Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

lcg caching #2

Open
wants to merge 3 commits into
base: master
Choose a base branch
from
Open

lcg caching #2

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c812b9f
lcg caching
DutChen18 Jun 11, 2020
fe54d84
fixed bug in generation of extra random numbers
DutChen18 Jun 12, 2020
66c3f3d
oops
DutChen18 Jun 12, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
107 changes: 80 additions & 27 deletions kaktoos.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,10 @@
#define WORK_UNIT_SIZE (1ULL << 23ULL)
#endif

#ifndef WORK_RANDOM_SIZE
#define WORK_RANDOM_SIZE (WORK_UNIT_SIZE + 1000)
#endif

#ifndef BLOCK_SIZE
#define BLOCK_SIZE 1024ULL
#endif
Expand All @@ -39,6 +43,7 @@
#define END (1ULL << 48ULL)
#endif


__device__ inline int8_t extract(const int8_t heightMap[], uint32_t i) {
return (int8_t) (heightMap[i >> 1ULL] >> ((i & 1ULL) << 2ULL)) & 0xF;
}
Expand All @@ -50,36 +55,71 @@ __device__ inline void increase(int8_t heightMap[], uint32_t i) {
namespace java_random {

// Random::next(bits)
__device__ inline uint32_t next(uint64_t *random, int32_t bits) {
*random = (*random * RANDOM_MULTIPLIER + RANDOM_ADDEND) & RANDOM_MASK;
return (uint32_t) (*random >> (48ULL - bits));
__device__ inline uint32_t next(uint64_t *random, size_t *i, int32_t bits) {
return (uint32_t) (random[++*i] >> (48ULL - bits));
}

__device__ inline int32_t next_int_unknown(uint64_t *seed, int16_t bound) {
__device__ inline int32_t next_int_unknown(uint64_t *random, size_t *i, int16_t bound) {
if ((bound & -bound) == bound) {
*seed = (*seed * RANDOM_MULTIPLIER + RANDOM_ADDEND) & RANDOM_MASK;
return (int32_t) ((bound * (*seed >> 17ULL)) >> 31ULL);
return (int32_t) ((bound * (random[++*i] >> 17ULL)) >> 31ULL);
}

int32_t bits, value;
do {
*seed = (*seed * RANDOM_MULTIPLIER + RANDOM_ADDEND) & RANDOM_MASK;
bits = *seed >> 17ULL;
bits = random[++*i] >> 17ULL;
value = bits % bound;
} while (bits - value + (bound - 1) < 0);
return value;
}

// Random::nextInt(bound)
__device__ inline uint32_t next_int(uint64_t *random) {
return java_random::next(random, 31) % 3;
__device__ inline uint32_t next_int(uint64_t *random, size_t *i) {
return java_random::next(random, i, 31) % 3;
}

}

__global__ __launch_bounds__(BLOCK_SIZE, 2) void crack(uint64_t seed_offset, int32_t *num_seeds, uint64_t *seeds) {
uint64_t originalSeed = blockIdx.x * blockDim.x + threadIdx.x + seed_offset;
uint64_t seed = originalSeed;
struct rand_params {
uint64_t multiplier;
uint64_t addend;
};

__host__ uint64_t get_start_seed(uint64_t offset) {
uint64_t seed = 0;
uint64_t i = 0;
for (; i + 1024 * 1024 <= offset; i += 1024 * 1024) {
seed = (seed * 280364631195649ULL + 215216710221824ULL) & RANDOM_MASK;
}
for (; i < offset; i++) {
seed = (seed * RANDOM_MULTIPLIER + RANDOM_ADDEND) & RANDOM_MASK;
}
return seed;
}

__host__ void gen_rand_params(rand_params *rp, size_t n) {
rp[0].multiplier = 1;
rp[0].addend = 0;
for (size_t i = 1; i < n; i++) {
rp[i].multiplier = (rp[i - 1].multiplier * RANDOM_MULTIPLIER) & RANDOM_MASK;
rp[i].addend = (rp[i - 1].addend * RANDOM_MULTIPLIER + RANDOM_ADDEND) & RANDOM_MASK;
}
}

// TODO: generate low bits seperately so crack() only has to index 32 bit values, ~3% speed improvement
__global__ __launch_bounds__(BLOCK_SIZE, 2) void init(rand_params *rp, uint64_t *random, uint64_t seed, uint64_t m, uint64_t a) {
size_t index = blockIdx.x * blockDim.x + threadIdx.x;
size_t stride = blockDim.x * gridDim.x;
seed = (seed * rp[index].multiplier + rp[index].addend) & RANDOM_MASK;
random[index] = seed;
for (size_t i = index + stride; i < WORK_RANDOM_SIZE; i += stride) {
seed = (seed * m + a) & RANDOM_MASK;
random[i] = seed;
}
}

__global__ __launch_bounds__(BLOCK_SIZE, 2) void crack(uint64_t *random, int32_t *num_seeds, uint64_t *seeds) {
size_t seedIndex = blockIdx.x * blockDim.x + threadIdx.x;
uint64_t originalSeed = random[seedIndex];

int8_t heightMap[512];

Expand All @@ -101,23 +141,23 @@ __global__ __launch_bounds__(BLOCK_SIZE, 2) void crack(uint64_t seed_offset, int
if (WANTED_CACTUS_HEIGHT - extract(heightMap, currentHighestPos) > 9 * (10 - i))
return;

initialPosX = java_random::next(&seed, 4) + 8;
initialPosZ = java_random::next(&seed, 4) + 8;
initialPosX = java_random::next(random, &seedIndex, 4) + 8;
initialPosZ = java_random::next(random, &seedIndex, 4) + 8;
terrainHeight = (extract(heightMap, initialPosX + initialPosZ * 32) + FLOOR_LEVEL + 1) * 2;

initialPosY = java_random::next_int_unknown(&seed, terrainHeight);
initialPosY = java_random::next_int_unknown(random, &seedIndex, terrainHeight);

for (a = 0; a < 10; a++) {
posX = initialPosX + java_random::next(&seed, 3) - java_random::next(&seed, 3);
posY = initialPosY + java_random::next(&seed, 2) - java_random::next(&seed, 2);
posZ = initialPosZ + java_random::next(&seed, 3) - java_random::next(&seed, 3);
posX = initialPosX + java_random::next(random, &seedIndex, 3) - java_random::next(random, &seedIndex, 3);
posY = initialPosY + java_random::next(random, &seedIndex, 2) - java_random::next(random, &seedIndex, 2);
posZ = initialPosZ + java_random::next(random, &seedIndex, 3) - java_random::next(random, &seedIndex, 3);

posMap = posX + posZ * 32;
// Keep
if (posY <= extract(heightMap, posMap) + FLOOR_LEVEL && posY >= 0)
continue;

offset = 1 + java_random::next_int_unknown(&seed, java_random::next_int(&seed) + 1);
offset = 1 + java_random::next_int_unknown(random, &seedIndex, java_random::next_int(random, &seedIndex) + 1);

for (j = 0; j < offset; j++) {
if ((posY + j - 1) > extract(heightMap, posMap) + FLOOR_LEVEL || posY < 0) continue;
Expand Down Expand Up @@ -145,32 +185,43 @@ __global__ __launch_bounds__(BLOCK_SIZE, 2) void crack(uint64_t seed_offset, int
struct GPU_Node {
int *num_seeds;
uint64_t *seeds;
uint64_t *random;
rand_params *rp;
};

void setup_gpu_node(GPU_Node *node, int32_t gpu) {
cudaSetDevice(gpu);
cudaMallocManaged(&node->num_seeds, sizeof(*node->num_seeds));
cudaMallocManaged(&node->seeds, 1ULL << 10ULL);
cudaMallocManaged(&node->random, WORK_RANDOM_SIZE * sizeof(*node->random));
cudaMallocManaged(&node->rp, WORK_RANDOM_SIZE * sizeof(*node->rp));
}

GPU_Node nodes[GPU_COUNT];
uint64_t offset = OFFSET;
uint64_t seed = get_start_seed(OFFSET);
uint64_t count = 0;
std::mutex info_lock;
std::vector<uint64_t> seeds;
rand_params *rp;

void gpu_manager(int32_t gpu_index) {
std::string fileName = "kaktoos_seeds" + std::to_string(gpu_index) + ".txt";
FILE *out_file = fopen(fileName.c_str(), "w");
cudaSetDevice(gpu_index);
cudaSetDevice(gpu_index);
for (size_t i = 0; i < WORK_RANDOM_SIZE; i++)
nodes[gpu_index].rp[i] = rp[i];
while (offset < END) {
*nodes[gpu_index].num_seeds = 0;
crack<<<WORK_UNIT_SIZE / BLOCK_SIZE, BLOCK_SIZE, 0>>>(offset, nodes[gpu_index].num_seeds,
nodes[gpu_index].seeds);
uint64_t m = rp[WORK_UNIT_SIZE].multiplier, a = rp[WORK_UNIT_SIZE].addend;
init<<<WORK_UNIT_SIZE / BLOCK_SIZE, BLOCK_SIZE, 0>>>(nodes[gpu_index].rp, nodes[gpu_index].random, seed, m, a);
info_lock.lock();
offset += WORK_UNIT_SIZE;
seed = (seed * m + a) & RANDOM_MASK;
offset += WORK_UNIT_SIZE;
info_lock.unlock();
cudaDeviceSynchronize();
crack<<<WORK_UNIT_SIZE / BLOCK_SIZE, BLOCK_SIZE, 0>>>(nodes[gpu_index].random, nodes[gpu_index].num_seeds, nodes[gpu_index].seeds);
cudaDeviceSynchronize();
for (int32_t i = 0, e = *nodes[gpu_index].num_seeds; i < e; i++) {
fprintf(out_file, "%lld\n", (long long int) nodes[gpu_index].seeds[i]);
seeds.push_back(nodes[gpu_index].seeds[i]);
Expand All @@ -183,9 +234,12 @@ void gpu_manager(int32_t gpu_index) {
fclose(out_file);
}

int main() {
printf("Searching %ld total seeds...\n", (long int) (END - OFFSET));

int main() {
printf("Searching %ld total seeds...\n", (long int) (END - OFFSET));

rp = new rand_params[WORK_RANDOM_SIZE];
gen_rand_params(rp, WORK_RANDOM_SIZE);
std::thread threads[GPU_COUNT];

time_t startTime = time(nullptr), currentTime;
Expand Down Expand Up @@ -224,5 +278,4 @@ int main() {

printf("Done!\n");
printf("But, verily, it be the nature of dreams to end.\n");

}