|
2
|
1
|
|
|
2 //
|
|
|
3 // File: fft_internal.h
|
|
|
4 //
|
|
|
5 // Version: <1.0>
|
|
|
6 //
|
|
|
7 // Disclaimer: IMPORTANT: This Apple software is supplied to you by Apple Inc. ("Apple")
|
|
|
8 // in consideration of your agreement to the following terms, and your use,
|
|
|
9 // installation, modification or redistribution of this Apple software
|
|
|
10 // constitutes acceptance of these terms. If you do not agree with these
|
|
|
11 // terms, please do not use, install, modify or redistribute this Apple
|
|
|
12 // software.
|
|
|
13 //
|
|
|
14 // In consideration of your agreement to abide by the following terms, and
|
|
|
15 // subject to these terms, Apple grants you a personal, non - exclusive
|
|
|
16 // license, under Apple's copyrights in this original Apple software ( the
|
|
|
17 // "Apple Software" ), to use, reproduce, modify and redistribute the Apple
|
|
|
18 // Software, with or without modifications, in source and / or binary forms;
|
|
|
19 // provided that if you redistribute the Apple Software in its entirety and
|
|
|
20 // without modifications, you must retain this notice and the following text
|
|
|
21 // and disclaimers in all such redistributions of the Apple Software. Neither
|
|
|
22 // the name, trademarks, service marks or logos of Apple Inc. may be used to
|
|
|
23 // endorse or promote products derived from the Apple Software without specific
|
|
|
24 // prior written permission from Apple. Except as expressly stated in this
|
|
|
25 // notice, no other rights or licenses, express or implied, are granted by
|
|
|
26 // Apple herein, including but not limited to any patent rights that may be
|
|
|
27 // infringed by your derivative works or by other works in which the Apple
|
|
|
28 // Software may be incorporated.
|
|
|
29 //
|
|
|
30 // The Apple Software is provided by Apple on an "AS IS" basis. APPLE MAKES NO
|
|
|
31 // WARRANTIES, EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION THE IMPLIED
|
|
|
32 // WARRANTIES OF NON - INFRINGEMENT, MERCHANTABILITY AND FITNESS FOR A
|
|
|
33 // PARTICULAR PURPOSE, REGARDING THE APPLE SOFTWARE OR ITS USE AND OPERATION
|
|
|
34 // ALONE OR IN COMBINATION WITH YOUR PRODUCTS.
|
|
|
35 //
|
|
|
36 // IN NO EVENT SHALL APPLE BE LIABLE FOR ANY SPECIAL, INDIRECT, INCIDENTAL OR
|
|
|
37 // CONSEQUENTIAL DAMAGES ( INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
|
|
|
38 // SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
|
|
|
39 // INTERRUPTION ) ARISING IN ANY WAY OUT OF THE USE, REPRODUCTION, MODIFICATION
|
|
|
40 // AND / OR DISTRIBUTION OF THE APPLE SOFTWARE, HOWEVER CAUSED AND WHETHER
|
|
|
41 // UNDER THEORY OF CONTRACT, TORT ( INCLUDING NEGLIGENCE ), STRICT LIABILITY OR
|
|
|
42 // OTHERWISE, EVEN IF APPLE HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
43 //
|
|
|
44 // Copyright ( C ) 2008 Apple Inc. All Rights Reserved.
|
|
|
45 //
|
|
|
46 ////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
47
|
|
|
48
|
|
|
49 #ifndef __CLFFT_INTERNAL_H
|
|
|
50 #define __CLFFT_INTERNAL_H
|
|
|
51
|
|
|
52 #include "clFFT.h"
|
|
|
53 #include <iostream>
|
|
|
54 #include <string>
|
|
|
55 #include <sstream>
|
|
|
56
|
|
|
57 using namespace std;
|
|
|
58
|
|
|
59 typedef enum kernel_dir_t
|
|
|
60 {
|
|
|
61 cl_fft_kernel_x,
|
|
|
62 cl_fft_kernel_y,
|
|
|
63 cl_fft_kernel_z
|
|
|
64 }cl_fft_kernel_dir;
|
|
|
65
|
|
|
66 typedef struct kernel_info_t
|
|
|
67 {
|
|
|
68 cl_kernel kernel;
|
|
|
69 char *kernel_name;
|
|
|
70 unsigned lmem_size;
|
|
|
71 unsigned num_workgroups;
|
|
|
72 unsigned num_xforms_per_workgroup;
|
|
|
73 unsigned num_workitems_per_workgroup;
|
|
|
74 cl_fft_kernel_dir dir;
|
|
|
75 int in_place_possible;
|
|
|
76 kernel_info_t *next;
|
|
|
77 }cl_fft_kernel_info;
|
|
|
78
|
|
|
79 typedef struct
|
|
|
80 {
|
|
|
81 // context in which fft resources are created and kernels are executed
|
|
|
82 cl_context context;
|
|
|
83
|
|
|
84 // size of signal
|
|
|
85 clFFT_Dim3 n;
|
|
|
86
|
|
|
87 // dimension of transform ... must be either 1D, 2D or 3D
|
|
|
88 clFFT_Dimension dim;
|
|
|
89
|
|
|
90 // data format ... must be either interleaved or plannar
|
|
|
91 clFFT_DataFormat format;
|
|
|
92
|
|
|
93 // string containing kernel source. Generated at runtime based on
|
|
|
94 // n, dim, format and other parameters
|
|
|
95 string *kernel_string;
|
|
|
96
|
|
|
97 // CL program containing source and kernel this particular
|
|
|
98 // n, dim, data format
|
|
|
99 cl_program program;
|
|
|
100
|
|
|
101 // linked list of kernels which needs to be executed for this fft
|
|
|
102 cl_fft_kernel_info *kernel_info;
|
|
|
103
|
|
|
104 // number of kernels
|
|
|
105 int num_kernels;
|
|
|
106
|
|
|
107 // twist kernel for virtualizing fft of very large sizes that do not
|
|
|
108 // fit in GPU global memory
|
|
|
109 cl_kernel twist_kernel;
|
|
|
110
|
|
|
111 // flag indicating if temporary intermediate buffer is needed or not.
|
|
|
112 // this depends on fft kernels being executed and if transform is
|
|
|
113 // in-place or out-of-place. e.g. Local memory fft (say 1D 1024 ...
|
|
|
114 // one that does not require global transpose do not need temporary buffer)
|
|
|
115 // 2D 1024x1024 out-of-place fft however do require intermediate buffer.
|
|
|
116 // If temp buffer is needed, its allocation is lazy i.e. its not allocated
|
|
|
117 // until its needed
|
|
|
118 cl_int temp_buffer_needed;
|
|
|
119
|
|
|
120 // Batch size is runtime parameter and size of temporary buffer (if needed)
|
|
|
121 // depends on batch size. Allocation of temporary buffer is lazy i.e. its
|
|
|
122 // only created when needed. Once its created at first call of clFFT_Executexxx
|
|
|
123 // it is not allocated next time if next time clFFT_Executexxx is called with
|
|
|
124 // batch size different than the first call. last_batch_size caches the last
|
|
|
125 // batch size with which this plan is used so that we dont keep allocating/deallocating
|
|
|
126 // temp buffer if same batch size is used again and again.
|
|
|
127 unsigned last_batch_size;
|
|
|
128
|
|
|
129 // temporary buffer for interleaved plan
|
|
|
130 cl_mem tempmemobj;
|
|
|
131
|
|
|
132 // temporary buffer for planner plan. Only one of tempmemobj or
|
|
|
133 // (tempmemobj_real, tempmemobj_imag) pair is valid (allocated) depending
|
|
|
134 // data format of plan (plannar or interleaved)
|
|
|
135 cl_mem tempmemobj_real, tempmemobj_imag;
|
|
|
136
|
|
|
137 // Maximum size of signal for which local memory transposed based
|
|
|
138 // fft is sufficient i.e. no global mem transpose (communication)
|
|
|
139 // is needed
|
|
|
140 unsigned max_localmem_fft_size;
|
|
|
141
|
|
|
142 // Maximum work items per work group allowed. This, along with max_radix below controls
|
|
|
143 // maximum local memory being used by fft kernels of this plan. Set to 256 by default
|
|
|
144 unsigned max_work_item_per_workgroup;
|
|
|
145
|
|
|
146 // Maximum base radix for local memory fft ... this controls the maximum register
|
|
|
147 // space used by work items. Currently defaults to 16
|
|
|
148 unsigned max_radix;
|
|
|
149
|
|
|
150 // Device depended parameter that tells how many work-items need to be read consecutive
|
|
|
151 // values to make sure global memory access by work-items of a work-group result in
|
|
|
152 // coalesced memory access to utilize full bandwidth e.g. on NVidia tesla, this is 16
|
|
|
153 unsigned min_mem_coalesce_width;
|
|
|
154
|
|
|
155 // Number of local memory banks. This is used to geneate kernel with local memory
|
|
|
156 // transposes with appropriate padding to avoid bank conflicts to local memory
|
|
|
157 // e.g. on NVidia it is 16.
|
|
|
158 unsigned num_local_mem_banks;
|
|
|
159 }cl_fft_plan;
|
|
|
160
|
|
|
161 void FFT1D(cl_fft_plan *plan, cl_fft_kernel_dir dir);
|
|
|
162
|
|
|
163 #endif
|
