GatedSpectrometer.cu 10.5 KB
Newer Older
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
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
#include "psrdada_cpp/effelsberg/edd/GatedSpectrometer.cuh"


namespace psrdada_cpp {
namespace effelsberg {
namespace edd {


__global__ void mergeSideChannels(uint64_t* __restrict__ A, uint64_t*
        __restrict__ B, size_t N)
{
  for (size_t i = blockIdx.x * blockDim.x + threadIdx.x; (i < N);
       i += blockDim.x * gridDim.x)
  {
    uint64_t v = A[i] || B[i];
    A[i] = v;
    B[i] = v;
  }
}


__global__ void gating(float* __restrict__ G0,
        float* __restrict__ G1,
        const uint64_t* __restrict__ sideChannelData,
        size_t N, size_t heapSize, size_t bitpos,
        size_t noOfSideChannels, size_t selectedSideChannel,
        const float*  __restrict__ _baseLineG0,
        const float*  __restrict__ _baseLineG1,
        float* __restrict__ baseLineNG0,
        float* __restrict__ baseLineNG1,
        uint64_cu* stats_G0, uint64_cu* stats_G1) {
  // statistics values for samopels to G0, G1
  uint32_t _G0stats = 0;
  uint32_t _G1stats = 0;

  const float baseLineG0 = _baseLineG0[0];
  const float baseLineG1 = _baseLineG1[0];

  float baselineUpdateG0 = 0;
  float baselineUpdateG1 = 0;

  for (size_t i = blockIdx.x * blockDim.x + threadIdx.x; (i < N);
       i += blockDim.x * gridDim.x) {
    const float v = G0[i];

    const uint64_t sideChannelItem = sideChannelData[((i / heapSize) * (noOfSideChannels)) +
                        selectedSideChannel];

    const unsigned int bit_set = TEST_BIT(sideChannelItem, bitpos);
    const unsigned int heap_lost = TEST_BIT(sideChannelItem, 63);
    G1[i] = (v - baseLineG1) * bit_set * (!heap_lost) + baseLineG1;
    G0[i] = (v - baseLineG0) * (!bit_set) *(!heap_lost) + baseLineG0;

    _G0stats += (!bit_set) *(!heap_lost);
    _G1stats += bit_set * (!heap_lost);

    baselineUpdateG1 += v * bit_set * (!heap_lost);
    baselineUpdateG0 += v * (!bit_set) *(!heap_lost);
  }

  __shared__ uint32_t x[1024];

  // Reduce G0, G1
  sum_reduce<uint32_t>(x, _G0stats);
  if(threadIdx.x == 0) {
    atomicAdd(stats_G0,  (uint64_cu) x[threadIdx.x]);
  }
  __syncthreads();

  sum_reduce<uint32_t>(x, _G1stats);
  if(threadIdx.x == 0) {
    atomicAdd(stats_G1,  (uint64_cu) x[threadIdx.x]);
  }
  __syncthreads();

  //reuse shared array
  float *y = (float*) x;
  //update the baseline array
  sum_reduce<float>(y, baselineUpdateG0);
  if(threadIdx.x == 0) {
    atomicAdd(baseLineNG0, y[threadIdx.x]);
  }
  __syncthreads();

  sum_reduce<float>(y, baselineUpdateG1);
  if(threadIdx.x == 0) {
    atomicAdd(baseLineNG1, y[threadIdx.x]);
  }
  __syncthreads();
}



// Updates the baselines of the gates for the polarization set for the next
// block
// only few output blocks per input block thus execution on only one thread.
// Important is that the execution is async on the GPU.
__global__ void update_baselines(float*  __restrict__ baseLineG0,
        float*  __restrict__ baseLineG1,
        float* __restrict__ baseLineNG0,
        float* __restrict__ baseLineNG1,
        uint64_cu* stats_G0, uint64_cu* stats_G1,
        size_t N)
{
    size_t NG0 = 0;
    size_t NG1 = 0;

    for (size_t i =0; i < N; i++)
    {
       NG0 += stats_G0[i];
       NG1 += stats_G1[i];
    }

    baseLineG0[0] = baseLineNG0[0] / NG0;
    baseLineG1[0] = baseLineNG1[0] / NG1;
    baseLineNG0[0] = 0;
    baseLineNG1[0] = 0;
}







PolarizationData::PolarizationData(size_t fft_length, size_t batch, const DadaBufferLayout
        &dadaBufferLayout) : _fft_length(fft_length), _batch(batch), _dadaBufferLayout(dadaBufferLayout)
{
    _raw_voltage.resize(_dadaBufferLayout.sizeOfData() / sizeof(uint64_t));
    BOOST_LOG_TRIVIAL(debug) << "  Input voltages size (in 64-bit words): " << _raw_voltage.size();

    _baseLineG0.resize(1);
    _baseLineG0_update.resize(1);
    _baseLineG1.resize(1);
    _baseLineG1_update.resize(1);
    _channelised_voltage_G0.resize((_fft_length / 2 + 1) * _batch);
    _channelised_voltage_G1.resize((_fft_length / 2 + 1) * _batch);
    _sideChannelData.resize(_dadaBufferLayout.getNSideChannels() * _dadaBufferLayout.getNHeaps());
    BOOST_LOG_TRIVIAL(debug) << "  Channelised voltages size: " << _channelised_voltage_G0.size();
};


void PolarizationData::swap()
{
    _raw_voltage.swap();
    _sideChannelData.swap();
}


void PolarizationData::getFromBlock(RawBytes &block, cudaStream_t &_h2d_stream)
{
  BOOST_LOG_TRIVIAL(debug) << "   block.used_bytes() = " << block.used_bytes()
                           << ", dataBlockBytes = " << _dadaBufferLayout.sizeOfData() << "\n";

  CUDA_ERROR_CHECK(cudaMemcpyAsync(static_cast<void *>(_raw_voltage.a_ptr()),
                                   static_cast<void *>(block.ptr()),
                                   _dadaBufferLayout.sizeOfData() , cudaMemcpyHostToDevice,
                                   _h2d_stream));
  CUDA_ERROR_CHECK(cudaMemcpyAsync(
      static_cast<void *>(_sideChannelData.a_ptr()),
      static_cast<void *>(block.ptr() + _dadaBufferLayout.sizeOfData() + _dadaBufferLayout.sizeOfGap()),
      _dadaBufferLayout.sizeOfSideChannelData(), cudaMemcpyHostToDevice, _h2d_stream));
  BOOST_LOG_TRIVIAL(debug) << "First side channel item: 0x" <<   std::setw(16)
      << std::setfill('0') << std::hex <<
      (reinterpret_cast<uint64_t*>(block.ptr() + _dadaBufferLayout.sizeOfData()
                                   + _dadaBufferLayout.sizeOfGap()))[0] <<
      std::dec;
}


DualPolarizationData::DualPolarizationData(size_t fft_length, size_t batch, const DadaBufferLayout
        &dadaBufferLayout) : polarization0(fft_length, batch, dadaBufferLayout),
                            polarization1(fft_length, batch, dadaBufferLayout),
                            _dadaBufferLayout(dadaBufferLayout)
{
};

void DualPolarizationData::swap()
{
    polarization0.swap(); polarization1.swap();
}

void DualPolarizationData::getFromBlock(RawBytes &block, cudaStream_t &_h2d_stream)
{
// Copy the data with stride to the GPU:
// CPU: P1P2P1P2P1P2 ...
// GPU: P1P1P1 ... P2P2P2 ...
// If this is a bottleneck the gating kernel could sort the layout out
// during copy
int heapsize_bytes =  _dadaBufferLayout.getHeapSize();
CUDA_ERROR_CHECK(cudaMemcpy2DAsync(
  static_cast<void *>(polarization0._raw_voltage.a_ptr()),
    heapsize_bytes,
    static_cast<void *>(block.ptr()),
    2 * heapsize_bytes,
    heapsize_bytes, _dadaBufferLayout.sizeOfData() / heapsize_bytes/ 2,
    cudaMemcpyHostToDevice, _h2d_stream));

CUDA_ERROR_CHECK(cudaMemcpy2DAsync(
  static_cast<void *>(polarization1._raw_voltage.a_ptr()),
    heapsize_bytes,
    static_cast<void *>(block.ptr()) + heapsize_bytes,
    2 * heapsize_bytes,
    heapsize_bytes, _dadaBufferLayout.sizeOfData() / heapsize_bytes/ 2,
    cudaMemcpyHostToDevice, _h2d_stream));

CUDA_ERROR_CHECK(cudaMemcpy2DAsync(
    static_cast<void *>(polarization0._sideChannelData.a_ptr()),
    sizeof(uint64_t),
    static_cast<void *>(block.ptr() + _dadaBufferLayout.sizeOfData() + _dadaBufferLayout.sizeOfGap()),
    2 * sizeof(uint64_t),
    sizeof(uint64_t),
    _dadaBufferLayout.sizeOfSideChannelData() / 2 / sizeof(uint64_t),
    cudaMemcpyHostToDevice, _h2d_stream));

CUDA_ERROR_CHECK(cudaMemcpy2DAsync(
    static_cast<void *>(polarization1._sideChannelData.a_ptr()),
    sizeof(uint64_t),
    static_cast<void *>(block.ptr() + _dadaBufferLayout.sizeOfData() + _dadaBufferLayout.sizeOfGap() + sizeof(uint64_t)),
    2 * sizeof(uint64_t),
    sizeof(uint64_t),
    _dadaBufferLayout.sizeOfSideChannelData() / 2 / sizeof(uint64_t), cudaMemcpyHostToDevice, _h2d_stream));

BOOST_LOG_TRIVIAL(debug) << "First side channel item: 0x" <<   std::setw(16)
    << std::setfill('0') << std::hex <<
    (reinterpret_cast<uint64_t*>(block.ptr() + _dadaBufferLayout.sizeOfData()
                                 + _dadaBufferLayout.sizeOfGap()))[0] << std::dec;
}



PowerSpectrumOutput::PowerSpectrumOutput(size_t size, size_t blocks)
{
    BOOST_LOG_TRIVIAL(debug) << "Setting size of power spectrum output size = " << size << ", blocks =  " << blocks;
   data.resize(size * blocks);
   _noOfBitSets.resize(blocks);
}


void PowerSpectrumOutput::reset(cudaStream_t &_proc_stream)
{
    thrust::fill(thrust::cuda::par.on(_proc_stream), data.a().begin(), data.a().end(), 0.);
    thrust::fill(thrust::cuda::par.on(_proc_stream), _noOfBitSets.a().begin(), _noOfBitSets.a().end(), 0L);
}


void PowerSpectrumOutput::swap()
{
    data.swap();
    _noOfBitSets.swap();
}


GatedPowerSpectrumOutput::GatedPowerSpectrumOutput(size_t nchans, size_t
        blocks) : OutputDataStream(nchans, blocks), G0(nchans, blocks),
G1(nchans, blocks)
{
  // on the host both power are stored in the same data buffer together with
  // the number of bit sets
  _host_power.resize( 2 * ( _nchans * sizeof(IntegratedPowerType) + sizeof(size_t) ) * G0._noOfBitSets.size());
}


void GatedPowerSpectrumOutput::reset(cudaStream_t &_proc_stream)
{
    G0.reset(_proc_stream);
    G1.reset(_proc_stream);
}


/// Swap output buffers
void GatedPowerSpectrumOutput::swap()
{
    G0.swap();
    G1.swap();
    _host_power.swap();
}


void GatedPowerSpectrumOutput::data2Host(cudaStream_t &_d2h_stream)
{
    // copy data to host if block is finished
  CUDA_ERROR_CHECK(cudaStreamSynchronize(_d2h_stream));

  for (size_t i = 0; i < G0._noOfBitSets.size(); i++)
  {
    // size of individual spectrum + meta
    size_t memslicesize = (_nchans * sizeof(IntegratedPowerType));
    // number of spectra per output
    size_t memOffset = 2 * i * (memslicesize +  + sizeof(size_t));

    // copy 2x channel data
    CUDA_ERROR_CHECK(
        cudaMemcpyAsync(static_cast<void *>(_host_power.a_ptr() + memOffset) ,
                        static_cast<void *>(G0.data.b_ptr() + i * memslicesize),
                        _nchans * sizeof(IntegratedPowerType),
                        cudaMemcpyDeviceToHost, _d2h_stream));

    CUDA_ERROR_CHECK(
        cudaMemcpyAsync(static_cast<void *>(_host_power.a_ptr() + memOffset + 1 * memslicesize) ,
                        static_cast<void *>(G1.data.b_ptr() + i * memslicesize),
                        _nchans * sizeof(IntegratedPowerType),
                        cudaMemcpyDeviceToHost, _d2h_stream));

    // copy noOf bit set data
    CUDA_ERROR_CHECK(
        cudaMemcpyAsync( static_cast<void *>(_host_power.a_ptr() + memOffset + 2 * _nchans * sizeof(IntegratedPowerType)),
          static_cast<void *>(G0._noOfBitSets.b_ptr() + i ),
            1 * sizeof(size_t),
            cudaMemcpyDeviceToHost, _d2h_stream));

    CUDA_ERROR_CHECK(
        cudaMemcpyAsync( static_cast<void *>(_host_power.a_ptr() + memOffset + 2 * _nchans * sizeof(IntegratedPowerType) + sizeof(size_t)),
          static_cast<void *>(G1._noOfBitSets.b_ptr() + i ),
            1 * sizeof(size_t),
            cudaMemcpyDeviceToHost, _d2h_stream));
  }
}





}}} // namespace