For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

bencher_iter_custom

def bencher_iter_custom[kernel_launch_fn: def(DeviceContext) raises capturing thin -> None](mut self_: Bencher, ctx: DeviceContext)

Times a target GPU function with custom number of iterations via DeviceContext ctx.

Parameters:

kernel_launch_fn (def(DeviceContext) raises capturing thin -> None): The target GPU kernel launch function to benchmark.

Args:

self_ (Bencher): The bencher state.
ctx (DeviceContext): The GPU DeviceContext for launching kernel.

def bencher_iter_custom[FuncType: def(DeviceContext) raises -> None](mut self_: Bencher, ref func: FuncType, ctx: DeviceContext)

Times a target GPU closure with custom number of iterations via DeviceContext ctx.

Notes:

This overload is intentionally separate from the parametric iter_custom[kernel_launch_fn](ctx) form. Nested launch closures that capture benchmark-local state are closure values, and the current closure typing rules do not let those values compose with a def(DeviceContext) raises capturing[_] compile-time parameter while preserving their capture object. This value-taking overload forwards the closure to DeviceContext.execution_time() so FuncType carries the captured state.

Parameters:

FuncType (def(DeviceContext) raises -> None): The target GPU kernel launch closure type.

Args:

self_ (Bencher): The bencher state.
func (FuncType): The closure carrying the captured state of the kernel launch.
ctx (DeviceContext): The GPU DeviceContext for launching kernel.

def bencher_iter_custom[kernel_launch_fn: def(DeviceContext, Int) raises capturing thin -> None](mut self_: Bencher, ctx: DeviceContext)

Times a target GPU function with custom number of iterations via DeviceContext ctx.

Parameters:

kernel_launch_fn (def(DeviceContext, Int) raises capturing thin -> None): The target GPU kernel launch function to benchmark.

Args:

self_ (Bencher): The bencher state.
ctx (DeviceContext): The GPU DeviceContext for launching kernel.

def bencher_iter_custom[FuncType: def(DeviceContext, Int) raises -> None](mut self_: Bencher, ref func: FuncType, ctx: DeviceContext)

Times a target GPU closure with custom number of iterations via DeviceContext ctx.

Notes:

This overload is intentionally separate from the parametric iter_custom[kernel_launch_fn](ctx) form. Nested launch closures that capture benchmark-local state are closure values, and the current closure typing rules do not let those values compose with a def(DeviceContext, Int) raises capturing[_] compile-time parameter while preserving their capture object. This value-taking overload forwards the closure to DeviceContext.execution_time_iter() so FuncType carries the captured state.

Parameters:

FuncType (def(DeviceContext, Int) raises -> None): The target GPU kernel launch closure type.

Args:

self_ (Bencher): The bencher state.
func (FuncType): The closure carrying the captured state of the kernel launch.
ctx (DeviceContext): The GPU DeviceContext for launching kernel.