!7767 GPU update resnet50 readme and add cast type

Merge pull request !7767 from VectorSL/readme

mindspore/ccsrc/backend/optimizer/gpu/reduce_precision_fusion.cc

@@ -86,8 +86,9 @@ bool ReducePrecisionFusion::Run(const FuncGraphPtr &graph) {
             auto used_node_index = used_node_list->at(j).second - 1;
             if (AnfAlgo::GetCNodeName(used_node) == prim::kPrimTupleGetItem->name()) {
               ProcessTupleGetItem(graph, used_node, used_node_index, deviceType, inferType);
+            } else {
+              ReducePrecision(graph, used_node, used_node_index, deviceType, inferType);
             }
-            ReducePrecision(graph, used_node, used_node_index, deviceType, inferType);
           }
         }
       }

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

@@ -65,6 +65,13 @@ cast_op_info = AkgGpuRegOp("Cast") \
     .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.F32_Default, DataType.I8_Default) \
+    .dtype_format(DataType.F32_Default, DataType.I16_Default) \
+    .dtype_format(DataType.F32_Default, DataType.I64_Default) \
+    .dtype_format(DataType.F32_Default, DataType.U8_Default) \
+    .dtype_format(DataType.F32_Default, DataType.U16_Default) \
+    .dtype_format(DataType.F32_Default, DataType.U32_Default) \
+    .dtype_format(DataType.F32_Default, DataType.U64_Default) \
     .dtype_format(DataType.F64_Default, DataType.BOOL_Default) \
     .dtype_format(DataType.F64_Default, DataType.F32_Default) \
     .dtype_format(DataType.F64_Default, DataType.F16_Default) \

model_zoo/official/cv/resnet/README.md

@@ -419,13 +419,13 @@ result: {'top_5_accuracy': 0.9342589628681178, 'top_1_accuracy': 0.7680657810499
 | uploaded Date              | 04/01/2020 (month/day/year)  ；                        | 08/01/2020 (month/day/year)
 | MindSpore Version          | 0.1.0-alpha                                                       |0.6.0-alpha   |
 | Dataset                    | ImageNet2012                                                    | ImageNet2012|
-| Training Parameters        | epoch=90, steps per epoch=626, batch_size = 256             |epoch=90, steps per epoch=5004, batch_size = 32  |
+| Training Parameters        | epoch=90, steps per epoch=626, batch_size = 256             |epoch=90, steps per epoch=626, batch_size = 256  |
 | Optimizer                  | Momentum                                                         |Momentum|
 | Loss Function              | Softmax Cross Entropy                                       |Softmax Cross Entropy           |
 | outputs                    | probability                                                 |  probability          |
 | Loss                       | 1.8464266                                                    | 1.9023  |
-| Speed                      | 118ms/step（8pcs）                     |67.1ms/step（8pcs）|
-| Total time                 | 114 mins                          | 500 mins|
+| Speed                      | 118ms/step（8pcs）                     |270ms/step（8pcs）|
+| Total time                 | 114 mins                          | 260 mins|
 | Parameters (M)             | 25.5                                                         | 25.5 |
 | Checkpoint for Fine tuning | 197M (.ckpt file)                                         |197M (.ckpt file)     |
 | Scripts                    | [Link](https://gitee.com/mindspore/mindspore/tree/master/model_zoo/official/cv/resnet) | [Link](https://gitee.com/mindspore/mindspore/tree/master/model_zoo/official/cv/resnet) |

tests/st/ops/gpu/test_cast_op.py

@@ -502,8 +502,8 @@ def test_cast26():
 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
+    x1 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.float32))
+    t1 = mstype.uint64
 
     context.set_context(mode=context.GRAPH_MODE, device_target='GPU')
     net = Net(t0, t1)
@@ -511,4 +511,55 @@ def test_cast27():
     type0 = output[0].asnumpy().dtype
     assert type0 == 'float64'
     type1 = output[1].asnumpy().dtype
-    assert type1 == 'float32'
+    assert type1 == 'uint64'
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_cast28():
+    x0 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.float32))
+    t0 = mstype.int8
+    x1 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.float32))
+    t1 = mstype.int16
+
+    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 == 'int8'
+    type1 = output[1].asnumpy().dtype
+    assert type1 == 'int16'
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_cast29():
+    x0 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.float32))
+    t0 = mstype.int64
+    x1 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.float32))
+    t1 = mstype.uint8
+
+    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 == 'uint8'
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_cast30():
+    x0 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.float32))
+    t0 = mstype.uint16
+    x1 = Tensor(np.arange(24).reshape((4, 3, 2)).astype(np.float32))
+    t1 = mstype.uint32
+
+    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 == 'uint16'
+    type1 = output[1].asnumpy().dtype
+    assert type1 == 'uint32'