| { |
| "tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] }, |
| "cases": [ |
| { |
| "name": "1m_f32", |
| "inputs": { |
| "input": { "dtype": "float32", "shape": [1048576] }, |
| "min": { "dtype": "float32", "shape": [], "dist": "constant", "value": -1 }, |
| "max": { "dtype": "float32", "shape": [], "dist": "constant", "value": 1 } |
| }, |
| "outputs": { "output": { "dtype": "float32", "shape": [1048576] } } |
| }, |
| { |
| "name": "clip-f32-8m", |
| "preset": "smoke", |
| "vars": { "dtype": "float32", "count": 8388608 }, |
| "inputs": { |
| "input": { "shape": [8388608], "dtype": "float32", "seed": 7025, "dist": "normal", "scale": 2 }, |
| "min": { "dtype": "float32", "shape": [], "dist": "constant", "value": -1 }, |
| "max": { "dtype": "float32", "shape": [], "dist": "constant", "value": 1 } |
| }, |
| "outputs": { "output": { "shape": [8388608], "dtype": "float32" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| }, |
| { |
| "name": "clip-f16-8m", |
| "preset": "smoke", |
| "vars": { "dtype": "float16", "count": 8388608 }, |
| "inputs": { |
| "input": { "shape": [8388608], "dtype": "float16", "seed": 7026, "dist": "normal", "scale": 2 }, |
| "min": { "dtype": "float16", "shape": [], "dist": "constant", "value": -1 }, |
| "max": { "dtype": "float16", "shape": [], "dist": "constant", "value": 1 } |
| }, |
| "outputs": { "output": { "shape": [8388608], "dtype": "float16" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| }, |
| { |
| "name": "clip-f32-input-bounds-large-scalar-path", |
| "provenance": { |
| "source": "local fixture", |
| "notes": "The 8,388,608-element float32 input is divisible by four and selects the vectorized input-bounds route." |
| }, |
| "preset": "stress", |
| "vars": { "dtype": "float32", "count": 8388608 }, |
| "inputs": { |
| "input": { "shape": [8388608], "dtype": "float32", "dist": "normal", "seed": 7101, "scale": 2 }, |
| "min": { "shape": [], "dtype": "float32", "dist": "constant", "value": -1 }, |
| "max": { "shape": [], "dtype": "float32", "dist": "constant", "value": 1 } |
| }, |
| "outputs": { "output": { "shape": [8388608], "dtype": "float32", "dist": "empty" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| }, |
| { |
| "name": "clip-f16-8m-default-bounds-vec4", |
| "provenance": { |
| "source": "local fixture", |
| "notes": "Clip without min or max inputs measures the vec4 unary path independently of scalar input-bound handling." |
| }, |
| "preset": "smoke", |
| "vars": { "dtype": "float16", "count": 8388608 }, |
| "inputs": { "input": { "shape": [8388608], "dtype": "float16", "seed": 7026, "dist": "normal", "scale": 2 } }, |
| "outputs": { "output": { "shape": [8388608], "dtype": "float16" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| } |
| ] |
| } |
|
|
