fft_execute.cpp 13.1 KB
Newer Older
Gaurav Khanna's avatar
Gaurav Khanna committed
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
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
#include "fft_internal.h" 
#include "clFFT.h" 
#include <stdlib.h> 
#include <stdio.h> 
#include <math.h> 
  
#define max(a,b) (((a)>(b)) ? (a) : (b)) 
#define min(a,b) (((a)<(b)) ? (a) : (b)) 
  
static cl_int 
allocateTemporaryBufferInterleaved(cl_fft_plan *plan, cl_uint batchSize) 
{ 
    cl_int err = CL_SUCCESS; 
    if(plan->temp_buffer_needed && plan->last_batch_size != batchSize)  
    { 
        plan->last_batch_size = batchSize;  
        size_t tmpLength = plan->n.x * plan->n.y * plan->n.z * batchSize * 2 * sizeof(cl_float); 
         
        if(plan->tempmemobj) 
            clReleaseMemObject(plan->tempmemobj); 
             
        plan->tempmemobj = clCreateBuffer(plan->context, CL_MEM_READ_WRITE, tmpLength, NULL, &err); 
    } 
    return err;  
} 
  
static cl_int 
allocateTemporaryBufferPlannar(cl_fft_plan *plan, cl_uint batchSize) 
{ 
    cl_int err = CL_SUCCESS; 
    cl_int terr; 
    if(plan->temp_buffer_needed && plan->last_batch_size != batchSize)  
    { 
        plan->last_batch_size = batchSize;  
        size_t tmpLength = plan->n.x * plan->n.y * plan->n.z * batchSize * sizeof(cl_float); 
         
        if(plan->tempmemobj_real) 
            clReleaseMemObject(plan->tempmemobj_real); 
  
        if(plan->tempmemobj_imag) 
            clReleaseMemObject(plan->tempmemobj_imag);           
             
        plan->tempmemobj_real = clCreateBuffer(plan->context, CL_MEM_READ_WRITE, tmpLength, NULL, &err); 
        plan->tempmemobj_imag = clCreateBuffer(plan->context, CL_MEM_READ_WRITE, tmpLength, NULL, &terr); 
        err |= terr; 
    }    
    return err; 
} 
  
void 
getKernelWorkDimensions(cl_fft_plan *plan, cl_fft_kernel_info *kernelInfo, cl_int *batchSize, size_t *gWorkItems, size_t *lWorkItems) 
{ 
    *lWorkItems = kernelInfo->num_workitems_per_workgroup; 
    int numWorkGroups = kernelInfo->num_workgroups; 
    int numXFormsPerWG = kernelInfo->num_xforms_per_workgroup; 
     
    switch(kernelInfo->dir) 
    { 
        case cl_fft_kernel_x: 
            *batchSize *= (plan->n.y * plan->n.z); 
            numWorkGroups = (*batchSize % numXFormsPerWG) ? (*batchSize/numXFormsPerWG + 1) : (*batchSize/numXFormsPerWG); 
            numWorkGroups *= kernelInfo->num_workgroups; 
            break; 
        case cl_fft_kernel_y: 
            *batchSize *= plan->n.z; 
            numWorkGroups *= *batchSize; 
            break; 
        case cl_fft_kernel_z: 
            numWorkGroups *= *batchSize; 
            break; 
    } 
     
    *gWorkItems = numWorkGroups * *lWorkItems; 
} 
  
cl_int  
clFFT_ExecuteInterleaved( cl_command_queue queue, clFFT_Plan Plan, cl_int batchSize, clFFT_Direction dir,  
                         cl_mem data_in, cl_mem data_out,  
                         cl_int num_events, cl_event *event_list, cl_event *event ) 
{    
    int s; 
    cl_fft_plan *plan = (cl_fft_plan *) Plan; 
    if(plan->format != clFFT_InterleavedComplexFormat) 
        return CL_INVALID_VALUE; 
     
    cl_int err; 
    size_t gWorkItems, lWorkItems; 
    int inPlaceDone; 
     
    cl_int isInPlace = data_in == data_out ? 1 : 0; 
     
    if((err = allocateTemporaryBufferInterleaved(plan, batchSize)) != CL_SUCCESS) 
        return err;  
     
    cl_mem memObj[3]; 
    memObj[0] = data_in; 
    memObj[1] = data_out; 
    memObj[2] = plan->tempmemobj; 
    cl_fft_kernel_info *kernelInfo = plan->kernel_info; 
    int numKernels = plan->num_kernels; 
     
    int numKernelsOdd = numKernels & 1; 
    int currRead  = 0; 
    int currWrite = 1; 
     
    // at least one external dram shuffle (transpose) required 
    if(plan->temp_buffer_needed)  
    { 
        // in-place transform 
        if(isInPlace)  
        { 
            inPlaceDone = 0; 
            currRead  = 1; 
            currWrite = 2; 
        } 
        else 
        { 
            currWrite = (numKernels & 1) ? 1 : 2; 
        } 
         
        while(kernelInfo)  
        { 
            if( isInPlace && numKernelsOdd && !inPlaceDone && kernelInfo->in_place_possible)  
            { 
                currWrite = currRead; 
                inPlaceDone = 1; 
            } 
             
            s = batchSize; 
            getKernelWorkDimensions(plan, kernelInfo, &s, &gWorkItems, &lWorkItems); 
            err |= clSetKernelArg(kernelInfo->kernel, 0, sizeof(cl_mem), &memObj[currRead]); 
            err |= clSetKernelArg(kernelInfo->kernel, 1, sizeof(cl_mem), &memObj[currWrite]); 
            err |= clSetKernelArg(kernelInfo->kernel, 2, sizeof(cl_int), &dir); 
            err |= clSetKernelArg(kernelInfo->kernel, 3, sizeof(cl_int), &s); 
             
            err |= clEnqueueNDRangeKernel(queue,  kernelInfo->kernel, 1, NULL, &gWorkItems, &lWorkItems, 0, NULL, NULL); 
            if(err) 
                return err; 
             
            currRead  = (currWrite == 1) ? 1 : 2; 
            currWrite = (currWrite == 1) ? 2 : 1;  
             
            kernelInfo = kernelInfo->next; 
        }            
    } 
    // no dram shuffle (transpose required) transform 
    // all kernels can execute in-place. 
    else { 
         
        while(kernelInfo) 
        { 
            s = batchSize; 
            getKernelWorkDimensions(plan, kernelInfo, &s, &gWorkItems, &lWorkItems); 
            err |= clSetKernelArg(kernelInfo->kernel, 0, sizeof(cl_mem), &memObj[currRead]); 
            err |= clSetKernelArg(kernelInfo->kernel, 1, sizeof(cl_mem), &memObj[currWrite]); 
            err |= clSetKernelArg(kernelInfo->kernel, 2, sizeof(cl_int), &dir); 
            err |= clSetKernelArg(kernelInfo->kernel, 3, sizeof(cl_int), &s); 
         
            err |= clEnqueueNDRangeKernel(queue,  kernelInfo->kernel, 1, NULL, &gWorkItems, &lWorkItems, 0, NULL, NULL); 
            if(err) 
                return err;      
             
            currRead  = 1; 
            currWrite = 1; 
             
            kernelInfo = kernelInfo->next; 
        } 
    } 
     
    return err; 
} 
  
cl_int  
clFFT_ExecutePlannar( cl_command_queue queue, clFFT_Plan Plan, cl_int batchSize, clFFT_Direction dir,  
                      cl_mem data_in_real, cl_mem data_in_imag, cl_mem data_out_real, cl_mem data_out_imag, 
                      cl_int num_events, cl_event *event_list, cl_event *event) 
{    
    int s; 
    cl_fft_plan *plan = (cl_fft_plan *) Plan; 
     
    if(plan->format != clFFT_SplitComplexFormat) 
        return CL_INVALID_VALUE; 
     
    cl_int err; 
    size_t gWorkItems, lWorkItems; 
    int inPlaceDone; 
     
    cl_int isInPlace = ((data_in_real == data_out_real) && (data_in_imag == data_out_imag)) ? 1 : 0; 
     
    if((err = allocateTemporaryBufferPlannar(plan, batchSize)) != CL_SUCCESS) 
        return err;  
     
    cl_mem memObj_real[3]; 
    cl_mem memObj_imag[3]; 
    memObj_real[0] = data_in_real; 
    memObj_real[1] = data_out_real; 
    memObj_real[2] = plan->tempmemobj_real; 
    memObj_imag[0] = data_in_imag; 
    memObj_imag[1] = data_out_imag; 
    memObj_imag[2] = plan->tempmemobj_imag; 
         
    cl_fft_kernel_info *kernelInfo = plan->kernel_info; 
    int numKernels = plan->num_kernels; 
     
    int numKernelsOdd = numKernels & 1; 
    int currRead  = 0; 
    int currWrite = 1; 
     
    // at least one external dram shuffle (transpose) required 
    if(plan->temp_buffer_needed)  
    { 
        // in-place transform 
        if(isInPlace)  
        { 
            inPlaceDone = 0; 
            currRead  = 1; 
            currWrite = 2; 
        } 
        else 
        { 
            currWrite = (numKernels & 1) ? 1 : 2; 
        } 
         
        while(kernelInfo)  
        { 
            if( isInPlace && numKernelsOdd && !inPlaceDone && kernelInfo->in_place_possible)  
            { 
                currWrite = currRead; 
                inPlaceDone = 1; 
            } 
             
            s = batchSize; 
            getKernelWorkDimensions(plan, kernelInfo, &s, &gWorkItems, &lWorkItems); 
            err |= clSetKernelArg(kernelInfo->kernel, 0, sizeof(cl_mem), &memObj_real[currRead]); 
            err |= clSetKernelArg(kernelInfo->kernel, 1, sizeof(cl_mem), &memObj_imag[currRead]); 
            err |= clSetKernelArg(kernelInfo->kernel, 2, sizeof(cl_mem), &memObj_real[currWrite]); 
            err |= clSetKernelArg(kernelInfo->kernel, 3, sizeof(cl_mem), &memObj_imag[currWrite]); 
            err |= clSetKernelArg(kernelInfo->kernel, 4, sizeof(cl_int), &dir); 
            err |= clSetKernelArg(kernelInfo->kernel, 5, sizeof(cl_int), &s); 
             
            err |= clEnqueueNDRangeKernel(queue,  kernelInfo->kernel, 1, NULL, &gWorkItems, &lWorkItems, 0, NULL, NULL); 
            if(err) 
                return err;          
             
            currRead  = (currWrite == 1) ? 1 : 2; 
            currWrite = (currWrite == 1) ? 2 : 1;  
             
            kernelInfo = kernelInfo->next; 
        }            
    } 
    // no dram shuffle (transpose required) transform 
    else { 
         
        while(kernelInfo) 
        { 
            s = batchSize; 
            getKernelWorkDimensions(plan, kernelInfo, &s, &gWorkItems, &lWorkItems); 
            err |= clSetKernelArg(kernelInfo->kernel, 0, sizeof(cl_mem), &memObj_real[currRead]); 
            err |= clSetKernelArg(kernelInfo->kernel, 1, sizeof(cl_mem), &memObj_imag[currRead]); 
            err |= clSetKernelArg(kernelInfo->kernel, 2, sizeof(cl_mem), &memObj_real[currWrite]); 
            err |= clSetKernelArg(kernelInfo->kernel, 3, sizeof(cl_mem), &memObj_imag[currWrite]); 
            err |= clSetKernelArg(kernelInfo->kernel, 4, sizeof(cl_int), &dir); 
            err |= clSetKernelArg(kernelInfo->kernel, 5, sizeof(cl_int), &s); 
         
            err |= clEnqueueNDRangeKernel(queue,  kernelInfo->kernel, 1, NULL, &gWorkItems, &lWorkItems, 0, NULL, NULL); 
            if(err) 
                return err;  
             
            currRead  = 1; 
            currWrite = 1; 
         
            kernelInfo = kernelInfo->next; 
        } 
    } 
     
    return err; 
} 
  
cl_int  
clFFT_1DTwistInterleaved(clFFT_Plan Plan, cl_command_queue queue, cl_mem array,  
                         size_t numRows, size_t numCols, size_t startRow, size_t rowsToProcess, clFFT_Direction dir) 
{ 
 puts ("X"); 
    cl_fft_plan *plan = (cl_fft_plan *) Plan; 
     
    unsigned int N = numRows*numCols; 
    unsigned int nCols = numCols; 
    unsigned int sRow = startRow; 
    unsigned int rToProcess = rowsToProcess; 
    int d = dir; 
    int err = 0; 
     
    cl_device_id device_id; 
    err = clGetCommandQueueInfo(queue, CL_QUEUE_DEVICE, sizeof(cl_device_id), &device_id, NULL); 
    if(err) 
        return err; 
     
    size_t gSize; 
    err = clGetKernelWorkGroupInfo(plan->twist_kernel, device_id, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &gSize, NULL); 
    if(err) 
        return err; 
           
    gSize = min(128, gSize); 
    size_t numGlobalThreads[1] = { max(numCols / gSize, 1)*gSize }; 
    size_t numLocalThreads[1]  = { gSize }; 
     
    err |= clSetKernelArg(plan->twist_kernel, 0, sizeof(cl_mem), &array); 
    err |= clSetKernelArg(plan->twist_kernel, 1, sizeof(unsigned int), &sRow); 
    err |= clSetKernelArg(plan->twist_kernel, 2, sizeof(unsigned int), &nCols); 
    err |= clSetKernelArg(plan->twist_kernel, 3, sizeof(unsigned int), &N); 
    err |= clSetKernelArg(plan->twist_kernel, 4, sizeof(unsigned int), &rToProcess); 
    err |= clSetKernelArg(plan->twist_kernel, 5, sizeof(int), &d); 
     
    err |= clEnqueueNDRangeKernel(queue, plan->twist_kernel, 1, NULL, numGlobalThreads, numLocalThreads, 0, NULL, NULL);             
     
    return err;  
} 
cl_int  
clFFT_1DTwistPlannar(clFFT_Plan Plan, cl_command_queue queue, cl_mem array_real, cl_mem array_imag,  
                     size_t numRows, size_t numCols, size_t startRow, size_t rowsToProcess, clFFT_Direction dir) 
{ 
 puts ("Y"); 
    cl_fft_plan *plan = (cl_fft_plan *) Plan; 
     
    unsigned int N = numRows*numCols; 
    unsigned int nCols = numCols; 
    unsigned int sRow = startRow; 
    unsigned int rToProcess = rowsToProcess; 
    int d = dir; 
    int err = 0; 
     
    cl_device_id device_id; 
    err = clGetCommandQueueInfo(queue, CL_QUEUE_DEVICE, sizeof(cl_device_id), &device_id, NULL); 
    if(err) 
        return err; 
     
    size_t gSize; 
    err = clGetKernelWorkGroupInfo(plan->twist_kernel, device_id, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &gSize, NULL); 
    if(err) 
        return err; 
           
    gSize = min(128, gSize); 
    size_t numGlobalThreads[1] = { max(numCols / gSize, 1)*gSize }; 
    size_t numLocalThreads[1]  = { gSize }; 
     
    err |= clSetKernelArg(plan->twist_kernel, 0, sizeof(cl_mem), &array_real); 
    err |= clSetKernelArg(plan->twist_kernel, 1, sizeof(cl_mem), &array_imag); 
    err |= clSetKernelArg(plan->twist_kernel, 2, sizeof(unsigned int), &sRow); 
    err |= clSetKernelArg(plan->twist_kernel, 3, sizeof(unsigned int), &nCols); 
    err |= clSetKernelArg(plan->twist_kernel, 4, sizeof(unsigned int), &N); 
    err |= clSetKernelArg(plan->twist_kernel, 5, sizeof(unsigned int), &rToProcess); 
    err |= clSetKernelArg(plan->twist_kernel, 6, sizeof(int), &d); 
     
    err |= clEnqueueNDRangeKernel(queue, plan->twist_kernel, 1, NULL, numGlobalThreads, numLocalThreads, 0, NULL, NULL);             
     
    return err;  
}