Subgroup built-ins

1 min read

Import: import system.gpu;

This reference lists the public declarations for this part of the library. See Reading API signatures for parameter and result notation. Shader-only declarations can be used only inside @compute, @vertex, @fragment, inferred helpers and ray tracing shader functions; using them elsewhere is a compile error.

subgroup_size

The number of threads in a subgroup, the group of threads the GPU runs in lockstep: for example 32 on NVIDIA, 32 or 64 on AMD and 32 on Apple GPUs. Subgroup functions need GpuFeature.Subgroups and are not available with GpuTarget.Dx11.

public extern fun subgroup_size(): uint

subgroup_invocation_id

This thread's lane within its subgroup, from 0 to subgroup_size() - 1.

public extern fun subgroup_invocation_id(): uint

subgroup_add

The sum of value over every active lane of the subgroup.

public extern fun subgroup_add<T>(value: T): T

subgroup_min

The smallest value among the active lanes.

public extern fun subgroup_min<T>(value: T): T

subgroup_max

The largest value among the active lanes.

public extern fun subgroup_max<T>(value: T): T

subgroup_prefix_add

The sum of value over the active lanes below this one.

public extern fun subgroup_prefix_add<T>(value: T): T

subgroup_prefix_count

The number of active lanes below this one where condition is true.

public extern fun subgroup_prefix_count(condition: bool): uint

subgroup_count

The number of active lanes where condition is true.

public extern fun subgroup_count(condition: bool): uint

subgroup_ballot

One bit per lane, set where condition is true.

public extern fun subgroup_ballot(condition: bool): GpuUVec4

subgroup_broadcast

value from lane lane, the same lane for every thread.

public extern fun subgroup_broadcast<T>(value: T, lane: uint): T

subgroup_shuffle

value from lane lane, which may differ per thread.

public extern fun subgroup_shuffle<T>(value: T, lane: uint): T

subgroup_all

True when condition is true on every active lane.

public extern fun subgroup_all(condition: bool): bool

subgroup_any

True when condition is true on any active lane.

public extern fun subgroup_any(condition: bool): bool

subgroup_elect

True on exactly one active lane, the lowest.

public extern fun subgroup_elect(): bool