR [rsp]
movdqa xmm3, xmm7
mov rdx, QWORD PTR [rsp+8]
movdqa xmm8, xmm6
mov r10, QWORD PTR [rsp+256]
movdqa xmm7, xmm9
mov r11, QWORD PTR [rsp+264]
movdqa xmm6, xmm10
mov r9, r15
dec r14d
jne rwz_main_loop_double
ldmxcsr DWORD PTR [rsp+272]
movaps xmm13, XMMWORD PTR [rsp+48]
lea r11, QWORD PTR [rsp+184]
movaps xmm6, XMMWORD PTR [r11-24]
movaps xmm7, XMMWORD PTR [r11-40]
movaps xmm8, XMMWORD PTR [r11-56]
movaps xmm9, XMMWORD PTR [r11-72]
movaps xmm10, XMMWORD PTR [r11-88]
movaps xmm11, XMMWORD PTR [r11-104]
movaps xmm12, XMMWORD PTR [r11-120]
movaps xmm14, XMMWORD PTR [rsp+32]
movaps xmm15, XMMWORD PTR [rsp+16]
mov rsp, r11
pop r15
pop r14
pop r13
pop r12
pop rdi
pop rsi
pop rbp
pop rbx
jmp rwz_cnv2_double_mainloop_asm_endp
rwz_div_fix_1:
dec rbx
add r11, rdx
jmp rwz_div_fix_1_ret
rwz_div_fix_2:
dec rdx
add r8, r9
jmp rwz_div_fix_2_ret
rwz_sqrt_fix_1:
movd r8, xmm3
movdqa xmm0, xmm5
psrldq xmm0, 8
dec r9
mov r11d, -1022
shl r11, 32
mov rax, r9
shr r9, 19
shr rax, 20
mov rdx, r9
sub rdx, rax
lea rdx, [rdx+r11+1]
add rax, r11
imul rdx, rax
sub rdx, r8
adc r9, 0
movd xmm5, r9
punpcklqdq xmm5, xmm0
jmp rwz_sqrt_fix_1_ret
rwz_sqrt_fix_2:
psrldq xmm3, 8
movd r11, xmm3
dec r8
mov ebx, -1022
shl rbx, 32
mov rax, r8
shr r8, 19
shr rax, 20
mov rdx, r8
sub rdx, rax
lea rdx, [rdx+rbx+1]
add rax, rbx
imul rdx, rax
sub rdx, r11
adc r8, 0
movd xmm0, r8
punpcklqdq xmm5, xmm0
jmp rwz_sqrt_fix_2_ret
rwz_cnv2_double_mainloop_asm_endp:

View file

@ -0,0 +1,186 @@
mov QWORD PTR [rsp+24], rbx
push rbp
push rsi
push rdi
push r12
push r13
push r14
push r15
sub rsp, 80
stmxcsr DWORD PTR [rsp]
mov DWORD PTR [rsp+4], 24448
ldmxcsr DWORD PTR [rsp+4]
mov rax, QWORD PTR [rcx+48]
mov r9, rcx
xor rax, QWORD PTR [rcx+16]
mov esi, 393216
mov r8, QWORD PTR [rcx+32]
mov r13d, -2147483647
xor r8, QWORD PTR [rcx]
mov r11, QWORD PTR [rcx+40]
mov r10, r8
mov rdx, QWORD PTR [rcx+56]
movd xmm4, rax
xor rdx, QWORD PTR [rcx+24]
xor r11, QWORD PTR [rcx+8]
mov rbx, QWORD PTR [rcx+224]
mov rax, QWORD PTR [r9+80]
xor rax, QWORD PTR [r9+64]
movd xmm0, rdx
mov rcx, QWORD PTR [rcx+88]
xor rcx, QWORD PTR [r9+72]
movd xmm3, QWORD PTR [r9+104]
movaps XMMWORD PTR [rsp+64], xmm6
movaps XMMWORD PTR [rsp+48], xmm7
movaps XMMWORD PTR [rsp+32], xmm8
and r10d, 2097136
movd xmm5, rax
xor eax, eax
mov QWORD PTR [rsp+16], rax
mov ax, 1023
shl rax, 52
movd xmm8, rax
mov r15, QWORD PTR [r9+96]
punpcklqdq xmm4, xmm0
movd xmm0, rcx
punpcklqdq xmm5, xmm0
movdqu xmm6, XMMWORD PTR [r10+rbx]
ALIGN(64)
rwz_main_loop:
lea rdx, QWORD PTR [r10+rbx]
mov ecx, r10d
mov eax, r10d
mov rdi, r15
xor ecx, 16
xor eax, 32
xor r10d, 48
movd xmm0, r11
movd xmm7, r8
punpcklqdq xmm7, xmm0
aesenc xmm6, xmm7
movd rbp, xmm6
mov r9, rbp
and r9d, 2097136
movdqu xmm0, XMMWORD PTR [rcx+rbx]
movdqu xmm1, XMMWORD PTR [rax+rbx]
movdqu xmm2, XMMWORD PTR [r10+rbx]
paddq xmm0, xmm5
paddq xmm1, xmm7
paddq xmm2, xmm4
movdqu XMMWORD PTR [rcx+rbx], xmm0
movdqu XMMWORD PTR [rax+rbx], xmm2
movdqu XMMWORD PTR [r10+rbx], xmm1
mov r10, r9
xor r10d, 32
movd rcx, xmm3
mov rax, rcx
shl rax, 32
xor rdi, rax
movdqa xmm0, xmm6
pxor xmm0, xmm4
movdqu XMMWORD PTR [rdx], xmm0
xor rdi, QWORD PTR [r9+rbx]
lea r14, QWORD PTR [r9+rbx]
mov r12, QWORD PTR [r14+8]
xor edx, edx
lea r9d, DWORD PTR [ecx+ecx]
add r9d, ebp
movdqa xmm0, xmm6
psrldq xmm0, 8
or r9d, r13d
movd rax, xmm0
div r9
xorps xmm3, xmm3
mov eax, eax
shl rdx, 32
add rdx, rax
lea r9, QWORD PTR [rdx+rbp]
mov r15, rdx
mov rax, r9
shr rax, 12
movd xmm0, rax
paddq xmm0, xmm8
sqrtsd xmm3, xmm0
psubq xmm3, XMMWORD PTR [rsp+16]
movd rdx, xmm3
test edx, 524287
je rwz_sqrt_fixup
psrlq xmm3, 19
rwz_sqrt_fixup_ret:
mov ecx, r10d
mov rax, rdi
mul rbp
movd xmm2, rdx
xor rdx, [rcx+rbx]
add r8, rdx
mov QWORD PTR [r14], r8
xor r8, rdi
mov edi, r8d
and edi, 2097136
movd xmm0, rax
xor rax, [rcx+rbx+8]
add r11, rax
mov QWORD PTR [r14+8], r11
punpcklqdq xmm2, xmm0
mov r9d, r10d
xor r9d, 48
xor r10d, 16
pxor xmm2, XMMWORD PTR [r9+rbx]
movdqu xmm0, XMMWORD PTR [r10+rbx]
paddq xmm0, xmm4
movdqu xmm1, XMMWORD PTR [rcx+rbx]
paddq xmm2, xmm5
paddq xmm1, xmm7
movdqa xmm5, xmm4
movdqu XMMWORD PTR [r9+rbx], xmm2
movdqa xmm4, xmm6
movdqu XMMWORD PTR [rcx+rbx], xmm0
movdqu XMMWORD PTR [r10+rbx], xmm1
movdqu xmm6, [rdi+rbx]
mov r10d, edi
xor r11, r12
dec rsi
jne rwz_main_loop
ldmxcsr DWORD PTR [rsp]
mov rbx, QWORD PTR [rsp+160]
movaps xmm6, XMMWORD PTR [rsp+64]
movaps xmm7, XMMWORD PTR [rsp+48]
movaps xmm8, XMMWORD PTR [rsp+32]
add rsp, 80
pop r15
pop r14
pop r13
pop r12
pop rdi
pop rsi
pop rbp
jmp cnv2_rwz_main_loop_endp
rwz_sqrt_fixup:
dec rdx
mov r13d, -1022
shl r13, 32
mov rax, rdx
shr rdx, 19
shr rax, 20
mov rcx, rdx
sub rcx, rax
add rax, r13
not r13
sub rcx, r13
mov r13d, -2147483647
imul rcx, rax
sub rcx, r9
adc rdx, 0
movd xmm3, rdx
jmp rwz_sqrt_fixup_ret
cnv2_rwz_main_loop_endp:

View file

@ -5,6 +5,8 @@
.global cnv2_mainloop_ryzen_asm
.global cnv2_mainloop_bulldozer_asm
.global cnv2_double_mainloop_sandybridge_asm
.global FN_PREFIX(cnv2_rwz_mainloop_asm)
.global FN_PREFIX(cnv2_rwz_double_mainloop_asm)
ALIGN(64)
cnv2_mainloop_ivybridge_asm:
@ -29,3 +31,21 @@ cnv2_double_mainloop_sandybridge_asm:
#include "../cn2/cnv2_double_main_loop_sandybridge.inc"
ret 0
mov eax, 3735929054
ALIGN(64)
FN_PREFIX(cnv2_rwz_mainloop_asm):
sub rsp, 48
mov rcx, rdi
#include "cn2/cnv2_rwz_main_loop.inc"
add rsp, 48
ret 0
mov eax, 3735929054
ALIGN(64)
FN_PREFIX(cnv2_rwz_double_mainloop_asm):
sub rsp, 48
mov rcx, rdi
#include "cn2/cnv2_rwz_double_main_loop.inc"
add rsp, 48
ret 0
mov eax, 3735929054

View file

@ -3,6 +3,8 @@ PUBLIC cnv2_mainloop_ivybridge_asm
PUBLIC cnv2_mainloop_ryzen_asm
PUBLIC cnv2_mainloop_bulldozer_asm
PUBLIC cnv2_double_mainloop_sandybridge_asm
PUBLIC cnv2_rwz_mainloop_asm
PUBLIC cnv2_rwz_double_mainloop_asm
ALIGN 64
cnv2_mainloop_ivybridge_asm PROC
@ -32,5 +34,19 @@ cnv2_double_mainloop_sandybridge_asm PROC
mov eax, 3735929054
cnv2_double_mainloop_sandybridge_asm ENDP
ALIGN(64)
cnv2_rwz_mainloop_asm PROC
INCLUDE cn2/cnv2_rwz_main_loop.inc
ret 0
mov eax, 3735929054
cnv2_rwz_mainloop_asm ENDP
ALIGN(64)
cnv2_rwz_double_mainloop_asm PROC
INCLUDE cn2/cnv2_rwz_double_main_loop.inc
ret 0
mov eax, 3735929054
cnv2_rwz_double_mainloop_asm ENDP
_TEXT_CNV2_MAINLOOP ENDS
END

View file

@ -180,6 +180,8 @@ xmrig::CpuThread::cn_hash_fun xmrig::CpuThread::fn(Algo algorithm, AlgoVariant a
add_asm_func<CRYPTONIGHT_PICO, VARIANT_TRTL>(asm_func_map);
# endif
add_asm_func<CRYPTONIGHT, VARIANT_RWZ>(asm_func_map);
asm_func_map_initialized = true;
}
@ -323,6 +325,17 @@ xmrig::CpuThread::cn_hash_fun xmrig::CpuThread::fn(Algo algorithm, AlgoVariant a
cryptonight_quad_hash<CRYPTONIGHT, true, VARIANT_4>,
cryptonight_penta_hash<CRYPTONIGHT, true, VARIANT_4>,
cryptonight_single_hash<CRYPTONIGHT, false, VARIANT_RWZ>,
cryptonight_double_hash<CRYPTONIGHT, false, VARIANT_RWZ>,
cryptonight_single_hash<CRYPTONIGHT, true, VARIANT_RWZ>,
cryptonight_double_hash<CRYPTONIGHT, true, VARIANT_RWZ>,
cryptonight_triple_hash<CRYPTONIGHT, false, VARIANT_RWZ>,
cryptonight_quad_hash<CRYPTONIGHT, false, VARIANT_RWZ>,
cryptonight_penta_hash<CRYPTONIGHT, false, VARIANT_RWZ>,
cryptonight_triple_hash<CRYPTONIGHT, true, VARIANT_RWZ>,
cryptonight_quad_hash<CRYPTONIGHT, true, VARIANT_RWZ>,
cryptonight_penta_hash<CRYPTONIGHT, true, VARIANT_RWZ>,
# ifndef XMRIG_NO_AEON
cryptonight_single_hash<CRYPTONIGHT_LITE, false, VARIANT_0>,
cryptonight_double_hash<CRYPTONIGHT_LITE, false, VARIANT_0>,
@ -358,6 +371,7 @@ xmrig::CpuThread::cn_hash_fun xmrig::CpuThread::fn(Algo algorithm, AlgoVariant a
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_GPU
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_WOW
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_4
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_RWZ
# else
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_0
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_1
@ -373,6 +387,7 @@ xmrig::CpuThread::cn_hash_fun xmrig::CpuThread::fn(Algo algorithm, AlgoVariant a
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_GPU
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_WOW
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_4
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_RWZ
# endif
# ifndef XMRIG_NO_SUMO
@ -422,6 +437,7 @@ xmrig::CpuThread::cn_hash_fun xmrig::CpuThread::fn(Algo algorithm, AlgoVariant a
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_GPU
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_WOW
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_4
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_RWZ
# else
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_0
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_1
@ -437,6 +453,7 @@ xmrig::CpuThread::cn_hash_fun xmrig::CpuThread::fn(Algo algorithm, AlgoVariant a
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_GPU
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_WOW
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_4
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_RWZ
# endif
# ifndef XMRIG_NO_CN_PICO
@ -465,6 +482,7 @@ xmrig::CpuThread::cn_hash_fun xmrig::CpuThread::fn(Algo algorithm, AlgoVariant a
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_GPU
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_WOW
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_4
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_RWZ
# else
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_0
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_1
@ -480,6 +498,7 @@ xmrig::CpuThread::cn_hash_fun xmrig::CpuThread::fn(Algo algorithm, AlgoVariant a
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_GPU
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_WOW
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_4
nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, // VARIANT_RWZ
# endif
};

View file

@ -64,7 +64,8 @@ bool MultiWorker<N>::selfTest()
verify(VARIANT_RTO, test_output_rto) &&
verify(VARIANT_HALF, test_output_half) &&
verify2(VARIANT_WOW, test_output_wow) &&
verify2(VARIANT_4, test_output_r);
verify2(VARIANT_4, test_output_r) &&
verify(VARIANT_RWZ, test_output_rwz);
# ifndef XMRIG_NO_CN_GPU
if (!rc || N > 1) {