1D Convolution Performance on Conv1d Production Shapes 1.0
0.016Latency (ms)KernelEvolve
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| KernelEvolvePrecision=FP16, Tensor Shape=64 × 96 × 96 × 2002025.12 | 0.016 | 1.91 | 1.26 | |
| KernelEvolvePrecision=FP16, Tensor Shape=128 × 96 × 96 × 2002025.12 | 0.0183 | 2.1 | 1.36 | |
| torch.conv2dPrecision=FP16, Tensor Shape=64 × 96 × 96 × 2002025.12 | 0.0202 | — | — | |
| KernelEvolvePrecision=FP32, Tensor Shape=64 × 96 × 96 × 2002025.12 | 0.0219 | 1.6 | 1.16 | |
| KernelEvolvePrecision=FP16, Tensor Shape=256 × 96 × 96 × 2002025.12 | 0.0284 | 1.87 | 1.18 | |
| torch.conv1dPrecision=FP16, Tensor Shape=64 × 96 × 96 × 2002025.12 | 0.0305 | — | — | |
| KernelEvolvePrecision=FP32, Tensor Shape=128 × 96 × 96 × 2002025.12 | 0.0351 | 1.32 | 0.93 | |
| KernelEvolvePrecision=FP16, Tensor Shape=512 × 96 × 96 × 2002025.12 | 0.0498 | 1.74 | 1.21 | |
| KernelEvolvePrecision=FP32, Tensor Shape=256 × 96 × 96 × 2002025.12 | 0.0579 | 1.24 | 0.99 | |
| KernelEvolvePrecision=FP16, Tensor Shape=1024 × 96 × 96 × 2002025.12 | 0.0841 | 2.05 | 1.34 | |
| KernelEvolvePrecision=FP32, Tensor Shape=512 × 96 × 96 × 2002025.12 | 0.1122 | 1.34 | 1.03 | |
| KernelEvolvePrecision=FP16, Tensor Shape=2048 × 96 × 96 × 2002025.12 | 0.1486 | 2.3 | 1.62 | |
| KernelEvolvePrecision=FP32, Tensor Shape=1024 × 96 × 96 × 2002025.12 | 0.1973 | 1.63 | 1.23 | |
| KernelEvolvePrecision=FP32, Tensor Shape=2048 × 96 × 96 × 2002025.12 | 0.3559 | 1.73 | 1.3 |