[XPU] sigmoid and sigmoid_grad support type of bfloat16 (#72406)

zhangyk0314 · web-flow · commit 9b4329cee0eb · 2025-04-23T10:34:46.000+08:00
diff --git a/paddle/phi/backends/xpu/xpu3_op_list.cc b/paddle/phi/backends/xpu/xpu3_op_list.cc
@@ -1329,8 +1329,13 @@ XPUOpMap& get_kl3_ops() {
                      phi::DataType::FLOAT16,
                      phi::DataType::BFLOAT16})},
       {"sigmoid",
-       XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
-      {"sigmoid_grad", XPUKernelSet({phi::DataType::FLOAT32})},
+       XPUKernelSet({phi::DataType::FLOAT32,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::BFLOAT16})},
+      {"sigmoid_grad",
+       XPUKernelSet({phi::DataType::FLOAT32,
+                     phi::DataType::FLOAT16,
+                     phi::DataType::BFLOAT16})},
       {"sign", XPUKernelSet({phi::DataType::FLOAT32})},
       {"slice_grad",
        XPUKernelSet({phi::DataType::FLOAT32,
diff --git a/paddle/phi/kernels/xpu/activation_grad_kernel.cc b/paddle/phi/kernels/xpu/activation_grad_kernel.cc
@@ -763,7 +763,8 @@ PD_REGISTER_KERNEL(sigmoid_grad,
                    ALL_LAYOUT,
                    phi::SigmoidGradKernel,
                    float,
-                   phi::dtype::float16) {}
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
 
 PD_REGISTER_KERNEL(pow_grad,
                    XPU,
diff --git a/paddle/phi/kernels/xpu/activation_kernel.cc b/paddle/phi/kernels/xpu/activation_kernel.cc
@@ -641,8 +641,13 @@ PD_REGISTER_KERNEL(silu,
                    phi::dtype::bfloat16) {}
 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(sigmoid,
+                   XPU,
+                   ALL_LAYOUT,
+                   phi::SigmoidKernel,
+                   float,
+                   phi::dtype::float16,
+                   phi::dtype::bfloat16) {}
 PD_REGISTER_KERNEL(swish,
                    XPU,
                    ALL_LAYOUT,
diff --git a/test/xpu/test_activation_op_xpu.py b/test/xpu/test_activation_op_xpu.py
@@ -305,36 +305,39 @@ def set_case(self):
             self.op_type = "sigmoid"
             self.dtype = self.in_type
             self.init_config()
+            if self.dtype == np.uint16:
+                # bfloat16 actually
+                new_x = convert_float_to_uint16(self.x)
+            else:
+                new_x = self.x.astype(self.dtype)
             out = 1 / (1 + np.exp(-self.x))
 
             self.attrs = {'use_xpu': True}
-            self.inputs = {'X': OpTest.np_dtype_to_base_dtype(self.x)}
+            self.inputs = {'X': OpTest.np_dtype_to_base_dtype(new_x)}
             self.outputs = {'Out': out}
 
         def init_config(self):
-            self.x = np.random.uniform(-1, 1, [11, 17]).astype(self.dtype)
+            self.x = np.random.uniform(-1, 1, [11, 17])
 
     class XPUTestSigmoid_ZeroDIm(XPUTestSigmoid):
         def init_config(self):
-            self.x = np.random.uniform(-2, 2, []).astype(self.dtype)
+            self.x = np.random.uniform(-2, 2, [])
 
     class XPUTestSigmoid2(XPUTestSigmoid):
         def init_config(self):
-            self.x = np.random.uniform(-2, 2, [100]).astype(self.dtype)
+            self.x = np.random.uniform(-2, 2, [100])
 
     class XPUTestSigmoid3(XPUTestSigmoid):
         def init_config(self):
-            self.x = np.random.uniform(-2, 2, [10, 12, 15]).astype(self.dtype)
+            self.x = np.random.uniform(-2, 2, [10, 12, 15])
 
     class XPUTestSigmoid4(XPUTestSigmoid):
         def init_config(self):
-            self.x = np.random.uniform(-2, 2, [19, 19]).astype(self.dtype)
+            self.x = np.random.uniform(-2, 2, [19, 19])
 
     class XPUTestSigmoid5(XPUTestSigmoid):
         def init_config(self):
-            self.x = np.random.uniform(-2, 2, [10, 20, 30, 40]).astype(
-                self.dtype
-            )
+            self.x = np.random.uniform(-2, 2, [10, 20, 30, 40])
 
 
 support_types = get_xpu_op_support_types('sigmoid')
