Cycles: Add more atomic operations
authorMai Lavelle <mai.lavelle@gmail.com>
Wed, 22 Feb 2017 12:41:18 +0000 (07:41 -0500)
committerMai Lavelle <mai.lavelle@gmail.com>
Wed, 8 Mar 2017 05:52:41 +0000 (00:52 -0500)
intern/cycles/util/util_atomic.h

index 433e41fbbb6fb1f4abbed2ef7dd80ad267e36018..6c52117ef9aa0c8ba2d0ff5225622daffa92ed8e 100644 (file)
@@ -32,6 +32,13 @@ ATOMIC_INLINE void atomic_update_max_z(size_t *maximum_value, size_t value)
        }
 }
 
        }
 }
 
+#define atomic_add_and_fetch_float(p, x) atomic_add_and_fetch_fl((p), (x))
+
+#define atomic_fetch_and_inc_uint32(p) atomic_fetch_and_add_uint32((p), 1)
+
+#define CCL_LOCAL_MEM_FENCE 0
+#define ccl_barrier(flags) (void)0
+
 #else  /* __KERNEL_GPU__ */
 
 #ifdef __KERNEL_OPENCL__
 #else  /* __KERNEL_GPU__ */
 
 #ifdef __KERNEL_OPENCL__
@@ -39,7 +46,7 @@ ATOMIC_INLINE void atomic_update_max_z(size_t *maximum_value, size_t value)
 /* Float atomics implementation credits:
  *   http://suhorukov.blogspot.in/2011/12/opencl-11-atomic-operations-on-floating.html
  */
 /* Float atomics implementation credits:
  *   http://suhorukov.blogspot.in/2011/12/opencl-11-atomic-operations-on-floating.html
  */
-ccl_device_inline void atomic_add_and_fetch_float(volatile ccl_global float *source,
+ccl_device_inline float atomic_add_and_fetch_float(volatile ccl_global float *source,
                                         const float operand)
 {
        union {
                                         const float operand)
 {
        union {
@@ -56,10 +63,29 @@ ccl_device_inline void atomic_add_and_fetch_float(volatile ccl_global float *sou
        } while(atomic_cmpxchg((volatile ccl_global unsigned int *)source,
                               prev_value.int_value,
                               new_value.int_value) != prev_value.int_value);
        } while(atomic_cmpxchg((volatile ccl_global unsigned int *)source,
                               prev_value.int_value,
                               new_value.int_value) != prev_value.int_value);
+       return new_value.float_value;
 }
 
 }
 
+#define atomic_fetch_and_add_uint32(p, x) atomic_add((p), (x))
+#define atomic_fetch_and_inc_uint32(p) atomic_inc((p))
+
+#define CCL_LOCAL_MEM_FENCE CLK_LOCAL_MEM_FENCE
+#define ccl_barrier(flags) barrier(flags)
+
 #endif  /* __KERNEL_OPENCL__ */
 
 #endif  /* __KERNEL_OPENCL__ */
 
+#ifdef __KERNEL_CUDA__
+
+#define atomic_add_and_fetch_float(p, x) (atomicAdd((float*)(p), (float)(x)) + (float)(x))
+
+#define atomic_fetch_and_add_uint32(p, x) atomicAdd((unsigned int*)(p), (unsigned int)(x))
+#define atomic_fetch_and_inc_uint32(p) atomic_fetch_and_add_uint32((p), 1)
+
+#define CCL_LOCAL_MEM_FENCE
+#define ccl_barrier(flags) __syncthreads()
+
+#endif  /* __KERNEL_CUDA__ */
+
 #endif  /* __KERNEL_GPU__ */
 
 #endif /* __UTIL_ATOMIC_H__ */
 #endif  /* __KERNEL_GPU__ */
 
 #endif /* __UTIL_ATOMIC_H__ */