GpuLaneBarrier¶

The lanes of an iteration wait for each other.

Declaration¶

Syntax¶

GpuLaneBarrier()

Arguments¶

None.

Return values¶

None.

Description¶

In a kernel whose layout has lanes (see gpu_kernel_layout), each iteration runs on several threads of one SIMD-group, its lanes. GpuLaneBarrier makes every lane of the iteration wait until all of them have reached it, and makes what each wrote to memory before it visible to the others after it. Every lane of the iteration reaches the same barriers: one is placed only where all of them go. With one lane per iteration it does nothing. Only such a kernel has one.

Examples¶

(GpuLaneBarrier)