!7545 GPU reshape add type

Merge pull request !7545 from VectorSL/int64

mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/slice_gpu_kernel.cc

@@ -24,5 +24,11 @@ MS_REG_GPU_KERNEL_ONE(Slice, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutp
                       SliceGpuFwdKernel, int)
 MS_REG_GPU_KERNEL_ONE(Slice, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
                       SliceGpuFwdKernel, half)
+MS_REG_GPU_KERNEL_ONE(Slice, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
+                      SliceGpuFwdKernel, int16_t)
+MS_REG_GPU_KERNEL_ONE(Slice, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
+                      SliceGpuFwdKernel, uchar)
+MS_REG_GPU_KERNEL_ONE(Slice, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool),
+                      SliceGpuFwdKernel, bool)
 }  // namespace kernel
 }  // namespace mindspore

mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/slice_grad_gpu_kernel.cc

@@ -29,5 +29,14 @@ MS_REG_GPU_KERNEL_ONE(
   SliceGrad,
   KernelAttr().AddInputAttr(kNumberTypeFloat16).AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
   SliceGradGpuKernel, half)
+MS_REG_GPU_KERNEL_ONE(
+  SliceGrad, KernelAttr().AddInputAttr(kNumberTypeInt16).AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
+  SliceGradGpuKernel, int16_t)
+MS_REG_GPU_KERNEL_ONE(
+  SliceGrad, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
+  SliceGradGpuKernel, uchar)
+MS_REG_GPU_KERNEL_ONE(
+  SliceGrad, KernelAttr().AddInputAttr(kNumberTypeBool).AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool),
+  SliceGradGpuKernel, bool)
 }  // namespace kernel
 }  // namespace mindspore

mindspore/ccsrc/backend/kernel_compiler/gpu/nn/flatten_gpu_kernel.cc

@@ -18,27 +18,68 @@
 
 namespace mindspore {
 namespace kernel {
+// float64
+MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeFloat64),
+                      FlattenGpuFwdKernel, double)
+MS_REG_GPU_KERNEL_ONE(Reshape, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeFloat64),
+                      FlattenGpuFwdKernel, double)
+MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeFloat64),
+                      FlattenGpuFwdKernel, double)
+// float32
 MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
                       FlattenGpuFwdKernel, float)
-MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
-                      FlattenGpuFwdKernel, int)
-MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
-                      FlattenGpuFwdKernel, half)
 MS_REG_GPU_KERNEL_ONE(Reshape, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
                       FlattenGpuFwdKernel, float)
+MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
+                      FlattenGpuFwdKernel, float)
+// float16
+MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
+                      FlattenGpuFwdKernel, half)
+MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
+                      FlattenGpuFwdKernel, half)
 MS_REG_GPU_KERNEL_ONE(Reshape, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
                       FlattenGpuFwdKernel, half)
+// int64
+MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64),
+                      FlattenGpuFwdKernel, int64_t)
+MS_REG_GPU_KERNEL_ONE(Reshape, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64),
+                      FlattenGpuFwdKernel, int64_t)
+MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64),
+                      FlattenGpuFwdKernel, int64_t)
+// int32
+MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
+                      FlattenGpuFwdKernel, int)
 MS_REG_GPU_KERNEL_ONE(Reshape, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
                       FlattenGpuFwdKernel, int)
+MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
+                      FlattenGpuFwdKernel, int)
+// int16
+MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
+                      FlattenGpuFwdKernel, int16_t)
+MS_REG_GPU_KERNEL_ONE(Reshape, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
+                      FlattenGpuFwdKernel, int16_t)
+MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
+                      FlattenGpuFwdKernel, int16_t)
+// uint8
+MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
+                      FlattenGpuFwdKernel, uchar)
 MS_REG_GPU_KERNEL_ONE(Reshape, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
                       FlattenGpuFwdKernel, uchar)
+MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
+                      FlattenGpuFwdKernel, uchar)
+// int8
+MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8),
+                      FlattenGpuFwdKernel, char)
+MS_REG_GPU_KERNEL_ONE(Reshape, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8),
+                      FlattenGpuFwdKernel, char)
+MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8),
+                      FlattenGpuFwdKernel, char)
+// bool
+MS_REG_GPU_KERNEL_ONE(Flatten, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool),
+                      FlattenGpuFwdKernel, bool)
 MS_REG_GPU_KERNEL_ONE(Reshape, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool),
                       FlattenGpuFwdKernel, bool)
-MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
-                      FlattenGpuFwdKernel, float)
-MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
-                      FlattenGpuFwdKernel, half)
-MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
-                      FlattenGpuFwdKernel, int)
+MS_REG_GPU_KERNEL_ONE(ExpandDims, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool),
+                      FlattenGpuFwdKernel, bool)
 }  // namespace kernel
 }  // namespace mindspore

mindspore/ops/_op_impl/akg/gpu/cast.py

@@ -26,6 +26,8 @@ cast_op_info = AkgGpuRegOp("Cast") \
     .dtype_format(DataType.F16_Default, DataType.BOOL_Default) \
     .dtype_format(DataType.I32_Default, DataType.F16_Default) \
     .dtype_format(DataType.I32_Default, DataType.F32_Default) \
+    .dtype_format(DataType.I32_Default, DataType.F64_Default) \
+    .dtype_format(DataType.I32_Default, DataType.I64_Default) \
     .dtype_format(DataType.I32_Default, DataType.I8_Default) \
     .dtype_format(DataType.I32_Default, DataType.U8_Default) \
     .dtype_format(DataType.I32_Default, DataType.BOOL_Default) \
@@ -61,6 +63,7 @@ cast_op_info = AkgGpuRegOp("Cast") \
     .dtype_format(DataType.I64_Default, DataType.BOOL_Default) \
     .dtype_format(DataType.F32_Default, DataType.I32_Default) \
     .dtype_format(DataType.F32_Default, DataType.F16_Default) \
+    .dtype_format(DataType.F32_Default, DataType.F64_Default) \
     .dtype_format(DataType.F32_Default, DataType.BOOL_Default) \
     .dtype_format(DataType.F64_Default, DataType.BOOL_Default) \
     .dtype_format(DataType.F64_Default, DataType.F32_Default) \

tests/st/ops/gpu/test_cast_op.py

@@ -478,3 +478,37 @@ def test_cast25():
     assert type0 == 'int16'
     type1 = output[1].asnumpy().dtype
     assert type1 == 'int8'
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_cast26():
+    x0 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.int32))
+    t0 = mstype.int64
+    x1 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.int32))
+    t1 = mstype.float64
+
+    context.set_context(mode=context.GRAPH_MODE, device_target='GPU')
+    net = Net(t0, t1)
+    output = net(x0, x1)
+    type0 = output[0].asnumpy().dtype
+    assert type0 == 'int64'
+    type1 = output[1].asnumpy().dtype
+    assert type1 == 'float64'
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_cast27():
+    x0 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.float32))
+    t0 = mstype.float64
+    x1 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.float64))
+    t1 = mstype.float32
+
+    context.set_context(mode=context.GRAPH_MODE, device_target='GPU')
+    net = Net(t0, t1)
+    output = net(x0, x1)
+    type0 = output[0].asnumpy().dtype
+    assert type0 == 'float64'
+    type1 = output[1].asnumpy().dtype
+    assert type1 == 'float32'