Skip to content

P
Paddle

机器未来 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
未验证 提交 b5dd12fb 编写于 作者: Y Yanxing Shi 提交者： GitHub
浏览文件
操作

fix softmax max dim (#37901)

上级 a8f009e4
隐藏空白更改
内联 并排
Showing with 2 addition and 2 deletion
paddle/fluid/operators/softmax_cudnn_op.cu.h
浏览文件 @ b5dd12fb
......@@ -449,7 +449,7 @@ void SoftmaxForwardCUDAKernelDriver(const platform::CUDADeviceContext& dev_ctx,
const int N = SizeToAxis(axis, dims);
const int D = SizeOutAxis(axis, dims);
constexpr int max_dim = 320;
constexpr int max_dim = 512;
constexpr int warps_per_block = 4;
if (D == 1 && dim <= max_dim && sizeof(T) <= 4) {
......@@ -540,7 +540,7 @@ void SoftmaxBackwardCUDAKernelDriver(const platform::CUDADeviceContext& dev_ctx,
const int N = SizeToAxis(axis, dims);
const int D = SizeOutAxis(axis, dims);
constexpr int max_dim = 320;
constexpr int max_dim = 512;
constexpr int warps_per_block = 4;
if (D == 1 && dim <= max_dim && sizeof(T) <= 4) {
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或