| { |
| "tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] }, |
| "cases": [ |
| { |
| "name": "conv-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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv1d-f32-b2c16m32-w256-k3", |
| "preset": "smoke", |
| "vars": { "batch": 2, "inChannels": 16, "outChannels": 32, "inW": 256, "kernelW": 3, "strideW": 1, "padW": 1 }, |
| "attrs": { "strides": [1], "pads": [1, 1] }, |
| "inputs": { |
| "x": { "shape": [2, 16, 256], "dtype": "float32", "dist": "normal", "seed": 362, "scale": 0.2 }, |
| "w": { "shape": [32, 16, 3], "dtype": "float32", "dist": "normal", "seed": 363, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [2, 32, 256], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * args.inChannels * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "whisper-stem1-1d-f32-b1c80m512-w3000-k3s1", |
| "preset": "smoke", |
| "provenance": { |
| "source": "Whisper encoder stem", |
| "notes": "Stage-1 stem conv (mel 80 -> 512, k=3, s=1, p=1, W=3000) covering the materialized subgroup-matrix Conv1d path and its implicit-GEMM fallbacks." |
| }, |
| "vars": { "batch": 1, "inChannels": 80, "outChannels": 512, "inW": 3000, "kernelW": 3, "strideW": 1, "padW": 1 }, |
| "attrs": { "strides": [1], "pads": [1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 80, 3000], "dtype": "float32", "dist": "normal", "seed": 371, "scale": 0.2 }, |
| "w": { "shape": [512, 80, 3], "dtype": "float32", "dist": "normal", "seed": 372, "scale": 0.1 }, |
| "bias": { "shape": [512], "dtype": "float32", "dist": "normal", "seed": 373, "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" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "whisper-stem2-1d-f32-b1c512m512-w3000-k3s2", |
| "preset": "smoke", |
| "provenance": { |
| "source": "Whisper encoder stem", |
| "notes": "Stage-2 stem conv (512 -> 512, k=3, s=2, p=1, W=3000 -> 1500) covering the materialized subgroup-matrix Conv1d path and its implicit-GEMM fallbacks." |
| }, |
| "vars": { "batch": 1, "inChannels": 512, "outChannels": 512, "inW": 3000, "kernelW": 3, "strideW": 2, "padW": 1 }, |
| "attrs": { "strides": [2], "pads": [1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 512, 3000], "dtype": "float32", "dist": "normal", "seed": 374, "scale": 0.2 }, |
| "w": { "shape": [512, 512, 3], "dtype": "float32", "dist": "normal", "seed": 375, "scale": 0.05 }, |
| "bias": { "shape": [512], "dtype": "float32", "dist": "normal", "seed": 376, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "shape": [1, 512, 1500], "dtype": "float32" } }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * args.inChannels * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "audio-feature-1d-f32-b1c256m256-w4096-k7-pathology", |
| "preset": "stress", |
| "tunableSpace": { "CONV1D_TILE_M": [4, 8], "CONV1D_TILE_N": [4, 8], "CONV1D_K_TILE": [16, 32] }, |
| "provenance": { |
| "source": "repository-authored benchmark", |
| "notes": "Realistic long-sequence audio feature block with K and output width aligned for the materialized subgroup-matrix Conv1d path; reduced-feature tiers retain the general implicit-GEMM and direct NCW implementations." |
| }, |
| "vars": { "batch": 1, "inChannels": 256, "outChannels": 256, "inW": 4096, "kernelW": 7, "strideW": 1, "padW": 3 }, |
| "attrs": { "strides": [1], "pads": [3, 3] }, |
| "inputs": { |
| "x": { "shape": [1, 256, 4096], "dtype": "float32", "dist": "normal", "seed": 368, "scale": 0.2 }, |
| "w": { "shape": [256, 256, 7], "dtype": "float32", "dist": "normal", "seed": 369, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "shape": [1, 256, 4096], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * args.inChannels * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv3d-f32-b1c4m8-8x8x8-k3", |
| "preset": "smoke", |
| "vars": { |
| "batch": 1, |
| "inChannels": 4, |
| "outChannels": 8, |
| "inD": 8, |
| "inH": 8, |
| "inW": 8, |
| "kernelD": 3, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideD": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padD": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1, 1], "pads": [1, 1, 1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 4, 8, 8, 8], "dtype": "float32", "dist": "normal", "seed": 364, "scale": 0.2 }, |
| "w": { "shape": [8, 4, 3, 3, 3], "dtype": "float32", "dist": "normal", "seed": 365, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 8, 8, 8, 8], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.y, 4) * args.inChannels * args.kernelD * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "medical-segmentation-3d-f32-b1c32m32-32x32x32-k3-pathology", |
| "preset": "stress", |
| "provenance": { |
| "source": "repository-authored benchmark", |
| "notes": "Realistic 3D medical-segmentation feature block for the direct NCDHW kernel. Each output voxel serially scans 32 input channels and a 3x3x3 kernel." |
| }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 32, |
| "inD": 32, |
| "inH": 32, |
| "inW": 32, |
| "kernelD": 3, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideD": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padD": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1, 1], "pads": [1, 1, 1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 32, 32, 32], "dtype": "float32", "dist": "normal", "seed": 366, "scale": 0.2 }, |
| "w": { "shape": [32, 32, 3, 3, 3], "dtype": "float32", "dist": "normal", "seed": 367, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "shape": [1, 32, 32, 32, 32], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * dim(shapes.y, 4) * args.inChannels * args.kernelD * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-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, "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) * args.inChannels * args.kernelH * args.kernelW / 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-group4-k5-64c-48x48-mid-kernel-band", |
| "provenance": { |
| "source": "repository-authored benchmark", |
| "notes": "A grouped 5x5 convolution exercises the wide-output grouped route at an intermediate kernel size between the 3x3 and 9x9 benchmark shapes." |
| }, |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 64, |
| "inH": 48, |
| "inW": 48, |
| "kernelH": 5, |
| "kernelW": 5, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 2, |
| "padW": 2 |
| }, |
| "attrs": { "group": 4, "strides": [1, 1], "pads": [2, 2, 2, 2] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 48, 48], "dtype": "float32", "dist": "normal", "seed": 4101, "scale": 0.2 }, |
| "w": { "shape": [64, 16, 5, 5], "dtype": "float32", "dist": "normal", "seed": 4102, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 48, 48], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW / 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-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": { "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-cout48-unaligned-28x28-k3", |
| "preset": "smoke", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 48, |
| "inH": 28, |
| "inW": 28, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 28, 28], "dtype": "float32", "dist": "normal", "seed": 701, "scale": 0.2 }, |
| "w": { "shape": [48, 32, 3, 3], "dtype": "float32", "dist": "normal", "seed": 702, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 48, 28, 28], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-cout64-aligned-28x28-k3", |
| "preset": "smoke", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 64, |
| "inH": 28, |
| "inW": 28, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 28, 28], "dtype": "float32", "dist": "normal", "seed": 703, "scale": 0.2 }, |
| "w": { "shape": [64, 32, 3, 3], "dtype": "float32", "dist": "normal", "seed": 704, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 28, 28], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f16-spatial196-unaligned-14x14-k3", |
| "preset": "smoke", |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 128, |
| "inH": 14, |
| "inW": 14, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 14, 14], "dtype": "float16", "dist": "normal", "seed": 711, "scale": 0.2 }, |
| "w": { "shape": [128, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 712, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 128, 14, 14], "dtype": "float16" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f16-spatial256-aligned-16x16-k3", |
| "preset": "smoke", |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 128, |
| "inH": 16, |
| "inW": 16, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 16, 16], "dtype": "float16", "dist": "normal", "seed": 713, "scale": 0.2 }, |
| "w": { "shape": [128, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 714, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 128, 16, 16], "dtype": "float16" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-depthwise-gcin-64c-56x56-k3", |
| "preset": "smoke", |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 64, |
| "inH": 56, |
| "inW": 56, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "group": 64, "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 56, 56], "dtype": "float32", "dist": "normal", "seed": 721, "scale": 0.2 }, |
| "w": { "shape": [64, 1, 3, 3], "dtype": "float32", "dist": "normal", "seed": 722, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 56, 56], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-dense-64c-56x56-k3", |
| "preset": "smoke", |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 64, |
| "inH": 56, |
| "inW": 56, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 56, 56], "dtype": "float32", "dist": "normal", "seed": 723, "scale": 0.2 }, |
| "w": { "shape": [64, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 724, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 56, 56], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "f16_grouped_large_kernel_k9_oc4_nchw_f16_fallback", |
| "preset": "smoke", |
| "vars": { |
| "batch": 1, |
| "inChannels": 4, |
| "outChannels": 16, |
| "inH": 12, |
| "inW": 12, |
| "kernelH": 9, |
| "kernelW": 9, |
| "group": 2 |
| }, |
| "attrs": { "group": 2, "strides": [1, 1], "pads": [4, 4, 4, 4] }, |
| "inputs": { |
| "x": { "shape": [1, 4, 12, 12], "dtype": "float16", "dist": "normal", "seed": 4091, "scale": 0.25 }, |
| "w": { "shape": [16, 2, 9, 9], "dtype": "float16", "dist": "normal", "seed": 4092, "scale": 0.2 } |
| }, |
| "outputs": { "y": { "shape": [1, 16, 12, 12], "dtype": "float16" } }, |
| "bench": { |
| "primary": true, |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * (args.inChannels / args.group) * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-group8-k9-256c-64x64-nchw-fallback", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 256, |
| "outChannels": 256, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 9, |
| "kernelW": 9, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 4, |
| "padW": 4, |
| "group": 8 |
| }, |
| "attrs": { "group": 8, "strides": [1, 1], "pads": [4, 4, 4, 4] }, |
| "inputs": { |
| "x": { "shape": [1, 256, 64, 64], "dtype": "float32", "dist": "normal", "seed": 4101, "scale": 0.2 }, |
| "w": { "shape": [256, 32, 9, 9], "dtype": "float32", "dist": "normal", "seed": 4102, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 256, 64, 64], "dtype": "float32", "dist": "empty" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "bandwidth", |
| "name": "one-touch lower-bound BW", |
| "value": "4 * (numel(shapes.x) + numel(shapes.w) + numel(shapes.y))" |
| }, |
| { |
| "type": "gflops", |
| "name": "nominal Conv FLOP/s", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW / args.group" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f16-group8-k9-256c-64x64-nchw-f16-fallback", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 256, |
| "outChannels": 256, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 9, |
| "kernelW": 9, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 4, |
| "padW": 4, |
| "group": 8 |
| }, |
| "attrs": { "group": 8, "strides": [1, 1], "pads": [4, 4, 4, 4] }, |
| "inputs": { |
| "x": { "shape": [1, 256, 64, 64], "dtype": "float16", "dist": "normal", "seed": 4111, "scale": 0.2 }, |
| "w": { "shape": [256, 32, 9, 9], "dtype": "float16", "dist": "normal", "seed": 4112, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 256, 64, 64], "dtype": "float16", "dist": "empty" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "bandwidth", |
| "name": "one-touch lower-bound BW", |
| "value": "2 * (numel(shapes.x) + numel(shapes.w) + numel(shapes.y))" |
| }, |
| { |
| "type": "gflops", |
| "name": "nominal Conv FLOP/s", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW / args.group" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-group4-k11-dilated-128c-48x48-nchw-fallback", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 128, |
| "outChannels": 128, |
| "inH": 48, |
| "inW": 48, |
| "kernelH": 11, |
| "kernelW": 11, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 10, |
| "padW": 10, |
| "dilationH": 2, |
| "dilationW": 2, |
| "group": 4 |
| }, |
| "attrs": { "group": 4, "strides": [1, 1], "dilations": [2, 2], "pads": [10, 10, 10, 10] }, |
| "inputs": { |
| "x": { "shape": [1, 128, 48, 48], "dtype": "float32", "dist": "normal", "seed": 4121, "scale": 0.2 }, |
| "w": { "shape": [128, 32, 11, 11], "dtype": "float32", "dist": "normal", "seed": 4122, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 128, 48, 48], "dtype": "float32", "dist": "empty" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "bandwidth", |
| "name": "one-touch lower-bound BW", |
| "value": "4 * (numel(shapes.x) + numel(shapes.w) + numel(shapes.y))" |
| }, |
| { |
| "type": "gflops", |
| "name": "nominal Conv FLOP/s", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW / args.group" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "manifest-gap-conv-f32-group8-k9-256c-64x63-w4-boundary", |
| "preset": "stress", |
| "provenance": { |
| "source": "repository-authored benchmark", |
| "notes": "A realistic grouped 9x9 convolution produces 63 columns, so the final four-lane output group contains three valid columns and uses masked tail stores." |
| }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 256, |
| "outChannels": 256, |
| "inH": 64, |
| "inW": 63, |
| "kernelH": 9, |
| "kernelW": 9, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 4, |
| "padW": 4, |
| "group": 8 |
| }, |
| "attrs": { "group": 8, "strides": [1, 1], "pads": [4, 4, 4, 4] }, |
| "inputs": { |
| "x": { "shape": [1, 256, 64, 63], "dtype": "float32", "dist": "normal", "seed": 9131, "scale": 0.2 }, |
| "w": { "shape": [256, 32, 9, 9], "dtype": "float32", "dist": "normal", "seed": 9132, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 256, 64, 63], "dtype": "float32", "dist": "empty" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "bandwidth", |
| "name": "one-touch lower-bound BW", |
| "value": "4 * (numel(shapes.x) + numel(shapes.w) + numel(shapes.y))" |
| }, |
| { |
| "type": "gflops", |
| "name": "nominal Conv FLOP/s", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW / args.group" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-group4-k11-dilated6-128c-48x48-register-span-cap", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 128, |
| "outChannels": 128, |
| "inH": 48, |
| "inW": 48, |
| "kernelH": 11, |
| "kernelW": 11, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 30, |
| "padW": 30, |
| "dilationH": 6, |
| "dilationW": 6, |
| "group": 4 |
| }, |
| "attrs": { "group": 4, "strides": [1, 1], "dilations": [6, 6], "pads": [30, 30, 30, 30] }, |
| "inputs": { |
| "x": { "shape": [1, 128, 48, 48], "dtype": "float32", "dist": "normal", "seed": 4121, "scale": 0.2 }, |
| "w": { "shape": [128, 32, 11, 11], "dtype": "float32", "dist": "normal", "seed": 4122, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 128, 48, 48], "dtype": "float32", "dist": "empty" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "bandwidth", |
| "name": "one-touch lower-bound BW", |
| "value": "4 * (numel(shapes.x) + numel(shapes.w) + numel(shapes.y))" |
| }, |
| { |
| "type": "gflops", |
| "name": "nominal Conv FLOP/s", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW / args.group" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-head-conv2-32c-518x686-k3", |
| "preset": "model", |
| "tunables": { "IMPLICIT_TILED_GATHER_MODE": 1 }, |
| "tunableSpace": { "IMPLICIT_TILED_GATHER_MODE": [0] }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 32, |
| "inH": 518, |
| "inW": 686, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 518, 686], "dtype": "float32", "dist": "normal", "seed": 9101, "scale": 0.2 }, |
| "w": { "shape": [32, 32, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9102, "scale": 0.1 }, |
| "bias": { "shape": [32], "dtype": "float32", "dist": "normal", "seed": 9103, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 32, 518, 686], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-implicit-m32-boundary-k72-n3072-bias", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 8, |
| "outChannels": 32, |
| "inH": 48, |
| "inW": 64, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 8, 48, 64], "dtype": "float32", "dist": "normal", "seed": 9151, "scale": 0.2 }, |
| "w": { "shape": [32, 8, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9152, "scale": 0.1 }, |
| "bias": { "shape": [32], "dtype": "float32", "dist": "normal", "seed": 9153, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 32, 48, 64], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f16-implicit-m32-boundary-k72-n3072-bias", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 8, |
| "outChannels": 32, |
| "inH": 48, |
| "inW": 64, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 8, 48, 64], "dtype": "float16", "dist": "normal", "seed": 9161, "scale": 0.2 }, |
| "w": { "shape": [32, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 9162, "scale": 0.1 }, |
| "bias": { "shape": [32], "dtype": "float16", "dist": "normal", "seed": 9163, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 32, 48, 64], "dtype": "float16" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-implicit-m32-large-boundary-k72-n32768-bias", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 8, |
| "outChannels": 32, |
| "inH": 128, |
| "inW": 256, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 8, 128, 256], "dtype": "float32", "dist": "normal", "seed": 9171, "scale": 0.2 }, |
| "w": { "shape": [32, 8, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9172, "scale": 0.1 }, |
| "bias": { "shape": [32], "dtype": "float32", "dist": "normal", "seed": 9173, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 32, 128, 256], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f16-implicit-m32-large-boundary-k72-n32768-bias", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 8, |
| "outChannels": 32, |
| "inH": 128, |
| "inW": 256, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 8, 128, 256], "dtype": "float16", "dist": "normal", "seed": 9181, "scale": 0.2 }, |
| "w": { "shape": [32, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 9182, "scale": 0.1 }, |
| "bias": { "shape": [32], "dtype": "float16", "dist": "normal", "seed": 9183, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 32, 128, 256], "dtype": "float16" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-head-conv1-64to32-296x392-k3", |
| "preset": "model", |
| "tunables": { "IMPLICIT_TILED_GATHER_MODE": 1 }, |
| "tunableSpace": { "IMPLICIT_TILED_GATHER_MODE": [0] }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 32, |
| "inH": 296, |
| "inW": 392, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 296, 392], "dtype": "float32", "dist": "normal", "seed": 9111, "scale": 0.2 }, |
| "w": { "shape": [32, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9112, "scale": 0.1 }, |
| "bias": { "shape": [32], "dtype": "float32", "dist": "normal", "seed": 9113, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 32, 296, 392], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-fusion-64c-148x196-k3", |
| "preset": "model", |
| "tunables": { "IMPLICIT_TILED_GATHER_MODE": 1 }, |
| "tunableSpace": { "IMPLICIT_TILED_GATHER_MODE": [0] }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 64, |
| "inH": 148, |
| "inW": 196, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 148, 196], "dtype": "float32", "dist": "normal", "seed": 9121, "scale": 0.2 }, |
| "w": { "shape": [64, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9122, "scale": 0.1 }, |
| "bias": { "shape": [64], "dtype": "float32", "dist": "normal", "seed": 9123, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 148, 196], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-neck-48to64-148x196-k3", |
| "preset": "model", |
| "tunables": { "IMPLICIT_TILED_GATHER_MODE": 1 }, |
| "tunableSpace": { "IMPLICIT_TILED_GATHER_MODE": [0] }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 48, |
| "outChannels": 64, |
| "inH": 148, |
| "inW": 196, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 48, 148, 196], "dtype": "float32", "dist": "normal", "seed": 9131, "scale": 0.2 }, |
| "w": { "shape": [64, 48, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9132, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 148, 196], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-neck-96to64-74x98-k3", |
| "preset": "model", |
| "tunables": { "IMPLICIT_TILED_GATHER_MODE": 1 }, |
| "tunableSpace": { "IMPLICIT_TILED_GATHER_MODE": [0] }, |
| "provenance": { "source": "Depth Anything V2 Small neck projection after the factor-2 reassemble branch" }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 96, |
| "outChannels": 64, |
| "inH": 74, |
| "inW": 98, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 96, 74, 98], "dtype": "float32", "dist": "normal", "seed": 9141, "scale": 0.2 }, |
| "w": { "shape": [64, 96, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9142, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 74, 98], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-implicit-boundary-oc64-spatial2048-k3", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 64, |
| "inH": 32, |
| "inW": 64, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 32, 64], "dtype": "float32", "dist": "normal", "seed": 8001, "scale": 0.2 }, |
| "w": { "shape": [64, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 8002, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 32, 64], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-implicit-mtiles-outside-oc256-spatial29008-k3", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 256, |
| "inH": 148, |
| "inW": 196, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 148, 196], "dtype": "float32", "dist": "normal", "seed": 8041, "scale": 0.2 }, |
| "w": { "shape": [256, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 8042, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 256, 148, 196], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-implicit-mtiles-inside-oc128-spatial4096-k3", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 128, |
| "outChannels": 128, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 128, 64, 64], "dtype": "float32", "dist": "normal", "seed": 8211, "scale": 0.2 }, |
| "w": { "shape": [128, 128, 3, 3], "dtype": "float32", "dist": "normal", "seed": 8212, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 128, 64, 64], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-implicit-mtiles-outside-oc512-spatial29008-k3", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 128, |
| "outChannels": 512, |
| "inH": 148, |
| "inW": 196, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 128, 148, 196], "dtype": "float32", "dist": "normal", "seed": 8231, "scale": 0.2 }, |
| "w": { "shape": [512, 128, 3, 3], "dtype": "float32", "dist": "normal", "seed": 8232, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 512, 148, 196], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-patchembed-3to384-518x686-k14s14", |
| "preset": "model", |
| "provenance": { |
| "notes": "Control for IMPLICIT_TILED_GATHER_MODE: stride equals kernel size, so this non-overlapping patch embed remains on the materialized im2col family." |
| }, |
| "tunables": { "IMPLICIT_TILED_GATHER_MODE": 1 }, |
| "tunableSpace": { "IMPLICIT_TILED_GATHER_MODE": [0] }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 3, |
| "outChannels": 384, |
| "inH": 518, |
| "inW": 686, |
| "kernelH": 14, |
| "kernelW": 14, |
| "strideH": 14, |
| "strideW": 14, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [14, 14], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 3, 518, 686], "dtype": "float32", "dist": "normal", "seed": 8001, "scale": 0.2 }, |
| "w": { "shape": [384, 3, 14, 14], "dtype": "float32", "dist": "normal", "seed": 8002, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 384, 37, 49], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-patchembed-3to384-518x924-k14s14", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 3, |
| "outChannels": 384, |
| "inH": 518, |
| "inW": 924, |
| "kernelH": 14, |
| "kernelW": 14, |
| "strideH": 14, |
| "strideW": 14, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [14, 14], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 3, 518, 924], "dtype": "float32", "dist": "normal", "seed": 8001, "scale": 0.2 }, |
| "w": { "shape": [384, 3, 14, 14], "dtype": "float32", "dist": "normal", "seed": 8002, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 384, 37, 66], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-neck-384to64-19x25-k3", |
| "preset": "model", |
| "tunableSpace": { "TILED_SPLIT_K_MODE": [0] }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 384, |
| "outChannels": 64, |
| "inH": 19, |
| "inW": 25, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 384, 19, 25], "dtype": "float32", "dist": "normal", "seed": 9141, "scale": 0.2 }, |
| "w": { "shape": [64, 384, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9142, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 19, 25], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-neck-192to64-37x49-k3", |
| "preset": "model", |
| "tunableSpace": { "TILED_SPLIT_K_MODE": [0] }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 192, |
| "outChannels": 64, |
| "inH": 37, |
| "inW": 49, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 192, 37, 49], "dtype": "float32", "dist": "normal", "seed": 9151, "scale": 0.2 }, |
| "w": { "shape": [64, 192, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9152, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 37, 49], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-fusion-64c-19x25-k3", |
| "preset": "model", |
| "tunableSpace": { "TILED_SPLIT_K_MODE": [0] }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 64, |
| "inH": 19, |
| "inW": 25, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 19, 25], "dtype": "float32", "dist": "normal", "seed": 9161, "scale": 0.2 }, |
| "w": { "shape": [64, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9162, "scale": 0.1 }, |
| "bias": { "shape": [64], "dtype": "float32", "dist": "normal", "seed": 9163, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 19, 25], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-fusion-64c-37x49-k3", |
| "preset": "model", |
| "tunableSpace": { "TILED_SPLIT_K_MODE": [0] }, |
| "vars": { |
| "batch": 1, |
| "inChannels": 64, |
| "outChannels": 64, |
| "inH": 37, |
| "inW": 49, |
| "kernelH": 3, |
| "kernelW": 3, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 1, |
| "padW": 1 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, |
| "inputs": { |
| "x": { "shape": [1, 64, 37, 49], "dtype": "float32", "dist": "normal", "seed": 9171, "scale": 0.2 }, |
| "w": { "shape": [64, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 9172, "scale": 0.1 }, |
| "bias": { "shape": [64], "dtype": "float32", "dist": "normal", "seed": 9173, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 37, 49], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-reassemble-proj-384to384-37x49-k1", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 384, |
| "outChannels": 384, |
| "inH": 37, |
| "inW": 49, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 384, 37, 49], "dtype": "float32", "dist": "normal", "seed": 9181, "scale": 0.2 }, |
| "w": { "shape": [384, 384, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9182, "scale": 0.1 }, |
| "bias": { "shape": [384], "dtype": "float32", "dist": "normal", "seed": 9183, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 384, 37, 49], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-reassemble-proj-384to96-37x49-k1", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 384, |
| "outChannels": 96, |
| "inH": 37, |
| "inW": 49, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 384, 37, 49], "dtype": "float32", "dist": "normal", "seed": 9191, "scale": 0.2 }, |
| "w": { "shape": [96, 384, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9192, "scale": 0.1 }, |
| "bias": { "shape": [96], "dtype": "float32", "dist": "normal", "seed": 9193, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 96, 37, 49], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "depthanything-head-conv3-32to1-518x686-k1", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 1, |
| "inH": 518, |
| "inW": 686, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 518, 686], "dtype": "float32", "dist": "normal", "seed": 9211, "scale": 0.2 }, |
| "w": { "shape": [1, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9212, "scale": 0.1 }, |
| "bias": { "shape": [1], "dtype": "float32", "dist": "normal", "seed": 9213, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 1, 518, 686], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-1x1-aligned-starved-256to64-32x32", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 256, |
| "outChannels": 64, |
| "inH": 32, |
| "inW": 32, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 256, 32, 32], "dtype": "float32", "dist": "normal", "seed": 9201, "scale": 0.2 }, |
| "w": { "shape": [64, 256, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9202, "scale": 0.1 }, |
| "bias": { "shape": [64], "dtype": "float32", "dist": "normal", "seed": 9203, "scale": 0.1 } |
| }, |
| "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to2-518x686-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 2, |
| "inH": 518, |
| "inW": 686, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 518, 686], "dtype": "float32", "dist": "normal", "seed": 9300, "scale": 0.2 }, |
| "w": { "shape": [2, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9301, "scale": 0.1 }, |
| "bias": { "shape": [2], "dtype": "float32", "dist": "normal", "seed": 9302, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 2, 518, 686], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to4-518x686-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 4, |
| "inH": 518, |
| "inW": 686, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 518, 686], "dtype": "float32", "dist": "normal", "seed": 9310, "scale": 0.2 }, |
| "w": { "shape": [4, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9311, "scale": 0.1 }, |
| "bias": { "shape": [4], "dtype": "float32", "dist": "normal", "seed": 9312, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 4, 518, 686], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to8-518x686-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 8, |
| "inH": 518, |
| "inW": 686, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 518, 686], "dtype": "float32", "dist": "normal", "seed": 9320, "scale": 0.2 }, |
| "w": { "shape": [8, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9321, "scale": 0.1 }, |
| "bias": { "shape": [8], "dtype": "float32", "dist": "normal", "seed": 9322, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 8, 518, 686], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to16-518x686-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 16, |
| "inH": 518, |
| "inW": 686, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 518, 686], "dtype": "float32", "dist": "normal", "seed": 9330, "scale": 0.2 }, |
| "w": { "shape": [16, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9331, "scale": 0.1 }, |
| "bias": { "shape": [16], "dtype": "float32", "dist": "normal", "seed": 9332, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 16, 518, 686], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to32-518x686-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 32, |
| "inH": 518, |
| "inW": 686, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 518, 686], "dtype": "float32", "dist": "normal", "seed": 9340, "scale": 0.2 }, |
| "w": { "shape": [32, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9341, "scale": 0.1 }, |
| "bias": { "shape": [32], "dtype": "float32", "dist": "normal", "seed": 9342, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 32, 518, 686], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to64-518x686-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 64, |
| "inH": 518, |
| "inW": 686, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 518, 686], "dtype": "float32", "dist": "normal", "seed": 9350, "scale": 0.2 }, |
| "w": { "shape": [64, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9351, "scale": 0.1 }, |
| "bias": { "shape": [64], "dtype": "float32", "dist": "normal", "seed": 9352, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 518, 686], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to1-32x32-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 1, |
| "inH": 32, |
| "inW": 32, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 32, 32], "dtype": "float32", "dist": "normal", "seed": 9360, "scale": 0.2 }, |
| "w": { "shape": [1, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9361, "scale": 0.1 }, |
| "bias": { "shape": [1], "dtype": "float32", "dist": "normal", "seed": 9362, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 1, 32, 32], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to1-64x64-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 1, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 64, 64], "dtype": "float32", "dist": "normal", "seed": 9370, "scale": 0.2 }, |
| "w": { "shape": [1, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9371, "scale": 0.1 }, |
| "bias": { "shape": [1], "dtype": "float32", "dist": "normal", "seed": 9372, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 1, 64, 64], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to1-128x128-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 1, |
| "inH": 128, |
| "inW": 128, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 128, 128], "dtype": "float32", "dist": "normal", "seed": 9380, "scale": 0.2 }, |
| "w": { "shape": [1, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9381, "scale": 0.1 }, |
| "bias": { "shape": [1], "dtype": "float32", "dist": "normal", "seed": 9382, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 1, 128, 128], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "sweep-pointwise-32to1-256x256-k1", |
| "preset": "stress", |
| "vars": { |
| "batch": 1, |
| "inChannels": 32, |
| "outChannels": 1, |
| "inH": 256, |
| "inW": 256, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 32, 256, 256], "dtype": "float32", "dist": "normal", "seed": 9390, "scale": 0.2 }, |
| "w": { "shape": [1, 32, 1, 1], "dtype": "float32", "dist": "normal", "seed": 9391, "scale": 0.1 }, |
| "bias": { "shape": [1], "dtype": "float32", "dist": "normal", "seed": 9392, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "shape": [1, 1, 256, 256], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| }, |
| { |
| "type": "bandwidth", |
| "value": "(args.inChannels * dim(shapes.y, 2) * dim(shapes.y, 3) + args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3)) * 4" |
| } |
| ] |
| } |
| }, |
| { |
| "name": "conv-f32-depthwise-96c-56x56-k3-bias", |
| "preset": "edge", |
| "attrs": { "group": 96, "kernel_shape": [3, 3], "pads": [1, 1, 1, 1], "strides": [1, 1] }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 96, 56, 56], "dist": "normal", "seed": 7, "scale": 1 }, |
| "w": { "dtype": "float32", "shape": [96, 1, 3, 3], "dist": "normal", "seed": 11, "scale": 0.2 }, |
| "bias": { "dtype": "float32", "shape": [96], "dist": "normal", "seed": 13, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 96, 56, 56] } }, |
| "bench": { "metrics": [{ "type": "gflops", "value": "2 * 96 * 56 * 56 * 9" }] } |
| }, |
| { |
| "name": "conv-f32-depthwise-96c-56x56-k3-no-bias-control", |
| "preset": "edge", |
| "attrs": { "group": 96, "kernel_shape": [3, 3], "pads": [1, 1, 1, 1], "strides": [1, 1] }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 96, 56, 56], "dist": "normal", "seed": 7, "scale": 1 }, |
| "w": { "dtype": "float32", "shape": [96, 1, 3, 3], "dist": "normal", "seed": 11, "scale": 0.2 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 96, 56, 56] } }, |
| "bench": { "metrics": [{ "type": "gflops", "value": "2 * 96 * 56 * 56 * 9" }] } |
| }, |
| { |
| "name": "conv-f32-depthwise-576c-14x14-k3-bias", |
| "preset": "edge", |
| "attrs": { "group": 576, "kernel_shape": [3, 3], "pads": [1, 1, 1, 1], "strides": [1, 1] }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [32, 576, 14, 14], "dist": "normal", "seed": 7, "scale": 1 }, |
| "w": { "dtype": "float32", "shape": [576, 1, 3, 3], "dist": "normal", "seed": 11, "scale": 0.2 }, |
| "bias": { "dtype": "float32", "shape": [576], "dist": "normal", "seed": 13, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [32, 576, 14, 14] } }, |
| "bench": { "metrics": [{ "type": "gflops", "value": "2 * 32 * 576 * 14 * 14 * 9" }] } |
| }, |
| { |
| "name": "conv-f32-depthwise-960c-7x7-k3-bias", |
| "preset": "edge", |
| "attrs": { "group": 960, "kernel_shape": [3, 3], "pads": [1, 1, 1, 1], "strides": [1, 1] }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [32, 960, 7, 7], "dist": "normal", "seed": 7, "scale": 1 }, |
| "w": { "dtype": "float32", "shape": [960, 1, 3, 3], "dist": "normal", "seed": 11, "scale": 0.2 }, |
| "bias": { "dtype": "float32", "shape": [960], "dist": "normal", "seed": 13, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [32, 960, 7, 7] } }, |
| "bench": { "metrics": [{ "type": "gflops", "value": "2 * 32 * 960 * 7 * 7 * 9" }] } |
| }, |
| { |
| "name": "conv-f32-depthwise-192c-28x28-k3-bias", |
| "preset": "edge", |
| "attrs": { "group": 192, "kernel_shape": [3, 3], "pads": [1, 1, 1, 1], "strides": [1, 1] }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [32, 192, 28, 28], "dist": "normal", "seed": 7, "scale": 1 }, |
| "w": { "dtype": "float32", "shape": [192, 1, 3, 3], "dist": "normal", "seed": 11, "scale": 0.2 }, |
| "bias": { "dtype": "float32", "shape": [192], "dist": "normal", "seed": 13, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [32, 192, 28, 28] } }, |
| "bench": { "metrics": [{ "type": "gflops", "value": "2 * 32 * 192 * 28 * 28 * 9" }] } |
| }, |
| { |
| "name": "conv-f16-64c-56x56-k3-bias-sgmat", |
| "preset": "edge", |
| "attrs": { "kernel_shape": [3, 3], "pads": [1, 1, 1, 1], "strides": [1, 1] }, |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 64, 56, 56], "dist": "normal", "seed": 5, "scale": 1 }, |
| "w": { "dtype": "float16", "shape": [64, 64, 3, 3], "dist": "normal", "seed": 9, "scale": 0.05 }, |
| "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 3, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 56, 56] } }, |
| "bench": { "metrics": [{ "type": "gflops", "value": "2 * 64 * 64 * 56 * 56 * 9" }] } |
| }, |
| { |
| "name": "conv-f32-64c-56x56-k3-bias-sgmat-control", |
| "preset": "edge", |
| "attrs": { "kernel_shape": [3, 3], "pads": [1, 1, 1, 1], "strides": [1, 1] }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 64, 56, 56], "dist": "normal", "seed": 5, "scale": 1 }, |
| "w": { "dtype": "float32", "shape": [64, 64, 3, 3], "dist": "normal", "seed": 9, "scale": 0.05 }, |
| "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 3, "scale": 0.1 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 56, 56] } }, |
| "bench": { "metrics": [{ "type": "gflops", "value": "2 * 64 * 64 * 56 * 56 * 9" }] } |
| }, |
| { |
| "name": "conv-reuse-float32-m32-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [32, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float32-m32-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [32, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float32-m64-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [64, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float32-m64-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [64, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float32-m96-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [96, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 96, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float32-m96-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [96, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 96, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float32-m128-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [128, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 128, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float32-m128-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [128, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 128, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float32-m256-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [256, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 256, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float32-m256-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float32", "shape": [256, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 256, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m32-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [32, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m32-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [32, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m64-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [64, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m64-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [64, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m96-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [96, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 96, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m96-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [96, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 96, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m128-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [128, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 128, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m128-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [128, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 128, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m256-c64-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 64, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [256, 64, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 256, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "conv-reuse-float16-m256-c256-n16384", |
| "preset": "edge", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [1, 256, 128, 128], "dist": "normal", "seed": 821, "scale": 0.2 }, |
| "w": { "dtype": "float16", "shape": [256, 256, 3, 3], "dist": "normal", "seed": 938, "scale": 0.05 } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [1, 256, 128, 128] } }, |
| "attrs": { "pads": [1, 1, 1, 1] }, |
| "bench": { |
| "metrics": [ |
| { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } |
| ] |
| } |
| }, |
| { |
| "name": "direct-1x1-float32-m63-k512-n4096", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 512, |
| "outChannels": 63, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 512, 64, 64], "dtype": "float32", "dist": "normal", "seed": 354, "scale": 0.2 }, |
| "w": { "shape": [63, 512, 1, 1], "dtype": "float32", "dist": "normal", "seed": 355, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 63, 64, 64], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| }, |
| "provenance": { |
| "source": "synthetic", |
| "notes": "Row-tile boundary sweep: full, partial, and device-dependent complete row bands share the same reduction and spatial extent." |
| } |
| }, |
| { |
| "name": "direct-1x1-float32-m64-k512-n4096", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 512, |
| "outChannels": 64, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 512, 64, 64], "dtype": "float32", "dist": "normal", "seed": 354, "scale": 0.2 }, |
| "w": { "shape": [64, 512, 1, 1], "dtype": "float32", "dist": "normal", "seed": 355, "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) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| }, |
| "provenance": { |
| "source": "synthetic", |
| "notes": "Row-tile boundary sweep: full, partial, and device-dependent complete row bands share the same reduction and spatial extent." |
| } |
| }, |
| { |
| "name": "direct-1x1-float32-m96-k512-n4096", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 512, |
| "outChannels": 96, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0] }, |
| "inputs": { |
| "x": { "shape": [1, 512, 64, 64], "dtype": "float32", "dist": "normal", "seed": 354, "scale": 0.2 }, |
| "w": { "shape": [96, 512, 1, 1], "dtype": "float32", "dist": "normal", "seed": 355, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 96, 64, 64], "dtype": "float32" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| }, |
| "provenance": { |
| "source": "synthetic", |
| "notes": "Row-tile boundary sweep: full, partial, and device-dependent complete row bands share the same reduction and spatial extent." |
| } |
| }, |
| { |
| "name": "direct-1x1-float16-m63-k512-n4096", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 512, |
| "outChannels": 63, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "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": [63, 512, 1, 1], "dtype": "float16", "dist": "normal", "seed": 355, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 63, 64, 64], "dtype": "float16" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| }, |
| "provenance": { |
| "source": "synthetic", |
| "notes": "Row-tile boundary sweep: full, partial, and device-dependent complete row bands share the same reduction and spatial extent." |
| } |
| }, |
| { |
| "name": "direct-1x1-float16-m64-k512-n4096", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 512, |
| "outChannels": 64, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "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": [64, 512, 1, 1], "dtype": "float16", "dist": "normal", "seed": 355, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 64, 64, 64], "dtype": "float16" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| }, |
| "provenance": { |
| "source": "synthetic", |
| "notes": "Row-tile boundary sweep: full, partial, and device-dependent complete row bands share the same reduction and spatial extent." |
| } |
| }, |
| { |
| "name": "direct-1x1-float16-m96-k512-n4096", |
| "preset": "model", |
| "vars": { |
| "batch": 1, |
| "inChannels": 512, |
| "outChannels": 96, |
| "inH": 64, |
| "inW": 64, |
| "kernelH": 1, |
| "kernelW": 1, |
| "strideH": 1, |
| "strideW": 1, |
| "padH": 0, |
| "padW": 0 |
| }, |
| "attrs": { "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": [96, 512, 1, 1], "dtype": "float16", "dist": "normal", "seed": 355, "scale": 0.02 } |
| }, |
| "outputs": { "y": { "shape": [1, 96, 64, 64], "dtype": "float16" } }, |
| "bench": { |
| "metrics": [ |
| { |
| "type": "gflops", |
| "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" |
| } |
| ] |
| }, |
| "provenance": { |
| "source": "synthetic", |
| "notes": "Row-tile boundary sweep: full, partial, and device-dependent complete row bands share the same reduction and spatial extent." |
| } |
| } |
| ] |
| } |
|
|
