Xenova's picture
Xenova HF Staff
sync 2e7068faf55e
6d3894e verified
Raw
History Blame
31.3 kB
{
"op": "com.microsoft.FusedConv",
"tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] },
"cases": [
{
"name": "fusedconv-f32-b1c16m32-32x32-k3",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 16,
"outChannels": 32,
"inH": 32,
"inW": 32,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 16, 32, 32], "dtype": "float32", "dist": "normal", "seed": 112, "scale": 0.2 },
"w": { "shape": [32, 16, 3, 3], "dtype": "float32", "dist": "normal", "seed": 113, "scale": 0.1 }
},
"outputs": { "y": { "shape": [1, 32, 32, 32], "dtype": "float32" } },
"bench": {
"primary": true,
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f32-bias-b1c16m32-32x32-k3",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 16,
"outChannels": 32,
"inH": 32,
"inW": 32,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 16, 32, 32], "dtype": "float32", "dist": "normal", "seed": 113, "scale": 0.2 },
"w": { "shape": [32, 16, 3, 3], "dtype": "float32", "dist": "normal", "seed": 114, "scale": 0.1 },
"bias": { "shape": [32], "dtype": "float32", "dist": "normal", "seed": 115, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 32, 32, 32], "dtype": "float32" } },
"bench": {
"primary": true,
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fused-flux2vae-dec-mid-512c-64x64-k3",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 512,
"outChannels": 512,
"inH": 64,
"inW": 64,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 512, 64, 64], "dtype": "float32", "dist": "normal", "seed": 340, "scale": 0.2 },
"w": { "shape": [512, 512, 3, 3], "dtype": "float32", "dist": "normal", "seed": 341, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 512, 64, 64], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fused-flux2vae-dec-up1-512c-128x128-k3",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 512,
"outChannels": 512,
"inH": 128,
"inW": 128,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 512, 128, 128], "dtype": "float32", "dist": "normal", "seed": 342, "scale": 0.2 },
"w": { "shape": [512, 512, 3, 3], "dtype": "float32", "dist": "normal", "seed": 343, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 512, 128, 128], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fused-flux2vae-attn-proj-512c-64x64-k1",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 512,
"outChannels": 512,
"inH": 64,
"inW": 64,
"kernelH": 1,
"kernelW": 1,
"strideH": 1,
"strideW": 1,
"padH": 0,
"padW": 0
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [0, 0, 0, 0] },
"inputs": {
"x": { "shape": [1, 512, 64, 64], "dtype": "float32", "dist": "normal", "seed": 344, "scale": 0.2 },
"w": { "shape": [512, 512, 1, 1], "dtype": "float32", "dist": "normal", "seed": 345, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 512, 64, 64], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fused-flux2vae-attn-proj-bias-512c-64x64-k1",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 512,
"outChannels": 512,
"inH": 64,
"inW": 64,
"kernelH": 1,
"kernelW": 1,
"strideH": 1,
"strideW": 1,
"padH": 0,
"padW": 0
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [0, 0, 0, 0] },
"inputs": {
"x": { "shape": [1, 512, 64, 64], "dtype": "float32", "dist": "normal", "seed": 346, "scale": 0.2 },
"w": { "shape": [512, 512, 1, 1], "dtype": "float32", "dist": "normal", "seed": 347, "scale": 0.02 },
"bias": { "shape": [512], "dtype": "float32", "dist": "normal", "seed": 348, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 512, 64, 64], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f32-1x1-b1c256m256-32x32",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 256,
"outChannels": 256,
"inH": 32,
"inW": 32,
"kernelH": 1,
"kernelW": 1,
"strideH": 1,
"strideW": 1,
"padH": 0,
"padW": 0
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [0, 0, 0, 0] },
"inputs": {
"x": { "shape": [1, 256, 32, 32], "dtype": "float32", "dist": "normal", "seed": 350, "scale": 0.2 },
"w": { "shape": [256, 256, 1, 1], "dtype": "float32", "dist": "normal", "seed": 351, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 256, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f32-1x1-bias-b1c256m256-32x32",
"preset": "smoke",
"provenance": {
"source": "authored for performance coverage",
"notes": "Bias twin of the paired non-bias case at the same 64 register tiles. No bias case sat in the 64-127 tile band, so the bias register-tile floor had no benchmark on either side of it."
},
"vars": {
"batch": 1,
"inChannels": 256,
"outChannels": 256,
"inH": 32,
"inW": 32,
"kernelH": 1,
"kernelW": 1,
"strideH": 1,
"strideW": 1,
"padH": 0,
"padW": 0
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [0, 0, 0, 0] },
"inputs": {
"x": { "shape": [1, 256, 32, 32], "dtype": "float32", "dist": "normal", "seed": 3521, "scale": 0.2 },
"w": { "shape": [256, 256, 1, 1], "dtype": "float32", "dist": "normal", "seed": 3522, "scale": 0.02 },
"bias": { "shape": [256], "dtype": "float32", "dist": "normal", "seed": 3523, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 256, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f32-im2col-b1c32m64-32x32-k3",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 32,
"outChannels": 64,
"inH": 32,
"inW": 32,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 32, 32, 32], "dtype": "float32", "dist": "normal", "seed": 352, "scale": 0.2 },
"w": { "shape": [64, 32, 3, 3], "dtype": "float32", "dist": "normal", "seed": 353, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 64, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f16-1x1-b1c512m512-64x64",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 512,
"outChannels": 512,
"inH": 64,
"inW": 64,
"kernelH": 1,
"kernelW": 1,
"strideH": 1,
"strideW": 1,
"padH": 0,
"padW": 0
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [0, 0, 0, 0] },
"inputs": {
"x": { "shape": [1, 512, 64, 64], "dtype": "float16", "dist": "normal", "seed": 354, "scale": 0.2 },
"w": { "shape": [512, 512, 1, 1], "dtype": "float16", "dist": "normal", "seed": 355, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 512, 64, 64], "dtype": "float16" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f16-b1c512m512-64x64-k3",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 512,
"outChannels": 512,
"inH": 64,
"inW": 64,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 512, 64, 64], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 },
"w": { "shape": [512, 512, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 512, 64, 64], "dtype": "float16" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f32-b1c64m128-56x56-k3s2",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 64,
"outChannels": 128,
"inH": 56,
"inW": 56,
"kernelH": 3,
"kernelW": 3,
"strideH": 2,
"strideW": 2,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [2, 2], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 64, 56, 56], "dtype": "float32", "dist": "normal", "seed": 358, "scale": 0.2 },
"w": { "shape": [128, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 359, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 128, 28, 28], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f32-group4-b1c64m64-32x32-k3",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 64,
"outChannels": 64,
"inH": 32,
"inW": 32,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "group": 4, "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 64, 32, 32], "dtype": "float32", "dist": "normal", "seed": 360, "scale": 0.2 },
"w": { "shape": [64, 16, 3, 3], "dtype": "float32", "dist": "normal", "seed": 361, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 64, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f32-stem-3ch-224-k7s2",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 3,
"outChannels": 64,
"inH": 224,
"inW": 224,
"kernelH": 7,
"kernelW": 7,
"strideH": 2,
"strideW": 2,
"padH": 3,
"padW": 3
},
"attrs": { "activation": "Relu", "strides": [2, 2], "pads": [3, 3, 3, 3] },
"inputs": {
"x": { "shape": [1, 3, 224, 224], "dtype": "float32", "dist": "normal", "seed": 1458, "scale": 0.2 },
"w": { "shape": [64, 3, 7, 7], "dtype": "float32", "dist": "normal", "seed": 1459, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 64, 112, 112], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-f32-resnet-stage-256c-14x14-k3",
"preset": "model",
"vars": {
"batch": 1,
"inChannels": 256,
"outChannels": 256,
"inH": 14,
"inW": 14,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 256, 14, 14], "dtype": "float32", "dist": "normal", "seed": 2458, "scale": 0.2 },
"w": { "shape": [256, 256, 3, 3], "dtype": "float32", "dist": "normal", "seed": 2459, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 256, 14, 14], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-1x1-tinyN-c256m256-8x8",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 256,
"outChannels": 256,
"inH": 8,
"inW": 8,
"kernelH": 1,
"kernelW": 1,
"strideH": 1,
"strideW": 1,
"padH": 0,
"padW": 0
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [0, 0, 0, 0] },
"inputs": {
"x": { "shape": [1, 256, 8, 8], "dtype": "float32", "dist": "normal", "seed": 951, "scale": 0.2 },
"w": { "shape": [256, 256, 1, 1], "dtype": "float32", "dist": "normal", "seed": 1642, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 256, 8, 8], "dtype": "float32" } },
"bench": {
"primary": true,
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-1x1-bigN-c256m256-64x64",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 256,
"outChannels": 256,
"inH": 64,
"inW": 64,
"kernelH": 1,
"kernelW": 1,
"strideH": 1,
"strideW": 1,
"padH": 0,
"padW": 0
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [0, 0, 0, 0] },
"inputs": {
"x": { "shape": [1, 256, 64, 64], "dtype": "float32", "dist": "normal", "seed": 7277, "scale": 0.2 },
"w": { "shape": [256, 256, 1, 1], "dtype": "float32", "dist": "normal", "seed": 1530, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 256, 64, 64], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-3x3-unalignedK-c48m64-32x32",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 48,
"outChannels": 64,
"inH": 32,
"inW": 32,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 48, 32, 32], "dtype": "float32", "dist": "normal", "seed": 3152, "scale": 0.2 },
"w": { "shape": [64, 48, 3, 3], "dtype": "float32", "dist": "normal", "seed": 7809, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 64, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-3x3-alignedK-c64m64-32x32",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 64,
"outChannels": 64,
"inH": 32,
"inW": 32,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 64, 32, 32], "dtype": "float32", "dist": "normal", "seed": 7770, "scale": 0.2 },
"w": { "shape": [64, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 2676, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 64, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-3x3-depthwise-c256-32x32",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 256,
"outChannels": 256,
"inH": 32,
"inW": 32,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "group": 256, "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 256, 32, 32], "dtype": "float32", "dist": "normal", "seed": 8532, "scale": 0.2 },
"w": { "shape": [256, 1, 3, 3], "dtype": "float32", "dist": "normal", "seed": 7192, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 256, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-3x3-dense-c256m256-32x32",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 256,
"outChannels": 256,
"inH": 32,
"inW": 32,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 256, 32, 32], "dtype": "float32", "dist": "normal", "seed": 839, "scale": 0.2 },
"w": { "shape": [256, 256, 3, 3], "dtype": "float32", "dist": "normal", "seed": 1375, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 256, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-3x3-dense-bias-c256m256-32x32",
"preset": "smoke",
"provenance": {
"source": "authored for performance coverage",
"notes": "Bias twin of the paired non-bias case at the same 64 register tiles. No bias case sat in the 64-127 tile band, so the bias register-tile floor had no benchmark on either side of it."
},
"vars": {
"batch": 1,
"inChannels": 256,
"outChannels": 256,
"inH": 32,
"inW": 32,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 256, 32, 32], "dtype": "float32", "dist": "normal", "seed": 3531, "scale": 0.2 },
"w": { "shape": [256, 256, 3, 3], "dtype": "float32", "dist": "normal", "seed": 3532, "scale": 0.05 },
"bias": { "shape": [256], "dtype": "float32", "dist": "normal", "seed": 3533, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 256, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-stem-cin3-k7s2-224",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 3,
"outChannels": 64,
"inH": 224,
"inW": 224,
"kernelH": 7,
"kernelW": 7,
"strideH": 2,
"strideW": 2,
"padH": 3,
"padW": 3
},
"attrs": { "activation": "Relu", "strides": [2, 2], "pads": [3, 3, 3, 3] },
"inputs": {
"x": { "shape": [1, 3, 224, 224], "dtype": "float32", "dist": "normal", "seed": 1633, "scale": 0.2 },
"w": { "shape": [64, 3, 7, 7], "dtype": "float32", "dist": "normal", "seed": 3717, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 64, 112, 112], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-mid-cin32-k3-112",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 32,
"outChannels": 64,
"inH": 112,
"inW": 112,
"kernelH": 3,
"kernelW": 3,
"strideH": 1,
"strideW": 1,
"padH": 1,
"padW": 1
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [1, 1, 1, 1] },
"inputs": {
"x": { "shape": [1, 32, 112, 112], "dtype": "float32", "dist": "normal", "seed": 895, "scale": 0.2 },
"w": { "shape": [64, 32, 3, 3], "dtype": "float32", "dist": "normal", "seed": 3583, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 64, 112, 112], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-1x1-oddM66-c64-32x32",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 64,
"outChannels": 66,
"inH": 32,
"inW": 32,
"kernelH": 1,
"kernelW": 1,
"strideH": 1,
"strideW": 1,
"padH": 0,
"padW": 0
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [0, 0, 0, 0] },
"inputs": {
"x": { "shape": [1, 64, 32, 32], "dtype": "float32", "dist": "normal", "seed": 7607, "scale": 0.2 },
"w": { "shape": [66, 64, 1, 1], "dtype": "float32", "dist": "normal", "seed": 4643, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 66, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-1x1-pow2M64-c64-32x32",
"preset": "smoke",
"vars": {
"batch": 1,
"inChannels": 64,
"outChannels": 64,
"inH": 32,
"inW": 32,
"kernelH": 1,
"kernelW": 1,
"strideH": 1,
"strideW": 1,
"padH": 0,
"padW": 0
},
"attrs": { "activation": "Relu", "strides": [1, 1], "pads": [0, 0, 0, 0] },
"inputs": {
"x": { "shape": [1, 64, 32, 32], "dtype": "float32", "dist": "normal", "seed": 8186, "scale": 0.2 },
"w": { "shape": [64, 64, 1, 1], "dtype": "float32", "dist": "normal", "seed": 7554, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 64, 32, 32], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-group2-k7-c64m64-64x64-direct-fallback",
"provenance": {
"notes": "Named for the direct_unrolled route it was authored to expose. It now selects grouped_large_kernel_w4, the wide-output grouped kernel FusedConv gained from ai.onnx.Conv; kept under its original name as the grouped route regression guard."
},
"preset": "stress",
"vars": {
"batch": 1,
"inChannels": 64,
"outChannels": 64,
"inH": 64,
"inW": 64,
"kernelH": 7,
"kernelW": 7,
"strideH": 1,
"strideW": 1,
"padH": 3,
"padW": 3,
"group": 2
},
"attrs": { "group": 2, "activation": "Relu", "strides": [1, 1], "pads": [3, 3, 3, 3] },
"inputs": {
"x": { "shape": [1, 64, 64, 64], "dtype": "float32", "dist": "normal", "seed": 4801, "scale": 0.2 },
"w": { "shape": [64, 32, 7, 7], "dtype": "float32", "dist": "normal", "seed": 4802, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 64, 64, 64], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-group4-k5-c128m128-48x48-direct-fallback",
"provenance": {
"notes": "Named for the direct_unrolled route it was authored to expose. It now selects grouped_large_kernel_w4, and its 5x5 kernel also sits in the band that kernel refused before its kernel-size floor was measured."
},
"preset": "stress",
"vars": {
"batch": 1,
"inChannels": 128,
"outChannels": 128,
"inH": 48,
"inW": 48,
"kernelH": 5,
"kernelW": 5,
"strideH": 1,
"strideW": 1,
"padH": 2,
"padW": 2,
"group": 4
},
"attrs": { "group": 4, "activation": "Relu", "strides": [1, 1], "pads": [2, 2, 2, 2] },
"inputs": {
"x": { "shape": [1, 128, 48, 48], "dtype": "float32", "dist": "normal", "seed": 4811, "scale": 0.2 },
"w": { "shape": [128, 32, 5, 5], "dtype": "float32", "dist": "normal", "seed": 4812, "scale": 0.02 }
},
"outputs": { "y": { "shape": [1, 128, 48, 48], "dtype": "float32" } },
"bench": {
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.w, 1) * args.kernelH * args.kernelW"
}
]
}
},
{
"name": "fusedconv-conv1d-hardswish-b1c80m512-w3000-k3s1",
"preset": "stress",
"provenance": {
"source": "ONNX Runtime FusedConv provider semantics",
"notes": "Large aligned temporal convolution with bias and provider-backed HardSwish on the register-tiled implicit-GEMM path."
},
"vars": { "batch": 1, "inChannels": 80, "outChannels": 512, "inW": 3000, "kernelW": 3, "strideW": 1, "padW": 1 },
"attrs": { "activation": "HardSwish", "strides": [1], "pads": [1, 1] },
"inputs": {
"x": { "shape": [1, 80, 3000], "dtype": "float32", "dist": "normal", "seed": 481, "scale": 0.2 },
"w": { "shape": [512, 80, 3], "dtype": "float32", "dist": "normal", "seed": 482, "scale": 0.1 },
"bias": { "shape": [512], "dtype": "float32", "dist": "normal", "seed": 483, "scale": 0.05 }
},
"outputs": { "y": { "shape": [1, 512, 3000], "dtype": "float32" } },
"bench": {
"primary": true,
"metrics": [
{
"type": "gflops",
"value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * args.inChannels * args.kernelW"
}
]
}
}
]
}