[bf16] add bf16 kernel: squeeze & unsqueeze & stack (#39402)

* add squeeze unsqueeze stack

* add unittest

* add cpu kernel

paddle/fluid/operators/squeeze_op.cc

@@ -393,7 +393,9 @@ REGISTER_OP_CPU_KERNEL(
     ops::SqueezeKernel<paddle::platform::CPUDeviceContext,
                        paddle::platform::complex<float>>,
     ops::SqueezeKernel<paddle::platform::CPUDeviceContext,
-                       paddle::platform::complex<double>>);
+                       paddle::platform::complex<double>>,
+    ops::SqueezeKernel<paddle::platform::CPUDeviceContext,
+                       paddle::platform::bfloat16>);
 REGISTER_OP_CPU_KERNEL(
     squeeze_grad,
     ops::SqueezeGradKernel<paddle::platform::CPUDeviceContext, float>,
@@ -406,7 +408,9 @@ REGISTER_OP_CPU_KERNEL(
     ops::SqueezeGradKernel<paddle::platform::CPUDeviceContext,
                            paddle::platform::complex<float>>,
     ops::SqueezeGradKernel<paddle::platform::CPUDeviceContext,
-                           paddle::platform::complex<double>>);
+                           paddle::platform::complex<double>>,
+    ops::SqueezeGradKernel<paddle::platform::CPUDeviceContext,
+                           paddle::platform::bfloat16>);
 
 REGISTER_OP_CPU_KERNEL(
     squeeze2, ops::Squeeze2Kernel<paddle::platform::CPUDeviceContext, float>,
@@ -419,7 +423,9 @@ REGISTER_OP_CPU_KERNEL(
     ops::Squeeze2Kernel<paddle::platform::CPUDeviceContext,
                         paddle::platform::complex<float>>,
     ops::Squeeze2Kernel<paddle::platform::CPUDeviceContext,
-                        paddle::platform::complex<double>>);
+                        paddle::platform::complex<double>>,
+    ops::Squeeze2Kernel<paddle::platform::CPUDeviceContext,
+                        paddle::platform::bfloat16>);
 
 REGISTER_OP_CPU_KERNEL(
     squeeze2_grad,
@@ -433,4 +439,6 @@ REGISTER_OP_CPU_KERNEL(
     ops::Squeeze2GradKernel<paddle::platform::CPUDeviceContext,
                             paddle::platform::complex<float>>,
     ops::Squeeze2GradKernel<paddle::platform::CPUDeviceContext,
-                            paddle::platform::complex<double>>);
+                            paddle::platform::complex<double>>,
+    ops::Squeeze2GradKernel<paddle::platform::CPUDeviceContext,
+                            paddle::platform::bfloat16>);

paddle/fluid/operators/squeeze_op.cu.cc

@@ -21,6 +21,7 @@ REGISTER_OP_CUDA_KERNEL(
     squeeze, ops::SqueezeKernel<paddle::platform::CUDADeviceContext, float>,
     ops::SqueezeKernel<paddle::platform::CUDADeviceContext, double>,
     ops::SqueezeKernel<paddle::platform::CUDADeviceContext, plat::float16>,
+    ops::SqueezeKernel<paddle::platform::CUDADeviceContext, plat::bfloat16>,
     ops::SqueezeKernel<paddle::platform::CUDADeviceContext, bool>,
     ops::SqueezeKernel<paddle::platform::CUDADeviceContext, int>,
     ops::SqueezeKernel<paddle::platform::CUDADeviceContext, uint8_t>,
@@ -35,6 +36,7 @@ REGISTER_OP_CUDA_KERNEL(
     ops::SqueezeGradKernel<paddle::platform::CUDADeviceContext, float>,
     ops::SqueezeGradKernel<paddle::platform::CUDADeviceContext, double>,
     ops::SqueezeGradKernel<paddle::platform::CUDADeviceContext, plat::float16>,
+    ops::SqueezeGradKernel<paddle::platform::CUDADeviceContext, plat::bfloat16>,
     ops::SqueezeGradKernel<paddle::platform::CUDADeviceContext, bool>,
     ops::SqueezeGradKernel<paddle::platform::CUDADeviceContext, int>,
     ops::SqueezeGradKernel<paddle::platform::CUDADeviceContext, uint8_t>,
@@ -48,6 +50,7 @@ REGISTER_OP_CUDA_KERNEL(
     squeeze2, ops::Squeeze2Kernel<paddle::platform::CUDADeviceContext, float>,
     ops::Squeeze2Kernel<paddle::platform::CUDADeviceContext, double>,
     ops::Squeeze2Kernel<paddle::platform::CUDADeviceContext, plat::float16>,
+    ops::Squeeze2Kernel<paddle::platform::CUDADeviceContext, plat::bfloat16>,
     ops::Squeeze2Kernel<paddle::platform::CUDADeviceContext, bool>,
     ops::Squeeze2Kernel<paddle::platform::CUDADeviceContext, int>,
     ops::Squeeze2Kernel<paddle::platform::CUDADeviceContext, int8_t>,
@@ -62,6 +65,8 @@ REGISTER_OP_CUDA_KERNEL(
     ops::Squeeze2GradKernel<paddle::platform::CUDADeviceContext, float>,
     ops::Squeeze2GradKernel<paddle::platform::CUDADeviceContext, double>,
     ops::Squeeze2GradKernel<paddle::platform::CUDADeviceContext, plat::float16>,
+    ops::Squeeze2GradKernel<paddle::platform::CUDADeviceContext,
+                            plat::bfloat16>,
     ops::Squeeze2GradKernel<paddle::platform::CUDADeviceContext, bool>,
     ops::Squeeze2GradKernel<paddle::platform::CUDADeviceContext, int>,
     ops::Squeeze2GradKernel<paddle::platform::CUDADeviceContext, int8_t>,

paddle/fluid/operators/stack_op.cc

@@ -173,13 +173,16 @@ REGISTER_OPERATOR(stack, ops::StackOp, ops::StackOpMaker,
                   ops::StackGradOpMaker<paddle::imperative::OpBase>);
 REGISTER_OPERATOR(stack_grad, ops::StackOpGrad);
 
-REGISTER_OP_CPU_KERNEL(stack, ops::StackKernel<plat::CPUDeviceContext, float>,
-                       ops::StackKernel<plat::CPUDeviceContext, double>,
-                       ops::StackKernel<plat::CPUDeviceContext, int>,
-                       ops::StackKernel<plat::CPUDeviceContext, int64_t>);
-
-REGISTER_OP_CPU_KERNEL(stack_grad,
-                       ops::StackGradKernel<plat::CPUDeviceContext, float>,
-                       ops::StackGradKernel<plat::CPUDeviceContext, double>,
-                       ops::StackGradKernel<plat::CPUDeviceContext, int>,
-                       ops::StackGradKernel<plat::CPUDeviceContext, int64_t>);
+REGISTER_OP_CPU_KERNEL(
+    stack, ops::StackKernel<plat::CPUDeviceContext, float>,
+    ops::StackKernel<plat::CPUDeviceContext, double>,
+    ops::StackKernel<plat::CPUDeviceContext, int>,
+    ops::StackKernel<plat::CPUDeviceContext, int64_t>,
+    ops::StackKernel<plat::CPUDeviceContext, paddle::platform::bfloat16>);
+
+REGISTER_OP_CPU_KERNEL(
+    stack_grad, ops::StackGradKernel<plat::CPUDeviceContext, float>,
+    ops::StackGradKernel<plat::CPUDeviceContext, double>,
+    ops::StackGradKernel<plat::CPUDeviceContext, int>,
+    ops::StackGradKernel<plat::CPUDeviceContext, int64_t>,
+    ops::StackGradKernel<plat::CPUDeviceContext, paddle::platform::bfloat16>);

paddle/fluid/operators/stack_op.cu

@@ -196,10 +196,12 @@ class StackGradGPUKernel : public framework::OpKernel<T> {
 REGISTER_OP_CUDA_KERNEL(stack, ops::StackGPUKernel<float>,
                         ops::StackGPUKernel<double>, ops::StackGPUKernel<int>,
                         ops::StackGPUKernel<int64_t>,
-                        ops::StackGPUKernel<plat::float16>);
+                        ops::StackGPUKernel<plat::float16>,
+                        ops::StackGPUKernel<plat::bfloat16>);
 
 REGISTER_OP_CUDA_KERNEL(stack_grad, ops::StackGradGPUKernel<float>,
                         ops::StackGradGPUKernel<double>,
                         ops::StackGradGPUKernel<int>,
                         ops::StackGradGPUKernel<int64_t>,
-                        ops::StackGradGPUKernel<plat::float16>);
+                        ops::StackGradGPUKernel<plat::float16>,
+                        ops::StackGradGPUKernel<plat::bfloat16>);

paddle/fluid/operators/unsqueeze_op.cc

@@ -366,7 +366,9 @@ REGISTER_OP_CPU_KERNEL(
     ops::UnsqueezeKernel<paddle::platform::CPUDeviceContext,
                          paddle::platform::complex<float>>,
     ops::UnsqueezeKernel<paddle::platform::CPUDeviceContext,
-                         paddle::platform::complex<double>>);
+                         paddle::platform::complex<double>>,
+    ops::UnsqueezeKernel<paddle::platform::CPUDeviceContext,
+                         paddle::platform::bfloat16>);
 REGISTER_OP_CPU_KERNEL(
     unsqueeze_grad,
     ops::UnsqueezeGradKernel<paddle::platform::CPUDeviceContext, float>,
@@ -379,7 +381,9 @@ REGISTER_OP_CPU_KERNEL(
     ops::UnsqueezeGradKernel<paddle::platform::CPUDeviceContext,
                              paddle::platform::complex<float>>,
     ops::UnsqueezeGradKernel<paddle::platform::CPUDeviceContext,
-                             paddle::platform::complex<double>>);
+                             paddle::platform::complex<double>>,
+    ops::UnsqueezeGradKernel<paddle::platform::CPUDeviceContext,
+                             paddle::platform::bfloat16>);
 REGISTER_OP_CPU_KERNEL(
     unsqueeze2, ops::UnsqueezeKernel<paddle::platform::CPUDeviceContext, float>,
     ops::UnsqueezeKernel<paddle::platform::CPUDeviceContext, double>,
@@ -391,7 +395,9 @@ REGISTER_OP_CPU_KERNEL(
     ops::UnsqueezeKernel<paddle::platform::CPUDeviceContext,
                          paddle::platform::complex<float>>,
     ops::UnsqueezeKernel<paddle::platform::CPUDeviceContext,
-                         paddle::platform::complex<double>>);
+                         paddle::platform::complex<double>>,
+    ops::UnsqueezeKernel<paddle::platform::CPUDeviceContext,
+                         paddle::platform::bfloat16>);
 REGISTER_OP_CPU_KERNEL(
     unsqueeze2_grad,
     ops::Unsqueeze2GradKernel<paddle::platform::CPUDeviceContext, float>,
@@ -404,4 +410,6 @@ REGISTER_OP_CPU_KERNEL(
     ops::Unsqueeze2GradKernel<paddle::platform::CPUDeviceContext,
                               paddle::platform::complex<float>>,
     ops::Unsqueeze2GradKernel<paddle::platform::CPUDeviceContext,
-                              paddle::platform::complex<double>>);
+                              paddle::platform::complex<double>>,
+    ops::Unsqueeze2GradKernel<paddle::platform::CPUDeviceContext,
+                              paddle::platform::bfloat16>);

paddle/fluid/operators/unsqueeze_op.cu.cc

@@ -21,6 +21,7 @@ REGISTER_OP_CUDA_KERNEL(
     unsqueeze, ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, float>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, double>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, plat::float16>,
+    ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, plat::bfloat16>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, bool>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, int>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, uint8_t>,
@@ -36,6 +37,8 @@ REGISTER_OP_CUDA_KERNEL(
     ops::UnsqueezeGradKernel<paddle::platform::CUDADeviceContext, double>,
     ops::UnsqueezeGradKernel<paddle::platform::CUDADeviceContext,
                              plat::float16>,
+    ops::UnsqueezeGradKernel<paddle::platform::CUDADeviceContext,
+                             plat::bfloat16>,
     ops::UnsqueezeGradKernel<paddle::platform::CUDADeviceContext, bool>,
     ops::UnsqueezeGradKernel<paddle::platform::CUDADeviceContext, int>,
     ops::UnsqueezeGradKernel<paddle::platform::CUDADeviceContext, int8_t>,
@@ -50,6 +53,7 @@ REGISTER_OP_CUDA_KERNEL(
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, float>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, double>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, plat::float16>,
+    ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, plat::bfloat16>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, bool>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, int>,
     ops::UnsqueezeKernel<paddle::platform::CUDADeviceContext, uint8_t>,
@@ -65,6 +69,8 @@ REGISTER_OP_CUDA_KERNEL(
     ops::Unsqueeze2GradKernel<paddle::platform::CUDADeviceContext, double>,
     ops::Unsqueeze2GradKernel<paddle::platform::CUDADeviceContext,
                               plat::float16>,
+    ops::Unsqueeze2GradKernel<paddle::platform::CUDADeviceContext,
+                              plat::bfloat16>,
     ops::Unsqueeze2GradKernel<paddle::platform::CUDADeviceContext, bool>,
     ops::Unsqueeze2GradKernel<paddle::platform::CUDADeviceContext, int>,
     ops::Unsqueeze2GradKernel<paddle::platform::CUDADeviceContext, uint8_t>,

python/paddle/fluid/tests/unittests/test_squeeze_op.py

@@ -20,7 +20,8 @@ import numpy as np
 import paddle
 import paddle.fluid as fluid
 from paddle.fluid import compiler, Program, program_guard
-from op_test import OpTest
+from op_test import OpTest, convert_float_to_uint16
+import paddle.fluid.core as core
 
 paddle.enable_static()
 
@@ -49,6 +50,32 @@ class TestSqueezeOp(OpTest):
         self.attrs = {"axes": self.axes}
 
 
+class TestSqueezeBF16Op(OpTest):
+    def setUp(self):
+        self.op_type = "squeeze"
+        self.dtype = np.uint16
+        self.init_test_case()
+        x = np.random.random(self.ori_shape).astype("float32")
+        out = x.reshape(self.new_shape)
+        self.inputs = {"X": convert_float_to_uint16(x)}
+        self.init_attrs()
+        self.outputs = {"Out": convert_float_to_uint16(out)}
+
+    def test_check_output(self):
+        self.check_output()
+
+    def test_check_grad(self):
+        self.check_grad(["X"], "Out")
+
+    def init_test_case(self):
+        self.ori_shape = (1, 3, 1, 40)
+        self.axes = (0, 2)
+        self.new_shape = (3, 40)
+
+    def init_attrs(self):
+        self.attrs = {"axes": self.axes}
+
+
 # Correct: There is mins axis.
 class TestSqueezeOp1(TestSqueezeOp):
     def init_test_case(self):

python/paddle/fluid/tests/unittests/test_stack_op.py

@@ -16,7 +16,8 @@ import numpy as np
 import unittest
 import paddle
 import paddle.fluid as fluid
-from op_test import OpTest
+from op_test import OpTest, convert_float_to_uint16
+import paddle.fluid.core as core
 
 
 class TestStackOpBase(OpTest):
@@ -90,6 +91,49 @@ class TestStackOp6(TestStackOpBase):
         self.axis = 3
 
 
+class TestStackBF16Op(OpTest):
+    def initDefaultParameters(self):
+        self.num_inputs = 4
+        self.input_dim = (5, 6, 7)
+        self.axis = 0
+        self.dtype = np.uint16
+
+    def initParameters(self):
+        pass
+
+    def get_x_names(self):
+        x_names = []
+        for i in range(self.num_inputs):
+            x_names.append('x{}'.format(i))
+        return x_names
+
+    def setUp(self):
+        self.initDefaultParameters()
+        self.initParameters()
+        self.op_type = 'stack'
+        self.x = []
+        for i in range(self.num_inputs):
+            self.x.append(
+                np.random.random(size=self.input_dim).astype(np.float32))
+
+        out = np.stack(self.x, axis=self.axis)
+
+        tmp = []
+        x_names = self.get_x_names()
+        for i in range(self.num_inputs):
+            tmp.append((x_names[i], convert_float_to_uint16(self.x[i])))
+
+        self.inputs = {'X': tmp}
+        self.outputs = {'Y': convert_float_to_uint16(out)}
+        self.attrs = {'axis': self.axis}
+
+    def test_check_output(self):
+        self.check_output()
+
+    def test_check_grad(self):
+        self.check_grad(self.get_x_names(), 'Y')
+
+
 class TestStackAPIWithLoDTensorArray(unittest.TestCase):
     """
     Test stack api when the input(x) is a LoDTensorArray.

python/paddle/fluid/tests/unittests/test_unsqueeze_op.py

@@ -19,7 +19,8 @@ import numpy as np
 
 import paddle
 import paddle.fluid as fluid
-from op_test import OpTest
+from op_test import OpTest, convert_float_to_uint16
+import paddle.fluid.core as core
 
 paddle.enable_static()
 
@@ -48,6 +49,32 @@ class TestUnsqueezeOp(OpTest):
         self.attrs = {"axes": self.axes}
 
 
+class TestUnsqueezeBF16Op(OpTest):
+    def setUp(self):
+        self.init_test_case()
+        self.op_type = "unsqueeze"
+        self.dtype = np.uint16
+        x = np.random.random(self.ori_shape).astype("float32")
+        out = x.reshape(self.new_shape)
+        self.inputs = {"X": convert_float_to_uint16(x)}
+        self.init_attrs()
+        self.outputs = {"Out": convert_float_to_uint16(out)}
+
+    def test_check_output(self):
+        self.check_output()
+
+    def test_check_grad(self):
+        self.check_grad(["X"], "Out")
+
+    def init_test_case(self):
+        self.ori_shape = (3, 40)
+        self.axes = (1, 2)
+        self.new_shape = (3, 1, 1, 40)
+
+    def init_attrs(self):
+        self.attrs = {"axes": self.axes}
+
+
 # Correct: Single input index.
 class TestUnsqueezeOp1(TestUnsqueezeOp):
     def init_test_case(self):