Fence Functions

Functions
__device__ void	__threadfence_block (void)
	threadfence_block makes writes visible to threads running in same block. More...
__device__ void	__threadfence (void)
	threadfence makes wirtes visible to other threads running on same GPU. More...
__device__ void	__threadfence_system (void)
	threadfence_system makes writes to pinned system memory visible on host CPU. More...
__device__ unsigned	__hip_ds_bpermute (int index, unsigned src)
__device__ float	__hip_ds_bpermutef (int index, float src)
__device__ unsigned	__hip_ds_permute (int index, unsigned src)
__device__ float	__hip_ds_permutef (int index, float src)
__device__ unsigned	__hip_ds_swizzle (unsigned int src, int pattern)
__device__ float	__hip_ds_swizzlef (float src, int pattern)
__device__ int	__hip_move_dpp (int src, int dpp_ctrl, int row_mask, int bank_mask, bool bound_ctrl)

Detailed Description

CUDA 8 device function features

Kernel launching

---

Warning

The HIP memory fence functions are currently not supported yet. If any of those threadfence stubs are reached by the application, you should set "export HSA_DISABLE_CACHE=1" to disable L1 and L2 caches.

On AMD platforms, the threadfence* routines are currently empty stubs.

Function Documentation

__device__ void __threadfence

(

void

)

threadfence makes wirtes visible to other threads running on same GPU.

void

Parameters

void

Warning

__threadfence is a stub and map to no-op, application should set "export HSA_DISABLE_CACHE=1" to disable both L1 and L2 caches.

__device__ void __threadfence_block

(

void

)

threadfence_block makes writes visible to threads running in same block.

void

Parameters

void

Warning

__threadfence_block is a stub and map to no-op.

__device__ void __threadfence_system

(

void

)

threadfence_system makes writes to pinned system memory visible on host CPU.

void

Parameters

void

Warning

__threadfence_system is a stub and map to no-op.