intern/cycles/util/util_atomic.h

   1 /*
   2  * Copyright 2014 Blender Foundation
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at
   7  *
   8  * http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #ifndef __UTIL_ATOMIC_H__
  18 #define __UTIL_ATOMIC_H__
  19
  20 #ifndef __KERNEL_GPU__
  21
  22 /* Using atomic ops header from Blender. */
  23 #include "atomic_ops.h"
  24
  25 #define atomic_add_and_fetch_float(p, x) atomic_add_and_fetch_fl((p), (x))
  26 #define atomic_compare_and_swap_float(p, old_val, new_val) atomic_cas_float((p), (old_val), (new_val))
  27
  28 #define atomic_fetch_and_inc_uint32(p) atomic_fetch_and_add_uint32((p), 1)
  29 #define atomic_fetch_and_dec_uint32(p) atomic_fetch_and_add_uint32((p), -1)
  30
  31 #define CCL_LOCAL_MEM_FENCE 0
  32 #define ccl_barrier(flags) (void)0
  33
  34 #else  /* __KERNEL_GPU__ */
  35
  36 #ifdef __KERNEL_OPENCL__
  37
  38 /* Float atomics implementation credits:
  39  *   http://suhorukov.blogspot.in/2011/12/opencl-11-atomic-operations-on-floating.html
  40  */
  41 ccl_device_inline float atomic_add_and_fetch_float(volatile ccl_global float *source,
  42                                         const float operand)
  43 {
  44         union {
  45                 unsigned int int_value;
  46                 float float_value;
  47         } new_value;
  48         union {
  49                 unsigned int int_value;
  50                 float float_value;
  51         } prev_value;
  52         do {
  53                 prev_value.float_value = *source;
  54                 new_value.float_value = prev_value.float_value + operand;
  55         } while(atomic_cmpxchg((volatile ccl_global unsigned int *)source,
  56                                prev_value.int_value,
  57                                new_value.int_value) != prev_value.int_value);
  58         return new_value.float_value;
  59 }
  60
  61 ccl_device_inline float atomic_compare_and_swap_float(volatile ccl_global float *dest,
  62                                                       const float old_val, const float new_val)
  63 {
  64         union {
  65                 unsigned int int_value;
  66                 float float_value;
  67         } new_value, prev_value, result;
  68         prev_value.float_value = old_val;
  69         new_value.float_value = new_val;
  70         result.int_value = atomic_cmpxchg((volatile ccl_global unsigned int *)dest,
  71                                        prev_value.int_value, new_value.int_value);
  72         return result.float_value;
  73 }
  74
  75 #define atomic_fetch_and_add_uint32(p, x) atomic_add((p), (x))
  76 #define atomic_fetch_and_inc_uint32(p) atomic_inc((p))
  77 #define atomic_fetch_and_dec_uint32(p) atomic_dec((p))
  78
  79 #define CCL_LOCAL_MEM_FENCE CLK_LOCAL_MEM_FENCE
  80 #define ccl_barrier(flags) barrier(flags)
  81
  82 #endif  /* __KERNEL_OPENCL__ */
  83
  84 #ifdef __KERNEL_CUDA__
  85
  86 #define atomic_add_and_fetch_float(p, x) (atomicAdd((float*)(p), (float)(x)) + (float)(x))
  87
  88 #define atomic_fetch_and_add_uint32(p, x) atomicAdd((unsigned int*)(p), (unsigned int)(x))
  89 #define atomic_fetch_and_sub_uint32(p, x) atomicSub((unsigned int*)(p), (unsigned int)(x))
  90 #define atomic_fetch_and_inc_uint32(p) atomic_fetch_and_add_uint32((p), 1)
  91 #define atomic_fetch_and_dec_uint32(p) atomic_fetch_and_sub_uint32((p), 1)
  92
  93 ccl_device_inline float atomic_compare_and_swap_float(volatile float *dest,
  94                                                       const float old_val, const float new_val)
  95 {
  96         union {
  97                 unsigned int int_value;
  98                 float float_value;
  99         } new_value, prev_value, result;
 100         prev_value.float_value = old_val;
 101         new_value.float_value = new_val;
 102         result.int_value = atomicCAS((unsigned int *)dest, prev_value.int_value,new_value.int_value);
 103         return result.float_value;
 104 }
 105
 106 #define CCL_LOCAL_MEM_FENCE
 107 #define ccl_barrier(flags) __syncthreads()
 108
 109 #endif  /* __KERNEL_CUDA__ */
 110
 111 #endif  /* __KERNEL_GPU__ */
 112
 113 #endif /* __UTIL_ATOMIC_H__ */