mirror of
https://github.com/monero-project/monero.git
synced 2024-12-23 12:09:54 +00:00
CNv4 JIT compiler for x86-64 and tests
Minimalistic JIT code generator for random math sequence in CryptonightR. Usage: - Allocate writable and executable memory - Call v4_generate_JIT_code with "buf" pointed to memory allocated on the previous step - Call the generated code instead of "v4_random_math(code, r)", omit the "code" parameter
This commit is contained in:
parent
cf27e26beb
commit
4b32859d2f
4 changed files with 2746 additions and 0 deletions
99
src/crypto/CryptonightR_JIT.c
Normal file
99
src/crypto/CryptonightR_JIT.c
Normal file
|
@ -0,0 +1,99 @@
|
|||
#include <assert.h>
|
||||
#include <stddef.h>
|
||||
#include <stdint.h>
|
||||
#include <string.h>
|
||||
#include <stdio.h>
|
||||
#include <unistd.h>
|
||||
|
||||
#include "common/int-util.h"
|
||||
#include "hash-ops.h"
|
||||
#include "variant4_random_math.h"
|
||||
#include "CryptonightR_JIT.h"
|
||||
#include "CryptonightR_template.h"
|
||||
|
||||
static const uint8_t prologue[] = {
|
||||
0x4C, 0x8B, 0xD7, // mov r10, rdi
|
||||
0x53, // push rbx
|
||||
0x55, // push rbp
|
||||
0x41, 0x57, // push r15
|
||||
0x4C, 0x8B, 0xDC, // mov r11, rsp
|
||||
0x41, 0x8B, 0x1A, // mov ebx, DWORD PTR [r10]
|
||||
0x41, 0x8B, 0x72, 0x04, // mov esi, DWORD PTR [r10+4]
|
||||
0x41, 0x8B, 0x7A, 0x08, // mov edi, DWORD PTR [r10+8]
|
||||
0x41, 0x8B, 0x6A, 0x0C, // mov ebp, DWORD PTR [r10+12]
|
||||
0x41, 0x8B, 0x62, 0x10, // mov esp, DWORD PTR [r10+16]
|
||||
0x45, 0x8B, 0x7A, 0x14, // mov r15d, DWORD PTR [r10+20]
|
||||
0x41, 0x8B, 0x42, 0x18, // mov eax, DWORD PTR [r10+24]
|
||||
0x41, 0x8B, 0x52, 0x1C, // mov edx, DWORD PTR [r10+28]
|
||||
0x45, 0x8B, 0x4A, 0x20, // mov r9d, DWORD PTR [r10+32]
|
||||
};
|
||||
|
||||
static const uint8_t epilogue[] = {
|
||||
0x49, 0x8B, 0xE3, // mov rsp, r11
|
||||
0x41, 0x89, 0x1A, // mov DWORD PTR [r10], ebx
|
||||
0x41, 0x89, 0x72, 0x04, // mov DWORD PTR [r10+4], esi
|
||||
0x41, 0x89, 0x7A, 0x08, // mov DWORD PTR [r10+8], edi
|
||||
0x41, 0x89, 0x6A, 0x0C, // mov DWORD PTR [r10+12], ebp
|
||||
0x41, 0x5F, // pop r15
|
||||
0x5D, // pop rbp
|
||||
0x5B, // pop rbx
|
||||
0xC3, // ret
|
||||
};
|
||||
|
||||
#define APPEND_CODE(src, size) \
|
||||
do { \
|
||||
if (JIT_code + (size) > JIT_code_end) \
|
||||
return -1; \
|
||||
memcpy(JIT_code, (src), (size)); \
|
||||
JIT_code += (size); \
|
||||
} while (0)
|
||||
|
||||
int v4_generate_JIT_code(const struct V4_Instruction* code, v4_random_math_JIT_func buf, const size_t buf_size)
|
||||
{
|
||||
uint8_t* JIT_code = (uint8_t*) buf;
|
||||
const uint8_t* JIT_code_end = JIT_code + buf_size;
|
||||
|
||||
APPEND_CODE(prologue, sizeof(prologue));
|
||||
|
||||
uint32_t prev_rot_src = 0xFFFFFFFFU;
|
||||
|
||||
for (int i = 0;; ++i)
|
||||
{
|
||||
const struct V4_Instruction inst = code[i];
|
||||
if (inst.opcode == RET)
|
||||
break;
|
||||
|
||||
const uint8_t opcode = (inst.opcode == MUL) ? inst.opcode : (inst.opcode + 2);
|
||||
|
||||
const uint32_t a = inst.dst_index;
|
||||
const uint32_t b = inst.src_index;
|
||||
const uint8_t c = opcode | (inst.dst_index << V4_OPCODE_BITS) | (((inst.src_index == 8) ? inst.dst_index : inst.src_index) << (V4_OPCODE_BITS + V4_DST_INDEX_BITS));
|
||||
|
||||
switch (inst.opcode)
|
||||
{
|
||||
case ROR:
|
||||
case ROL:
|
||||
if (b != prev_rot_src)
|
||||
{
|
||||
prev_rot_src = b;
|
||||
const uint8_t* p1 = (const uint8_t*) instructions_mov[c];
|
||||
const uint8_t* p2 = (const uint8_t*) instructions_mov[c + 1];
|
||||
APPEND_CODE(p1, p2 - p1);
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
if (a == prev_rot_src)
|
||||
prev_rot_src = 0xFFFFFFFFU;
|
||||
|
||||
const uint8_t* p1 = (const uint8_t*) instructions[c];
|
||||
const uint8_t* p2 = (const uint8_t*) instructions[c + 1];
|
||||
APPEND_CODE(p1, p2 - p1);
|
||||
|
||||
if (inst.opcode == ADD)
|
||||
*(uint32_t*)(JIT_code - 4) = inst.C;
|
||||
}
|
||||
|
||||
APPEND_CODE(epilogue, sizeof(epilogue));
|
||||
return 0;
|
||||
}
|
18
src/crypto/CryptonightR_JIT.h
Normal file
18
src/crypto/CryptonightR_JIT.h
Normal file
|
@ -0,0 +1,18 @@
|
|||
#ifndef CRYPTONIGHTR_JIT_H
|
||||
#define CRYPTONIGHTR_JIT_H
|
||||
|
||||
// Minimalistic JIT code generator for random math sequence in CryptonightR
|
||||
//
|
||||
// Usage:
|
||||
// - Allocate writable and executable memory
|
||||
// - Call v4_generate_JIT_code with "buf" pointed to memory allocated on previous step
|
||||
// - Call the generated code instead of "v4_random_math(code, r)", omit the "code" parameter
|
||||
|
||||
typedef void (*v4_random_math_JIT_func)(uint32_t* r) __attribute__((sysv_abi));
|
||||
|
||||
// Given the random math sequence, generates machine code (x86-64) for it
|
||||
// Returns 0 if code was generated successfully
|
||||
// Returns -1 if provided buffer was too small
|
||||
int v4_generate_JIT_code(const struct V4_Instruction* code, v4_random_math_JIT_func buf, const size_t buf_size);
|
||||
|
||||
#endif // CRYPTONIGHTR_JIT_H
|
1590
src/crypto/CryptonightR_template.S
Normal file
1590
src/crypto/CryptonightR_template.S
Normal file
File diff suppressed because it is too large
Load diff
1039
src/crypto/CryptonightR_template.h
Normal file
1039
src/crypto/CryptonightR_template.h
Normal file
File diff suppressed because it is too large
Load diff
Loading…
Reference in a new issue