Thanks for sending an issue! Please fill in the following template to help quickly solve your problem.
CheckFullLoadEnable()通过修改 curEnableFullLoad_,但conv3d_bp_func.h 和 conv3d_bp_kernel_split_func.h 中有 3 处仍直接读 self->ctx.tiling_->enableFullLoad(从未被修改,恒为初始值 1)
A5
核数在1核下执行用例, fmap:(4, 512, 1, 20, 16) filter:(512, 1, 2, 2, 256) bias:(256,) {'strides': (1, 1, 1, 2, 2), 'pads': [0, 0, 0, 0, 0, 0], 'dilations': (1, 1, 1, 1, 1), 'groups': 1}
精度正常
部分case精度只有80%多
/assign
Thanks for sending an issue! Please fill in the following template to help quickly solve your problem.
Describe the current behavior / 问题描述 (Mandatory / 必填)
CheckFullLoadEnable()通过修改 curEnableFullLoad_,但conv3d_bp_func.h 和 conv3d_bp_kernel_split_func.h 中有 3 处仍直接读 self->ctx.tiling_->enableFullLoad(从未被修改,恒为初始值 1)
Environment / 环境信息 (Mandatory / 必填)
A5
Steps to reproduce the issue / 重现步骤 (Mandatory / 必填)
核数在1核下执行用例,
fmap:(4, 512, 1, 20, 16)
filter:(512, 1, 2, 2, 256)
bias:(256,)
{'strides': (1, 1, 1, 2, 2), 'pads': [0, 0, 0, 0, 0, 0], 'dilations': (1, 1, 1, 1, 1), 'groups': 1}
Describe the expected behavior / 预期结果 (Mandatory / 必填)
精度正常
Related log / screenshot / 日志 / 截图 (Mandatory / 必填)
部分case精度只有80%多
Special notes for this issue/备注 (Optional / 选填)