{ "tile_config": { "tile_m": { "max": 256, "min": 64, "step": 64 }, "tile_n": { "max": 256, "min": 64, "step": 64 }, "tile_k": { "max": 256, "min": 64, "step": 64 }, "warp_m": { "values": [ 4, 2, 1 ] }, "warp_n": { "values": [ 4, 2, 1 ] }, "warp_k": { "values": [ 1 ] }, "warp_tile_m": { "values": [ 4, 16, 32 ] }, "warp_tile_n": { "values": [ 16, 32, 64 ] }, "warp_tile_k": { "values": [ 8, 16, 32, 64, 128 ] } }, "trait_config": { "pipeline": { "values": [ "compv3", "compv4", "mem" ] }, "scheduler": { "values": [ "intrawave", "interwave" ] }, "epilogue": { "values": [ "cshuffle", "default" ] }, "pad_m": { "values": [ false ] }, "pad_n": { "values": [ false ] }, "pad_k": { "values": [ false ] }, "persistent": { "values": [ false, true ] } }, "k_block_per_cu": 1 }