Age | Commit message (Expand) | Author |
---|---|---|
2021-11-29 | Use loop unrolling only when the kernel height is less than 5 | Gian Marco Iodice |
2021-11-17 | Improve start-up time for depthwise convolution | Sheri Zhang |
2021-10-14 | Implement CLDirectConv3D f32/f16 | Giorgio Arena |
2021-07-25 | Reorganize the kernels into nhwc, nchw and common folders | Adnan AlSinan |