Skip to content

Instantly share code, notes, and snippets.

Show Gist options
  • Save iree-github-actions-bot/1feefec4b28fab0448428171461deaab to your computer and use it in GitHub Desktop.
Save iree-github-actions-bot/1feefec4b28fab0448428171461deaab to your computer and use it in GitHub Desktop.

Full Benchmark Summary

Regressed Latencies 🚩

Benchmark Name Average Latency (ms) Median Latency (ms) Latency Standard Deviation (ms)
MobileNetV3Small [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-VMVX @ Pixel-6-Pro (CPU-ARMv8.2-A) 2246 (vs. 1839, 22.13%↑) 2221 92
MobileNetV3Small [fp32,imagenet] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 14 (vs. 13, 7.69%↑) 14 0

Improved Latencies 🎉

Benchmark Name Average Latency (ms) Median Latency (ms) Latency Standard Deviation (ms)
DeepLabV3 [fp32] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 134 (vs. 152, 11.84%↓) 135 8
MobileSSD [fp32] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 64 (vs. 69, 7.25%↓) 64 0

Similar Latencies

Benchmark Name Average Latency (ms) Median Latency (ms) Latency Standard Deviation (ms)
MobileNetV3Small [fp32,imagenet] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 5 (vs. 4, 25.00%↑) 5 1
MobileNetV2 [fp32,imagenet] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 9 (vs. 8, 12.50%↑) 8 0
MobileNetV3Small [fp32,imagenet] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 68 (vs. 77, 11.69%↓) 65 6
MobileNetV2 [fp32,imagenet] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 10 (vs. 9, 11.11%↑) 10 1
MobileNetV3Small [fp32,imagenet] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 32 (vs. 36, 11.11%↓) 33 5
MobileNetV3Small [fp32,imagenet] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 9 (vs. 10, 10.00%↓) 9 0
MobileSSD [fp32] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 643 (vs. 589, 9.17%↑) 642 9
DeepLabV3 [fp32] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 13 (vs. 12, 8.33%↑) 13 0
MobileNetV2 [fp32,imagenet] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 23 (vs. 25, 8.00%↓) 23 0
MobileSSD [fp32] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 590 (vs. 641, 7.96%↓) 590 1
DeepLabV3 [fp32] (TFLite) full-inference,default-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 15 (vs. 14, 7.14%↑) 15 0
DeepLabV3 [fp32] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 42 (vs. 45, 6.67%↓) 42 1
MobileNetV2 [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 15 (vs. 16, 6.25%↓) 15 0
DeepLabV3 [fp32] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 31 (vs. 33, 6.06%↓) 31 0
MobileNetV2 [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 17 (vs. 18, 5.56%↓) 17 0
MobileNetV2 [fp32,imagenet] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 41 (vs. 43, 4.65%↓) 41 0
MobileNetV2 [fp32,imagenet] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 227 (vs. 217, 4.61%↑) 227 3
MobileSSD [fp32] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 43 (vs. 45, 4.44%↓) 43 1
MobileNetV2 [fp32,imagenet] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 22 (vs. 23, 4.35%↓) 22 0
DeepLabV3 [fp32] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 45 (vs. 47, 4.26%↓) 45 0
MobileSSD [fp32] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 68 (vs. 71, 4.23%↓) 68 1
DeepLabV3 [fp32] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 47 (vs. 49, 4.08%↓) 47 1
MobileSSD [fp32] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 49 (vs. 51, 3.92%↓) 49 1
MobileNetV2 [fp32,imagenet] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 25 (vs. 26, 3.85%↓) 25 0
MobileNetV2 [fp32,imagenet] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 25 (vs. 26, 3.85%↓) 25 0
MobileBertSquad [fp32] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 279 (vs. 269, 3.72%↑) 281 9
MobileBertSquad [fp32] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 480 (vs. 498, 3.61%↓) 480 3
MobileNetV2 [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 27 (vs. 28, 3.57%↓) 27 0
MobileSSD [fp32] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 55 (vs. 57, 3.51%↓) 55 1
MobileNetV2 [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-VMVX @ Pixel-6-Pro (CPU-ARMv8.2-A) 8038 (vs. 8326, 3.46%↓) 8090 206
MobileSSD [fp32] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 56 (vs. 58, 3.45%↓) 56 1
PoseNet [fp32] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 30 (vs. 29, 3.45%↑) 30 1
MobileSSD [fp32] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 58 (vs. 60, 3.33%↓) 58 0
MobileBertSquad [int8] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 192 (vs. 186, 3.23%↑) 193 5
MobileBertSquad [fp32] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 274 (vs. 283, 3.18%↓) 276 8
DeepLabV3 [fp32] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 31 (vs. 32, 3.12%↓) 31 0
PoseNet [fp32] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 34 (vs. 35, 2.86%↓) 35 1
MobileNetV3Small [fp32,imagenet] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 72 (vs. 70, 2.86%↑) 72 0
DeepLabV3 [fp32] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 409 (vs. 421, 2.85%↓) 409 0
MobileBertSquad [fp32] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 341 (vs. 351, 2.85%↓) 343 7
MobileNetV2 [fp32,imagenet] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 35 (vs. 36, 2.78%↓) 35 0
MobileBertSquad [int8] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 560 (vs. 576, 2.78%↓) 560 1
MobileBertSquad [fp32] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 413 (vs. 402, 2.74%↑) 443 88
MobileSSD [fp32] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 39 (vs. 38, 2.63%↑) 39 1
PoseNet [fp32] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 39 (vs. 38, 2.63%↑) 38 2
PoseNet [fp32] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 80 (vs. 78, 2.56%↑) 79 2
MobileBertSquad [int8] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 575 (vs. 561, 2.50%↑) 569 19
MobileBertSquad [fp32] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 280 (vs. 287, 2.44%↓) 283 15
DeepLabV3 [fp32] (TFLite) full-inference,default-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 41 (vs. 42, 2.38%↓) 42 1
MobileNetV2 [fp32,imagenet] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 43 (vs. 42, 2.38%↑) 43 0
MobileBertSquad [fp32] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 336 (vs. 344, 2.33%↓) 337 6
PoseNet [fp32] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 1341 (vs. 1311, 2.29%↑) 1337 20
MobileBertSquad [int8] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 218 (vs. 223, 2.24%↓) 217 7
DeepLabV3 [fp32] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 44 (vs. 45, 2.22%↓) 44 1
MobileSSD [fp32] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 93 (vs. 91, 2.20%↑) 93 0
MobileSSD [fp32] (TFLite) full-inference,default-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 45 (vs. 46, 2.17%↓) 45 0
DeepLabV3 [fp32] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 46 (vs. 47, 2.13%↓) 46 1
PoseNet [fp32] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 1600 (vs. 1567, 2.11%↑) 1603 11
MobileNetV2 [fp32,imagenet] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 93 (vs. 95, 2.11%↓) 94 3
PoseNet [fp32] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 96 (vs. 98, 2.04%↓) 95 3
MobileSSD [fp32] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 102 (vs. 100, 2.00%↑) 102 0
DeepLabV3 [fp32] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 50 (vs. 51, 1.96%↓) 50 1
MobileNetV3Small [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-VMVX @ Pixel-4 (CPU-ARMv8.2-A) 1792 (vs. 1759, 1.88%↑) 1802 48
PoseNet [fp32] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 470 (vs. 462, 1.73%↑) 469 3
MobileNetV2 [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-VMVX @ Pixel-4 (CPU-ARMv8.2-A) 9553 (vs. 9402, 1.61%↑) 9636 283
PoseNet [fp32] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 186 (vs. 189, 1.59%↓) 186 0
MobileBertSquad [fp32] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 129 (vs. 127, 1.57%↑) 129 2
MobileSSD [fp32] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 684 (vs. 674, 1.48%↑) 683 3
MobileBertSquad [fp32] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 837 (vs. 825, 1.45%↑) 836 2
MobileNetV3Small [fp32,imagenet] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 70 (vs. 69, 1.45%↑) 70 1
DeepLabV3 [fp32] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 139 (vs. 141, 1.42%↓) 139 0
MobileNetV2 [fp32,imagenet] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 73 (vs. 74, 1.35%↓) 73 1
MobileBertSquad [int8] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 226 (vs. 229, 1.31%↓) 227 2
MobileNetV3Small [fp32,imagenet] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 78 (vs. 77, 1.30%↑) 78 1
DeepLabV3 [fp32] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 86 (vs. 85, 1.18%↑) 86 0
DeepLabV3 [fp32] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 425 (vs. 430, 1.16%↓) 426 8
PoseNet [fp32] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 278 (vs. 275, 1.09%↑) 278 0
DeepLabV3 [fp32] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 94 (vs. 93, 1.08%↑) 94 0
MobileSSD [fp32] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 94 (vs. 93, 1.08%↑) 94 0
MobileBertSquad [fp16] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 94 (vs. 93, 1.08%↑) 94 1
PoseNet [fp32] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 184 (vs. 186, 1.08%↓) 184 1
MobileBertSquad [fp32] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 624 (vs. 618, 0.97%↑) 624 1
MobileBertSquad [fp32] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 4356 (vs. 4315, 0.95%↑) 4357 10
MobileSSD [fp32] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 210 (vs. 212, 0.94%↓) 209 5
MobileBertSquad [int8] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 317 (vs. 320, 0.94%↓) 317 3
MobileSSD [fp32] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 111 (vs. 110, 0.91%↑) 111 0
MobileBertSquad [fp32] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 4084 (vs. 4050, 0.84%↑) 4119 68
MobileNetV2 [fp32,imagenet] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 242 (vs. 240, 0.83%↑) 242 1
DeepLabV3 [fp32] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 494 (vs. 490, 0.82%↑) 494 2
MobileBertSquad [fp32] (TFLite) full-inference,default-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 128 (vs. 127, 0.79%↑) 128 1
PoseNet [fp32] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 1559 (vs. 1571, 0.76%↓) 1559 3
MobileBertSquad [fp16] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 133 (vs. 132, 0.76%↑) 133 2
PoseNet [fp32] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 284 (vs. 282, 0.71%↑) 283 3
MobileBertSquad [int8] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 903 (vs. 897, 0.67%↑) 902 48
MobileBertSquad [fp32] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 628 (vs. 624, 0.64%↑) 628 2
MobileBertSquad [fp32] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 2914 (vs. 2930, 0.55%↓) 2921 20
DeepLabV3 [fp32] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 376 (vs. 374, 0.53%↑) 372 6
PoseNet [fp32] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 197 (vs. 196, 0.51%↑) 197 0
MobileBertSquad [int8] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 199 (vs. 198, 0.51%↑) 199 0
MobileBertSquad [fp32] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 4144 (vs. 4165, 0.50%↓) 4145 19
PoseNet [fp32] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 422 (vs. 420, 0.48%↑) 422 2
MobileSSD [fp32] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 461 (vs. 459, 0.44%↑) 461 1
MobileNetV2 [fp32,imagenet] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 229 (vs. 230, 0.43%↓) 228 1
MobileSSD [fp32] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 467 (vs. 465, 0.43%↑) 466 4
MobileBertSquad [fp32] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 726 (vs. 723, 0.41%↑) 726 2
PoseNet [fp32] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 1357 (vs. 1352, 0.37%↑) 1358 6
MobileBertSquad [fp32] (TFLite) full-inference,default-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 283 (vs. 284, 0.35%↓) 294 16
MobileBertSquad [int8] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 2883 (vs. 2873, 0.35%↑) 2884 10
MobileBertSquad [int8] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 1451 (vs. 1456, 0.34%↓) 1445 12
PoseNet [fp32] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 317 (vs. 316, 0.32%↑) 317 3
MobileBertSquad [fp32] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 1343 (vs. 1347, 0.30%↓) 1335 22
MobileBertSquad [fp32] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 1366 (vs. 1362, 0.29%↑) 1365 6
MobileSSD [fp32] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 680 (vs. 682, 0.29%↓) 679 4
MobileBertSquad [fp32] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 4360 (vs. 4348, 0.28%↑) 4361 11
MobileBertSquad [int8] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 798 (vs. 800, 0.25%↓) 797 5
MobileBertSquad [int8] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 1634 (vs. 1630, 0.25%↑) 1636 6
DeepLabV3 [fp32] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 437 (vs. 436, 0.23%↑) 437 3
MobileBertSquad [int8] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 2613 (vs. 2619, 0.23%↓) 2613 4
MobileBertSquad [fp32] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 3144 (vs. 3137, 0.22%↑) 3140 7
MobileBertSquad [fp32] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 469 (vs. 470, 0.21%↓) 452 59
MobileBertSquad [int8] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 1014 (vs. 1012, 0.20%↑) 1014 1
MobileBertSquad [fp32] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 525 (vs. 526, 0.19%↓) 527 18
MobileBertSquad [int8] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 2860 (vs. 2856, 0.14%↑) 2860 8
MobileBertSquad [int8] (TFLite) 1-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 2622 (vs. 2619, 0.11%↑) 2621 5
MobileBertSquad [int8] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 1187 (vs. 1186, 0.08%↑) 1188 1
DeepLabV3 [fp32] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 40 (vs. 40, 0.00%) 40 1
MobileNetV2 [fp32,imagenet] (TFLite) full-inference,default-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 11 (vs. 11, 0.00%) 11 1
MobileNetV2 [fp32,imagenet] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 7 (vs. 7, 0.00%) 7 0
MobileNetV3Small [fp32,imagenet] (TFLite) full-inference,default-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 5 (vs. 5, 0.00%) 5 1
MobileNetV3Small [fp32,imagenet] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 3 (vs. 3, 0.00%) 3 0
MobileSSD [fp32] (TFLite) full-inference,default-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 29 (vs. 29, 0.00%) 30 2
MobileSSD [fp32] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 26 (vs. 26, 0.00%) 26 1
MobileSSD [fp32] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 23 (vs. 23, 0.00%) 23 0
PoseNet [fp32] (TFLite) full-inference,default-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 37 (vs. 37, 0.00%) 38 2
PoseNet [fp32] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ XT2201-2 (GPU-Adreno-730) 31 (vs. 31, 0.00%) 31 0
DeepLabV3 [fp32] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 75 (vs. 75, 0.00%) 75 0
DeepLabV3 [fp32] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 89 (vs. 89, 0.00%) 89 0
DeepLabV3 [fp32] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 483 (vs. 483, 0.00%) 483 2
MobileBertSquad [int8] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 366 (vs. 366, 0.00%) 366 1
MobileBertSquad [int8] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 336 (vs. 336, 0.00%) 336 0
MobileBertSquad [int8] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 429 (vs. 429, 0.00%) 430 1
MobileNetV2 [fp32,imagenet] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 38 (vs. 38, 0.00%) 38 0
MobileNetV2 [fp32,imagenet] (TFLite) little-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 241 (vs. 241, 0.00%) 240 1
MobileNetV2 [fp32,imagenet] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 188 (vs. 188, 0.00%) 188 1
MobileNetV3Small [fp32,imagenet] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 12 (vs. 12, 0.00%) 12 0
MobileNetV3Small [fp32,imagenet] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 12 (vs. 12, 0.00%) 12 0
MobileNetV3Small [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 6 (vs. 6, 0.00%) 6 0
MobileNetV3Small [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 10 (vs. 10, 0.00%) 10 0
MobileNetV3Small [fp32,imagenet] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 24 (vs. 24, 0.00%) 24 0
MobileNetV3Small [fp32,imagenet] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 13 (vs. 13, 0.00%) 13 0
MobileNetV3Small [fp32,imagenet] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 63 (vs. 63, 0.00%) 63 0
MobileSSD [fp32] (TFLite) 4-thread,little-core,full-inference,default-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 183 (vs. 183, 0.00%) 183 1
PoseNet [fp32] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 63 (vs. 63, 0.00%) 63 0
PoseNet [fp32] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-4 (CPU-ARMv8.2-A) 38 (vs. 38, 0.00%) 38 0
PoseNet [fp32] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-4 (CPU-ARMv8.2-A) 68 (vs. 68, 0.00%) 68 0
DeepLabV3 [fp32] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 12 (vs. 12, 0.00%) 12 0
MobileBertSquad [fp16] (TFLite) full-inference,default-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 133 (vs. 133, 0.00%) 132 3
MobileBertSquad [fp32] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 143 (vs. 143, 0.00%) 143 1
MobileNetV2 [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 25 (vs. 25, 0.00%) 25 1
MobileNetV2 [fp32,imagenet] (TFLite) full-inference,default-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 10 (vs. 10, 0.00%) 10 0
MobileNetV2 [fp32,imagenet] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 9 (vs. 9, 0.00%) 9 0
MobileNetV2 [fp32,imagenet] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 183 (vs. 183, 0.00%) 182 3
MobileNetV3Small [fp32,imagenet] (TFLite) 1-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 7 (vs. 7, 0.00%) 7 0
MobileNetV3Small [fp32,imagenet] (TFLite) 1-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 8 (vs. 8, 0.00%) 8 0
MobileNetV3Small [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,default-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 6 (vs. 6, 0.00%) 6 0
MobileNetV3Small [fp32,imagenet] (TFLite) 4-thread,big-core,full-inference,experimental-flags with IREE-Dylib @ Pixel-6-Pro (CPU-ARMv8.2-A) 9 (vs. 9, 0.00%) 9 0
MobileNetV3Small [fp32,imagenet] (TFLite) big-core,full-inference,default-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 7 (vs. 7, 0.00%) 7 1
MobileNetV3Small [fp32,imagenet] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 8 (vs. 8, 0.00%) 8 0
MobileNetV3Small [fp32,imagenet] (TFLite) full-inference,default-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 10 (vs. 10, 0.00%) 10 0
MobileNetV3Small [fp32,imagenet] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 5 (vs. 5, 0.00%) 5 0
MobileNetV3Small [fp32,imagenet] (TFLite) little-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 64 (vs. 64, 0.00%) 64 0
MobileSSD [fp32] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 44 (vs. 44, 0.00%) 44 1
MobileSSD [fp32] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 40 (vs. 40, 0.00%) 40 0
PoseNet [fp32] (TFLite) big-core,full-inference,experimental-flags with IREE-Dylib-Sync @ Pixel-6-Pro (CPU-ARMv8.2-A) 35 (vs. 35, 0.00%) 35 1
PoseNet [fp32] (TFLite) full-inference,default-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 16 (vs. 16, 0.00%) 16 0
PoseNet [fp32] (TFLite) full-inference,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 15 (vs. 15, 0.00%) 15 0
PoseNet [fp32] (TFLite) kernel-execution,experimental-flags with IREE-Vulkan @ Pixel-6-Pro (GPU-Mali-G78) 14 (vs. 14, 0.00%) 14 0
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment