oidn-web 0.4.0 → 0.5.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +45 -0
- package/README.md +83 -19
- package/dist/oidn.js +3239 -2642
- package/dist/oidn.umd.cjs +512 -296
- package/lib/UNet.d.ts +54 -20
- package/lib/UNet.js +194 -118
- package/lib/UNet.js.map +1 -1
- package/lib/backend.d.ts +1 -8
- package/lib/backend.js +1 -9
- package/lib/backend.js.map +1 -1
- package/lib/finalRgbShader.d.ts +13 -0
- package/lib/finalRgbShader.js +160 -0
- package/lib/finalRgbShader.js.map +1 -0
- package/lib/graphOptimizer.js +1 -2
- package/lib/graphOptimizer.js.map +1 -1
- package/lib/hdrTransfer.d.ts +14 -0
- package/lib/hdrTransfer.js +61 -0
- package/lib/hdrTransfer.js.map +1 -0
- package/lib/main.d.ts +16 -7
- package/lib/main.js +4 -5
- package/lib/main.js.map +1 -1
- package/lib/nativeUNet.d.ts +39 -3
- package/lib/nativeUNet.js +449 -120
- package/lib/nativeUNet.js.map +1 -1
- package/lib/process.d.ts +5 -11
- package/lib/process.js +35 -49
- package/lib/process.js.map +1 -1
- package/lib/tileScheduler.d.ts +32 -4
- package/lib/tileScheduler.js +133 -20
- package/lib/tileScheduler.js.map +1 -1
- package/package.json +9 -2
- package/src/UNet.ts +287 -158
- package/src/backend.ts +1 -14
- package/src/finalRgbShader.ts +186 -0
- package/src/graphOptimizer.ts +1 -2
- package/src/hdrTransfer.ts +88 -0
- package/src/main.ts +28 -13
- package/src/nativeUNet.ts +515 -116
- package/src/process.ts +43 -70
- package/src/tileScheduler.ts +216 -24
- package/benchmarks/compare.mjs +0 -651
- package/benchmarks/leak.mjs +0 -255
- package/benchmarks/results/before-spatial.json +0 -391
- package/benchmarks/results/before-spatial.md +0 -47
- package/benchmarks/results/int8-scan.json +0 -2007
- package/benchmarks/results/int8-scan.md +0 -160
- package/benchmarks/results/int8-w8a8-scan.json +0 -2007
- package/benchmarks/results/int8-w8a8-scan.md +0 -160
- package/benchmarks/results/int8-weight-channel.json +0 -1413
- package/benchmarks/results/int8-weight-channel.md +0 -118
- package/benchmarks/results/int8-weight-only.json +0 -1437
- package/benchmarks/results/int8-weight-only.md +0 -118
- package/benchmarks/results/kernel-webnn-final.json +0 -1115
- package/benchmarks/results/kernel-webnn-final.md +0 -104
- package/benchmarks/results/latest-optimized.json +0 -375
- package/benchmarks/results/latest-optimized.md +0 -47
- package/benchmarks/results/latest.json +0 -391
- package/benchmarks/results/latest.md +0 -47
- package/benchmarks/results/profile-baseline.json +0 -331
- package/benchmarks/results/profile-baseline.md +0 -12
- package/benchmarks/results/profile-conv2x.json +0 -331
- package/benchmarks/results/profile-conv2x.md +0 -12
- package/benchmarks/results/profile-fast-init.json +0 -385
- package/benchmarks/results/profile-fast-init.md +0 -47
- package/benchmarks/results/profile-fp16-fma.json +0 -369
- package/benchmarks/results/profile-fp16-fma.md +0 -47
- package/benchmarks/results/profile-fp16-tiled-decoder.json +0 -369
- package/benchmarks/results/profile-fp16-tiled-decoder.md +0 -47
- package/benchmarks/results/profile-fp16-tiled-encoder.json +0 -369
- package/benchmarks/results/profile-fp16-tiled-encoder.md +0 -47
- package/benchmarks/results/profile-fp16-unfused-pool.json +0 -385
- package/benchmarks/results/profile-fp16-unfused-pool.md +0 -47
- package/benchmarks/results/profile-input-major.json +0 -347
- package/benchmarks/results/profile-input-major.md +0 -12
- package/benchmarks/results/profile-k16.json +0 -347
- package/benchmarks/results/profile-k16.md +0 -12
- package/benchmarks/results/profile-k4.json +0 -347
- package/benchmarks/results/profile-k4.md +0 -12
- package/benchmarks/results/profile-pool-reuse.json +0 -331
- package/benchmarks/results/profile-pool-reuse.md +0 -12
- package/benchmarks/results/profile-precompiled.json +0 -385
- package/benchmarks/results/profile-precompiled.md +0 -47
- package/benchmarks/results/profile-static-channels.json +0 -385
- package/benchmarks/results/profile-static-channels.md +0 -47
- package/benchmarks/results/profile-static-io.json +0 -385
- package/benchmarks/results/profile-static-io.md +0 -47
- package/benchmarks/results/profile-tiled-conv.json +0 -331
- package/benchmarks/results/profile-tiled-conv.md +0 -12
- package/benchmarks/results/profile-tiled-decoder.json +0 -347
- package/benchmarks/results/profile-tiled-decoder.md +0 -12
- package/benchmarks/results/profile-tiled-matmul.json +0 -331
- package/benchmarks/results/profile-tiled-matmul.md +0 -12
- package/benchmarks/results/profile-unfused-decoder.json +0 -379
- package/benchmarks/results/profile-unfused-decoder.md +0 -12
- package/benchmarks/results/profile-unfused-pool.json +0 -347
- package/benchmarks/results/profile-unfused-pool.md +0 -12
- package/benchmarks/results/spatial-auto.json +0 -575
- package/benchmarks/results/spatial-auto.md +0 -61
- package/benchmarks/results/subgroup-smoke.json +0 -1094
- package/benchmarks/results/subgroup-smoke.md +0 -104
- package/benchmarks/results/webnn-smoke.json +0 -739
- package/benchmarks/results/webnn-smoke.md +0 -76
- package/scripts/inspect-model.mjs +0 -64
- package/tests/modelSpec.test.mjs +0 -128
- package/tests/resourceLifecycle.test.mjs +0 -383
- package/tests/tileScheduler.test.mjs +0 -90
|
@@ -1,385 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"generatedAt": "2026-08-20T04:08:18.670Z",
|
|
3
|
-
"currentCommit": "5d9c1c8771e2a03c90bfadafec4b5dd28fada446",
|
|
4
|
-
"baselineCommit": "136c00c34aa5cf4561b4137034dbd643177f0a5f",
|
|
5
|
-
"settings": {
|
|
6
|
-
"width": 512,
|
|
7
|
-
"height": 512,
|
|
8
|
-
"tileSize": 512,
|
|
9
|
-
"warmup": 1,
|
|
10
|
-
"runs": 3,
|
|
11
|
-
"model": "rt_hdr_calb_cnrm_large.tza",
|
|
12
|
-
"gpuQueueCompletionIncluded": true
|
|
13
|
-
},
|
|
14
|
-
"browser": "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome",
|
|
15
|
-
"adapter": {
|
|
16
|
-
"vendor": "apple",
|
|
17
|
-
"architecture": "metal-3",
|
|
18
|
-
"device": "",
|
|
19
|
-
"description": ""
|
|
20
|
-
},
|
|
21
|
-
"results": [
|
|
22
|
-
{
|
|
23
|
-
"label": "TFJS (136c00c)",
|
|
24
|
-
"bundle": "baseline",
|
|
25
|
-
"baseline": true,
|
|
26
|
-
"initializationMs": 9.100000001490116,
|
|
27
|
-
"timesMs": [
|
|
28
|
-
34.899999998509884,
|
|
29
|
-
35,
|
|
30
|
-
35
|
|
31
|
-
],
|
|
32
|
-
"supportsFP16": true,
|
|
33
|
-
"adapter": {
|
|
34
|
-
"vendor": "apple",
|
|
35
|
-
"architecture": "metal-3",
|
|
36
|
-
"device": "",
|
|
37
|
-
"description": ""
|
|
38
|
-
},
|
|
39
|
-
"summary": {
|
|
40
|
-
"minMs": 34.899999998509884,
|
|
41
|
-
"medianMs": 35,
|
|
42
|
-
"meanMs": 34.966666666169964,
|
|
43
|
-
"p95Ms": 35,
|
|
44
|
-
"maxMs": 35
|
|
45
|
-
},
|
|
46
|
-
"messages": []
|
|
47
|
-
},
|
|
48
|
-
{
|
|
49
|
-
"label": "WGSL FP32",
|
|
50
|
-
"bundle": "current",
|
|
51
|
-
"baseline": false,
|
|
52
|
-
"precision": "fp32",
|
|
53
|
-
"initializationMs": 349.20000000298023,
|
|
54
|
-
"timesMs": [
|
|
55
|
-
35.899999998509884,
|
|
56
|
-
35.70000000298023,
|
|
57
|
-
35.79999999701977
|
|
58
|
-
],
|
|
59
|
-
"executionProfile": {
|
|
60
|
-
"totalMs": 35.147932000000004,
|
|
61
|
-
"layers": [
|
|
62
|
-
{
|
|
63
|
-
"id": "input-pack",
|
|
64
|
-
"durationMs": 0.050706
|
|
65
|
-
},
|
|
66
|
-
{
|
|
67
|
-
"id": "enc_conv1a",
|
|
68
|
-
"durationMs": 0.839336
|
|
69
|
-
},
|
|
70
|
-
{
|
|
71
|
-
"id": "enc_conv1b",
|
|
72
|
-
"durationMs": 3.830283
|
|
73
|
-
},
|
|
74
|
-
{
|
|
75
|
-
"id": "pool1",
|
|
76
|
-
"durationMs": 0.413356
|
|
77
|
-
},
|
|
78
|
-
{
|
|
79
|
-
"id": "enc_conv2a",
|
|
80
|
-
"durationMs": 1.413145
|
|
81
|
-
},
|
|
82
|
-
{
|
|
83
|
-
"id": "enc_conv2b",
|
|
84
|
-
"durationMs": 2.162776
|
|
85
|
-
},
|
|
86
|
-
{
|
|
87
|
-
"id": "pool2",
|
|
88
|
-
"durationMs": 0.058206
|
|
89
|
-
},
|
|
90
|
-
{
|
|
91
|
-
"id": "enc_conv3a",
|
|
92
|
-
"durationMs": 0.759923
|
|
93
|
-
},
|
|
94
|
-
{
|
|
95
|
-
"id": "enc_conv3b",
|
|
96
|
-
"durationMs": 0.987371
|
|
97
|
-
},
|
|
98
|
-
{
|
|
99
|
-
"id": "pool3",
|
|
100
|
-
"durationMs": 0.018124
|
|
101
|
-
},
|
|
102
|
-
{
|
|
103
|
-
"id": "enc_conv4a",
|
|
104
|
-
"durationMs": 0.401273
|
|
105
|
-
},
|
|
106
|
-
{
|
|
107
|
-
"id": "enc_conv4b",
|
|
108
|
-
"durationMs": 0.713717
|
|
109
|
-
},
|
|
110
|
-
{
|
|
111
|
-
"id": "pool4",
|
|
112
|
-
"durationMs": 0.008457
|
|
113
|
-
},
|
|
114
|
-
{
|
|
115
|
-
"id": "enc_conv5a",
|
|
116
|
-
"durationMs": 0.257156
|
|
117
|
-
},
|
|
118
|
-
{
|
|
119
|
-
"id": "enc_conv5b",
|
|
120
|
-
"durationMs": 0.335484
|
|
121
|
-
},
|
|
122
|
-
{
|
|
123
|
-
"id": "dec_conv4a",
|
|
124
|
-
"durationMs": 1.246193
|
|
125
|
-
},
|
|
126
|
-
{
|
|
127
|
-
"id": "dec_conv4b",
|
|
128
|
-
"durationMs": 0.610722
|
|
129
|
-
},
|
|
130
|
-
{
|
|
131
|
-
"id": "dec_conv3a",
|
|
132
|
-
"durationMs": 2.332436
|
|
133
|
-
},
|
|
134
|
-
{
|
|
135
|
-
"id": "dec_conv3b",
|
|
136
|
-
"durationMs": 0.991038
|
|
137
|
-
},
|
|
138
|
-
{
|
|
139
|
-
"id": "dec_conv2a",
|
|
140
|
-
"durationMs": 4.446797
|
|
141
|
-
},
|
|
142
|
-
{
|
|
143
|
-
"id": "dec_conv2b",
|
|
144
|
-
"durationMs": 2.161401
|
|
145
|
-
},
|
|
146
|
-
{
|
|
147
|
-
"id": "dec_conv1a",
|
|
148
|
-
"durationMs": 6.894392
|
|
149
|
-
},
|
|
150
|
-
{
|
|
151
|
-
"id": "dec_conv1b",
|
|
152
|
-
"durationMs": 3.709913
|
|
153
|
-
},
|
|
154
|
-
{
|
|
155
|
-
"id": "dec_conv1c",
|
|
156
|
-
"durationMs": 0.505727
|
|
157
|
-
}
|
|
158
|
-
]
|
|
159
|
-
},
|
|
160
|
-
"runtimeInfo": {
|
|
161
|
-
"configuredEngine": "wgsl",
|
|
162
|
-
"gpuEngine": "wgsl",
|
|
163
|
-
"precision": "fp32",
|
|
164
|
-
"model": "oidn-unet-large-v1",
|
|
165
|
-
"modelFamily": "oidn-unet-large",
|
|
166
|
-
"inputChannels": 9,
|
|
167
|
-
"dynamicTile": {
|
|
168
|
-
"enabled": false,
|
|
169
|
-
"currentTileSize": 512,
|
|
170
|
-
"minTileSize": 256,
|
|
171
|
-
"maxTileSize": 512,
|
|
172
|
-
"targetTileTimeMs": 16
|
|
173
|
-
},
|
|
174
|
-
"lastExecution": {
|
|
175
|
-
"width": 512,
|
|
176
|
-
"height": 512,
|
|
177
|
-
"tileWidth": 512,
|
|
178
|
-
"tileHeight": 512,
|
|
179
|
-
"tileCount": 1,
|
|
180
|
-
"durationMs": 37,
|
|
181
|
-
"tileTimeMs": {
|
|
182
|
-
"min": 37,
|
|
183
|
-
"median": 37,
|
|
184
|
-
"mean": 37,
|
|
185
|
-
"max": 37
|
|
186
|
-
}
|
|
187
|
-
}
|
|
188
|
-
},
|
|
189
|
-
"supportsFP16": true,
|
|
190
|
-
"adapter": {
|
|
191
|
-
"vendor": "apple",
|
|
192
|
-
"architecture": "metal-3",
|
|
193
|
-
"device": "",
|
|
194
|
-
"description": ""
|
|
195
|
-
},
|
|
196
|
-
"summary": {
|
|
197
|
-
"minMs": 35.70000000298023,
|
|
198
|
-
"medianMs": 35.79999999701977,
|
|
199
|
-
"meanMs": 35.79999999950329,
|
|
200
|
-
"p95Ms": 35.899999998509884,
|
|
201
|
-
"maxMs": 35.899999998509884
|
|
202
|
-
},
|
|
203
|
-
"messages": [],
|
|
204
|
-
"validation": {
|
|
205
|
-
"sampleCount": 6144,
|
|
206
|
-
"meanAbsoluteError": 0.0000011096847932397698,
|
|
207
|
-
"rootMeanSquaredError": 0.0000018071588418628667,
|
|
208
|
-
"maxAbsoluteError": 0.000013589859008789062,
|
|
209
|
-
"passed": true,
|
|
210
|
-
"thresholds": {
|
|
211
|
-
"maxMeanError": 0.0001,
|
|
212
|
-
"maxAbsoluteError": 0.001
|
|
213
|
-
}
|
|
214
|
-
}
|
|
215
|
-
},
|
|
216
|
-
{
|
|
217
|
-
"label": "WGSL FP16",
|
|
218
|
-
"bundle": "current",
|
|
219
|
-
"baseline": false,
|
|
220
|
-
"precision": "fp16",
|
|
221
|
-
"initializationMs": 23.100000001490116,
|
|
222
|
-
"timesMs": [
|
|
223
|
-
27.80000000447035,
|
|
224
|
-
27.599999994039536,
|
|
225
|
-
27.5
|
|
226
|
-
],
|
|
227
|
-
"executionProfile": {
|
|
228
|
-
"totalMs": 26.778235000000006,
|
|
229
|
-
"layers": [
|
|
230
|
-
{
|
|
231
|
-
"id": "input-pack",
|
|
232
|
-
"durationMs": 0.044664
|
|
233
|
-
},
|
|
234
|
-
{
|
|
235
|
-
"id": "enc_conv1a",
|
|
236
|
-
"durationMs": 0.618878
|
|
237
|
-
},
|
|
238
|
-
{
|
|
239
|
-
"id": "enc_conv1b",
|
|
240
|
-
"durationMs": 3.089854
|
|
241
|
-
},
|
|
242
|
-
{
|
|
243
|
-
"id": "pool1",
|
|
244
|
-
"durationMs": 0.34252
|
|
245
|
-
},
|
|
246
|
-
{
|
|
247
|
-
"id": "enc_conv2a",
|
|
248
|
-
"durationMs": 1.194968
|
|
249
|
-
},
|
|
250
|
-
{
|
|
251
|
-
"id": "enc_conv2b",
|
|
252
|
-
"durationMs": 1.96863
|
|
253
|
-
},
|
|
254
|
-
{
|
|
255
|
-
"id": "pool2",
|
|
256
|
-
"durationMs": 0.053871
|
|
257
|
-
},
|
|
258
|
-
{
|
|
259
|
-
"id": "enc_conv3a",
|
|
260
|
-
"durationMs": 0.674834
|
|
261
|
-
},
|
|
262
|
-
{
|
|
263
|
-
"id": "enc_conv3b",
|
|
264
|
-
"durationMs": 0.840991
|
|
265
|
-
},
|
|
266
|
-
{
|
|
267
|
-
"id": "pool3",
|
|
268
|
-
"durationMs": 0.016415
|
|
269
|
-
},
|
|
270
|
-
{
|
|
271
|
-
"id": "enc_conv4a",
|
|
272
|
-
"durationMs": 0.340396
|
|
273
|
-
},
|
|
274
|
-
{
|
|
275
|
-
"id": "enc_conv4b",
|
|
276
|
-
"durationMs": 0.52851
|
|
277
|
-
},
|
|
278
|
-
{
|
|
279
|
-
"id": "pool4",
|
|
280
|
-
"durationMs": 0.01129
|
|
281
|
-
},
|
|
282
|
-
{
|
|
283
|
-
"id": "enc_conv5a",
|
|
284
|
-
"durationMs": 0.183405
|
|
285
|
-
},
|
|
286
|
-
{
|
|
287
|
-
"id": "enc_conv5b",
|
|
288
|
-
"durationMs": 0.231277
|
|
289
|
-
},
|
|
290
|
-
{
|
|
291
|
-
"id": "dec_conv4a",
|
|
292
|
-
"durationMs": 0.943984
|
|
293
|
-
},
|
|
294
|
-
{
|
|
295
|
-
"id": "dec_conv4b",
|
|
296
|
-
"durationMs": 0.53001
|
|
297
|
-
},
|
|
298
|
-
{
|
|
299
|
-
"id": "dec_conv3a",
|
|
300
|
-
"durationMs": 1.699229
|
|
301
|
-
},
|
|
302
|
-
{
|
|
303
|
-
"id": "dec_conv3b",
|
|
304
|
-
"durationMs": 0.840032
|
|
305
|
-
},
|
|
306
|
-
{
|
|
307
|
-
"id": "dec_conv2a",
|
|
308
|
-
"durationMs": 2.969859
|
|
309
|
-
},
|
|
310
|
-
{
|
|
311
|
-
"id": "dec_conv2b",
|
|
312
|
-
"durationMs": 1.968963
|
|
313
|
-
},
|
|
314
|
-
{
|
|
315
|
-
"id": "dec_conv1a",
|
|
316
|
-
"durationMs": 4.403064
|
|
317
|
-
},
|
|
318
|
-
{
|
|
319
|
-
"id": "dec_conv1b",
|
|
320
|
-
"durationMs": 3.08827
|
|
321
|
-
},
|
|
322
|
-
{
|
|
323
|
-
"id": "dec_conv1c",
|
|
324
|
-
"durationMs": 0.194321
|
|
325
|
-
}
|
|
326
|
-
]
|
|
327
|
-
},
|
|
328
|
-
"runtimeInfo": {
|
|
329
|
-
"configuredEngine": "wgsl",
|
|
330
|
-
"gpuEngine": "wgsl",
|
|
331
|
-
"precision": "fp16",
|
|
332
|
-
"model": "oidn-unet-large-v1",
|
|
333
|
-
"modelFamily": "oidn-unet-large",
|
|
334
|
-
"inputChannels": 9,
|
|
335
|
-
"dynamicTile": {
|
|
336
|
-
"enabled": false,
|
|
337
|
-
"currentTileSize": 512,
|
|
338
|
-
"minTileSize": 256,
|
|
339
|
-
"maxTileSize": 512,
|
|
340
|
-
"targetTileTimeMs": 16
|
|
341
|
-
},
|
|
342
|
-
"lastExecution": {
|
|
343
|
-
"width": 512,
|
|
344
|
-
"height": 512,
|
|
345
|
-
"tileWidth": 512,
|
|
346
|
-
"tileHeight": 512,
|
|
347
|
-
"tileCount": 1,
|
|
348
|
-
"durationMs": 28.400000005960464,
|
|
349
|
-
"tileTimeMs": {
|
|
350
|
-
"min": 28.400000005960464,
|
|
351
|
-
"median": 28.400000005960464,
|
|
352
|
-
"mean": 28.400000005960464,
|
|
353
|
-
"max": 28.400000005960464
|
|
354
|
-
}
|
|
355
|
-
}
|
|
356
|
-
},
|
|
357
|
-
"supportsFP16": true,
|
|
358
|
-
"adapter": {
|
|
359
|
-
"vendor": "apple",
|
|
360
|
-
"architecture": "metal-3",
|
|
361
|
-
"device": "",
|
|
362
|
-
"description": ""
|
|
363
|
-
},
|
|
364
|
-
"summary": {
|
|
365
|
-
"minMs": 27.5,
|
|
366
|
-
"medianMs": 27.599999994039536,
|
|
367
|
-
"meanMs": 27.633333332836628,
|
|
368
|
-
"p95Ms": 27.80000000447035,
|
|
369
|
-
"maxMs": 27.80000000447035
|
|
370
|
-
},
|
|
371
|
-
"messages": [],
|
|
372
|
-
"validation": {
|
|
373
|
-
"sampleCount": 6144,
|
|
374
|
-
"meanAbsoluteError": 0.0010210334021394374,
|
|
375
|
-
"rootMeanSquaredError": 0.001609680379854656,
|
|
376
|
-
"maxAbsoluteError": 0.012395620346069336,
|
|
377
|
-
"passed": true,
|
|
378
|
-
"thresholds": {
|
|
379
|
-
"maxMeanError": 0.005,
|
|
380
|
-
"maxAbsoluteError": 0.05
|
|
381
|
-
}
|
|
382
|
-
}
|
|
383
|
-
}
|
|
384
|
-
]
|
|
385
|
-
}
|
|
@@ -1,47 +0,0 @@
|
|
|
1
|
-
# oidn-web benchmark
|
|
2
|
-
|
|
3
|
-
- Current: `5d9c1c8771e2a03c90bfadafec4b5dd28fada446`
|
|
4
|
-
- TFJS baseline: `136c00c34aa5cf4561b4137034dbd643177f0a5f`
|
|
5
|
-
- Input: 512x512, fixed tile 512, 3 runs after 1 warmup(s)
|
|
6
|
-
- Adapter: apple
|
|
7
|
-
|
|
8
|
-
| Runtime | Init ms | Median ms | P95 ms | Speedup | Precision |
|
|
9
|
-
| --- | ---: | ---: | ---: | ---: | --- |
|
|
10
|
-
| TFJS (136c00c) | 9.10 | 35.00 | 35.00 | 1.00x | fp32 |
|
|
11
|
-
| WGSL FP32 | 349.20 | 35.80 | 35.90 | 0.98x | fp32 |
|
|
12
|
-
| WGSL FP16 | 23.10 | 27.60 | 27.80 | 1.27x | fp16 |
|
|
13
|
-
|
|
14
|
-
## Output validation
|
|
15
|
-
|
|
16
|
-
Compared against sampled TFJS FP32 output.
|
|
17
|
-
|
|
18
|
-
| Runtime | Status | MAE | RMSE | Max error | Samples |
|
|
19
|
-
| --- | --- | ---: | ---: | ---: | ---: |
|
|
20
|
-
| WGSL FP32 | pass | 1.11e-6 | 1.81e-6 | 1.36e-5 | 6144 |
|
|
21
|
-
| WGSL FP16 | pass | 1.02e-3 | 1.61e-3 | 1.24e-2 | 6144 |
|
|
22
|
-
|
|
23
|
-
## GPU hot layers
|
|
24
|
-
|
|
25
|
-
### WGSL FP32
|
|
26
|
-
|
|
27
|
-
Profiled GPU total: 35.15 ms
|
|
28
|
-
|
|
29
|
-
| Node | GPU ms |
|
|
30
|
-
| --- | ---: |
|
|
31
|
-
| dec_conv1a | 6.89 |
|
|
32
|
-
| dec_conv2a | 4.45 |
|
|
33
|
-
| enc_conv1b | 3.83 |
|
|
34
|
-
| dec_conv1b | 3.71 |
|
|
35
|
-
| dec_conv3a | 2.33 |
|
|
36
|
-
|
|
37
|
-
### WGSL FP16
|
|
38
|
-
|
|
39
|
-
Profiled GPU total: 26.78 ms
|
|
40
|
-
|
|
41
|
-
| Node | GPU ms |
|
|
42
|
-
| --- | ---: |
|
|
43
|
-
| dec_conv1a | 4.40 |
|
|
44
|
-
| enc_conv1b | 3.09 |
|
|
45
|
-
| dec_conv1b | 3.09 |
|
|
46
|
-
| dec_conv2a | 2.97 |
|
|
47
|
-
| dec_conv2b | 1.97 |
|