| import torch | |
| from kernels.benchmark import Benchmark | |
| class FusedSplitBenchmark(Benchmark): | |
| def setup(self): | |
| self.size = 1024 | |
| self.input = torch.randn(self.size, self.size, device=self.device) | |
| self.out = torch.empty_like(self.input) | |
| def benchmark_base(self): | |
| self.kernel.fused_split(self.out, self.input) | |
| def verify_base(self) -> torch.Tensor: | |
| # Reference implementation | |
| return self.input.clone() |