5 #ifndef MAPREDUCE_KERNELS_H
6 #define MAPREDUCE_KERNELS_H
33 "__kernel void UnaryMapReduceKernel_KERNELNAME(__global TYPE* input, __global TYPE* output, size_t n, __local TYPE* sdata, CONST_TYPE const1)\n"
35 " size_t blockSize = get_local_size(0);\n"
36 " size_t tid = get_local_id(0);\n"
37 " size_t i = get_group_id(0)*blockSize + get_local_id(0);\n"
38 " size_t gridSize = blockSize*get_num_groups(0);\n"
42 " result = FUNCTIONNAME_MAP(input[i], const1);\n"
48 " tempMap = FUNCTIONNAME_MAP(input[i], const1);\n"
49 " result = FUNCTIONNAME_REDUCE(result, tempMap, (TYPE)0);\n"
52 " sdata[tid] = result;\n"
53 " barrier(CLK_LOCAL_MEM_FENCE);\n"
54 " if(blockSize >= 512) { if (tid < 256 && tid + 256 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 256], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
55 " if(blockSize >= 256) { if (tid < 128 && tid + 128 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 128], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
56 " if(blockSize >= 128) { if (tid < 64 && tid + 64 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 64], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
57 " if(blockSize >= 64) { if (tid < 32 && tid + 32 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 32], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
58 " if(blockSize >= 32) { if (tid < 16 && tid + 16 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 16], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
59 " if(blockSize >= 16) { if (tid < 8 && tid + 8 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 8], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
60 " if(blockSize >= 8) { if (tid < 4 && tid + 4 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 4], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
61 " if(blockSize >= 4) { if (tid < 2 && tid + 2 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 2], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
62 " if(blockSize >= 2) { if (tid < 1 && tid + 1 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 1], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
65 " output[get_group_id(0)] = sdata[tid];\n"
77 "__kernel void BinaryMapReduceKernel_KERNELNAME(__global TYPE* input1, __global TYPE* input2, __global TYPE* output, size_t n, __local TYPE* sdata, CONST_TYPE const1)\n"
79 " size_t blockSize = get_local_size(0);\n"
80 " size_t tid = get_local_id(0);\n"
81 " size_t i = get_group_id(0)*blockSize + get_local_id(0);\n"
82 " size_t gridSize = blockSize*get_num_groups(0);\n"
86 " result = FUNCTIONNAME_MAP(input1[i], input2[i], const1);\n"
92 " tempMap = FUNCTIONNAME_MAP(input1[i], input2[i], const1);\n"
93 " result = FUNCTIONNAME_REDUCE(result, tempMap, (TYPE)0);\n"
96 " sdata[tid] = result;\n"
97 " barrier(CLK_LOCAL_MEM_FENCE);\n"
98 " if(blockSize >= 512) { if (tid < 256 && tid + 256 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 256], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
99 " if(blockSize >= 256) { if (tid < 128 && tid + 128 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 128], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
100 " if(blockSize >= 128) { if (tid < 64 && tid + 64 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 64], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
101 " if(blockSize >= 64) { if (tid < 32 && tid + 32 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 32], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
102 " if(blockSize >= 32) { if (tid < 16 && tid + 16 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 16], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
103 " if(blockSize >= 16) { if (tid < 8 && tid + 8 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 8], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
104 " if(blockSize >= 8) { if (tid < 4 && tid + 4 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 4], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
105 " if(blockSize >= 4) { if (tid < 2 && tid + 2 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 2], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
106 " if(blockSize >= 2) { if (tid < 1 && tid + 1 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 1], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
109 " output[get_group_id(0)] = sdata[tid];\n"
121 "__kernel void TrinaryMapReduceKernel_KERNELNAME(__global TYPE* input1, __global TYPE* input2, __global TYPE* input3, __global TYPE* output, size_t n, __local TYPE* sdata, CONST_TYPE const1)\n"
123 " size_t blockSize = get_local_size(0);\n"
124 " size_t tid = get_local_id(0);\n"
125 " size_t i = get_group_id(0)*blockSize + get_local_id(0);\n"
126 " size_t gridSize = blockSize*get_num_groups(0);\n"
127 " TYPE result = 0;\n"
130 " result = FUNCTIONNAME_MAP(input1[i], input2[i], input3[i], const1);\n"
136 " tempMap = FUNCTIONNAME_MAP(input1[i], input2[i], input3[i], const1);\n"
137 " result = FUNCTIONNAME_REDUCE(result, tempMap, (TYPE)0);\n"
140 " sdata[tid] = result;\n"
141 " barrier(CLK_LOCAL_MEM_FENCE);\n"
142 " if(blockSize >= 512) { if (tid < 256 && tid + 256 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 256], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
143 " if(blockSize >= 256) { if (tid < 128 && tid + 128 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 128], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
144 " if(blockSize >= 128) { if (tid < 64 && tid + 64 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 64], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
145 " if(blockSize >= 64) { if (tid < 32 && tid + 32 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 32], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
146 " if(blockSize >= 32) { if (tid < 16 && tid + 16 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 16], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
147 " if(blockSize >= 16) { if (tid < 8 && tid + 8 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 8], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
148 " if(blockSize >= 8) { if (tid < 4 && tid + 4 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 4], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
149 " if(blockSize >= 4) { if (tid < 2 && tid + 2 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 2], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
150 " if(blockSize >= 2) { if (tid < 1 && tid + 1 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 1], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"
153 " output[get_group_id(0)] = sdata[tid];\n"
187 template <
typename T,
typename UnaryFunc,
typename BinaryFunc>
188 __global__
void MapReduceKernel1_CU(UnaryFunc mapFunc, BinaryFunc reduceFunc, T* input, T* output,
size_t n)
190 extern __shared__
char _sdata[];
191 T* sdata =
reinterpret_cast<T*
>(_sdata);
193 size_t blockSize = blockDim.x;
194 size_t tid = threadIdx.x;
195 size_t i = blockIdx.x * blockSize + tid;
196 size_t gridSize = blockSize*gridDim.x;
201 result = mapFunc.CU(input[i]);
208 tempMap = mapFunc.CU(input[i]);
209 result = reduceFunc.CU(result, tempMap);
219 if (tid < 256 && tid + 256 < n)
221 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 256]);
227 if (tid < 128 && tid + 128 < n)
229 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 128]);
235 if (tid < 64 && tid + 64 < n)
237 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 64]);
243 if (tid < 32 && tid + 32 < n)
245 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 32]);
251 if (tid < 16 && tid + 16 < n)
253 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 16]);
259 if (tid < 8 && tid + 8 < n)
261 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 8]);
267 if (tid < 4 && tid + 4 < n)
269 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 4]);
275 if (tid < 2 && tid + 2 < n)
277 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 2]);
283 if (tid < 1 && tid + 1 < n)
285 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 1]);
292 output[blockIdx.x] = sdata[tid];
301 template <
typename T,
typename BinaryFunc1,
typename BinaryFunc2>
302 __global__
void MapReduceKernel2_CU(BinaryFunc1 mapFunc, BinaryFunc2 reduceFunc, T* input1, T* input2, T* output,
size_t n)
304 extern __shared__
char _sdata[];
305 T* sdata =
reinterpret_cast<T*
>(_sdata);
307 size_t blockSize = blockDim.x;
308 size_t tid = threadIdx.x;
309 size_t i = blockIdx.x * blockSize + tid;
310 size_t gridSize = blockSize*gridDim.x;
315 result = mapFunc.CU(input1[i], input2[i]);
322 tempMap = mapFunc.CU(input1[i], input2[i]);
323 result = reduceFunc.CU(result, tempMap);
333 if (tid < 256 && tid + 256 < n)
335 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 256]);
341 if (tid < 128 && tid + 128 < n)
343 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 128]);
349 if (tid < 64 && tid + 64 < n)
351 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 64]);
357 if (tid < 32 && tid + 32 < n)
359 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 32]);
365 if (tid < 16 && tid + 16 < n)
367 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 16]);
373 if (tid < 8 && tid + 8 < n)
375 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 8]);
381 if (tid < 4 && tid + 4 < n)
383 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 4]);
389 if (tid < 2 && tid + 2 < n)
391 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 2]);
397 if (tid < 1 && tid + 1 < n)
399 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 1]);
406 output[blockIdx.x] = sdata[tid];
415 template <
typename T,
typename TrinaryFunc,
typename BinaryFunc>
416 __global__
void MapReduceKernel3_CU(TrinaryFunc mapFunc, BinaryFunc reduceFunc, T* input1, T* input2, T* input3, T* output,
size_t n)
418 extern __shared__
char _sdata[];
419 T* sdata =
reinterpret_cast<T*
>(_sdata);
421 size_t blockSize = blockDim.x;
422 size_t tid = threadIdx.x;
423 size_t i = blockIdx.x * blockSize + tid;
424 size_t gridSize = blockSize*gridDim.x;
429 result = mapFunc.CU(input1[i], input2[i], input3[i]);
436 tempMap = mapFunc.CU(input1[i], input2[i], input3[i]);
437 result = reduceFunc.CU(result, tempMap);
447 if (tid < 256 && tid + 256 < n)
449 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 256]);
455 if (tid < 128 && tid + 128 < n)
457 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 128]);
463 if (tid < 64 && tid + 64 < n)
465 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 64]);
471 if (tid < 32 && tid + 32 < n)
473 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 32]);
479 if (tid < 16 && tid + 16 < n)
481 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 16]);
487 if (tid < 8 && tid + 8 < n)
489 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 8]);
495 if (tid < 4 && tid + 4 < n)
497 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 4]);
503 if (tid < 2 && tid + 2 < n)
505 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 2]);
511 if (tid < 1 && tid + 1 < n)
513 sdata[tid] = reduceFunc.CU(sdata[tid], sdata[tid + 1]);
520 output[blockIdx.x] = sdata[tid];
__global__ void MapReduceKernel1_CU(UnaryFunc mapFunc, BinaryFunc reduceFunc, T *input, T *output, size_t n)
Definition: mapreduce_kernels.h:188
static std::string TrinaryMapReduceKernel_CL("__kernel void TrinaryMapReduceKernel_KERNELNAME(__global TYPE* input1, __global TYPE* input2, __global TYPE* input3, __global TYPE* output, size_t n, __local TYPE* sdata, CONST_TYPE const1)\n""{\n"" size_t blockSize = get_local_size(0);\n"" size_t tid = get_local_id(0);\n"" size_t i = get_group_id(0)*blockSize + get_local_id(0);\n"" size_t gridSize = blockSize*get_num_groups(0);\n"" TYPE result = 0;\n"" if(i < n)\n"" {\n"" result = FUNCTIONNAME_MAP(input1[i], input2[i], input3[i], const1);\n"" i += gridSize;\n"" }\n"" while(i < n)\n"" {\n"" TYPE tempMap;\n"" tempMap = FUNCTIONNAME_MAP(input1[i], input2[i], input3[i], const1);\n"" result = FUNCTIONNAME_REDUCE(result, tempMap, (TYPE)0);\n"" i += gridSize;\n"" }\n"" sdata[tid] = result;\n"" barrier(CLK_LOCAL_MEM_FENCE);\n"" if(blockSize >= 512) { if (tid < 256 && tid + 256 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 256], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 256) { if (tid < 128 && tid + 128 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 128], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 128) { if (tid < 64 && tid + 64 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 64], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 64) { if (tid < 32 && tid + 32 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 32], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 32) { if (tid < 16 && tid + 16 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 16], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 16) { if (tid < 8 && tid + 8 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 8], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 8) { if (tid < 4 && tid + 4 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 4], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 4) { if (tid < 2 && tid + 2 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 2], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 2) { if (tid < 1 && tid + 1 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 1], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(tid == 0)\n"" {\n"" output[get_group_id(0)] = sdata[tid];\n"" }\n""}\n")
static std::string UnaryMapReduceKernel_CL("__kernel void UnaryMapReduceKernel_KERNELNAME(__global TYPE* input, __global TYPE* output, size_t n, __local TYPE* sdata, CONST_TYPE const1)\n""{\n"" size_t blockSize = get_local_size(0);\n"" size_t tid = get_local_id(0);\n"" size_t i = get_group_id(0)*blockSize + get_local_id(0);\n"" size_t gridSize = blockSize*get_num_groups(0);\n"" TYPE result = 0;\n"" if(i < n)\n"" {\n"" result = FUNCTIONNAME_MAP(input[i], const1);\n"" i += gridSize;\n"" }\n"" while(i < n)\n"" {\n"" TYPE tempMap;\n"" tempMap = FUNCTIONNAME_MAP(input[i], const1);\n"" result = FUNCTIONNAME_REDUCE(result, tempMap, (TYPE)0);\n"" i += gridSize;\n"" }\n"" sdata[tid] = result;\n"" barrier(CLK_LOCAL_MEM_FENCE);\n"" if(blockSize >= 512) { if (tid < 256 && tid + 256 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 256], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 256) { if (tid < 128 && tid + 128 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 128], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 128) { if (tid < 64 && tid + 64 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 64], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 64) { if (tid < 32 && tid + 32 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 32], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 32) { if (tid < 16 && tid + 16 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 16], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 16) { if (tid < 8 && tid + 8 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 8], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 8) { if (tid < 4 && tid + 4 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 4], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 4) { if (tid < 2 && tid + 2 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 2], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 2) { if (tid < 1 && tid + 1 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 1], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(tid == 0)\n"" {\n"" output[get_group_id(0)] = sdata[tid];\n"" }\n""}\n")
__global__ void MapReduceKernel3_CU(TrinaryFunc mapFunc, BinaryFunc reduceFunc, T *input1, T *input2, T *input3, T *output, size_t n)
Definition: mapreduce_kernels.h:416
static std::string BinaryMapReduceKernel_CL("__kernel void BinaryMapReduceKernel_KERNELNAME(__global TYPE* input1, __global TYPE* input2, __global TYPE* output, size_t n, __local TYPE* sdata, CONST_TYPE const1)\n""{\n"" size_t blockSize = get_local_size(0);\n"" size_t tid = get_local_id(0);\n"" size_t i = get_group_id(0)*blockSize + get_local_id(0);\n"" size_t gridSize = blockSize*get_num_groups(0);\n"" TYPE result = 0;\n"" if(i < n)\n"" {\n"" result = FUNCTIONNAME_MAP(input1[i], input2[i], const1);\n"" i += gridSize;\n"" }\n"" while(i < n)\n"" {\n"" TYPE tempMap;\n"" tempMap = FUNCTIONNAME_MAP(input1[i], input2[i], const1);\n"" result = FUNCTIONNAME_REDUCE(result, tempMap, (TYPE)0);\n"" i += gridSize;\n"" }\n"" sdata[tid] = result;\n"" barrier(CLK_LOCAL_MEM_FENCE);\n"" if(blockSize >= 512) { if (tid < 256 && tid + 256 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 256], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 256) { if (tid < 128 && tid + 128 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 128], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 128) { if (tid < 64 && tid + 64 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 64], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 64) { if (tid < 32 && tid + 32 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 32], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 32) { if (tid < 16 && tid + 16 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 16], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 16) { if (tid < 8 && tid + 8 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 8], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 8) { if (tid < 4 && tid + 4 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 4], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 4) { if (tid < 2 && tid + 2 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 2], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(blockSize >= 2) { if (tid < 1 && tid + 1 < n) { sdata[tid] = FUNCTIONNAME_REDUCE(sdata[tid], sdata[tid + 1], (TYPE)0); } barrier(CLK_LOCAL_MEM_FENCE); }\n"" if(tid == 0)\n"" {\n"" output[get_group_id(0)] = sdata[tid];\n"" }\n""}\n")
__global__ void MapReduceKernel2_CU(BinaryFunc1 mapFunc, BinaryFunc2 reduceFunc, T *input1, T *input2, T *output, size_t n)
Definition: mapreduce_kernels.h:302