File size: 17,891 Bytes
be11144
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
/******************************************************************************
 * Copyright (c) 2011, Duane Merrill.  All rights reserved.
 * Copyright (c) 2011-2018, NVIDIA CORPORATION.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *     * Redistributions of source code must retain the above copyright
 *       notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copyright
 *       notice, this list of conditions and the following disclaimer in the
 *       documentation and/or other materials provided with the distribution.
 *     * Neither the name of the NVIDIA CORPORATION nor the
 *       names of its contributors may be used to endorse or promote products
 *       derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE FOR ANY
 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 ******************************************************************************/

/**
 * \file
 * Thread utilities for writing memory using PTX cache modifiers.
 */

#pragma once

#include "../config.cuh"
#include "../util_ptx.cuh"
#include "../util_type.cuh"

/// Optional outer namespace(s)
CUB_NS_PREFIX

/// CUB namespace
namespace cub {

/**
 * \addtogroup UtilIo
 * @{
 */


//-----------------------------------------------------------------------------
// Tags and constants
//-----------------------------------------------------------------------------

/**
 * \brief Enumeration of cache modifiers for memory store operations.
 */
enum CacheStoreModifier
{
    STORE_DEFAULT,              ///< Default (no modifier)
    STORE_WB,                   ///< Cache write-back all coherent levels
    STORE_CG,                   ///< Cache at global level
    STORE_CS,                   ///< Cache streaming (likely to be accessed once)
    STORE_WT,                   ///< Cache write-through (to system memory)
    STORE_VOLATILE,             ///< Volatile shared (any memory space)
};


/**
 * \name Thread I/O (cache modified)
 * @{
 */

/**
 * \brief Thread utility for writing memory using cub::CacheStoreModifier cache modifiers.  Can be used to store any data type.
 *
 * \par Example
 * \code
 * #include <cub/cub.cuh>   // or equivalently <cub/thread/thread_store.cuh>
 *
 * // 32-bit store using cache-global modifier:
 * int *d_out;
 * int val;
 * cub::ThreadStore<cub::STORE_CG>(d_out + threadIdx.x, val);
 *
 * // 16-bit store using default modifier
 * short *d_out;
 * short val;
 * cub::ThreadStore<cub::STORE_DEFAULT>(d_out + threadIdx.x, val);
 *
 * // 256-bit store using write-through modifier
 * double4 *d_out;
 * double4 val;
 * cub::ThreadStore<cub::STORE_WT>(d_out + threadIdx.x, val);
 *
 * // 96-bit store using cache-streaming cache modifier
 * struct TestFoo { bool a; short b; };
 * TestFoo *d_struct;
 * TestFoo val;
 * cub::ThreadStore<cub::STORE_CS>(d_out + threadIdx.x, val);
 * \endcode
 *
 * \tparam MODIFIER             <b>[inferred]</b> CacheStoreModifier enumeration
 * \tparam InputIteratorT       <b>[inferred]</b> Output iterator type \iterator
 * \tparam T                    <b>[inferred]</b> Data type of output value
 */
template <
    CacheStoreModifier  MODIFIER,
    typename            OutputIteratorT,
    typename            T>
__device__ __forceinline__ void ThreadStore(OutputIteratorT itr, T val);


//@}  end member group


#ifndef DOXYGEN_SHOULD_SKIP_THIS    // Do not document


/// Helper structure for templated store iteration (inductive case)
template <int COUNT, int MAX>
struct IterateThreadStore
{
    template <CacheStoreModifier MODIFIER, typename T>
    static __device__ __forceinline__ void Store(T *ptr, T *vals)
    {
        ThreadStore<MODIFIER>(ptr + COUNT, vals[COUNT]);
        IterateThreadStore<COUNT + 1, MAX>::template Store<MODIFIER>(ptr, vals);
    }

    template <typename OutputIteratorT, typename T>
    static __device__ __forceinline__ void Dereference(OutputIteratorT ptr, T *vals)
    {
        ptr[COUNT] = vals[COUNT];
        IterateThreadStore<COUNT + 1, MAX>::Dereference(ptr, vals);
    }

};

/// Helper structure for templated store iteration (termination case)
template <int MAX>
struct IterateThreadStore<MAX, MAX>
{
    template <CacheStoreModifier MODIFIER, typename T>
    static __device__ __forceinline__ void Store(T * /*ptr*/, T * /*vals*/) {}

    template <typename OutputIteratorT, typename T>
    static __device__ __forceinline__ void Dereference(OutputIteratorT /*ptr*/, T * /*vals*/) {}
};


/**
 * Define a uint4 (16B) ThreadStore specialization for the given Cache load modifier
 */
#define _CUB_STORE_16(cub_modifier, ptx_modifier)                                            \
    template<>                                                                              \
    __device__ __forceinline__ void ThreadStore<cub_modifier, uint4*, uint4>(uint4* ptr, uint4 val)                         \
    {                                                                                       \
        asm volatile ("st."#ptx_modifier".v4.u32 [%0], {%1, %2, %3, %4};" : :               \
            _CUB_ASM_PTR_(ptr),                                                             \
            "r"(val.x),                                                                     \
            "r"(val.y),                                                                     \
            "r"(val.z),                                                                     \
            "r"(val.w));                                                                    \
    }                                                                                       \
    template<>                                                                              \
    __device__ __forceinline__ void ThreadStore<cub_modifier, ulonglong2*, ulonglong2>(ulonglong2* ptr, ulonglong2 val)     \
    {                                                                                       \
        asm volatile ("st."#ptx_modifier".v2.u64 [%0], {%1, %2};" : :                       \
            _CUB_ASM_PTR_(ptr),                                                             \
            "l"(val.x),                                                                     \
            "l"(val.y));                                                                    \
    }


/**
 * Define a uint2 (8B) ThreadStore specialization for the given Cache load modifier
 */
#define _CUB_STORE_8(cub_modifier, ptx_modifier)                                             \
    template<>                                                                              \
    __device__ __forceinline__ void ThreadStore<cub_modifier, ushort4*, ushort4>(ushort4* ptr, ushort4 val)                 \
    {                                                                                       \
        asm volatile ("st."#ptx_modifier".v4.u16 [%0], {%1, %2, %3, %4};" : :               \
            _CUB_ASM_PTR_(ptr),                                                             \
            "h"(val.x),                                                                     \
            "h"(val.y),                                                                     \
            "h"(val.z),                                                                     \
            "h"(val.w));                                                                    \
    }                                                                                       \
    template<>                                                                              \
    __device__ __forceinline__ void ThreadStore<cub_modifier, uint2*, uint2>(uint2* ptr, uint2 val)                         \
    {                                                                                       \
        asm volatile ("st."#ptx_modifier".v2.u32 [%0], {%1, %2};" : :                       \
            _CUB_ASM_PTR_(ptr),                                                             \
            "r"(val.x),                                                                     \
            "r"(val.y));                                                                    \
    }                                                                                       \
    template<>                                                                              \
    __device__ __forceinline__ void ThreadStore<cub_modifier, unsigned long long*, unsigned long long>(unsigned long long* ptr, unsigned long long val)     \
    {                                                                                       \
        asm volatile ("st."#ptx_modifier".u64 [%0], %1;" : :                                \
            _CUB_ASM_PTR_(ptr),                                                             \
            "l"(val));                                                                      \
    }

/**
 * Define a unsigned int (4B) ThreadStore specialization for the given Cache load modifier
 */
#define _CUB_STORE_4(cub_modifier, ptx_modifier)                                             \
    template<>                                                                              \
    __device__ __forceinline__ void ThreadStore<cub_modifier, unsigned int*, unsigned int>(unsigned int* ptr, unsigned int val)                             \
    {                                                                                       \
        asm volatile ("st."#ptx_modifier".u32 [%0], %1;" : :                                \
            _CUB_ASM_PTR_(ptr),                                                             \
            "r"(val));                                                                      \
    }


/**
 * Define a unsigned short (2B) ThreadStore specialization for the given Cache load modifier
 */
#define _CUB_STORE_2(cub_modifier, ptx_modifier)                                             \
    template<>                                                                              \
    __device__ __forceinline__ void ThreadStore<cub_modifier, unsigned short*, unsigned short>(unsigned short* ptr, unsigned short val)                     \
    {                                                                                       \
        asm volatile ("st."#ptx_modifier".u16 [%0], %1;" : :                                \
            _CUB_ASM_PTR_(ptr),                                                             \
            "h"(val));                                                                      \
    }


/**
 * Define a unsigned char (1B) ThreadStore specialization for the given Cache load modifier
 */
#define _CUB_STORE_1(cub_modifier, ptx_modifier)                                             \
    template<>                                                                              \
    __device__ __forceinline__ void ThreadStore<cub_modifier, unsigned char*, unsigned char>(unsigned char* ptr, unsigned char val)                         \
    {                                                                                       \
        asm volatile (                                                                      \
        "{"                                                                                 \
        "   .reg .u8 datum;"                                                                \
        "   cvt.u8.u16 datum, %1;"                                                          \
        "   st."#ptx_modifier".u8 [%0], datum;"                                             \
        "}" : :                                                                             \
            _CUB_ASM_PTR_(ptr),                                                             \
            "h"((unsigned short) val));                                                               \
    }

/**
 * Define powers-of-two ThreadStore specializations for the given Cache load modifier
 */
#define _CUB_STORE_ALL(cub_modifier, ptx_modifier)                                           \
    _CUB_STORE_16(cub_modifier, ptx_modifier)                                                \
    _CUB_STORE_8(cub_modifier, ptx_modifier)                                                 \
    _CUB_STORE_4(cub_modifier, ptx_modifier)                                                 \
    _CUB_STORE_2(cub_modifier, ptx_modifier)                                                 \
    _CUB_STORE_1(cub_modifier, ptx_modifier)                                                 \


/**
 * Define ThreadStore specializations for the various Cache load modifiers
 */
#if CUB_PTX_ARCH >= 200
    _CUB_STORE_ALL(STORE_WB, wb)
    _CUB_STORE_ALL(STORE_CG, cg)
    _CUB_STORE_ALL(STORE_CS, cs)
    _CUB_STORE_ALL(STORE_WT, wt)
#else
    _CUB_STORE_ALL(STORE_WB, global)
    _CUB_STORE_ALL(STORE_CG, global)
    _CUB_STORE_ALL(STORE_CS, global)
    _CUB_STORE_ALL(STORE_WT, volatile.global)
#endif


// Macro cleanup
#undef _CUB_STORE_ALL
#undef _CUB_STORE_1
#undef _CUB_STORE_2
#undef _CUB_STORE_4
#undef _CUB_STORE_8
#undef _CUB_STORE_16


/**
 * ThreadStore definition for STORE_DEFAULT modifier on iterator types
 */
template <typename OutputIteratorT, typename T>
__device__ __forceinline__ void ThreadStore(
    OutputIteratorT             itr,
    T                           val,
    Int2Type<STORE_DEFAULT>     /*modifier*/,
    Int2Type<false>             /*is_pointer*/)
{
    *itr = val;
}


/**
 * ThreadStore definition for STORE_DEFAULT modifier on pointer types
 */
template <typename T>
__device__ __forceinline__ void ThreadStore(
    T                           *ptr,
    T                           val,
    Int2Type<STORE_DEFAULT>     /*modifier*/,
    Int2Type<true>              /*is_pointer*/)
{
    *ptr = val;
}


/**
 * ThreadStore definition for STORE_VOLATILE modifier on primitive pointer types
 */
template <typename T>
__device__ __forceinline__ void ThreadStoreVolatilePtr(
    T                           *ptr,
    T                           val,
    Int2Type<true>              /*is_primitive*/)
{
    *reinterpret_cast<volatile T*>(ptr) = val;
}


/**
 * ThreadStore definition for STORE_VOLATILE modifier on non-primitive pointer types
 */
template <typename T>
__device__ __forceinline__ void ThreadStoreVolatilePtr(
    T                           *ptr,
    T                           val,
    Int2Type<false>             /*is_primitive*/)
{
    // Create a temporary using shuffle-words, then store using volatile-words
    typedef typename UnitWord<T>::VolatileWord  VolatileWord;
    typedef typename UnitWord<T>::ShuffleWord   ShuffleWord;

    const int VOLATILE_MULTIPLE = sizeof(T) / sizeof(VolatileWord);
    const int SHUFFLE_MULTIPLE  = sizeof(T) / sizeof(ShuffleWord);

    VolatileWord words[VOLATILE_MULTIPLE];

    #pragma unroll
    for (int i = 0; i < SHUFFLE_MULTIPLE; ++i)
        reinterpret_cast<ShuffleWord*>(words)[i] = reinterpret_cast<ShuffleWord*>(&val)[i];

    IterateThreadStore<0, VOLATILE_MULTIPLE>::template Dereference(
        reinterpret_cast<volatile VolatileWord*>(ptr),
        words);
}


/**
 * ThreadStore definition for STORE_VOLATILE modifier on pointer types
 */
template <typename T>
__device__ __forceinline__ void ThreadStore(
    T                           *ptr,
    T                           val,
    Int2Type<STORE_VOLATILE>    /*modifier*/,
    Int2Type<true>              /*is_pointer*/)
{
    ThreadStoreVolatilePtr(ptr, val, Int2Type<Traits<T>::PRIMITIVE>());
}


/**
 * ThreadStore definition for generic modifiers on pointer types
 */
template <typename T, int MODIFIER>
__device__ __forceinline__ void ThreadStore(
    T                           *ptr,
    T                           val,
    Int2Type<MODIFIER>          /*modifier*/,
    Int2Type<true>              /*is_pointer*/)
{
    // Create a temporary using shuffle-words, then store using device-words
    typedef typename UnitWord<T>::DeviceWord    DeviceWord;
    typedef typename UnitWord<T>::ShuffleWord   ShuffleWord;

    const int DEVICE_MULTIPLE   = sizeof(T) / sizeof(DeviceWord);
    const int SHUFFLE_MULTIPLE  = sizeof(T) / sizeof(ShuffleWord);

    DeviceWord words[DEVICE_MULTIPLE];

    #pragma unroll
    for (int i = 0; i < SHUFFLE_MULTIPLE; ++i)
        reinterpret_cast<ShuffleWord*>(words)[i] = reinterpret_cast<ShuffleWord*>(&val)[i];

    IterateThreadStore<0, DEVICE_MULTIPLE>::template Store<CacheStoreModifier(MODIFIER)>(
        reinterpret_cast<DeviceWord*>(ptr),
        words);
}


/**
 * ThreadStore definition for generic modifiers
 */
template <CacheStoreModifier MODIFIER, typename OutputIteratorT, typename T>
__device__ __forceinline__ void ThreadStore(OutputIteratorT itr, T val)
{
    ThreadStore(
        itr,
        val,
        Int2Type<MODIFIER>(),
        Int2Type<IsPointer<OutputIteratorT>::VALUE>());
}



#endif // DOXYGEN_SHOULD_SKIP_THIS


/** @} */       // end group UtilIo


}               // CUB namespace
CUB_NS_POSTFIX  // Optional outer namespace(s)