| { |
| "op": "ai.onnx.DynamicQuantizeLinear", |
| "cases": [ |
| { |
| "name": "f32_64k", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [65536], |
| "data": { "kind": "fillFloat32", "sinStep": 0.013, "cosStep": 0.017, "scale": 3.0 } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "uint8", "shape": [65536] }, |
| "y_scale": { "dtype": "float32", "shape": [1] }, |
| "y_zero_point": { "dtype": "uint8", "shape": [1] } |
| } |
| }, |
| { |
| "name": "f32_4m", |
| "preset": "smoke", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2048, 2048], |
| "data": { "kind": "fillFloat32", "sinStep": 0.013, "cosStep": 0.017, "scale": 3.0 } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "uint8", "shape": [2048, 2048] }, |
| "y_scale": { "dtype": "float32", "shape": [1] }, |
| "y_zero_point": { "dtype": "uint8", "shape": [1] } |
| }, |
| "bench": { |
| "metrics": [ |
| { "type": "bandwidth", "name": "useful_onnx_payload", "value": "5 * numel(shapes.x) + 5" }, |
| { |
| "type": "bandwidth", |
| "name": "physical_3pass_storage", |
| "value": "12 * numel(shapes.x) + 16 * min(256, ceilDiv(numel(shapes.x), 1024)) + 16" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "alignment_cliff_non_vec4_4194303", |
| "preset": "smoke", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [4194303], |
| "data": { "kind": "fillFloat32", "sinStep": 0.013, "cosStep": 0.017, "scale": 3.0 } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "uint8", "shape": [4194303] }, |
| "y_scale": { "dtype": "float32", "shape": [1] }, |
| "y_zero_point": { "dtype": "uint8", "shape": [1] } |
| }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { "type": "bandwidth", "name": "useful_onnx_payload", "value": "5 * numel(shapes.x) + 5" }, |
| { |
| "type": "bandwidth", |
| "name": "physical_3pass_storage", |
| "value": "12 * numel(shapes.x) + 16 * min(256, ceilDiv(numel(shapes.x), 1024)) + 16" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "alignment_healthy_vec4_4194304", |
| "preset": "smoke", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [4194304], |
| "data": { "kind": "fillFloat32", "sinStep": 0.013, "cosStep": 0.017, "scale": 3.0 } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "uint8", "shape": [4194304] }, |
| "y_scale": { "dtype": "float32", "shape": [1] }, |
| "y_zero_point": { "dtype": "uint8", "shape": [1] } |
| }, |
| "bench": { |
| "metrics": [ |
| { "type": "bandwidth", "name": "useful_onnx_payload", "value": "5 * numel(shapes.x) + 5" }, |
| { |
| "type": "bandwidth", |
| "name": "physical_3pass_storage", |
| "value": "12 * numel(shapes.x) + 16 * min(256, ceilDiv(numel(shapes.x), 1024)) + 16" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "serial_finalize_uncapped_non_vec4_2m", |
| "preset": "smoke", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2097153], |
| "data": { "kind": "fillFloat32", "sinStep": 0.013, "cosStep": 0.017, "scale": 3.0 } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "uint8", "shape": [2097153] }, |
| "y_scale": { "dtype": "float32", "shape": [1] }, |
| "y_zero_point": { "dtype": "uint8", "shape": [1] } |
| }, |
| "bench": { |
| "metrics": [ |
| { "type": "bandwidth", "name": "useful_onnx_payload", "value": "5 * numel(shapes.x) + 5" }, |
| { |
| "type": "bandwidth", |
| "name": "physical_3pass_storage", |
| "value": "12 * numel(shapes.x) + 16 * min(256, ceilDiv(numel(shapes.x), 1024)) + 16" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "grid_stride_capped_finalize_2m", |
| "preset": "smoke", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2097152], |
| "data": { "kind": "fillFloat32", "sinStep": 0.013, "cosStep": 0.017, "scale": 3.0 } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "uint8", "shape": [2097152] }, |
| "y_scale": { "dtype": "float32", "shape": [1] }, |
| "y_zero_point": { "dtype": "uint8", "shape": [1] } |
| }, |
| "bench": { |
| "metrics": [ |
| { "type": "bandwidth", "name": "useful_onnx_payload", "value": "5 * numel(shapes.x) + 5" }, |
| { |
| "type": "bandwidth", |
| "name": "physical_3pass_storage", |
| "value": "12 * numel(shapes.x) + 16 * min(256, ceilDiv(numel(shapes.x), 1024)) + 16" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "below_gridcap_threshold_non_vec4_262143", |
| "preset": "smoke", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [262143], |
| "data": { "kind": "fillFloat32", "sinStep": 0.013, "cosStep": 0.017, "scale": 3.0 } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "uint8", "shape": [262143] }, |
| "y_scale": { "dtype": "float32", "shape": [1] }, |
| "y_zero_point": { "dtype": "uint8", "shape": [1] } |
| }, |
| "bench": { |
| "metrics": [ |
| { "type": "bandwidth", "name": "useful_onnx_payload", "value": "5 * numel(shapes.x) + 5" }, |
| { |
| "type": "bandwidth", |
| "name": "physical_3pass_storage", |
| "value": "12 * numel(shapes.x) + 16 * min(256, ceilDiv(numel(shapes.x), 1024)) + 16" |
| } |
| ] |
| } |
| } |
| ] |
| } |
|
|
