[PIR] rename data_layout (#60678)

* rename data_layout
PaddlePaddle · Jan 11, 2024 · 3179fdf · 3179fdf
1 parent 55558f1
commit 3179fdf
Show file tree

Hide file tree

Showing 11 changed files with 101 additions and 44 deletions.
diff --git a/paddle/fluid/operators/generator/generate_sparse_op.py b/paddle/fluid/operators/generator/generate_sparse_op.py
@@ -108,12 +108,49 @@ def main(op_yaml_path, backward_yaml_path, output_op_path, output_arg_map_path):
         op['name'] = op['op_name']
         if op["backward"] is not None:
             op["backward"] = SPARSE_OP_PREFIX + op["backward"]
+        if op['name'] in [
+            SPARSE_OP_PREFIX + "batch_norm",
+            SPARSE_OP_PREFIX + "sync_batch_norm",
+        ]:
+            for item in op["attrs"]:
+                if item["name"] == "data_format":
+                    item["name"] = "data_layout"
+            value = op["attr_dict"].pop('data_format')
+            op["attr_dict"]['data_layout'] = value
+            for i in range(len(op["kernel"]["param"])):
+                if op["kernel"]["param"][i] == "data_format":
+                    op["kernel"]["param"][i] = "data_layout"
+            for i in range(len(op["infer_meta"]["param"])):
+                if op["infer_meta"]["param"][i] == "data_format":
+                    op["infer_meta"]["param"][i] = "data_layout"
         add_fluid_name(op["inputs"])
         add_fluid_name(op["attrs"])
         add_fluid_name(op["outputs"])
+
     for bw_op in backward_ops:
         bw_op['op_name'] = SPARSE_OP_PREFIX + bw_op['name']
         bw_op['name'] = bw_op['op_name']
+
+        if bw_op['name'] in [
+            SPARSE_OP_PREFIX + "batch_norm_grad",
+            SPARSE_OP_PREFIX + "sync_batch_norm_grad",
+        ]:
+            for item in bw_op["attrs"]:
+                if item["name"] == "data_format":
+                    item["name"] = "data_layout"
+            for item in bw_op["forward"]["attrs"]:
+                if item["name"] == "data_format":
+                    item["name"] = "data_layout"
+                    item["fluid_name"] = "data_layout"
+            value = bw_op["attr_dict"].pop('data_format')
+            bw_op["attr_dict"]['data_layout'] = value
+            for i in range(len(bw_op["kernel"]["param"])):
+                if bw_op["kernel"]["param"][i] == "data_format":
+                    bw_op["kernel"]["param"][i] = "data_layout"
+            for i in range(len(bw_op["infer_meta"]["param"])):
+                if bw_op["infer_meta"]["param"][i] == "data_format":
+                    bw_op["infer_meta"]["param"][i] = "data_layout"
+
         add_fluid_name(bw_op["inputs"])
         add_fluid_name(bw_op["attrs"])
         add_fluid_name(bw_op["outputs"])

diff --git a/paddle/fluid/pir/dialect/operator/ir/ops.yaml b/paddle/fluid/pir/dialect/operator/ir/ops.yaml
@@ -134,7 +134,7 @@
     backend : place > output
 
 - op : batch_norm
-  args : (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_layout, bool use_global_stats, bool trainable_statistics)
+  args : (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_format, bool use_global_stats, bool trainable_statistics)
   output : Tensor(out), Tensor(mean_out), Tensor(variance_out), Tensor(saved_mean), Tensor(saved_variance), Tensor(reserve_space)
   infer_meta:
     func : BatchNormInferMeta
@@ -1275,7 +1275,7 @@
   backward : swish_grad
 
 - op : sync_batch_norm_
-  args : (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_layout, bool use_global_stats, bool trainable_statistics)
+  args : (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_format, bool use_global_stats, bool trainable_statistics)
   output : Tensor(out), Tensor(mean_out), Tensor(variance_out), Tensor(saved_mean), Tensor(saved_variance), Tensor(reserve_space)
   infer_meta :
     func : BatchNormInferMeta

diff --git a/paddle/fluid/pir/dialect/operator/ir/ops_backward.yaml b/paddle/fluid/pir/dialect/operator/ir/ops_backward.yaml
@@ -82,8 +82,8 @@
   inplace : (out_grad -> x_grad)
 
 - backward_op : batch_norm_double_grad
-  forward : batch_norm_grad (Tensor x, Tensor scale, Tensor bias, Tensor out_mean, Tensor out_variance, Tensor saved_mean, Tensor saved_variance, Tensor reserve_space, Tensor grad_out, float momentum, float epsilon, str data_layout, bool is_test, bool use_global_stats, bool trainable_statistics) -> Tensor(grad_x), Tensor(grad_scale), Tensor(grad_bias)
-  args : (Tensor x, Tensor scale, Tensor out_mean, Tensor out_variance, Tensor saved_mean, Tensor saved_variance, Tensor grad_out,  Tensor grad_x_grad, Tensor grad_scale_grad, Tensor grad_bias_grad, float momentum, float epsilon, str data_layout, bool is_test, bool use_global_stats, bool trainable_statistics)
+  forward : batch_norm_grad (Tensor x, Tensor scale, Tensor bias, Tensor out_mean, Tensor out_variance, Tensor saved_mean, Tensor saved_variance, Tensor reserve_space, Tensor grad_out, float momentum, float epsilon, str data_format, bool is_test, bool use_global_stats, bool trainable_statistics) -> Tensor(grad_x), Tensor(grad_scale), Tensor(grad_bias)
+  args : (Tensor x, Tensor scale, Tensor out_mean, Tensor out_variance, Tensor saved_mean, Tensor saved_variance, Tensor grad_out,  Tensor grad_x_grad, Tensor grad_scale_grad, Tensor grad_bias_grad, float momentum, float epsilon, str data_format, bool is_test, bool use_global_stats, bool trainable_statistics)
   output : Tensor(x_grad), Tensor(scale_grad), Tensor(grad_out_grad)
   infer_meta :
     func : GeneralTernaryGradInferMeta
@@ -95,8 +95,8 @@
   inplace : (grad_out -> grad_out_grad)
 
 - backward_op : batch_norm_grad
-  forward : batch_norm (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_layout, bool use_global_stats, bool trainable_statistics) -> Tensor(out), Tensor(mean_out), Tensor(variance_out), Tensor(saved_mean), Tensor(saved_variance), Tensor(reserve_space)
-  args : (Tensor x, Tensor scale, Tensor bias, Tensor mean_out, Tensor variance_out, Tensor saved_mean, Tensor saved_variance, Tensor reserve_space, Tensor out_grad, float momentum, float epsilon, str data_layout, bool is_test, bool use_global_stats, bool trainable_statistics)
+  forward : batch_norm (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_format, bool use_global_stats, bool trainable_statistics) -> Tensor(out), Tensor(mean_out), Tensor(variance_out), Tensor(saved_mean), Tensor(saved_variance), Tensor(reserve_space)
+  args : (Tensor x, Tensor scale, Tensor bias, Tensor mean_out, Tensor variance_out, Tensor saved_mean, Tensor saved_variance, Tensor reserve_space, Tensor out_grad, float momentum, float epsilon, str data_format, bool is_test, bool use_global_stats, bool trainable_statistics)
   output : Tensor(x_grad), Tensor(scale_grad), Tensor(bias_grad)
   infer_meta :
     func : GeneralTernaryGradInferMeta
@@ -105,7 +105,7 @@
     func : batch_norm_grad
     data_type : out_grad
   optional : scale, bias, mean_out, variance_out, reserve_space
-  composite: batch_norm_grad(x, scale, bias, mean_out, variance_out, saved_mean, saved_variance, reserve_space, out_grad, momentum, epsilon, data_layout, is_test, use_global_stats, trainable_statistics)
+  composite: batch_norm_grad(x, scale, bias, mean_out, variance_out, saved_mean, saved_variance, reserve_space, out_grad, momentum, epsilon, data_format, is_test, use_global_stats, trainable_statistics)
   backward : batch_norm_double_grad
 
 - backward_op : c_embedding_grad
@@ -837,8 +837,8 @@
   inplace : (out_grad -> x_grad)
 
 - backward_op : sync_batch_norm_grad
-  forward : sync_batch_norm_ (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_layout, bool use_global_stats, bool trainable_statistics) -> Tensor(out), Tensor(mean_out), Tensor(variance_out), Tensor(saved_mean), Tensor(saved_variance), Tensor(reserve_space)
-  args : (Tensor x, Tensor scale, Tensor bias, Tensor saved_mean, Tensor saved_variance, Tensor reserve_space, Tensor out_grad, float momentum, float epsilon, str data_layout, bool is_test, bool use_global_stats, bool trainable_statistics)
+  forward : sync_batch_norm_ (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_format, bool use_global_stats, bool trainable_statistics) -> Tensor(out), Tensor(mean_out), Tensor(variance_out), Tensor(saved_mean), Tensor(saved_variance), Tensor(reserve_space)
+  args : (Tensor x, Tensor scale, Tensor bias, Tensor saved_mean, Tensor saved_variance, Tensor reserve_space, Tensor out_grad, float momentum, float epsilon, str data_format, bool is_test, bool use_global_stats, bool trainable_statistics)
   output : Tensor(x_grad), Tensor(scale_grad), Tensor(bias_grad)
   infer_meta :
     func : GeneralTernaryGradInferMeta

diff --git a/paddle/phi/api/yaml/backward.yaml b/paddle/phi/api/yaml/backward.yaml
@@ -180,8 +180,8 @@
   inplace : (out_grad -> input_grad)
 
 - backward_op : bicubic_interp_grad
-  forward : bicubic_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_layout="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
-  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_layout, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
+  forward : bicubic_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_format="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
+  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_format, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
   output : Tensor(x_grad)
   infer_meta :
     func : UnchangedInferMeta
@@ -204,8 +204,8 @@
     func : bilinear_grad
 
 - backward_op : bilinear_interp_grad
-  forward : bilinear_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_layout="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
-  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_layout, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
+  forward : bilinear_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_format="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
+  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_format, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
   output : Tensor(x_grad)
   infer_meta :
     func : UnchangedInferMeta
@@ -992,16 +992,16 @@
     data_type : x
 
 - backward_op : group_norm_grad
-  forward : group_norm (Tensor x, Tensor scale, Tensor bias, float epsilon = 1e-5, int groups = -1, str data_layout = "NCHW") -> Tensor(y), Tensor(mean), Tensor(variance)
-  args : (Tensor x, Tensor scale, Tensor bias, Tensor y, Tensor mean, Tensor variance, Tensor y_grad, float epsilon, int groups, str data_layout)
+  forward : group_norm (Tensor x, Tensor scale, Tensor bias, float epsilon = 1e-5, int groups = -1, str data_format = "NCHW") -> Tensor(y), Tensor(mean), Tensor(variance)
+  args : (Tensor x, Tensor scale, Tensor bias, Tensor y, Tensor mean, Tensor variance, Tensor y_grad, float epsilon, int groups, str data_format)
   output : Tensor(x_grad), Tensor(scale_grad), Tensor(bias_grad)
   infer_meta :
     func : GeneralTernaryGradInferMeta
     param : [y, scale, bias]
   kernel :
     func : group_norm_grad
     data_type : y_grad
-  composite : group_norm_grad(x, scale, bias, y, mean, variance, y_grad, epsilon, groups, data_layout, x_grad, scale_grad, bias_grad)
+  composite : group_norm_grad(x, scale, bias, y, mean, variance, y_grad, epsilon, groups, data_format, x_grad, scale_grad, bias_grad)
   optional: scale, bias
   inplace : (y_grad -> x_grad)
 
@@ -1328,8 +1328,8 @@
     func : lgamma_grad
 
 - backward_op : linear_interp_grad
-  forward : linear_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_layout="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
-  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_layout, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
+  forward : linear_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_format="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
+  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_format, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
   output : Tensor(x_grad)
   infer_meta :
     func : UnchangedInferMeta
@@ -1617,8 +1617,8 @@
     func : nanmedian_grad
 
 - backward_op : nearest_interp_grad
-  forward : nearest_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_layout="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
-  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_layout, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
+  forward : nearest_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_format="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
+  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_format, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
   output : Tensor(x_grad)
   infer_meta :
     func : UnchangedInferMeta
@@ -2484,8 +2484,8 @@
     func : triangular_solve_grad
 
 - backward_op : trilinear_interp_grad
-  forward : trilinear_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_layout="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
-  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_layout, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
+  forward : trilinear_interp (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, str data_format="NCHW", int out_d=0, int out_h=0, int out_w=0, float[] scale={}, str interp_method="bilinear", bool align_corners=true, int align_mode=1) -> Tensor(output)
+  args : (Tensor x, Tensor out_size, Tensor[] size_tensor, Tensor scale_tensor, Tensor output_grad, str data_format, int out_d, int out_h, int out_w, float[] scale, str interp_method, bool align_corners, int align_mode)
   output : Tensor(x_grad)
   infer_meta :
     func : UnchangedInferMeta

diff --git a/paddle/phi/api/yaml/fused_ops.yaml b/paddle/phi/api/yaml/fused_ops.yaml
@@ -45,7 +45,7 @@
   support_dygraph_mode : true
 
 - op : bn_act_xpu
-  args : (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, float momentum, float epsilon, str data_layout, int act_type)
+  args : (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, float momentum, float epsilon, str data_format, int act_type)
   output : Tensor(out)
   infer_meta :
     func : BNActXPUInferMeta