From 64e14b7d821db7d1e08df2eca6a47559c44b6c21 Mon Sep 17 00:00:00 2001 From: Tanguy Pruvot Date: Fri, 6 Nov 2015 14:55:43 +0100 Subject: [PATCH] quark: final cleanup for the 1.7 --- quark/cuda_quark_compactionTest.cu | 58 ++++++++++++++---------------- 1 file changed, 27 insertions(+), 31 deletions(-) diff --git a/quark/cuda_quark_compactionTest.cu b/quark/cuda_quark_compactionTest.cu index 2f19753..96c0afd 100644 --- a/quark/cuda_quark_compactionTest.cu +++ b/quark/cuda_quark_compactionTest.cu @@ -1,25 +1,23 @@ -/* - * REQUIRE SM 3.0 arch! +/** + * __shfl_up require SM 3.0 arch! + * + * SM 2 alternative method by tpruvot@github 2015 */ - #include #include #include "cuda_helper.h" -#include + +#ifdef __INTELLISENSE__ +/* just for vstudio code colors */ +#define __CUDA_ARCH__ 300 +#define __shfl_up(var, delta, width) (0) +#endif static uint32_t *h_numValid[MAX_GPUS]; static uint32_t *d_tempBranch1Nonces[MAX_GPUS]; static uint32_t *d_partSum[2][MAX_GPUS]; // für bis zu vier partielle Summen -#if __CUDA_ARCH__ < 300 -/** - * __shfl_up() calculates a source lane ID by subtracting delta from the caller's lane ID, and clamping to the range 0..width-1 - */ -#undef __shfl_up -#define __shfl_up(var, delta, width) (0) -#endif - // True/False tester typedef uint32_t(*cuda_compactTestFunction_t)(uint32_t *inpHash); @@ -80,6 +78,7 @@ __global__ void quark_compactTest_gpu_SCAN(uint32_t *data, const int width, uint32_t *partial_sums=NULL, cuda_compactTestFunction_t testFunc=NULL, uint32_t threads=0, uint32_t startNounce=0, uint32_t *inpHashes=NULL, uint32_t *d_validNonceTable=NULL) { +#if __CUDA_ARCH__ >= 300 extern __shared__ uint32_t sums[]; int id = ((blockIdx.x * blockDim.x) + threadIdx.x); //int lane_id = id % warpSize; @@ -182,6 +181,7 @@ void quark_compactTest_gpu_SCAN(uint32_t *data, const int width, uint32_t *parti { partial_sums[blockIdx.x] = value; } +#endif // SM3+ } // Uniform add: add partial sums array @@ -202,9 +202,9 @@ void quark_compactTest_gpu_ADD(uint32_t *data, uint32_t *partial_sums, int len) data[id] += buf; } -// Der Scatter __global__ -void quark_compactTest_gpu_SCATTER(uint32_t *sum, uint32_t *outp, cuda_compactTestFunction_t testFunc, uint32_t threads=0, uint32_t startNounce=0, uint32_t *inpHashes=NULL, uint32_t *d_validNonceTable=NULL) +void quark_compactTest_gpu_SCATTER(uint32_t *sum, uint32_t *outp, cuda_compactTestFunction_t testFunc, + uint32_t threads=0, uint32_t startNounce=0, uint32_t *inpHashes=NULL, uint32_t *d_validNonceTable=NULL) { int id = ((blockIdx.x * blockDim.x) + threadIdx.x); uint32_t actNounce = id; @@ -254,7 +254,7 @@ void quark_compactTest_cpu_singleCompaction(int thr_id, uint32_t threads, uint32 cuda_compactTestFunction_t function, uint32_t startNounce, uint32_t *inpHashes, uint32_t *d_validNonceTable) { int orgThreads = threads; - threads = (int)quark_compactTest_roundUpExp((uint32_t)threads); + threads = quark_compactTest_roundUpExp(threads); // threadsPerBlock ausrechnen int blockSize = 256; int nSummen = threads / blockSize; @@ -286,14 +286,11 @@ void quark_compactTest_cpu_singleCompaction(int thr_id, uint32_t threads, uint32 else cudaMemcpy(nrm, &(d_partSum[0][thr_id])[nSummen-1], sizeof(uint32_t), cudaMemcpyDeviceToHost); - - // Addieren if(callThrid) { quark_compactTest_gpu_ADD<<>>(d_partSum[0][thr_id]+blockSize, d_partSum[1][thr_id], blockSize*thr2); } quark_compactTest_gpu_ADD<<>>(d_tempBranch1Nonces[thr_id]+blockSize, d_partSum[0][thr_id], threads); - - // Scatter + quark_compactTest_gpu_SCATTER<<>>(d_tempBranch1Nonces[thr_id], d_nonces1, function, orgThreads, startNounce, inpHashes, d_validNonceTable); @@ -301,12 +298,9 @@ void quark_compactTest_cpu_singleCompaction(int thr_id, uint32_t threads, uint32 cudaStreamSynchronize(NULL); } -#ifdef __INTELLISENSE__ -#define atomicAdd(x,n) ( *(x)+=n ) -#endif - +#if __CUDA_ARCH__ < 300 __global__ __launch_bounds__(128, 8) -void quark_filter_gpu_sm2(const uint32_t threads, const uint32_t* d_hash, uint32_t* d_branch2, uint32_t* d_NonceBranch, uint32_t &count) +void quark_filter_gpu_sm2(const uint32_t threads, const uint32_t* d_hash, uint32_t* d_branch2, uint32_t* d_NonceBranch) { const uint32_t thread = (blockDim.x * blockIdx.x + threadIdx.x); if (thread < threads) @@ -315,7 +309,6 @@ void quark_filter_gpu_sm2(const uint32_t threads, const uint32_t* d_hash, uint32 uint4 *psrc = (uint4*) (&d_hash[offset]); d_NonceBranch[thread] = ((uint8_t*)psrc)[0] & 0x8; if (d_NonceBranch[thread]) return; - //uint32_t off_br = atomicAdd(&count, 1) * 16U; // uint4 = 4x uint32_t = 16 bytes uint4 *pdst = (uint4*) (&d_branch2[offset]); pdst[0] = psrc[0]; @@ -331,7 +324,7 @@ void quark_merge_gpu_sm2(const uint32_t threads, uint32_t* d_hash, uint32_t* d_b const uint32_t thread = (blockDim.x * blockIdx.x + threadIdx.x); if (thread < threads && !d_NonceBranch[thread]) { - const uint32_t offset = thread * 16U; // 64U / sizeof(uint32_t); + const uint32_t offset = thread * 16U; uint4 *pdst = (uint4*) (&d_hash[offset]); uint4 *psrc = (uint4*) (&d_branch2[offset]); pdst[0] = psrc[0]; @@ -340,17 +333,20 @@ void quark_merge_gpu_sm2(const uint32_t threads, uint32_t* d_hash, uint32_t* d_b pdst[3] = psrc[3]; } } +#else +__global__ void quark_filter_gpu_sm2(const uint32_t threads, const uint32_t* d_hash, uint32_t* d_branch2, uint32_t* d_NonceBranch) {} +__global__ void quark_merge_gpu_sm2(const uint32_t threads, uint32_t* d_hash, uint32_t* d_branch2, uint32_t* const d_NonceBranch) {} +#endif __host__ uint32_t quark_filter_cpu_sm2(const int thr_id, const uint32_t threads, const uint32_t *inpHashes, uint32_t* d_branch2) { - uint32_t branch2_nonces = 0; const uint32_t threadsperblock = 128; dim3 grid((threads + threadsperblock - 1) / threadsperblock); dim3 block(threadsperblock); - // copy all hashes in the right branch buffer - quark_filter_gpu_sm2 <<>> (threads, inpHashes, d_branch2, d_tempBranch1Nonces[thr_id], branch2_nonces); - return branch2_nonces; + // extract algo permution hashes to a second branch buffer + quark_filter_gpu_sm2 <<>> (threads, inpHashes, d_branch2, d_tempBranch1Nonces[thr_id]); + return threads; } __host__ @@ -359,7 +355,7 @@ void quark_merge_cpu_sm2(const int thr_id, const uint32_t threads, uint32_t *out const uint32_t threadsperblock = 128; dim3 grid((threads + threadsperblock - 1) / threadsperblock); dim3 block(threadsperblock); - // copy second branch hashes to d_hash + // put back second branch hashes to the common buffer d_hash quark_merge_gpu_sm2 <<>> (threads, outpHashes, d_branch2, d_tempBranch1Nonces[thr_id]); }