[CINN] Refine infer_symbol_shape of data, slice op (#72418)

zyfncg · web-flow · commit ed3f6363cbdf · 2025-04-29T10:20:19.000+08:00
* refine infer_symbol_shape of data, slice op

* add file

* delete check

* fix bug

* fix bug
diff --git a/paddle/cinn/common/dim_expr_converter.cc b/paddle/cinn/common/dim_expr_converter.cc
@@ -143,9 +143,43 @@ struct DimExprConverterWithSymbolBindings::
     if (std::holds_alternative<ShapeSymbolBinding>(symbol_binding)) {
       return inputs_[input_idx]->sym_shape[input_dim_idx]->GetDimExpr();
     }
+
+    auto LinearToMultiDim = [](int64_t index,
+                               const std::vector<int64_t>& dimensions) {
+      std::vector<int64_t> result(dimensions.size(), 0);
+      std::vector<int64_t> strides(dimensions.size(), 1);
+      for (int64_t i = dimensions.size() - 2; i >= 0; --i) {
+        strides[i] = strides[i + 1] * dimensions[i + 1];
+      }
+      int64_t cur_index = index;
+      for (int64_t i = 0; i < dimensions.size(); ++i) {
+        result[i] = cur_index / strides[i];
+        cur_index %= strides[i];
+      }
+      return result;
+    };
     // for data binding [S0, a, b], inputs[a] is Tensor A, return A(b)
-    return ir::Cast::Make(cinn::common::I64(),
-                          inputs_[input_idx](cinn::ir::Expr(input_dim_idx)));
+    PADDLE_ENFORCE_LE(inputs_[input_idx].ndims(),
+                      9,
+                      ::common::errors::InvalidArgument(
+                          "The rank of the input tensor must be less than or "
+                          "equal to 9, but got %d",
+                          inputs_[input_idx].ndims()));
+    const std::vector<ir::Expr> indices = [&]() -> std::vector<ir::Expr> {
+      const auto& dimensions = inputs_[input_idx]->shape;
+      std::vector<ir::Expr> result(dimensions.size(), 0);
+      std::vector<int64_t> strides(dimensions.size(), 1);
+      for (int64_t i = dimensions.size() - 2; i >= 0; --i) {
+        strides[i] = strides[i + 1] * dimensions[i + 1].as_int64();
+      }
+      int64_t cur_index = input_dim_idx;
+      for (int64_t i = 0; i < dimensions.size(); ++i) {
+        result[i] = ir::Expr(cur_index / strides[i]);
+        cur_index %= strides[i];
+      }
+      return result;
+    }();
+    return ir::Cast::Make(cinn::common::I64(), inputs_[input_idx](indices));
   }
 
   DimExprToIrExprVisitorWithSymbolBinding(
diff --git a/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/element_wise_binary.cc b/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/element_wise_binary.cc
@@ -60,18 +60,6 @@ bool InferSymbolicShapeElementWiseBinary(
 
   if (x_shape.data() && y_shape.data() &&
       x_shape.data()->size() == y_shape.data()->size() && DataComputeFunc) {
-    PADDLE_ENFORCE_LE(
-        x_shape.shape().size(),
-        1,
-        common::errors::InvalidArgument("When compute data, the rank of x "
-                                        "should be 0 or 1, but now received %d",
-                                        x_shape.shape().size()));
-    PADDLE_ENFORCE_LE(
-        y_shape.shape().size(),
-        1,
-        common::errors::InvalidArgument("When compute data, the rank of y "
-                                        "should be 0 or 1, but now received %d",
-                                        y_shape.shape().size()));
     std::vector<symbol::DimExpr> out_data;
     for (size_t i = 0; i < x_shape.data()->size(); ++i) {
       out_data.emplace_back(
diff --git a/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/infer_sym_slice_utils.cc b/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/infer_sym_slice_utils.cc
@@ -0,0 +1,65 @@
+// Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/infer_sym_slice_utils.h"
+
+namespace paddle::dialect::slice_utils {
+
+void SliceDfsImpl(const ExprVec &datas,
+                  const std::vector<int64_t> &shape,
+                  int64_t axis,
+                  int64_t start,
+                  int64_t end,
+                  int64_t cur_visit_axis,
+                  int offset,
+                  ExprVec *result) {
+  int64_t begin = 0;
+  int64_t stop = shape.at(cur_visit_axis);
+  if (cur_visit_axis == axis) {
+    begin = start;
+    stop = end;
+  }
+  const int64_t cur_stride = std::accumulate(shape.begin() + cur_visit_axis + 1,
+                                             shape.end(),
+                                             1,
+                                             std::multiplies<int64_t>());
+  for (int64_t i = begin; i < stop; ++i) {
+    const int64_t cur_offset = offset + i * cur_stride;
+    // last dim
+    if (cur_visit_axis == static_cast<int64_t>(shape.size() - 1)) {
+      result->push_back(datas[cur_offset]);
+    } else {
+      SliceDfsImpl(datas,
+                   shape,
+                   axis,
+                   start,
+                   end,
+                   cur_visit_axis + 1,
+                   cur_offset,
+                   result);
+    }
+  }
+}
+
+ExprVec SimpleSlice(const ExprVec &datas,
+                    const std::vector<int64_t> &shape,
+                    int64_t axis,
+                    int64_t start,
+                    int64_t end) {
+  ExprVec result;
+  SliceDfsImpl(datas, shape, axis, start, end, 0, 0, &result);
+  return result;
+}
+
+}  // namespace paddle::dialect::slice_utils
diff --git a/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/infer_sym_slice_utils.h b/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/infer_sym_slice_utils.h
@@ -195,6 +195,23 @@ inline std::vector<int64_t> FormatSliceAxes(
   return axes_vec;
 }
 
+/**
+ * @brief Simple slice function like paddle.slice for a given the data vector.
+ *
+ * @param datas Input dataset of type ExprVec.
+ * @param shape The shape of datas
+ * @param axis Axis along which to perform the slicing action.
+ * @param start Starting index for the slice.
+ * @param end Ending index for the slice.
+ *
+ * @return Returns the result after slicing the input data.
+ */
+ExprVec SimpleSlice(const ExprVec &datas,
+                    const std::vector<int64_t> &shape,
+                    int64_t axis,
+                    int64_t start,
+                    int64_t end);
+
 inline ShapeOrData SliceRawInferSymbolicShape(
     const pir::Value x,
     const pir::Value out,
@@ -205,20 +222,19 @@ inline ShapeOrData SliceRawInferSymbolicShape(
     const std::vector<int64_t> &decrease_axis,
     pir::InferSymbolicShapeContext *infer_context) {
   const auto &in_shapeordata = infer_context->GetShapeOrDataForValue(x);
+  const ExprVec &in_dims = in_shapeordata.shape();
   ExprVec starts = starts_expr;
   ExprVec ends = ends_expr;
   std::vector<int64_t> infer_flags = [&infer_flags_raw, &axes_raw] {
     return infer_flags_raw.empty() ? std::vector<int64_t>(axes_raw.size(), 1)
                                    : infer_flags_raw;
   }();
+  const std::vector<int64_t> axes = FormatSliceAxes(axes_raw, in_dims.size());
+  const ExprVec slice_dims =
+      GetSliceDims(in_dims, axes, starts, ends, &infer_flags);
+  const ExprVec out_dims = GetDecreasedDims(slice_dims, decrease_axis);
 
   const auto &GetShapeDimExprs = [&]() -> symbol::ShapeOrDataDimExprs {
-    const ExprVec &in_dims = in_shapeordata.shape();
-    std::vector<int64_t> axes = FormatSliceAxes(axes_raw, in_dims.size());
-    ExprVec slice_dims =
-        GetSliceDims(in_dims, axes, starts, ends, &infer_flags);
-    ExprVec out_dims = GetDecreasedDims(slice_dims, decrease_axis);
-
     auto IsOne = [](const symbol::DimExpr &expr) {
       return expr.isa<int64_t>() && expr.dyn_cast<int64_t>() == 1;
     };
@@ -240,8 +256,6 @@ inline ShapeOrData SliceRawInferSymbolicShape(
   // When `pd.slice` is operating on a tensor which is produced by a `pd.shape`
   // op, the result should be written into data.
   const auto &GetDataDimExprs = [&]() -> symbol::ShapeOrDataDimExprs {
-    std::vector<symbol::DimExpr> out_data;
-
     // Currently, we DO NOT support the case that any element in `axes` `starts`
     // or `ends` is a Symbol.
     auto vec_int64 = details::VecExpr2Int64(starts);
@@ -263,15 +277,13 @@ inline ShapeOrData SliceRawInferSymbolicShape(
       }
       return ends_int[0];
     }();
+    const std::vector<int64_t> in_shape =
+        details::VecExpr2Int64(in_dims).value();
+    std::vector<symbol::DimExpr> out_data = SimpleSlice(
+        in_shapeordata.data().value(), in_shape, axes.at(0), start, end);
 
-    for (int64_t i = start; i < end; i++) {
-      out_data.push_back(in_shapeordata.data().value().at(i));
-    }
-
-    const ExprVec shape = GetDecreasedDims(
-        ExprVec{static_cast<int64_t>(out_data.size())}, decrease_axis);
     return symbol::ShapeOrDataDimExprs{
-        symbol::TensorShapeOrDataDimExprs(shape, out_data)};
+        symbol::TensorShapeOrDataDimExprs(out_dims, out_data)};
   };
   bool starts_ends_all_int =
       std::all_of(starts_expr.begin(),
@@ -281,10 +293,10 @@ inline ShapeOrData SliceRawInferSymbolicShape(
                   ends_expr.end(),
                   [](const symbol::DimExpr &e) { return e.isa<int64_t>(); });
 
-  const auto &out_shape =
-      in_shapeordata.data().has_value() && starts_ends_all_int
-          ? GetDataDimExprs()
-          : GetShapeDimExprs();
+  const auto &out_shape = in_shapeordata.data().has_value() &&
+                                  starts_ends_all_int && axes_raw.size() == 1
+                              ? GetDataDimExprs()
+                              : GetShapeDimExprs();
   if (out_shape.data().has_value() && out_shape.shape().empty()) {  // 0D tensor
     const paddle::dialect::DenseTensorType &tensor_type =
         out.type().dyn_cast<paddle::dialect::DenseTensorType>();
diff --git a/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/nullary_infer_sym.cc b/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/nullary_infer_sym.cc
@@ -193,11 +193,23 @@ bool DataOpInferSymbolicShape(pir::Operation *op,
         value.type().dyn_cast<pir::DenseTensorType>();
     const auto &dims = tensor_type.dims();
     if (dims.size() == 0) return true;
-    if (dims.size() != 1) return false;
-    if (dims[0] >= 1 && dims[0] <= ::common::DDim::kMaxRank) {
-      return true;
+    if (dims.size() == 1) {
+      if (dims[0] >= 1 && dims[0] <= ::common::DDim::kMaxRank) {
+        return true;
+      }
+      return false;
     }
-    return false;
+    if (common::contain_unknown_dim(dims)) return false;
+    if (common::product(dims) > ::common::DDim::kMaxRank) return false;
+
+    // only one dim is greater than one, and the other dims are 1
+    int gt_one_dim_count = 0;
+    for (int i = 0; i < dims.size(); ++i) {
+      if (dims[i] > 1) {
+        gt_one_dim_count++;
+      }
+    }
+    return gt_one_dim_count <= 1;
   };
 
   auto IsIntType = [&](pir::Value value) {
