Support some ops for full quantization (#40083)

* add some op for full_quantization

python/paddle/fluid/contrib/slim/quantization/post_training_quantization.py

@@ -979,8 +979,6 @@ class PostTrainingQuantization(object):
                 if op.type in (
                         self._quantizable_op_type + self._out_scale_op_list):
                     out_var_names = _get_op_output_var_names(op)
-                    assert len(out_var_names) == 1, "Post training " + \
-                        "quantization only support one output for " + op.type
                     for var_name in out_var_names:
                         analysis_and_save_info(op, var_name)
 

python/paddle/fluid/contrib/slim/quantization/quantization_pass.py

@@ -59,6 +59,7 @@ _out_scale_op_list = [
     "tanh",
     "prelu",
     "swish",
+    "dropout",
     "softmax",
     "batch_norm",
     "layer_norm",
@@ -68,6 +69,8 @@ _out_scale_op_list = [
     "transpose2",
     "concat",
     "elementwise_mul",
+    "elementwise_pow",
+    "elementwise_sub",
     "scale",
     "slice",
     "hard_swish",
@@ -81,8 +84,54 @@ _out_scale_op_list = [
     "flatten2",
     "transpose",
     "pad2d",
+    "pad3d",
     "reshape",
-    "layer_norm",
+    "split",
+    "flatten_contiguous_range",
+    "squeeze",
+    "squeeze2",
+    "nearest_interp_v2",
+    "fill_constant_batch_size_like",
+    "bilinear_interp",
+    "bilinear_interp_v2",
+    "arg_max",
+    "abs",
+    "assign",
+    "cast",
+    "clip",
+    "box_coder",
+    "crop",
+    "cumsum",
+    "equal",
+    "expand_v2",
+    "fill_any_like",
+    "fill_constant",
+    "gelu",
+    "instance_norm",
+    "lookup_table",
+    "lookup_table_v2",
+    "norm",
+    "p_norm",
+    "pow",
+    "reduce_mean",
+    "stack",
+    "top_k_v2",
+    "unsqueeze",
+    "unsqueeze2",
+    "logical_and",
+    "logical_not",
+    "meshgrid",
+    "roi_align",
+    "strided_slice",
+    "where",
+    "grid_sampler",
+    "tile",
+    "group_norm",
+    "reduce_sum",
+    "square",
+    "softplus",
+    "gather",
+    "shuffle_channel",
 ]
 
 # list op real input and output names, to avoid processing input such as AxisTensor.
@@ -119,7 +168,7 @@ _op_real_in_out_name = {
     "relu": [["X"], ["Out"]],
     "relu6": [["X"], ["Out"]],
     "leaky_relu": [["X"], ["Out"]],
-    "prelu": [["X"], ["Out"]],
+    "prelu": [["X", "Alpha"], ["Out"]],
     "tanh": [["X"], ["Out"]],
     "swish": [["X"], ["Out"]],
     "dropout": [["X"], ["Out"]],
@@ -127,16 +176,59 @@ _op_real_in_out_name = {
     "layer_norm": [["X"], ["Y"]],
     "sigmoid": [["X"], ["Out"]],
     "elementwise_mul": [["X", "Y"], ["Out"]],
+    "elementwise_pow": [["X", "Y"], ["Out"]],
     "scale": [["X"], ["Out"]],
     "hard_swish": [["X"], ["Out"]],
     "hard_sigmoid": [["X"], ["Out"]],
     "gru": [["Input", "Weight"], ["Hidden"]],
     "lstm": [["Input", "Weight"], ["Hidden"]],
     "pad2d": [["X"], ["Out"]],
+    "pad3d": [["X"], ["Out"]],
     "flatten": [["X"], ["Out"]],
     "flatten2": [["X"], ["Out"]],
     "unsqueeze2": [["X"], ["Out"]],
-    "flatten_contiguous_range": [['X'], ["Out"]],
+    "unsqueeze2": [["X"], ["Out"]],
+    "flatten_contiguous_range": [["X"], ["Out"]],
+    "split": [["X"], ["Out"]],
+    "squeeze2": [["X"], ["Out"]],
+    "nearest_interp_v2": [["X"], ["Out"]],
+    "bilinear_interp": [["X"], ["Out"]],
+    "bilinear_interp_v2": [["X"], ["Out"]],
+    "fill_constant_batch_size_like": [["Input"], ["Out"]],
+    "arg_max": [["X"], ["Out"]],
+    "abs": [["X"], ["Out"]],
+    "assign": [["X"], ["Out"]],
+    "cast": [["X"], ["Out"]],
+    "clip": [["X"], ["Out"]],
+    "box_coder": [["PriorBox"], ["OutputBox"]],
+    "crop": [["X"], ["Out"]],
+    "cumsum": [["X"], ["Out"]],
+    "expand_v2": [["X"], ["Out"]],
+    "fill_any_like": [["X"], ["Out"]],
+    "fill_constant": [[], ["Out"]],
+    "gelu": [["X"], ["Out"]],
+    "instance_norm": [["X"], ["Out"]],
+    "lookup_table": [["W", "Ids"], ["Out"]],
+    "lookup_table_v2": [["W", "Ids"], ["Out"]],
+    "norm": [["X"], ["Norm"]],
+    "p_norm": [["X"], ["Out"]],
+    "pow": [["X"], ["Out"]],
+    "reduce_mean": [["X"], ["Out"]],
+    "stack": [["X"], ["Y"]],
+    "top_k_v2": [["X"], ["Out", "Indices"]],
+    "logical_and": [["X", "Y"], ["Out"]],
+    "logical_not": [["X"], ["Out"]],
+    "meshgrid": [["X"], ["Out"]],
+    "roi_align": [["X", "ROIs"], ["Out"]],
+    "strided_slice": [["Input"], ["Out"]],
+    "where": [["Condition", "X", "Y"], ["Out"]],
+    "grid_sampler": [["X", "Grid"], ["Output"]],
+    "tile": [["X"], ["Out"]],
+    "group_norm": [["X"], ["Y", "Mean", "Variance"]],
+    "reduce_sum": [["X"], ["Out"]],
+    "square": [["X"], ["Out"]],
+    "softplus": [["X"], ["Out"]],
+    "shuffle_channel": [["X"], ["Out"]],
 }
 
 _conv_ops = ['conv2d', 'depthwise_conv2d', 'conv2d_transpose']
@@ -1797,14 +1889,93 @@ class AddQuantDequantPass(object):
     quantized ops's inputs.
     """
     _supported_quantizable_op_type = [
-        "pool2d", "elementwise_add", "concat", "softmax", "argmax", "transpose",
-        "equal", "gather", "greater_equal", "greater_than", "less_equal",
-        "less_than", "mean", "not_equal", "reshape", "reshape2",
-        "bilinear_interp", "nearest_interp", "trilinear_interp", "slice",
-        "squeeze", "elementwise_sub", "mul", "matmul", "relu", "relu6",
-        "leaky_relu", "tanh", "swish", "scale", "transpose", "transpose2",
-        "sigmoid", "pad2d", "flatten", "flatten2", "batch_norm", "layer_norm",
-        "matmul_v2"
+        "pool2d",
+        "elementwise_add",
+        "concat",
+        "softmax",
+        "argmax",
+        "transpose",
+        "equal",
+        "gather",
+        "greater_equal",
+        "greater_than",
+        "less_equal",
+        "less_than",
+        "mean",
+        "not_equal",
+        "reshape",
+        "reshape2",
+        "dropout",
+        "bilinear_interp",
+        "nearest_interp",
+        "trilinear_interp",
+        "slice",
+        "squeeze",
+        "elementwise_sub",
+        "mul",
+        "matmul",
+        "relu",
+        "relu6",
+        "leaky_relu",
+        "tanh",
+        "swish",
+        "scale",
+        "transpose",
+        "transpose2",
+        "sigmoid",
+        "pad2d",
+        "flatten",
+        "flatten2",
+        "batch_norm",
+        "layer_norm",
+        "matmul_v2",
+        "split",
+        "flatten_contiguous_range",
+        "squeeze2",
+        "nearest_interp_v2",
+        "bilinear_interp",
+        "bilinear_interp_v2",
+        "fill_constant_batch_size_like",
+        "arg_max",
+        "abs",
+        "assign",
+        "cast",
+        "clip",
+        "box_coder",
+        "crop",
+        "cumsum",
+        "elementwise_mul",
+        "elementwise_pow",
+        "expand_v2",
+        "fill_any_like",
+        "fill_constant",
+        "gelu",
+        "hard_sigmoid",
+        "hard_swish",
+        "instance_norm",
+        "lookup_table",
+        "lookup_table_v2",
+        "norm",
+        "p_norm",
+        "pad3d",
+        "pow",
+        "prelu",
+        "reduce_mean",
+        "unsqueeze",
+        "unsqueeze2",
+        "logical_and",
+        "logical_not",
+        "meshgrid",
+        "roi_align",
+        "strided_slice",
+        "where",
+        "grid_sampler",
+        "tile",
+        "group_norm",
+        "reduce_sum",
+        "square",
+        "softplus",
+        "shuffle_channel",
     ]
 
     # To be compatible with PaddleSlim, not remove _activation_type for now