| { |
| "op": "ai.onnx.LpNormalization", |
| "tunableSpace": { |
| "WORKGROUP_SIZE": [64, 128, 256], |
| "SPLIT_TARGET_DIM": [128, 256, 512, 1024], |
| "MAX_SPLITS": [8, 16, 32, 64, 128, 256] |
| }, |
| "cases": [ |
| { |
| "name": "4096x4096_last_axis_p2", |
| "preset": "smoke", |
| "vars": { "dtype": "float32", "rows": 4096, "dim": 4096 }, |
| "attrs": { "axis": -1, "p": 2 }, |
| "inputs": { "input": { "dtype": "float32", "shape": [4096, 4096], "dist": "normal", "seed": 760, "scale": 0.5 } }, |
| "outputs": { "output": { "dtype": "float32", "shape": [4096, 4096] } }, |
| "bench": { |
| "primary": true, |
| "metrics": [{ "type": "bandwidth", "value": "args.rows * args.dim * 2 * dtypeBytes(args.dtype)" }] |
| } |
| }, |
| { |
| "name": "4096x4096_axis0_p2_nonlast", |
| "preset": "smoke", |
| "vars": { "dtype": "float32", "rows": 4096, "dim": 4096 }, |
| "attrs": { "axis": 0, "p": 2 }, |
| "inputs": { "input": { "dtype": "float32", "shape": [4096, 4096], "dist": "normal", "seed": 761, "scale": 0.5 } }, |
| "outputs": { "output": { "dtype": "float32", "shape": [4096, 4096] } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "bandwidth", |
| "name": "reduce-read plus divide read-write traffic", |
| "value": "args.rows * args.dim * 3 * dtypeBytes(args.dtype)" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "1024x1024_axis1", |
| "preset": "smoke", |
| "attrs": { "axis": 1, "p": 2 }, |
| "inputs": { "input": { "dtype": "float32", "shape": [1024, 1024] } }, |
| "outputs": { "output": { "dtype": "float32", "shape": [1024, 1024] } } |
| }, |
| { |
| "name": "axis0_fewrows_hugedim_p2_f32_starved_reduce", |
| "preset": "stress", |
| "vars": { "dtype": "float32", "rows": 128, "dim": 16384 }, |
| "attrs": { "axis": 0, "p": 2 }, |
| "inputs": { "input": { "dtype": "float32", "shape": [16384, 128], "dist": "normal", "seed": 911, "scale": 0.5 } }, |
| "outputs": { "output": { "dtype": "float32", "shape": [16384, 128], "dist": "empty" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "bandwidth", |
| "name": "reduce-read plus divide read-write traffic", |
| "value": "args.rows * args.dim * 3 * dtypeBytes(args.dtype)" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "axis0_64rows_dim32k_p1_f32_sharper_starvation", |
| "preset": "stress", |
| "vars": { "dtype": "float32", "rows": 64, "dim": 32768 }, |
| "attrs": { "axis": 0, "p": 1 }, |
| "inputs": { "input": { "dtype": "float32", "shape": [32768, 64], "dist": "normal", "seed": 912, "scale": 0.5 } }, |
| "outputs": { "output": { "dtype": "float32", "shape": [32768, 64], "dist": "empty" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "bandwidth", |
| "name": "reduce-read plus divide read-write traffic", |
| "value": "args.rows * args.dim * 3 * dtypeBytes(args.dtype)" |
| } |
| ] |
| } |
| } |
| ] |
| } |
|
|