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 module

debug

Diagnostic tooling for compiled pipeline schedules.

Compiled schedules are comptime values (List[ScheduleEntry]), and List is not ImplicitlyCopyable, so the only way to walk the entries in a printer is via a comptime for i in range(N) loop with materialize[entries[i]]() per access. The helpers below provide the ingredients for that pattern; callers wire up the comptime loops.

Usage pattern:

comptime n_krn = len(schedule.kernel)
print_phase_header("KERNEL", n_krn)
comptime for i in range(n_krn):
    print_entry[my_op_name_fn]("KRN", i, materialize[schedule.kernel[i]]())

var counts = WaitCounts()
comptime for i in range(n_krn):
    counts.add(materialize[schedule.kernel[i]](), i, n_krn)
print_wait_counts("kernel  ", counts)

Structs

WaitCounts: Per-phase wait/barrier counts, populated by repeated add calls.

Functions

default_op_name: Returns the default name for a framework op tag.
print_entry: Prints one ScheduleEntry on a single line, with the caller's pre-resolved tag name.
print_phase_header: Prints a phase section header with separator rule.
print_wait_counts: Prints a one-line summary of a WaitCounts value.
update_wait_counts: Updates counts for one entry; next_tag is the next entry's tag (or _Ops.NONE.value for the last entry).

Structs​

Functions​

Structs

Functions