wownero
/
wownero
7
13
Fork 13
Code Issues 8 Pull Requests Releases 41 Wiki Activity

mod variant4_random_math

Browse Source
pull/461/head
wowario 1 year ago
parent a636325df3
commit 6aaee726f1
No known key found for this signature in database
GPG Key ID: 24DCBE762DE9C111
1 changed files with 7 additions and 14 deletions
Show Stats Download Patch File Download Diff File

21
src/crypto/variant4_random_math.h
Unescape View File

@ -200,7 +200,6 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
memset(data, 0, sizeof(data)); memset(data, 0, sizeof(data));
uint64_t tmp = SWAP64LE(height); uint64_t tmp = SWAP64LE(height);
memcpy(data, &tmp, sizeof(uint64_t)); memcpy(data, &tmp, sizeof(uint64_t));
data[20] = -38; // change seed
// Set data_index past the last byte in data // Set data_index past the last byte in data
// to trigger full data update with blake hash // to trigger full data update with blake hash
@ -211,10 +210,9 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
// There is a small chance (1.8%) that register R8 won't be used in the generated program // There is a small chance (1.8%) that register R8 won't be used in the generated program
// So we keep track of it and try again if it's not used // So we keep track of it and try again if it's not used
bool r8_used;
do { do {
int latency[9]; int latency[8];
int asic_latency[9]; int asic_latency[8];
// Tracks previous instruction and value of the source operand for registers R0-R3 throughout code execution // Tracks previous instruction and value of the source operand for registers R0-R3 throughout code execution
// byte 0: current value of the destination register // byte 0: current value of the destination register
@ -223,7 +221,7 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
// //
// Registers R4-R8 are constant and are treated as having the same value because when we do // Registers R4-R8 are constant and are treated as having the same value because when we do
// the same operation twice with two constant source registers, it can be optimized into a single operation // the same operation twice with two constant source registers, it can be optimized into a single operation
uint32_t inst_data[9] = { 0, 1, 2, 3, 0xFFFFFF, 0xFFFFFF, 0xFFFFFF, 0xFFFFFF, 0xFFFFFF }; uint32_t inst_data[8] = { 0, 1, 2, 3, 0xFFFFFF, 0xFFFFFF, 0xFFFFFF, 0xFFFFFF };
bool alu_busy[TOTAL_LATENCY + 1][ALU_COUNT]; bool alu_busy[TOTAL_LATENCY + 1][ALU_COUNT];
bool is_rotation[V4_INSTRUCTION_COUNT]; bool is_rotation[V4_INSTRUCTION_COUNT];
@ -242,7 +240,6 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
code_size = 0; code_size = 0;
int total_iterations = 0; int total_iterations = 0;
r8_used = false;
// Generate random code to achieve minimal required latency for our abstract CPU // Generate random code to achieve minimal required latency for our abstract CPU
// Try to get this latency for all 4 registers // Try to get this latency for all 4 registers
@ -287,8 +284,8 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
if (((opcode == ADD) || (opcode == SUB) || (opcode == XOR)) && (a == b)) if (((opcode == ADD) || (opcode == SUB) || (opcode == XOR)) && (a == b))
{ {
// Use register R8 as source instead // Use register R8 as source instead
b = 8; b = a + 4;
src_index = 8; src_index = b;
} }
// Don't do rotation with the same destination twice because it's equal to a single rotation // Don't do rotation with the same destination twice because it's equal to a single rotation
@ -368,10 +365,6 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
code[code_size].src_index = src_index; code[code_size].src_index = src_index;
code[code_size].C = 0; code[code_size].C = 0;
if (src_index == 8)
{
r8_used = true;
}
if (opcode == ADD) if (opcode == ADD)
{ {
@ -402,7 +395,7 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
// We need to add a few more MUL and ROR instructions to achieve minimal required latency for ASIC // We need to add a few more MUL and ROR instructions to achieve minimal required latency for ASIC
// Get this latency for at least 1 of the 4 registers // Get this latency for at least 1 of the 4 registers
const int prev_code_size = code_size; const int prev_code_size = code_size;
while ((code_size < NUM_INSTRUCTIONS_MAX) && (asic_latency[0] < TOTAL_LATENCY) && (asic_latency[1] < TOTAL_LATENCY) && (asic_latency[2] < TOTAL_LATENCY) && (asic_latency[3] < TOTAL_LATENCY)) while ((asic_latency[0] < TOTAL_LATENCY) && (asic_latency[1] < TOTAL_LATENCY) && (asic_latency[2] < TOTAL_LATENCY) && (asic_latency[3] < TOTAL_LATENCY))
{ {
int min_idx = 0; int min_idx = 0;
int max_idx = 0; int max_idx = 0;
@ -426,7 +419,7 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_
// There is ~98.15% chance that loop condition is false, so this loop will execute only 1 iteration most of the time // There is ~98.15% chance that loop condition is false, so this loop will execute only 1 iteration most of the time
// It never does more than 4 iterations for all block heights < 10,000,000 // It never does more than 4 iterations for all block heights < 10,000,000
} while (!r8_used || (code_size < NUM_INSTRUCTIONS_MIN) || (code_size > NUM_INSTRUCTIONS_MAX)); } while (code_size < NUM_INSTRUCTIONS_MIN);
// It's guaranteed that NUM_INSTRUCTIONS_MIN <= code_size <= NUM_INSTRUCTIONS_MAX here // It's guaranteed that NUM_INSTRUCTIONS_MIN <= code_size <= NUM_INSTRUCTIONS_MAX here
// Add final instruction to stop the interpreter // Add final instruction to stop the interpreter

Write Preview
Loading…
Cancel
Save