You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
103 lines
3.4 KiB
103 lines
3.4 KiB
#include <stdio.h> |
|
|
|
#include "cuda_helper.h" |
|
|
|
static uint32_t *d_offsets1[MAX_GPUS] = { 0 }; |
|
static uint32_t *d_offsets2[MAX_GPUS] = { 0 }; |
|
|
|
static uint32_t *d_brcount1[MAX_GPUS] = { 0 }; |
|
static uint32_t *d_brcount2[MAX_GPUS] = { 0 }; |
|
|
|
__global__ __launch_bounds__(128, 6) |
|
void bastion_filter2_gpu(const uint32_t threads, const uint32_t* d_hash, uint32_t* d_hash1, uint32_t* d_hash2, uint32_t* d_br_ofts1, uint32_t* d_count1, uint32_t* d_br_ofts2, uint32_t* d_count2) |
|
{ |
|
const uint32_t thread = (blockDim.x * blockIdx.x + threadIdx.x); |
|
if (thread < threads) |
|
{ |
|
const uint32_t offset = thread * 16U; // 64U / sizeof(uint32_t); |
|
uint4 *psrc = (uint4*) (&d_hash[offset]); |
|
uint4 *pdst; |
|
d_br_ofts1[thread] = 0; |
|
d_br_ofts2[thread] = 0; |
|
if (((uint8_t*)psrc)[0] & 0x8) { |
|
// uint4 = 4x uint32_t = 16 bytes |
|
uint32_t oft = atomicAdd(d_count1, 1U) * 16U; |
|
d_br_ofts1[thread] = oft + 16U; |
|
pdst = (uint4*) (&d_hash1[oft]); |
|
} else { |
|
uint32_t oft = atomicAdd(d_count2, 1U) * 16U; |
|
d_br_ofts2[thread] = oft + 16U; |
|
pdst = (uint4*) (&d_hash2[oft]); |
|
} |
|
pdst[0] = psrc[0]; |
|
pdst[1] = psrc[1]; |
|
pdst[2] = psrc[2]; |
|
pdst[3] = psrc[3]; |
|
} |
|
} |
|
|
|
__global__ __launch_bounds__(128, 6) |
|
void bastion_merge2_gpu(const uint32_t threads, uint32_t* d_hash, uint32_t* d_hash1, uint32_t* d_hash2, uint32_t* d_br_ofts1, uint32_t* d_br_ofts2) |
|
{ |
|
const uint32_t thread = (blockDim.x * blockIdx.x + threadIdx.x); |
|
if (thread < threads) |
|
{ |
|
const uint32_t offset = thread * 16U; |
|
uint4 *pdst = (uint4*) (&d_hash[offset]); |
|
uint4 *psrc; |
|
if (d_br_ofts1[thread]) { |
|
const uint32_t oft = d_br_ofts1[thread] - 16U; |
|
psrc = (uint4*) (&d_hash1[oft]); |
|
} else { |
|
const uint32_t oft = d_br_ofts2[thread] - 16U; |
|
psrc = (uint4*) (&d_hash2[oft]); |
|
} |
|
pdst[0] = psrc[0]; |
|
pdst[1] = psrc[1]; |
|
pdst[2] = psrc[2]; |
|
pdst[3] = psrc[3]; |
|
} |
|
} |
|
|
|
|
|
__host__ |
|
void bastion_init(const int thr_id, const uint32_t threads) |
|
{ |
|
CUDA_SAFE_CALL(cudaMalloc(&d_offsets1[thr_id], sizeof(uint32_t) * threads)); |
|
CUDA_SAFE_CALL(cudaMalloc(&d_offsets2[thr_id], sizeof(uint32_t) * threads)); |
|
CUDA_SAFE_CALL(cudaMalloc(&d_brcount1[thr_id], sizeof(uint32_t))); |
|
CUDA_SAFE_CALL(cudaMalloc(&d_brcount2[thr_id], sizeof(uint32_t))); |
|
} |
|
|
|
__host__ |
|
void bastion_free(const int thr_id) |
|
{ |
|
cudaFree(d_offsets1[thr_id]); |
|
cudaFree(d_offsets2[thr_id]); |
|
cudaFree(d_brcount1[thr_id]); |
|
cudaFree(d_brcount2[thr_id]); |
|
} |
|
|
|
__host__ |
|
uint32_t bastion_filter2(const int thr_id, const uint32_t threads, const uint32_t *inpHashes, uint32_t* d_hash1, uint32_t* d_hash2) |
|
{ |
|
uint32_t num = 0; |
|
cudaMemset(d_brcount1[thr_id], 0, 4); |
|
cudaMemset(d_brcount2[thr_id], 0, 4); |
|
const uint32_t threadsperblock = 128; |
|
dim3 grid((threads + threadsperblock - 1) / threadsperblock); |
|
dim3 block(threadsperblock); |
|
bastion_filter2_gpu <<<grid, block>>> (threads, inpHashes, d_hash1, d_hash2, d_offsets1[thr_id], d_brcount1[thr_id], d_offsets2[thr_id], d_brcount2[thr_id]); |
|
cudaMemcpy(&num, d_brcount1[thr_id], 4, cudaMemcpyDeviceToHost); |
|
return num; |
|
} |
|
|
|
__host__ |
|
void bastion_merge2(const int thr_id, const uint32_t threads, uint32_t *outpHashes, uint32_t* d_hash1, uint32_t* d_hash2) |
|
{ |
|
const uint32_t threadsperblock = 128; |
|
dim3 grid((threads + threadsperblock - 1) / threadsperblock); |
|
dim3 block(threadsperblock); |
|
// put back branch hashes to the common buffer d_hash |
|
bastion_merge2_gpu <<<grid, block>>> (threads, outpHashes, d_hash1, d_hash2, d_offsets1[thr_id], d_offsets2[thr_id]); |
|
}
|
|
|