-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathFFT_OpenCL_Contiguous.h
194 lines (156 loc) · 10.2 KB
/
FFT_OpenCL_Contiguous.h
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
/* The information in this file is
* Copyright (C) 2011, Sven De Smet <[email protected]>
* and is subject to the terms and conditions of the
* GNU Lesser General Public License Version 2.1
* The license text is available from
* http://www.gnu.org/licenses/lgpl.html
*/
#ifndef FFT_OPENCL_CONTIGUOUS_H
#define FFT_OPENCL_CONTIGUOUS_H
#include <vector>
#include <sstream>
#include <string>
#include <exception>
#include <sys/types.h>
#include "tests/Timer.h"
#include "KernelGenerator.h"
#include "OpenCLFFTAlgorithm.h"
#define DBG(a) a
#include "SimpleMath.h"
template <class D> class FFT_OpenCL_Contiguous : public OpenCLFFTAlgorithm<D> {
public:
typedef PlannarizedDataInterface<D> DataInterfaceType;
//typedef SplitInterleavedDataInterface<D> DataInterfaceType;
private:
ComplexArrayCL<D>* data[2];
bool forward, debug;
std::vector<cl::Kernel*> kernels;
std::vector<cl::Event> kernelEvents;
CLProgram* program;
std::vector<streng> src;
DataInterfaceType* arrayDataInterface;
public:
static streng getName() { return "Contiguous OpenCL FFT"; }
std::vector<int> AGs, BGs, NGs;
static int getRequiredMemory(int n) { return 3*sizeof(Complex<D>)*n; }
static int getMaxBatchSize(int n) { return OpenCLAlgorithm::getGlobalMemory()/getRequiredMemory(n); }
FFT_OpenCL_Contiguous(int iSize, bool iForward = true, int iBatchCount = 1) : OpenCLFFTAlgorithm<D>(iSize, iBatchCount), forward(iForward) {
this->dataInterface = arrayDataInterface = new DataInterfaceType(this->batchCount*this->size);
if (getMaxBatchSize(this->size) < this->batchCount) { printf("Insufficient global memory"); } // throw exception
data[0] = new ComplexArrayCL<float>(*this->context, arrayDataInterface->in);
data[1] = new ComplexArrayCL<float>(*this->context, arrayDataInterface->out);
if (this->size > 1) {
int defaultBGl2 = 4;
int log2Size = ceilog(this->size, 2);
int stepsG = (log2Size + (defaultBGl2 - 1))/defaultBGl2;
BGs = std::vector<int>(log2Size/defaultBGl2, 1 << defaultBGl2);
if ((log2Size % defaultBGl2) > 0) BGs.push_back(1 << (log2Size % defaultBGl2));
this->computeParameters(AGs, BGs, NGs);
for (int qG = 1; qG <= stepsG; ++qG) src.push_back(this->generateKernel(qG - 1));
program = new CLProgram(*this->context, src, this->devicesToUse);
for (int qG = 1; qG <= stepsG; ++qG) kernels.push_back(program->getKernel(streng("contiguousFFT_step") + intToStr(qG)));
}
debug = false;
this->kernelTimers.resize(kernels.size()); // printf("Initialization complete"); fflush(stdout);
}
double getTime(int first, bool firstStart, int last, bool lastStart) {
cl_ulong start, end;
clGetEventProfilingInfo(kernelEvents[last](), lastStart ? CL_PROFILING_COMMAND_START : CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &end, NULL);
clGetEventProfilingInfo(kernelEvents[first](), firstStart ? CL_PROFILING_COMMAND_START : CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &start, NULL);
return (end - start) * 1.0e-9f;
}
void printGPUDebugData(int qG) {
data[(qG + 1) % 2]->enqueueReadArray(*this->commandQueue, *arrayDataInterface->out);
printf("\nqG = %i", qG); for (int q = 0; q < this->size * this->batchCount; ++q) { if ((q % this->size) == 0) printf("\n");
arrayDataInterface->out->getElement(q).print();
}
//data[1]->enqueueReadArray(*commandQueue, *arrayDataInterface->in);
//printf("1 qG = %i", qG); for (int q = 0; q < this->size; ++q) arrayDataInterface->in->getElement(q).print();
}
int getSwarmSize(int qG) { return 1; } //mAx(1, mIn(16, mIn(this->size/NGs[qG - 1], getButterflyCount(qG)))); }
int getButterflyCount(int qG) { return this->size/BGs[qG - 1]; }
int getSwarmCount(int qG) { return getButterflyCount(qG)/getSwarmSize(qG); }
int getWorkItemsPerKernel(int qG) { return getSwarmCount(qG) * this->batchCount; }
virtual double getTotalComputationFlops(int kernel) { return (5.0*this->size)*(log((double) BGs[kernel])/log(2.0)) * this->batchCount; }
virtual streng getKernelInfo(int kernel) { return intToStr(ceilog(BGs[kernel], 2)); }
virtual void execute() { //printf("Starting execution with %i kernels", (int) kernels.size()); fflush(stdout);
if (this->size == 1) { arrayDataInterface->out->setElement(0, arrayDataInterface->in->getElement(0)); }
else { if (kernels.size() > 0) {
if (!forward) {
for (int d = 0; d < this->size; ++d) arrayDataInterface->in->setElement(d, arrayDataInterface->in->getElement(d).getConjugate());
}
if (debug) for (int q = 0; q < this->size * this->batchCount; ++q) arrayDataInterface->out->setElement(q, 888); data[1]->enqueueWriteArray(*this->commandQueue, *arrayDataInterface->out);
if (this->timerTotal) this->timerTotal->resume();
data[0]->enqueueWriteArray(*this->commandQueue, *arrayDataInterface->in);
kernelEvents.clear();
for (int qG = 1; qG <= (int) kernels.size(); qG++) { cl::Kernel* kernel = kernels[qG - 1]; //printf("kernel q = %i", qG);
size_t workGroupSize;
kernel->getWorkGroupInfo<size_t>(this->devicesToUse[0], CL_KERNEL_WORK_GROUP_SIZE, &workGroupSize); // printf("[workGroupSize = %i]", workGroupSize);
int localSize = mIn(getWorkItemsPerKernel(qG), workGroupSize);
// printf("<%i>", getWorkItemsPerKernel(qG));
//printf("wgs = %i --- size = %i -- s = %i -- localSize == %i -- butterFlyCount = %i", workGroupSize, this->size, s, localSize, butterflyCount);
cl::KernelFunctor func = kernel->bind(*this->commandQueue, cl::NDRange(ceilint(getWorkItemsPerKernel(qG), localSize)), cl::NDRange(localSize));
kernel->setArg<cl_mem>(0, data[(qG ^ 1) & 1]->getData());
kernel->setArg<cl_mem>(1, data[(qG ^ 0) & 1]->getData());
// printGPUDebugData(qG);
try { kernelEvents.push_back(func()); } // enqueue kernel
catch (cl::Error e) { printf("CL Exception"); CLException cle = CLException(e); cle.handle(); fflush(stdout); }
catch (...) { printf("Unknown exception"); fflush(stdout); }
// kernelEvents[kernelEvents.size() - 1].wait(); printf("After execution"); printGPUDebugData(qG); fflush(stdout);
} // printf("Computation ends..."); fflush(stdout);
//printGPUDebugData(kernels.size() + 1);
data[kernels.size() & 1]->enqueueReadArray(*this->commandQueue, *arrayDataInterface->out);
kernelEvents[kernels.size() - 1].wait();
if (this->timerComputation) this->timerComputation->addRun(getTime(0, true, kernels.size() - 1, false));
for (int qG = 1; qG <= (int) kernels.size(); ++qG) this->kernelTimers[qG - 1].addRun(getTime(qG - 1, true, qG - 1, false));
if (this->timerTotal) this->timerTotal->suspend();
if (!forward) {
for (int d = 0; d < this->size; ++d) arrayDataInterface->out->setElement(d, arrayDataInterface->out->getElement(d).getConjugate());
}
// printf("out"); for (int q = 0; q < this->size; ++q) this->out->getElement(q).print();
// data[(kernels.size() & 1) ^ 1]->enqueueReadArray(*commandQueue, *this->in);
// printf("in"); for (int q = 0; q < this->size; ++q) this->in->getElement(q).print();
} }
}
virtual streng generateKernel(int q_G) { int qG = q_G + 1;
//int swarmStrideLevel = 5;
int LG = this->size;
std::vector<int> AL, NL, BL = std::vector<int>(ceilog(BGs[q_G], 2), 2);
this->computeParameters(AL, BL, NL);
int kL = BL.size();
std::stringstream result;
if (qG == 1) result << KomplexMath::getDeclarations();
result << "__kernel void contiguousFFT_step" << qG << "(__global float *in, __global float *out) {\n";
//int swarmIxOffset = globid >> " << swarmStrideLevel << ";\n
//int subSwarmIx = globid & " << ((1 << swarmStrideLevel) - 1) << ";\n
//for (int swarmIx = 0; swarmIx < " << swarmSize << "; ++swarmIx) {\n
//int j = subSwarmIx + ((swarmIxOffset*" << swarmSize << " + swarmIx) << " << swarmStrideLevel << ");\n";
result << "int globid = get_global_id(0);"; // = phi*w + v \n
Array buff0 = Array("K", BGs[q_G], true, "buff0_");
Array buff1 = Array("K", BGs[q_G], true, "buff1_");
PlannarizedComplexCLArray in = PlannarizedComplexCLArray(GlobalPlannarLevel, "in", LG);
PlannarizedComplexCLArray out = PlannarizedComplexCLArray(GlobalPlannarLevel, "out", LG);
Array* buffs[2] = { &buff0, &buff1 };
result << buff0.getDeclaration() << "\n" << buff1.getDeclaration() << "\n";
result << " if (globid < " << getWorkItemsPerKernel(qG) << ") {\n\
int batchIx = (globid/" << getSwarmCount(qG) << ");\
int j = globid - batchIx*" << getSwarmCount(qG) << ";\
int gG = " << IntegerDivision("j", LG/NGs[q_G])() << ";\n\
int zG = j - " << LG/NGs[q_G] << "*gG + batchIx*" << LG << ";\n"; // zG = j % " << LG/NGs[q_G] << ";\n";
result << "int readStartOffset = zG + " << (LG/NGs[q_G]) * BGs[q_G] << "*gG;";
for (int sG = 0; sG < BGs[q_G]; ++sG) result << in.assignToItem("readStartOffset + " + intToStr(sG*(LG/NGs[q_G])), Array(buff0[sG]()))(); // Load data
if (qG > 1) for (int sG = 0; sG < BGs[q_G]; ++sG) result << buff0[sG]() << " = mul(" << buff0[sG]() << ", unit(" << -sG << "*gG, "<< NGs[q_G] << ")" << ");\n";
result << this->generateLocalFFTKernel(buffs, AL, BL, NL, BGs[q_G]); // Local FFT
result << "int writeStartOffset = zG + " << LG/NGs[q_G] << "*gG;";
for (int h = 0; h < BGs[q_G]; ++h) result << out.assignFromItem("writeStartOffset + " + intToStr(h*AGs[q_G]*(LG/NGs[q_G])), Array((*buffs[kL & 1])[h]()))(); // Write results
result << "}}";
return result.str();
}
virtual ~FFT_OpenCL_Contiguous() {
for (int d = 0; d < 2; ++d) delete data[d];
kernelEvents.clear();
for (int k = 0; k < (int) kernels.size(); ++k) delete kernels[k];
delete program;
}
};
#endif // FFT_OPENCL_CONTIGUOUS_H