Xenova's picture
Xenova HF Staff
sync 2e7068faf55e
758a8fa verified
Raw
History Blame
7.74 kB
{
"op": "ai.onnx.ReduceMean",
"tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256], "MIDDLE_AXIS_TILE_COLS": [16, 32, 64] },
"cases": [
{
"name": "reduce-mean-axis0-1024x512",
"preset": "smoke",
"vars": { "dtype": "float32", "rows": 1024, "cols": 512 },
"attrs": { "axes": [0], "keepdims": 0 },
"inputs": { "x": { "shape": [1024, 512], "dtype": "float32", "dist": "normal", "seed": 881, "scale": 1 } },
"outputs": { "y": { "shape": [512], "dtype": "float32" } },
"bench": {
"primary": true,
"metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }]
}
},
{
"name": "reduce-mean-axis1-1024x4096",
"preset": "smoke",
"vars": { "dtype": "float32", "rows": 1024, "cols": 4096 },
"attrs": { "axes": [1], "keepdims": 0 },
"inputs": { "x": { "shape": [1024, 4096], "dtype": "float32", "dist": "normal", "seed": 880, "scale": 1 } },
"outputs": { "y": { "shape": [1024], "dtype": "float32" } },
"bench": {
"primary": true,
"metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }]
}
},
{
"name": "reducemean-lastaxis-f32-4096x4096",
"preset": "smoke",
"vars": { "rows": 4096, "cols": 4096 },
"attrs": { "axes": [1], "keepdims": 0 },
"inputs": { "x": { "shape": [4096, 4096], "dtype": "float32", "dist": "normal", "seed": 117, "scale": 0.2 } },
"outputs": { "y": { "shape": [4096], "dtype": "float32" } },
"bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * 4" }] }
},
{
"name": "reducemean-rank3-lastaxis-f32-64x64x4096",
"preset": "smoke",
"vars": { "rows": 4096, "cols": 4096 },
"attrs": { "axes": [2], "keepdims": 0 },
"inputs": { "x": { "shape": [64, 64, 4096], "dtype": "float32", "dist": "normal", "seed": 218, "scale": 0.2 } },
"outputs": { "y": { "shape": [64, 64], "dtype": "float32" } },
"bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * 4" }] }
},
{
"name": "reducemean-rank3-lastaxis-smallcols-f32-1024x32x16",
"preset": "smoke",
"vars": { "rows": 32768, "cols": 16 },
"attrs": { "axes": [2], "keepdims": 0 },
"inputs": { "x": { "shape": [1024, 32, 16], "dtype": "float32", "dist": "normal", "seed": 412, "scale": 0.2 } },
"outputs": { "y": { "shape": [1024, 32], "dtype": "float32" } },
"bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * 4" }] }
},
{
"name": "axis0-large-65536x512",
"preset": "smoke",
"vars": { "rows": 65536, "cols": 512 },
"attrs": { "axes": [0], "keepdims": 0 },
"inputs": { "x": { "shape": [65536, 512], "dtype": "float32", "dist": "normal", "scale": 0.2, "seed": 311 } },
"outputs": { "y": { "shape": [512], "dtype": "float32" } },
"bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * 4" }] }
},
{
"name": "axis0-tall-16384x1024",
"preset": "smoke",
"vars": { "rows": 16384, "cols": 1024 },
"attrs": { "axes": [0], "keepdims": 0 },
"inputs": { "x": { "shape": [16384, 1024], "dtype": "float32", "dist": "normal", "scale": 0.2, "seed": 312 } },
"outputs": { "y": { "shape": [1024], "dtype": "float32" } },
"bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * 4" }] }
},
{
"name": "reducemean-rank4-axis1-serial-32x256x28x28",
"preset": "smoke",
"vars": { "dtype": "float32" },
"attrs": { "axes": [1], "keepdims": 0 },
"inputs": {
"x": { "shape": [32, 256, 28, 28], "dtype": "float32", "dist": "normal", "scale": 0.2, "seed": 5511 }
},
"outputs": { "y": { "shape": [32, 28, 28], "dtype": "float32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "32 * 256 * 28 * 28 * dtypeBytes(args.dtype)" }] }
},
{
"name": "reducemean-rank3-axis1-serial-8x1024x768",
"preset": "smoke",
"vars": { "dtype": "float32" },
"attrs": { "axes": [1], "keepdims": 0 },
"inputs": { "x": { "shape": [8, 1024, 768], "dtype": "float32", "dist": "normal", "scale": 0.2, "seed": 5512 } },
"outputs": { "y": { "shape": [8, 768], "dtype": "float32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "8 * 1024 * 768 * dtypeBytes(args.dtype)" }] }
},
{
"name": "reducemean-lastaxis-scalar-subgroup-cols-unaligned-4096x4095",
"preset": "smoke",
"vars": { "dtype": "float32", "rows": 4096, "cols": 4095 },
"attrs": { "axes": [1], "keepdims": 0 },
"inputs": { "x": { "shape": [4096, 4095], "dtype": "float32", "dist": "normal", "scale": 0.2, "seed": 5513 } },
"outputs": { "y": { "shape": [4096], "dtype": "float32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }] }
},
{
"name": "reducemean-fullreduce-serial-numel-not-mul4-r3",
"preset": "stress",
"attrs": { "keepdims": 0 },
"inputs": { "x": { "shape": [1001, 1001, 1], "dtype": "float32", "dist": "normal", "seed": 601, "scale": 1 } },
"outputs": { "y": { "shape": [], "dtype": "float32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "1001 * 1001 * 1 * 4" }] }
},
{
"name": "reducemean-int32-lastaxis-serial-4096x1024",
"preset": "stress",
"attrs": { "axes": [1], "keepdims": 0 },
"inputs": {
"x": { "shape": [4096, 1024], "dtype": "int32", "dist": "randint", "seed": 602, "min": -1000, "max": 1000 }
},
"outputs": { "y": { "shape": [4096], "dtype": "int32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "4096 * 1024 * 4" }] }
},
{
"name": "reducemean-axis0-f32-131072x15-narrow-splitk-boundary",
"preset": "stress",
"vars": { "rows": 131072, "cols": 15 },
"attrs": { "axes": [0], "keepdims": 0 },
"inputs": { "x": { "shape": [131072, 15], "dtype": "float32", "dist": "normal", "seed": 603, "scale": 0.2 } },
"outputs": { "y": { "shape": [15], "dtype": "float32", "dist": "empty" } },
"bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * 4" }] }
},
{
"name": "reducemean-spatial-axes23-f32-2x64x256x256-globalpool-pattern",
"preset": "stress",
"provenance": {
"source": "authored for variant coverage",
"notes": "Rank-4 contiguous-suffix reduction that verifies the vec4 subgroup reducer and portable workgroup-tree fallback."
},
"vars": { "batch": 2, "channels": 64, "height": 256, "width": 256 },
"attrs": { "axes": [2, 3], "keepdims": 1 },
"inputs": { "x": { "shape": [2, 64, 256, 256], "dtype": "float32", "dist": "normal", "seed": 909, "scale": 0.2 } },
"outputs": { "y": { "shape": [2, 64, 1, 1], "dtype": "float32", "dist": "empty" } },
"bench": {
"metrics": [{ "type": "bandwidth", "value": "args.batch * args.channels * args.height * args.width * 4" }]
}
},
{
"name": "reducemean-rank4-axis1-largeaxis-2x131072x1x8",
"preset": "smoke",
"vars": { "numel": 2097152 },
"attrs": { "axes": [1], "keepdims": 0 },
"inputs": { "x": { "shape": [2, 131072, 1, 8], "dtype": "float32", "dist": "normal", "seed": 120, "scale": 0.2 } },
"outputs": { "y": { "shape": [2, 1, 8], "dtype": "float32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.numel * 4" }] }
}
]
}