| { |
| "op": "ai.onnx.CumProd", |
| "cases": [ |
| { |
| "name": "rows_1024_cols_1024_axis1", |
| "args": { "axis": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [1024, 1024], "dist": "uniform", "seed": 914, "scale": 1 } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1024, 1024] } } |
| }, |
| { |
| "name": "rows_256_cols_4096_axis1", |
| "preset": "smoke", |
| "args": { "axis": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [256, 4096], "dist": "uniform", "seed": 915, "scale": 1 } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [256, 4096], "dist": "empty" } }, |
| "bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "2 * 4 * numel(shapes.x)" }] } |
| }, |
| { |
| "name": "axis0_rows_16_inner_65536", |
| "preset": "smoke", |
| "args": { "axis": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [16, 65536], "dist": "uniform", "seed": 916, "scale": 1 } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [16, 65536], "dist": "empty" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "2 * 4 * numel(shapes.x)" }] } |
| }, |
| { |
| "name": "dispatch_cliff_rows_131072_axis1_short_rows", |
| "preset": "smoke", |
| "args": { "axis": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [131072, 64], "dist": "uniform", "seed": 917, "scale": 1 } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [131072, 64], "dist": "empty" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "2 * 4 * numel(shapes.x)" }] } |
| }, |
| { |
| "name": "rows_65535_axis1_short_rows", |
| "preset": "smoke", |
| "args": { "axis": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [65535, 64], "dist": "uniform", "seed": 918, "scale": 1 } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [65535, 64], "dist": "empty" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "2 * 4 * numel(shapes.x)" }] } |
| }, |
| { |
| "name": "short_dim_floor_dim16_axis1_short_rows", |
| "preset": "smoke", |
| "args": { "axis": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [65536, 16], "dist": "uniform", "seed": 919, "scale": 1 } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [65536, 16], "dist": "empty" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "2 * 4 * numel(shapes.x)" }] } |
| }, |
| { |
| "name": "alignment_mismatch_dim4094_axis1_scalar", |
| "preset": "smoke", |
| "args": { "axis": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [256, 4094], "dist": "uniform", "seed": 920, "scale": 1 } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [256, 4094], "dist": "empty" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "2 * 4 * numel(shapes.x)" }] } |
| }, |
| { |
| "name": "f16_rows_256_cols_4096_axis1_wg_row_scan", |
| "preset": "smoke", |
| "args": { "axis": 1 }, |
| "inputs": { "x": { "dtype": "float16", "shape": [256, 4096], "dist": "uniform", "seed": 921, "scale": 1 } }, |
| "outputs": { "y": { "dtype": "float16", "shape": [256, 4096], "dist": "empty" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "2 * 2 * numel(shapes.x)" }] } |
| }, |
| { |
| "name": "reverse_axis1_large_dim4096_wg_row_scan_scalar", |
| "preset": "stress", |
| "args": { "axis": 1 }, |
| "attrs": { "reverse": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [256, 4096], "dist": "uniform", "seed": 930, "scale": 1 } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [256, 4096], "dist": "empty" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "2 * 4 * numel(shapes.x)" }] } |
| }, |
| { |
| "name": "exclusive_nonmult4_dim_large_wg_row_scan", |
| "preset": "stress", |
| "provenance": { |
| "source": "authored for variant coverage", |
| "notes": "Non-multiple-of-four exclusive scan that verifies the cooperative row-scan route; the serial path remains reserved for shapes outside its domain." |
| }, |
| "attrs": { "exclusive": 1 }, |
| "args": { "axis": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [4, 4098], "dist": "constant", "value": 1 } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [4, 4098], "dist": "empty" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "2 * 4 * numel(shapes.x)" }] } |
| } |
| ] |
| } |
|
|