| { |
| "tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] }, |
| "cases": [ |
| { |
| "name": "gelu-f32-1m", |
| "preset": "smoke", |
| "vars": { "dtype": "float32", "count": 1048576 }, |
| "attrs": { "approximate": "tanh" }, |
| "inputs": { "x": { "shape": [1048576], "dtype": "float32", "dist": "normal", "seed": 650, "scale": 0.5 } }, |
| "outputs": { "y": { "shape": [1048576], "dtype": "float32" } }, |
| "bench": { |
| "primary": true, |
| "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] |
| } |
| }, |
| { |
| "name": "gelu-f32-8m", |
| "preset": "smoke", |
| "vars": { "dtype": "float32", "count": 8388608 }, |
| "attrs": { "approximate": "tanh" }, |
| "inputs": { "x": { "shape": [8388608], "dtype": "float32", "seed": 7001, "dist": "normal", "scale": 0.5 } }, |
| "outputs": { "y": { "shape": [8388608], "dtype": "float32" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| }, |
| { |
| "name": "gelu-f32-16m", |
| "preset": "smoke", |
| "vars": { "dtype": "float32", "count": 16777216 }, |
| "attrs": { "approximate": "tanh" }, |
| "inputs": { "x": { "shape": [16777216], "dtype": "float32", "seed": 7011, "dist": "normal", "scale": 0.5 } }, |
| "outputs": { "y": { "shape": [16777216], "dtype": "float32" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| }, |
| { |
| "name": "gelu-f16-8m", |
| "preset": "smoke", |
| "vars": { "dtype": "float16", "count": 8388608 }, |
| "attrs": { "approximate": "tanh" }, |
| "inputs": { "x": { "shape": [8388608], "dtype": "float16", "seed": 7002, "dist": "normal", "scale": 0.5 } }, |
| "outputs": { "y": { "shape": [8388608], "dtype": "float16" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| }, |
| { |
| "name": "gelu-f32-1m-erf", |
| "preset": "smoke", |
| "vars": { "dtype": "float32", "count": 1048576 }, |
| "attrs": { "approximate": "none" }, |
| "inputs": { "x": { "shape": [1048576], "dtype": "float32", "dist": "normal", "seed": 651, "scale": 0.5 } }, |
| "outputs": { "y": { "shape": [1048576], "dtype": "float32" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| }, |
| { |
| "name": "gelu-f16-8m-erf", |
| "preset": "smoke", |
| "vars": { "dtype": "float16", "count": 8388608 }, |
| "attrs": { "approximate": "none" }, |
| "inputs": { "x": { "shape": [8388608], "dtype": "float16", "dist": "normal", "seed": 7003, "scale": 0.5 } }, |
| "outputs": { "y": { "shape": [8388608], "dtype": "float16" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| }, |
| { |
| "name": "gelu-f32-1m-scalar-fallback", |
| "preset": "smoke", |
| "vars": { "dtype": "float32", "count": 1048575 }, |
| "attrs": { "approximate": "tanh" }, |
| "inputs": { "x": { "shape": [1048575], "dtype": "float32", "dist": "normal", "seed": 652, "scale": 0.5 } }, |
| "outputs": { "y": { "shape": [1048575], "dtype": "float32" } }, |
| "bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] } |
| } |
| ] |
| } |
|
|
