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
amd_target
AMD GPU target definitions for the pipeline scheduling framework.
Platform-specific hardware descriptions for AMD CDNA/RDNA GPUs. Provides:
- MI355X cost model: per-op latencies, resources, and roles
- MI355X pipeline configurations: double-buffer and single-buffer
- MI355X target profiles: unified cost model + pipeline config
- AMD schedule_group_barrier hints for instruction interleaving
The generic framework (schedule_framework.mojo) has no AMD-specific references. All hardware knowledge lives here.
Structs
-
AMDScheduleHints: Hardware expansion factors for AMD schedule_group_barrier hints.
Functions
-
append_amd_hints: Append AMD schedule_group_barrier hints to a kernel entry list. -
mi355x_cost_model: MI355X cost model: production-tuned latencies. -
mi355x_double_buffer: MI355X ping-pong: double buffer, 2 warp groups, 2x2 MMA grid. -
mi355x_single_buffer: MI355X default matmul: single buffer, barrier-gated. -
mi355x_single_buffer_cost_model: MI355X cost model for single-buffer matmul (DefaultMatmulOps tags). -
mi355x_single_buffer_target: MI355X target: single-buffer with production-tuned costs. -
mi355x_target: MI355X target: ping-pong double-buffer with production-tuned costs.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!