swish, swish_grad add bfloat16 datatype support for XPU (#58440)

ykkk2333 · web-flow · commit 61df3fe7c421 · 2023-11-02T10:30:40.000+08:00
diff --git a/cmake/external/xpu.cmake b/cmake/external/xpu.cmake
@@ -24,7 +24,7 @@ set(XPU_XFT_LIB_NAME "libxft.so")
 set(XPU_XPTI_LIB_NAME "libxpti.so")
 
 if(NOT DEFINED XPU_BASE_DATE)
-  set(XPU_BASE_DATE "20231023")
+  set(XPU_BASE_DATE "20231025")
 endif()
 set(XPU_XCCL_BASE_VERSION "1.0.53.6")
 if(NOT DEFINED XPU_XFT_BASE_VERSION)
diff --git a/paddle/phi/backends/xpu/xpu2_op_list.cc b/paddle/phi/backends/xpu/xpu2_op_list.cc
@@ -864,7 +864,8 @@ XPUOpMap& get_kl2_ops() {
                      phi::DataType::INT32})},
       {"sum", XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
       {"swish", XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
-      {"swish_grad", XPUKernelSet({phi::DataType::FLOAT32})},
+      {"swish_grad",
+       XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
       {"take_along_axis",
        XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
       {"tanh_grad",
diff --git a/paddle/phi/backends/xpu/xpu3_op_list.cc b/paddle/phi/backends/xpu/xpu3_op_list.cc
@@ -839,8 +839,14 @@ XPUOpMap& get_kl3_ops() {
                      phi::DataType::INT16,
                      phi::DataType::INT32})},
       {"sum", XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
-      {"swish", XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
-      {"swish_grad", XPUKernelSet({phi::DataType::FLOAT32})},
+      {"swish",
+       XPUKernelSet({phi::DataType::FLOAT32,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::BFLOAT16})},
+      {"swish_grad",
+       XPUKernelSet({phi::DataType::FLOAT32,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::BFLOAT16})},
       {"take_along_axis",
        XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
       {"tanh_grad",
diff --git a/paddle/phi/kernels/xpu/activation_grad_kernel.cc b/paddle/phi/kernels/xpu/activation_grad_kernel.cc
@@ -700,6 +700,14 @@ PD_REGISTER_KERNEL(square_grad,
                    float,
                    phi::dtype::float16) {}
 
+PD_REGISTER_KERNEL(swish_grad,
+                   XPU,
+                   ALL_LAYOUT,
+                   phi::SwishGradKernel,
+                   float,
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
+
 PD_REGISTER_ACTIVATION_GRAD_KERNEL(exp_grad, ExpGradKernel)
 PD_REGISTER_ACTIVATION_GRAD_KERNEL(log_grad, LogGradKernel)
 PD_REGISTER_ACTIVATION_GRAD_KERNEL(leaky_relu_grad, LeakyReluGradKernel)
@@ -710,7 +718,6 @@ PD_REGISTER_ACTIVATION_GRAD_KERNEL(relu6_grad, Relu6GradKernel)
 PD_REGISTER_ACTIVATION_GRAD_KERNEL(sigmoid_grad, SigmoidGradKernel)
 PD_REGISTER_ACTIVATION_GRAD_KERNEL(sqrt_grad, SqrtGradKernel)
 PD_REGISTER_ACTIVATION_GRAD_KERNEL(mish_grad, MishGradKernel)
-PD_REGISTER_ACTIVATION_GRAD_KERNEL(swish_grad, SwishGradKernel)
 PD_REGISTER_ACTIVATION_GRAD_KERNEL(softplus_grad, SoftplusGradKernel)
 PD_REGISTER_ACTIVATION_GRAD_KERNEL(sin_grad, SinGradKernel)
 PD_REGISTER_ACTIVATION_GRAD_KERNEL(cos_grad, CosGradKernel)
diff --git a/paddle/phi/kernels/xpu/activation_kernel.cc b/paddle/phi/kernels/xpu/activation_kernel.cc
@@ -566,8 +566,13 @@ PD_REGISTER_KERNEL(
     elu, XPU, ALL_LAYOUT, phi::EluKernel, float, phi::dtype::float16) {}
 PD_REGISTER_KERNEL(
     sigmoid, XPU, ALL_LAYOUT, phi::SigmoidKernel, float, phi::dtype::float16) {}
-PD_REGISTER_KERNEL(
-    swish, XPU, ALL_LAYOUT, phi::SwishKernel, float, phi::dtype::float16) {}
+PD_REGISTER_KERNEL(swish,
+                   XPU,
+                   ALL_LAYOUT,
+                   phi::SwishKernel,
+                   float,
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
 PD_REGISTER_KERNEL(hardsigmoid,
                    XPU,
                    ALL_LAYOUT,
