Mojo function

mi355x_single_buffer_cost_model

mi355x_single_buffer_cost_model() -> TargetCostModel

MI355X cost model for single-buffer matmul (DefaultMatmulOps tags).

Tag mapping (from DefaultMatmulOps): 0=LOAD_DRAM: GLOBAL_MEM, 200 cycles, GLOBAL_LOAD 1=STORE_SMEM: LDS, 20 cycles, SHARED_STORE 2=LOAD_FRAG: LDS, 20 cycles, FRAGMENT_LOAD 3=COMPUTE: MMA_UNIT, 64 cycles, COMPUTE

VGPR liveness hints (BF16 32x32x16 MFMA on MI355X): LOAD_DRAM: 0 (buffer_load_lds writes directly to LDS) STORE_SMEM: 0 (ds_write consumes existing register data) LOAD_FRAG: vgpr_def=4 (one 32x32 frag is 4 VGPRs/lane) COMPUTE: vgpr_def=16 (32x32 fp32 accumulator = 16 VGPRs/lane)

Returns:

TargetCostModel

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit