RISC-V: auto-detect and use vector code for all RandomX AES functions
What changed, and why it matters
This commit is a performance optimization for XMRig on RISC-V processors. It adds new code paths that use RISC-V vector instructions (RVV) to speed up the AES-based hashing routines used by the RandomX mining algorithm. There is no indication this change fixes a security bug or introduces a known vulnerability; it is purely about making mining faster on RISC-V hardware.
No security action required; treat as a normal performance/architecture-specific feature update. Standard regression testing on RISC-V builds is sufficient.
Security signals we found
No strong security signals were identified.
Evidence from the diff
The change refactors RandomX AES helper functions (hashAes1Rx4, fillAes1Rx4, fillAes4Rx4, hashAndFillAes1Rx4) so that on RISC-V builds they dispatch to new RVV vectorized implementations when the CPU reports vector support. It removes the compile-time XMRIG_RVV_ENABLED macro in favor of runtime detection via Cpu::info()->hasRISCV_Vector(), moves the vectorized soft-AES helper into a new translation unit (aes_hash_rv64_vector.cpp), and makes a small JIT compiler optimization for RISC-V immediate materialization. No security-relevant bug fixes or memory-safety issues are visible in the diff.
Changed components
src/crypto/randomx/aes_hash.cppsrc/crypto/randomx/aes_hash_rv64_vector.cppsrc/crypto/randomx/aes_hash_rv64_vector.hppsrc/crypto/randomx/jit_compiler_rv64_vector.cppsrc/crypto/randomx/soft_aes.hcmake/cpu.cmakeInspect captured patch +463 / −199
diff --git a/cmake/cpu.cmake b/cmake/cpu.cmake
index e9f1cf8..380db8a 100644
--- a/cmake/cpu.cmake
+++ b/cmake/cpu.cmake
@@ -109,7 +109,6 @@ if (XMRIG_RISCV)
if (ARCH STREQUAL "native")
if (RVARCH_V)
set(RVARCH "${RVARCH}v")
- add_definitions(-DXMRIG_RVV_ENABLED)
endif()
if (RVARCH_ZICBOP)
set(RVARCH "${RVARCH}_zicbop")
diff --git a/src/crypto/randomx/aes_hash.cpp b/src/crypto/randomx/aes_hash.cpp
index 04b813b..9efde1e 100644
--- a/src/crypto/randomx/aes_hash.cpp
+++ b/src/crypto/randomx/aes_hash.cpp
@@ -38,6 +38,11 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "crypto/randomx/common.hpp"
#include "crypto/rx/Profiler.h"
+#ifdef XMRIG_RISCV
+#include "backend/cpu/Cpu.h"
+#include "crypto/randomx/aes_hash_rv64_vector.hpp"
+#endif
+
#define AES_HASH_1R_STATE0 0xd7983aad, 0xcc82db47, 0x9fa856de, 0x92b52c0d
#define AES_HASH_1R_STATE1 0xace78057, 0xf59e125a, 0x15c7b798, 0x338d996e
#define AES_HASH_1R_STATE2 0xe8a07ce4, 0x5079506b, 0xae62c7d0, 0x6a770017
@@ -59,7 +64,15 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Hashing throughput: >20 GiB/s per CPU core with hardware AES
*/
template<int softAes>
-void hashAes1Rx4(const void *input, size_t inputSize, void *hash) {
+void hashAes1Rx4(const void *input, size_t inputSize, void *hash)
+{
+#ifdef XMRIG_RISCV
+ if (xmrig::Cpu::info()->hasRISCV_Vector()) {
+ hashAes1Rx4_RVV<softAes>(input, inputSize, hash);
+ return;
+ }
+#endif
+
const uint8_t* inptr = (uint8_t*)input;
const uint8_t* inputEnd = inptr + inputSize;
@@ -127,7 +140,15 @@ template void hashAes1Rx4<true>(const void *input, size_t inputSize, void *hash)
calls to this function.
*/
template<int softAes>
-void fillAes1Rx4(void *state, size_t outputSize, void *buffer) {
+void fillAes1Rx4(void *state, size_t outputSize, void *buffer)
+{
+#ifdef XMRIG_RISCV
+ if (xmrig::Cpu::info()->hasRISCV_Vector()) {
+ fillAes1Rx4_RVV<softAes>(state, outputSize, buffer);
+ return;
+ }
+#endif
+
const uint8_t* outptr = (uint8_t*)buffer;
const uint8_t* outputEnd = outptr + outputSize;
@@ -171,7 +192,15 @@ static constexpr randomx::Instruction inst{ 0xFF, 7, 7, 0xFF, 0xFFFFFFFFU };
alignas(16) static const randomx::Instruction inst_mask[2] = { inst, inst };
template<int softAes>
-void fillAes4Rx4(void *state, size_t outputSize, void *buffer) {
+void fillAes4Rx4(void *state, size_t outputSize, void *buffer)
+{
+#ifdef XMRIG_RISCV
+ if (xmrig::Cpu::info()->hasRISCV_Vector()) {
+ fillAes4Rx4_RVV<softAes>(state, outputSize, buffer);
+ return;
+ }
+#endif
+
const uint8_t* outptr = (uint8_t*)buffer;
const uint8_t* outputEnd = outptr + outputSize;
@@ -235,134 +264,17 @@ void fillAes4Rx4(void *state, size_t outputSize, void *buffer) {
template void fillAes4Rx4<true>(void *state, size_t outputSize, void *buffer);
template void fillAes4Rx4<false>(void *state, size_t outputSize, void *buffer);
-#if defined(XMRIG_RISCV) && defined(XMRIG_RVV_ENABLED)
-static constexpr uint32_t AES_HASH_1R_STATE02[8] = { 0x92b52c0d, 0x9fa856de, 0xcc82db47, 0xd7983aad, 0x6a770017, 0xae62c7d0, 0x5079506b, 0xe8a07ce4 };
-static constexpr uint32_t AES_HASH_1R_STATE13[8] = { 0x338d996e, 0x15c7b798, 0xf59e125a, 0xace78057, 0x630a240c, 0x07ad828d, 0x79a10005, 0x7e994948 };
-
-static constexpr uint32_t AES_GEN_1R_KEY02[8] = { 0x6daca553, 0x62716609, 0xdbb5552b, 0xb4f44917, 0x3f1262f1, 0x9f947ec6, 0xf4c0794f, 0x3e20e345 };
-static constexpr uint32_t AES_GEN_1R_KEY13[8] = { 0x6d7caf07, 0x846a710d, 0x1725d378, 0x0da1dc4e, 0x6aef8135, 0xb1ba317c, 0x16314c88, 0x49169154 };
-
-static constexpr uint32_t AES_HASH_1R_XKEY00[8] = { 0xf6fa8389, 0x8b24949f, 0x90dc56bf, 0x06890201, 0xf6fa8389, 0x8b24949f, 0x90dc56bf, 0x06890201 };
-static constexpr uint32_t AES_HASH_1R_XKEY11[8] = { 0x61b263d1, 0x51f4e03c, 0xee1043c6, 0xed18f99b, 0x61b263d1, 0x51f4e03c, 0xee1043c6, 0xed18f99b };
-
-static constexpr uint32_t AES_HASH_STRIDE[8] = { 0, 4, 8, 12, 32, 36, 40, 44 };
-
template<int softAes, int unroll>
-void hashAndFillAes1Rx4(void *scratchpad, size_t scratchpadSize, void *hash, void* fill_state) {
+void hashAndFillAes1Rx4(void *scratchpad, size_t scratchpadSize, void *hash, void* fill_state)
+{
PROFILE_SCOPE(RandomX_AES);
- uint8_t* scratchpadPtr = (uint8_t*)scratchpad;
- const uint8_t* scratchpadEnd = scratchpadPtr + scratchpadSize;
-
- vuint32m1_t hash_state02 = __riscv_vle32_v_u32m1(AES_HASH_1R_STATE02, 8);
- vuint32m1_t hash_state13 = __riscv_vle32_v_u32m1(AES_HASH_1R_STATE13, 8);
-
- const vuint32m1_t key02 = __riscv_vle32_v_u32m1(AES_GEN_1R_KEY02, 8);
- const vuint32m1_t key13 = __riscv_vle32_v_u32m1(AES_GEN_1R_KEY13, 8);
-
- const vuint32m1_t stride = __riscv_vle32_v_u32m1(AES_HASH_STRIDE, 8);
-
- vuint32m1_t fill_state02 = __riscv_vluxei32_v_u32m1((uint32_t*)fill_state + 0, stride, 8);
- vuint32m1_t fill_state13 = __riscv_vluxei32_v_u32m1((uint32_t*)fill_state + 4, stride, 8);
-
- const vuint8m1_t lutenc_index0 = __riscv_vle8_v_u8m1(lutEncIndex[0], 32);
- const vuint8m1_t lutenc_index1 = __riscv_vle8_v_u8m1(lutEncIndex[1], 32);
- const vuint8m1_t lutenc_index2 = __riscv_vle8_v_u8m1(lutEncIndex[2], 32);
- const vuint8m1_t lutenc_index3 = __riscv_vle8_v_u8m1(lutEncIndex[3], 32);
-
- const vuint8m1_t& lutdec_index0 = lutenc_index0;
- const vuint8m1_t lutdec_index1 = __riscv_vle8_v_u8m1(lutDecIndex[1], 32);
- const vuint8m1_t& lutdec_index2 = lutenc_index2;
- const vuint8m1_t lutdec_index3 = __riscv_vle8_v_u8m1(lutDecIndex[3], 32);
-
- //process 64 bytes at a time in 4 lanes
- while (scratchpadPtr < scratchpadEnd) {
-#define HASH_STATE(k) \
- hash_state02 = softaes_vector_double(hash_state02, __riscv_vluxei32_v_u32m1((uint32_t*)scratchpadPtr + k * 16 + 0, stride, 8), lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3); \
- hash_state13 = softaes_vector_double(hash_state13, __riscv_vluxei32_v_u32m1((uint32_t*)scratchpadPtr + k * 16 + 4, stride, 8), lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
-
-#define FILL_STATE(k) \
- fill_state02 = softaes_vector_double(fill_state02, key02, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3); \
- fill_state13 = softaes_vector_double(fill_state13, key13, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3); \
- __riscv_vsuxei32_v_u32m1((uint32_t*)scratchpadPtr + k * 16 + 0, stride, fill_state02, 8); \
- __riscv_vsuxei32_v_u32m1((uint32_t*)scratchpadPtr + k * 16 + 4, stride, fill_state13, 8);
-
- switch (softAes) {
- case 0:
- HASH_STATE(0);
- HASH_STATE(1);
-
- FILL_STATE(0);
- FILL_STATE(1);
-
- scratchpadPtr += 128;
- break;
-
- default:
- switch (unroll) {
- case 4:
- HASH_STATE(0);
- FILL_STATE(0);
-
- HASH_STATE(1);
- FILL_STATE(1);
-
- HASH_STATE(2);
- FILL_STATE(2);
-
- HASH_STATE(3);
- FILL_STATE(3);
-
- scratchpadPtr += 64 * 4;
- break;
-
- case 2:
- HASH_STATE(0);
- FILL_STATE(0);
-
- HASH_STATE(1);
- FILL_STATE(1);
-
- scratchpadPtr += 64 * 2;
- break;
-
- default:
- HASH_STATE(0);
- FILL_STATE(0);
-
- scratchpadPtr += 64;
- break;
- }
- break;
- }
+#ifdef XMRIG_RISCV
+ if (xmrig::Cpu::info()->hasRISCV_Vector()) {
+ hashAndFillAes1Rx4_RVV<softAes, unroll>(scratchpad, scratchpadSize, hash, fill_state);
+ return;
}
-
-#undef HASH_STATE
-#undef FILL_STATE
-
- __riscv_vsuxei32_v_u32m1((uint32_t*)fill_state + 0, stride, fill_state02, 8);
- __riscv_vsuxei32_v_u32m1((uint32_t*)fill_state + 4, stride, fill_state13, 8);
-
- //two extra rounds to achieve full diffusion
- const vuint32m1_t xkey00 = __riscv_vle32_v_u32m1(AES_HASH_1R_XKEY00, 8);
- const vuint32m1_t xkey11 = __riscv_vle32_v_u32m1(AES_HASH_1R_XKEY11, 8);
-
- hash_state02 = softaes_vector_double(hash_state02, xkey00, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
- hash_state13 = softaes_vector_double(hash_state13, xkey00, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
-
- hash_state02 = softaes_vector_double(hash_state02, xkey11, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
- hash_state13 = softaes_vector_double(hash_state13, xkey11, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
-
- //output hash
- __riscv_vsuxei32_v_u32m1((uint32_t*)hash + 0, stride, hash_state02, 8);
- __riscv_vsuxei32_v_u32m1((uint32_t*)hash + 4, stride, hash_state13, 8);
-}
-
-#else // defined(XMRIG_RISCV) && defined(XMRIG_RVV_ENABLED)
-
-template<int softAes, int unroll>
-void hashAndFillAes1Rx4(void *scratchpad, size_t scratchpadSize, void *hash, void* fill_state) {
- PROFILE_SCOPE(RandomX_AES);
+#endif
uint8_t* scratchpadPtr = (uint8_t*)scratchpad;
const uint8_t* scratchpadEnd = scratchpadPtr + scratchpadSize;
@@ -500,7 +412,6 @@ void hashAndFillAes1Rx4(void *scratchpad, size_t scratchpadSize, void *hash, voi
rx_store_vec_i128((rx_vec_i128*)hash + 2, hash_state2);
rx_store_vec_i128((rx_vec_i128*)hash + 3, hash_state3);
}
-#endif // defined(XMRIG_RISCV) && defined(XMRIG_RVV_ENABLED)
template void hashAndFillAes1Rx4<0,2>(void* scratchpad, size_t scratchpadSize, void* hash, void* fill_state);
template void hashAndFillAes1Rx4<1,1>(void* scratchpad, size_t scratchpadSize, void* hash, void* fill_state);
@@ -512,43 +423,54 @@ hashAndFillAes1Rx4_impl* softAESImpl = &hashAndFillAes1Rx4<1,1>;
void SelectSoftAESImpl(size_t threadsCount)
{
- constexpr uint64_t test_length_ms = 100;
- const std::array<hashAndFillAes1Rx4_impl *, 4> impl = {
- &hashAndFillAes1Rx4<1,1>,
- &hashAndFillAes1Rx4<2,1>,
- &hashAndFillAes1Rx4<2,2>,
- &hashAndFillAes1Rx4<2,4>,
- };
- size_t fast_idx = 0;
- double fast_speed = 0.0;
- for (size_t run = 0; run < 3; ++run) {
- for (size_t i = 0; i < impl.size(); ++i) {
- const double t1 = xmrig::Chrono::highResolutionMSecs();
- std::vector<uint32_t> count(threadsCount, 0);
- std::vector<std::thread> threads;
- for (size_t t = 0; t < threadsCount; ++t) {
- threads.emplace_back([&, t]() {
- std::vector<uint8_t> scratchpad(10 * 1024);
- alignas(16) uint8_t hash[64] = {};
- alignas(16) uint8_t state[64] = {};
- do {
- (*impl[i])(scratchpad.data(), scratchpad.size(), hash, state);
- ++count[t];
- } while (xmrig::Chrono::highResolutionMSecs() - t1 < test_length_ms);
- });
- }
- uint32_t total = 0;
- for (size_t t = 0; t < threadsCount; ++t) {
- threads[t].join();
- total += count[t];
- }
- const double t2 = xmrig::Chrono::highResolutionMSecs();
- const double speed = total * 1e3 / (t2 - t1);
- if (speed > fast_speed) {
- fast_idx = i;
- fast_speed = speed;
- }
- }
- }
- softAESImpl = impl[fast_idx];
+ constexpr uint64_t test_length_ms = 100;
+
+ const std::array<hashAndFillAes1Rx4_impl *, 4> impl = {
+ &hashAndFillAes1Rx4<1,1>,
+ &hashAndFillAes1Rx4<2,1>,
+ &hashAndFillAes1Rx4<2,2>,
+ &hashAndFillAes1Rx4<2,4>,
+ };
+
+ size_t fast_idx = 0;
+ double fast_speed = 0.0;
+
+ for (size_t run = 0; run < 3; ++run) {
+ for (size_t i = 0; i < impl.size(); ++i) {
+ const double t1 = xmrig::Chrono::highResolutionMSecs();
+
+ std::vector<uint32_t> count(threadsCount, 0);
+ std::vector<std::thread> threads;
+
+ for (size_t t = 0; t < threadsCount; ++t) {
+ threads.emplace_back([&, t]() {
+ std::vector<uint8_t> scratchpad(10 * 1024);
+
+ alignas(16) uint8_t hash[64] = {};
+ alignas(16) uint8_t state[64] = {};
+
+ do {
+ (*impl[i])(scratchpad.data(), scratchpad.size(), hash, state);
+ ++count[t];
+ } while (xmrig::Chrono::highResolutionMSecs() - t1 < test_length_ms);
+ });
+ }
+
+ uint32_t total = 0;
+
+ for (size_t t = 0; t < threadsCount; ++t) {
+ threads[t].join();
+ total += count[t];
+ }
+
+ const double t2 = xmrig::Chrono::highResolutionMSecs();
+ const double speed = total * 1e3 / (t2 - t1);
+
+ if (speed > fast_speed) {
+ fast_idx = i;
+ fast_speed = speed;
+ }
+ }
+ }
+ softAESImpl = impl[fast_idx];
}
diff --git a/src/crypto/randomx/aes_hash_rv64_vector.cpp b/src/crypto/randomx/aes_hash_rv64_vector.cpp
new file mode 100644
index 0000000..76401f7
--- /dev/null
+++ b/src/crypto/randomx/aes_hash_rv64_vector.cpp
@@ -0,0 +1,322 @@
+/*
+Copyright (c) 2025 SChernykh <https://github.com/SChernykh>
+Copyright (c) 2025 XMRig <support@xmrig.com>
+
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+ * Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ * Neither the name of the copyright holder nor the
+ names of its contributors may be used to endorse or promote products
+ derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#include <riscv_vector.h>
+
+#include "crypto/randomx/soft_aes.h"
+#include "crypto/randomx/randomx.h"
+
+static FORCE_INLINE vuint32m1_t softaes_vector_double(
+ vuint32m1_t in,
+ vuint32m1_t key,
+ vuint8m1_t i0, vuint8m1_t i1, vuint8m1_t i2, vuint8m1_t i3,
+ const uint32_t* lut0, const uint32_t* lut1, const uint32_t *lut2, const uint32_t* lut3)
+{
+ const vuint8m1_t in8 = __riscv_vreinterpret_v_u32m1_u8m1(in);
+
+ const vuint32m1_t index0 = __riscv_vreinterpret_v_u8m1_u32m1(__riscv_vrgather_vv_u8m1(in8, i0, 32));
+ const vuint32m1_t index1 = __riscv_vreinterpret_v_u8m1_u32m1(__riscv_vrgather_vv_u8m1(in8, i1, 32));
+ const vuint32m1_t index2 = __riscv_vreinterpret_v_u8m1_u32m1(__riscv_vrgather_vv_u8m1(in8, i2, 32));
+ const vuint32m1_t index3 = __riscv_vreinterpret_v_u8m1_u32m1(__riscv_vrgather_vv_u8m1(in8, i3, 32));
+
+ vuint32m1_t s0 = __riscv_vluxei32_v_u32m1(lut0, __riscv_vsll_vx_u32m1(index0, 2, 8), 8);
+ vuint32m1_t s1 = __riscv_vluxei32_v_u32m1(lut1, __riscv_vsll_vx_u32m1(index1, 2, 8), 8);
+ vuint32m1_t s2 = __riscv_vluxei32_v_u32m1(lut2, __riscv_vsll_vx_u32m1(index2, 2, 8), 8);
+ vuint32m1_t s3 = __riscv_vluxei32_v_u32m1(lut3, __riscv_vsll_vx_u32m1(index3, 2, 8), 8);
+
+ s0 = __riscv_vxor_vv_u32m1(s0, s1, 8);
+ s2 = __riscv_vxor_vv_u32m1(s2, s3, 8);
+ s0 = __riscv_vxor_vv_u32m1(s0, s2, 8);
+
+ return __riscv_vxor_vv_u32m1(s0, key, 8);
+}
+
+static constexpr uint32_t AES_HASH_1R_STATE02[8] = { 0x92b52c0d, 0x9fa856de, 0xcc82db47, 0xd7983aad, 0x6a770017, 0xae62c7d0, 0x5079506b, 0xe8a07ce4 };
+static constexpr uint32_t AES_HASH_1R_STATE13[8] = { 0x338d996e, 0x15c7b798, 0xf59e125a, 0xace78057, 0x630a240c, 0x07ad828d, 0x79a10005, 0x7e994948 };
+
+static constexpr uint32_t AES_GEN_1R_KEY02[8] = { 0x6daca553, 0x62716609, 0xdbb5552b, 0xb4f44917, 0x3f1262f1, 0x9f947ec6, 0xf4c0794f, 0x3e20e345 };
+static constexpr uint32_t AES_GEN_1R_KEY13[8] = { 0x6d7caf07, 0x846a710d, 0x1725d378, 0x0da1dc4e, 0x6aef8135, 0xb1ba317c, 0x16314c88, 0x49169154 };
+
+static constexpr uint32_t AES_HASH_1R_XKEY00[8] = { 0xf6fa8389, 0x8b24949f, 0x90dc56bf, 0x06890201, 0xf6fa8389, 0x8b24949f, 0x90dc56bf, 0x06890201 };
+static constexpr uint32_t AES_HASH_1R_XKEY11[8] = { 0x61b263d1, 0x51f4e03c, 0xee1043c6, 0xed18f99b, 0x61b263d1, 0x51f4e03c, 0xee1043c6, 0xed18f99b };
+
+static constexpr uint32_t AES_HASH_STRIDE_X2[8] = { 0, 4, 8, 12, 32, 36, 40, 44 };
+static constexpr uint32_t AES_HASH_STRIDE_X4[8] = { 0, 4, 8, 12, 64, 68, 72, 76 };
+
+template<int softAes>
+void hashAes1Rx4_RVV(const void *input, size_t inputSize, void *hash) {
+ const uint8_t* inptr = (const uint8_t*)input;
+ const uint8_t* inputEnd = inptr + inputSize;
+
+ //intial state
+ vuint32m1_t state02 = __riscv_vle32_v_u32m1(AES_HASH_1R_STATE02, 8);
+ vuint32m1_t state13 = __riscv_vle32_v_u32m1(AES_HASH_1R_STATE13, 8);
+
+ const vuint32m1_t stride = __riscv_vle32_v_u32m1(AES_HASH_STRIDE_X2, 8);
+
+ const vuint8m1_t lutenc_index0 = __riscv_vle8_v_u8m1(lutEncIndex[0], 32);
+ const vuint8m1_t lutenc_index1 = __riscv_vle8_v_u8m1(lutEncIndex[1], 32);
+ const vuint8m1_t lutenc_index2 = __riscv_vle8_v_u8m1(lutEncIndex[2], 32);
+ const vuint8m1_t lutenc_index3 = __riscv_vle8_v_u8m1(lutEncIndex[3], 32);
+
+ const vuint8m1_t& lutdec_index0 = lutenc_index0;
+ const vuint8m1_t lutdec_index1 = __riscv_vle8_v_u8m1(lutDecIndex[1], 32);
+ const vuint8m1_t& lutdec_index2 = lutenc_index2;
+ const vuint8m1_t lutdec_index3 = __riscv_vle8_v_u8m1(lutDecIndex[3], 32);
+
+ //process 64 bytes at a time in 4 lanes
+ while (inptr < inputEnd) {
+ state02 = softaes_vector_double(state02, __riscv_vluxei32_v_u32m1((uint32_t*)inptr + 0, stride, 8), lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+ state13 = softaes_vector_double(state13, __riscv_vluxei32_v_u32m1((uint32_t*)inptr + 4, stride, 8), lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+
+ inptr += 64;
+ }
+
+ //two extra rounds to achieve full diffusion
+ const vuint32m1_t xkey00 = __riscv_vle32_v_u32m1(AES_HASH_1R_XKEY00, 8);
+ const vuint32m1_t xkey11 = __riscv_vle32_v_u32m1(AES_HASH_1R_XKEY11, 8);
+
+ state02 = softaes_vector_double(state02, xkey00, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+ state13 = softaes_vector_double(state13, xkey00, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+
+ state02 = softaes_vector_double(state02, xkey11, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+ state13 = softaes_vector_double(state13, xkey11, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+
+ //output hash
+ __riscv_vsuxei32_v_u32m1((uint32_t*)hash + 0, stride, state02, 8);
+ __riscv_vsuxei32_v_u32m1((uint32_t*)hash + 4, stride, state13, 8);
+}
+
+template void hashAes1Rx4_RVV<false>(const void *input, size_t inputSize, void *hash);
+template void hashAes1Rx4_RVV<true>(const void *input, size_t inputSize, void *hash);
+
+template<int softAes>
+void fillAes1Rx4_RVV(void *state, size_t outputSize, void *buffer) {
+ const uint8_t* outptr = (uint8_t*)buffer;
+ const uint8_t* outputEnd = outptr + outputSize;
+
+ const vuint32m1_t key02 = __riscv_vle32_v_u32m1(AES_GEN_1R_KEY02, 8);
+ const vuint32m1_t key13 = __riscv_vle32_v_u32m1(AES_GEN_1R_KEY13, 8);
+
+ const vuint32m1_t stride = __riscv_vle32_v_u32m1(AES_HASH_STRIDE_X2, 8);
+
+ vuint32m1_t state02 = __riscv_vluxei32_v_u32m1((uint32_t*)state + 0, stride, 8);
+ vuint32m1_t state13 = __riscv_vluxei32_v_u32m1((uint32_t*)state + 4, stride, 8);
+
+ const vuint8m1_t lutenc_index0 = __riscv_vle8_v_u8m1(lutEncIndex[0], 32);
+ const vuint8m1_t lutenc_index1 = __riscv_vle8_v_u8m1(lutEncIndex[1], 32);
+ const vuint8m1_t lutenc_index2 = __riscv_vle8_v_u8m1(lutEncIndex[2], 32);
+ const vuint8m1_t lutenc_index3 = __riscv_vle8_v_u8m1(lutEncIndex[3], 32);
+
+ const vuint8m1_t& lutdec_index0 = lutenc_index0;
+ const vuint8m1_t lutdec_index1 = __riscv_vle8_v_u8m1(lutDecIndex[1], 32);
+ const vuint8m1_t& lutdec_index2 = lutenc_index2;
+ const vuint8m1_t lutdec_index3 = __riscv_vle8_v_u8m1(lutDecIndex[3], 32);
+
+ while (outptr < outputEnd) {
+ state02 = softaes_vector_double(state02, key02, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+ state13 = softaes_vector_double(state13, key13, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+
+ __riscv_vsuxei32_v_u32m1((uint32_t*)outptr + 0, stride, state02, 8);
+ __riscv_vsuxei32_v_u32m1((uint32_t*)outptr + 4, stride, state13, 8);
+
+ outptr += 64;
+ }
+
+ __riscv_vsuxei32_v_u32m1((uint32_t*)state + 0, stride, state02, 8);
+ __riscv_vsuxei32_v_u32m1((uint32_t*)state + 4, stride, state13, 8);
+}
+
+template void fillAes1Rx4_RVV<false>(void *state, size_t outputSize, void *buffer);
+template void fillAes1Rx4_RVV<true>(void *state, size_t outputSize, void *buffer);
+
+template<int softAes>
+void fillAes4Rx4_RVV(void *state, size_t outputSize, void *buffer) {
+ const uint8_t* outptr = (uint8_t*)buffer;
+ const uint8_t* outputEnd = outptr + outputSize;
+
+ const vuint32m1_t stride4 = __riscv_vle32_v_u32m1(AES_HASH_STRIDE_X4, 8);
+
+ const vuint32m1_t key04 = __riscv_vluxei32_v_u32m1((uint32_t*)(RandomX_CurrentConfig.fillAes4Rx4_Key + 0), stride4, 8);
+ const vuint32m1_t key15 = __riscv_vluxei32_v_u32m1((uint32_t*)(RandomX_CurrentConfig.fillAes4Rx4_Key + 1), stride4, 8);
+ const vuint32m1_t key26 = __riscv_vluxei32_v_u32m1((uint32_t*)(RandomX_CurrentConfig.fillAes4Rx4_Key + 2), stride4, 8);
+ const vuint32m1_t key37 = __riscv_vluxei32_v_u32m1((uint32_t*)(RandomX_CurrentConfig.fillAes4Rx4_Key + 3), stride4, 8);
+
+ const vuint32m1_t stride = __riscv_vle32_v_u32m1(AES_HASH_STRIDE_X2, 8);
+
+ vuint32m1_t state02 = __riscv_vluxei32_v_u32m1((uint32_t*)state + 0, stride, 8);
+ vuint32m1_t state13 = __riscv_vluxei32_v_u32m1((uint32_t*)state + 4, stride, 8);
+
+ const vuint8m1_t lutenc_index0 = __riscv_vle8_v_u8m1(lutEncIndex[0], 32);
+ const vuint8m1_t lutenc_index1 = __riscv_vle8_v_u8m1(lutEncIndex[1], 32);
+ const vuint8m1_t lutenc_index2 = __riscv_vle8_v_u8m1(lutEncIndex[2], 32);
+ const vuint8m1_t lutenc_index3 = __riscv_vle8_v_u8m1(lutEncIndex[3], 32);
+
+ const vuint8m1_t& lutdec_index0 = lutenc_index0;
+ const vuint8m1_t lutdec_index1 = __riscv_vle8_v_u8m1(lutDecIndex[1], 32);
+ const vuint8m1_t& lutdec_index2 = lutenc_index2;
+ const vuint8m1_t lutdec_index3 = __riscv_vle8_v_u8m1(lutDecIndex[3], 32);
+
+ while (outptr < outputEnd) {
+ state02 = softaes_vector_double(state02, key04, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+ state13 = softaes_vector_double(state13, key04, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+
+ state02 = softaes_vector_double(state02, key15, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+ state13 = softaes_vector_double(state13, key15, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+
+ state02 = softaes_vector_double(state02, key26, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+ state13 = softaes_vector_double(state13, key26, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+
+ state02 = softaes_vector_double(state02, key37, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+ state13 = softaes_vector_double(state13, key37, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+
+ __riscv_vsuxei32_v_u32m1((uint32_t*)outptr + 0, stride, state02, 8);
+ __riscv_vsuxei32_v_u32m1((uint32_t*)outptr + 4, stride, state13, 8);
+
+ outptr += 64;
+ }
+}
+
+template void fillAes4Rx4_RVV<false>(void *state, size_t outputSize, void *buffer);
+template void fillAes4Rx4_RVV<true>(void *state, size_t outputSize, void *buffer);
+
+template<int softAes, int unroll>
+void hashAndFillAes1Rx4_RVV(void *scratchpad, size_t scratchpadSize, void *hash, void* fill_state) {
+ uint8_t* scratchpadPtr = (uint8_t*)scratchpad;
+ const uint8_t* scratchpadEnd = scratchpadPtr + scratchpadSize;
+
+ vuint32m1_t hash_state02 = __riscv_vle32_v_u32m1(AES_HASH_1R_STATE02, 8);
+ vuint32m1_t hash_state13 = __riscv_vle32_v_u32m1(AES_HASH_1R_STATE13, 8);
+
+ const vuint32m1_t key02 = __riscv_vle32_v_u32m1(AES_GEN_1R_KEY02, 8);
+ const vuint32m1_t key13 = __riscv_vle32_v_u32m1(AES_GEN_1R_KEY13, 8);
+
+ const vuint32m1_t stride = __riscv_vle32_v_u32m1(AES_HASH_STRIDE_X2, 8);
+
+ vuint32m1_t fill_state02 = __riscv_vluxei32_v_u32m1((uint32_t*)fill_state + 0, stride, 8);
+ vuint32m1_t fill_state13 = __riscv_vluxei32_v_u32m1((uint32_t*)fill_state + 4, stride, 8);
+
+ const vuint8m1_t lutenc_index0 = __riscv_vle8_v_u8m1(lutEncIndex[0], 32);
+ const vuint8m1_t lutenc_index1 = __riscv_vle8_v_u8m1(lutEncIndex[1], 32);
+ const vuint8m1_t lutenc_index2 = __riscv_vle8_v_u8m1(lutEncIndex[2], 32);
+ const vuint8m1_t lutenc_index3 = __riscv_vle8_v_u8m1(lutEncIndex[3], 32);
+
+ const vuint8m1_t& lutdec_index0 = lutenc_index0;
+ const vuint8m1_t lutdec_index1 = __riscv_vle8_v_u8m1(lutDecIndex[1], 32);
+ const vuint8m1_t& lutdec_index2 = lutenc_index2;
+ const vuint8m1_t lutdec_index3 = __riscv_vle8_v_u8m1(lutDecIndex[3], 32);
+
+ //process 64 bytes at a time in 4 lanes
+ while (scratchpadPtr < scratchpadEnd) {
+#define HASH_STATE(k) \
+ hash_state02 = softaes_vector_double(hash_state02, __riscv_vluxei32_v_u32m1((uint32_t*)scratchpadPtr + k * 16 + 0, stride, 8), lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3); \
+ hash_state13 = softaes_vector_double(hash_state13, __riscv_vluxei32_v_u32m1((uint32_t*)scratchpadPtr + k * 16 + 4, stride, 8), lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+
+#define FILL_STATE(k) \
+ fill_state02 = softaes_vector_double(fill_state02, key02, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3); \
+ fill_state13 = softaes_vector_double(fill_state13, key13, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3); \
+ __riscv_vsuxei32_v_u32m1((uint32_t*)scratchpadPtr + k * 16 + 0, stride, fill_state02, 8); \
+ __riscv_vsuxei32_v_u32m1((uint32_t*)scratchpadPtr + k * 16 + 4, stride, fill_state13, 8);
+
+ switch (softAes) {
+ case 0:
+ HASH_STATE(0);
+ HASH_STATE(1);
+
+ FILL_STATE(0);
+ FILL_STATE(1);
+
+ scratchpadPtr += 128;
+ break;
+
+ default:
+ switch (unroll) {
+ case 4:
+ HASH_STATE(0);
+ FILL_STATE(0);
+
+ HASH_STATE(1);
+ FILL_STATE(1);
+
+ HASH_STATE(2);
+ FILL_STATE(2);
+
+ HASH_STATE(3);
+ FILL_STATE(3);
+
+ scratchpadPtr += 64 * 4;
+ break;
+
+ case 2:
+ HASH_STATE(0);
+ FILL_STATE(0);
+
+ HASH_STATE(1);
+ FILL_STATE(1);
+
+ scratchpadPtr += 64 * 2;
+ break;
+
+ default:
+ HASH_STATE(0);
+ FILL_STATE(0);
+
+ scratchpadPtr += 64;
+ break;
+ }
+ break;
+ }
+ }
+
+#undef HASH_STATE
+#undef FILL_STATE
+
+ __riscv_vsuxei32_v_u32m1((uint32_t*)fill_state + 0, stride, fill_state02, 8);
+ __riscv_vsuxei32_v_u32m1((uint32_t*)fill_state + 4, stride, fill_state13, 8);
+
+ //two extra rounds to achieve full diffusion
+ const vuint32m1_t xkey00 = __riscv_vle32_v_u32m1(AES_HASH_1R_XKEY00, 8);
+ const vuint32m1_t xkey11 = __riscv_vle32_v_u32m1(AES_HASH_1R_XKEY11, 8);
+
+ hash_state02 = softaes_vector_double(hash_state02, xkey00, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+ hash_state13 = softaes_vector_double(hash_state13, xkey00, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+
+ hash_state02 = softaes_vector_double(hash_state02, xkey11, lutenc_index0, lutenc_index1, lutenc_index2, lutenc_index3, lutEnc0, lutEnc1, lutEnc2, lutEnc3);
+ hash_state13 = softaes_vector_double(hash_state13, xkey11, lutdec_index0, lutdec_index1, lutdec_index2, lutdec_index3, lutDec0, lutDec1, lutDec2, lutDec3);
+
+ //output hash
+ __riscv_vsuxei32_v_u32m1((uint32_t*)hash + 0, stride, hash_state02, 8);
+ __riscv_vsuxei32_v_u32m1((uint32_t*)hash + 4, stride, hash_state13, 8);
+}
+
+template void hashAndFillAes1Rx4_RVV<0,2>(void* scratchpad, size_t scratchpadSize, void* hash, void* fill_state);
+template void hashAndFillAes1Rx4_RVV<1,1>(void* scratchpad, size_t scratchpadSize, void* hash, void* fill_state);
+template void hashAndFillAes1Rx4_RVV<2,1>(void* scratchpad, size_t scratchpadSize, void* hash, void* fill_state);
+template void hashAndFillAes1Rx4_RVV<2,2>(void* scratchpad, size_t scratchpadSize, void* hash, void* fill_state);
+template void hashAndFillAes1Rx4_RVV<2,4>(void* scratchpad, size_t scratchpadSize, void* hash, void* fill_state);
diff --git a/src/crypto/randomx/aes_hash_rv64_vector.hpp b/src/crypto/randomx/aes_hash_rv64_vector.hpp
new file mode 100644
index 0000000..d0101d6
--- /dev/null
+++ b/src/crypto/randomx/aes_hash_rv64_vector.hpp
@@ -0,0 +1,42 @@
+/*
+Copyright (c) 2025 SChernykh <https://github.com/SChernykh>
+Copyright (c) 2025 XMRig <support@xmrig.com>
+
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+ * Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+ * Neither the name of the copyright holder nor the
+ names of its contributors may be used to endorse or promote products
+ derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#pragma once
+
+template<int softAes>
+void hashAes1Rx4_RVV(const void *input, size_t inputSize, void *hash);
+
+template<int softAes>
+void fillAes1Rx4_RVV(void *state, size_t outputSize, void *buffer);
+
+template<int softAes>
+void fillAes4Rx4_RVV(void *state, size_t outputSize, void *buffer);
+
+template<int softAes, int unroll>
+void hashAndFillAes1Rx4_RVV(void *scratchpad, size_t scratchpadSize, void *hash, void* fill_state);
diff --git a/src/crypto/randomx/jit_compiler_rv64_vector.cpp b/src/crypto/randomx/jit_compiler_rv64_vector.cpp
index e7a9c4d..f101095 100644
--- a/src/crypto/randomx/jit_compiler_rv64_vector.cpp
+++ b/src/crypto/randomx/jit_compiler_rv64_vector.cpp
@@ -235,9 +235,17 @@ static void imm_to_x5(uint32_t imm, uint8_t*& p)
return;
}
- // lui x5, imm_hi
+ if (imm_hi < (32 << 12)) {
+ //c.lui x5, imm_hi
+ emit16(0x6281 + (imm_hi >> 10));
+ }
+ else {
+ // lui x5, imm_hi
+ emit32(0x000002B7 + imm_hi);
+ }
+
// addiw x5, x5, imm_lo
- emit64(0x0002829B000002B7ULL | imm_hi | (static_cast<uint64_t>(imm_lo) << 52))
+ emit32(0x0002829B | (imm_lo << 20));
}
static void loadFromScratchpad(uint32_t src, uint32_t dst, uint32_t mod, uint32_t imm, uint8_t*& p)
diff --git a/src/crypto/randomx/soft_aes.h b/src/crypto/randomx/soft_aes.h
index 4e13391..a2fecc2 100644
--- a/src/crypto/randomx/soft_aes.h
+++ b/src/crypto/randomx/soft_aes.h
@@ -150,32 +150,3 @@ template<>
FORCE_INLINE rx_vec_i128 aesdec<0>(rx_vec_i128 in, rx_vec_i128 key) {
return rx_aesdec_vec_i128(in, key);
}
-
-#if defined(XMRIG_RISCV) && defined(XMRIG_RVV_ENABLED)
-#include <riscv_vector.h>
-
-FORCE_INLINE vuint32m1_t softaes_vector_double(
- vuint32m1_t in,
- vuint32m1_t key,
- vuint8m1_t i0, vuint8m1_t i1, vuint8m1_t i2, vuint8m1_t i3,
- const uint32_t* lut0, const uint32_t* lut1, const uint32_t *lut2, const uint32_t* lut3)
-{
- const vuint8m1_t in8 = __riscv_vreinterpret_v_u32m1_u8m1(in);
-
- const vuint32m1_t index0 = __riscv_vreinterpret_v_u8m1_u32m1(__riscv_vrgather_vv_u8m1(in8, i0, 32));
- const vuint32m1_t index1 = __riscv_vreinterpret_v_u8m1_u32m1(__riscv_vrgather_vv_u8m1(in8, i1, 32));
- const vuint32m1_t index2 = __riscv_vreinterpret_v_u8m1_u32m1(__riscv_vrgather_vv_u8m1(in8, i2, 32));
- const vuint32m1_t index3 = __riscv_vreinterpret_v_u8m1_u32m1(__riscv_vrgather_vv_u8m1(in8, i3, 32));
-
- vuint32m1_t s0 = __riscv_vluxei32_v_u32m1(lut0, __riscv_vsll_vx_u32m1(index0, 2, 8), 8);
- vuint32m1_t s1 = __riscv_vluxei32_v_u32m1(lut1, __riscv_vsll_vx_u32m1(index1, 2, 8), 8);
- vuint32m1_t s2 = __riscv_vluxei32_v_u32m1(lut2, __riscv_vsll_vx_u32m1(index2, 2, 8), 8);
- vuint32m1_t s3 = __riscv_vluxei32_v_u32m1(lut3, __riscv_vsll_vx_u32m1(index3, 2, 8), 8);
-
- s0 = __riscv_vxor_vv_u32m1(s0, s1, 8);
- s2 = __riscv_vxor_vv_u32m1(s2, s3, 8);
- s0 = __riscv_vxor_vv_u32m1(s0, s2, 8);
-
- return __riscv_vxor_vv_u32m1(s0, key, 8);
-}
-#endif // defined(XMRIG_RISCV) && defined(XMRIG_RVV_ENABLED)
Why this scored 15/100
Community notes
Notes can correct, qualify, or add evidence to the AI analysis. Every note shown here has been validated by a human moderator.
The AI analysis stands alone for now. Submit a note if you can add evidence or important context.