[XPU] remove cast for elementwise_add(float, bf16/fp16) #64289
Conversation
dynamicheart
requested review from
XiaoguangHu01,
zhiqiu,
Xreki,
qili93 and
Aurelius84
as code owners
|
你的PR提交成功，感谢你对开源项目的贡献! |
dynamicheart
force-pushed
the
xpu_fp32_fp16_add
branch
from
e547462
to
229f75c
Compare
| ret = xpu::add_mul_type<float, XPUYType, float>( | ||
| dev_ctx.x_context(), x_data, y_data, z_data, x.numel()); | ||
| } | ||
| PADDLE_ENFORCE_XDNN_SUCCESS(ret, "add_mul_type"); |
There was a problem hiding this comment.
这里的报错信息支持common::errors::XX 类型提示么？如果支持的话，建议加一下报错类型
There was a problem hiding this comment.
目前原本的XPUElementwise里面也是调用的PADDLE_ENFORCE_XDNN_SUCCESS，包括其它XPU算子也是用的PADDLE_ENFORCE_XDNN_SUCCESS，至于common::errors::XX报错信息的建议后续可以专门批量去改，看看是否能加到PADDLE_ENFORCE_XDNN_SUCCESS宏里面
| @@ -145,6 +145,11 @@ inline bool NeedTypePromotion(const std::string& op_name, | |||
| (y == phi::DataType::BFLOAT16 || y == phi::DataType::FLOAT16)) { | |||
| return false; | |||
| } | |||
| #elif defined(PADDLE_WITH_XPU) | |||
| if ((op_name == "add" || op_name == "add_") && x == DataType::FLOAT32 && | |||
There was a problem hiding this comment.
@Aurelius84 @zxcd 麻烦看看CUDA版本是否是漏了add_算子，这样XPU和CUDA可以合并，add_(fp32， bf16/fp16)最一开始的需求是来自于main_grad的inplace_add：https://github.com/PaddlePaddle/Paddle/pull/54415/files#diff-1c8fea91efa1fa5f00f4b766fb11723120b74ec26365724cca4be8a0792946d9R62
PR Category
Custom Device
PR Types
Performance
Description
Reference PR: