refactoring dynamic slice

masahi · May 27, 2021 · 510bce6 · 510bce6
1 parent 1b3969a
commit 510bce6
Show file tree

Hide file tree

Showing 2 changed files with 53 additions and 40 deletions.
diff --git a/include/tvm/topi/transform.h b/include/tvm/topi/transform.h
@@ -570,6 +570,50 @@ inline te::Tensor strided_slice_compute_common(const te::Tensor& x,
       name, tag);
 }
 
+inline Tensor dynamic_strided_slice(const Tensor& x, const Array<PrimExpr>& begin,
+                                    const Array<PrimExpr>& end, const Array<PrimExpr>& strides,
+                                    std::string name = "T_dynamic_strided_slice",
+                                    std::string tag = kInjective) {
+  const size_t src_tensor_dim = static_cast<size_t>(x->shape.size());
+  ICHECK_LE(begin.size(), src_tensor_dim);
+  ICHECK_LE(end.size(), src_tensor_dim);
+  ICHECK_LE(strides.size(), src_tensor_dim);
+  ICHECK_EQ(begin.size(), end.size());
+  ICHECK_EQ(begin.size(), strides.size());
+
+  const size_t num_slice_axes = begin.size();
+  Array<PrimExpr> out_shape;
+
+  for (size_t i = 0; i < num_slice_axes; ++i) {
+    auto d = indexdiv(end[i] - begin[i], strides[i]);
+    if (d->IsInstance<tvm::IntImmNode>()) {
+      // Preserve static dimension if possible
+      out_shape.push_back(d);
+    } else {
+      out_shape.push_back(tvm::tir::Var("dim"));
+    }
+  }
+
+  for (size_t i = num_slice_axes; i < src_tensor_dim; ++i) {
+    out_shape.push_back(x->shape[i]);
+  }
+
+  return te::compute(
+      out_shape,
+      [&](const Array<tvm::tir::Var>& indices) {
+        Array<PrimExpr> real_indices;
+        for (int32_t i = 0; i < num_slice_axes; ++i) {
+          real_indices.push_back(indices[i] * strides[i] + tvm::min(begin[i], x->shape[i] - 1));
+        }
+        // keep input dim
+        for (int32_t i = num_slice_axes; i < src_tensor_dim; ++i) {
+          real_indices.push_back(indices[i]);
+        }
+        return x(real_indices);
+      },
+      name, tag);
+}
+
 /*!
  * \brief strided_slice of a tensor with dynamic begin/end/stride
  *
@@ -587,48 +631,18 @@ inline te::Tensor dynamic_strided_slice(const te::Tensor& x, const te::Tensor& b
                                         const te::Tensor& end, const te::Tensor& strides,
                                         std::string name = "T_strided_slice_dynamic",
                                         std::string tag = topi::kInjective) {
-  int64_t src_tensor_dim = x->shape.size();
-  Array<PrimExpr> out_shape;
   const int64_t num_dynamic_axes = begin->shape[0].as<IntImmNode>()->value;
-  for (int64_t i = 0; i < num_dynamic_axes; ++i) {
-    out_shape.push_back(tvm::tir::Var("dim"));
-  }
-  for (int64_t i = num_dynamic_axes; i < src_tensor_dim; ++i) {
-    out_shape.push_back(x->shape[i]);
-  }
-  return te::compute(
-      out_shape,
-      [&](const Array<tvm::tir::Var>& indices) {
-        Array<PrimExpr> real_indices;
-        // dynamic slicing
-        for (int32_t i = 0; i < num_dynamic_axes; ++i) {
-          real_indices.push_back(indices[i] * strides(i) + tvm::min(begin(i), x->shape[i] - 1));
-        }
-        // keep input dim
-        for (int32_t i = num_dynamic_axes; i < src_tensor_dim; ++i) {
-          real_indices.push_back(indices[i]);
-        }
-        return x(real_indices);
-      },
-      name, tag);
-}
-
-inline Tensor dynamic_strided_slice(const Tensor& x, const Array<PrimExpr>& begin,
-                                    const Array<PrimExpr>& end, const Array<PrimExpr>& strides,
-                                    std::string name = "T_dynamic_strided_slice",
-                                    std::string tag = kInjective) {
-  size_t src_tensor_dim = static_cast<size_t>(x->shape.size());
-  ICHECK_EQ(begin.size(), src_tensor_dim);
-  ICHECK_EQ(end.size(), src_tensor_dim);
-  ICHECK_EQ(strides.size(), src_tensor_dim);
+  ICHECK_EQ(end->shape[0].as<IntImmNode>()->value, num_dynamic_axes);
+  ICHECK_EQ(strides->shape[0].as<IntImmNode>()->value, num_dynamic_axes);
 
-  Array<PrimExpr> out_shape;
-  Array<Integer> axes;
-  for (size_t i = 0; i < src_tensor_dim; ++i) {
-    out_shape.push_back(indexdiv(end[i] - begin[i], strides[i]));
-    axes.push_back(i);
+  Array<PrimExpr> begin_expr, end_expr, strides_expr;
+  for (int64_t i = 0; i < num_dynamic_axes; ++i) {
+    auto i64_ind = IntImm(DataType::Int(64), i);
+    begin_expr.push_back(begin(i64_ind));
+    end_expr.push_back(end(i64_ind));
+    strides_expr.push_back(strides(i64_ind));
   }
-  return strided_slice_compute_common(x, out_shape, begin, strides, axes, name, tag);
+  return dynamic_strided_slice(x, begin_expr, end_expr, strides_expr, name, tag);
 }
 
 inline Tensor strided_slice_dynamic_input(const Tensor& x, const Array<Integer>& begin,

diff --git a/python/tvm/topi/cuda/sort.py b/python/tvm/topi/cuda/sort.py
@@ -962,7 +962,6 @@ def topk(data, k=1, axis=-1, ret_type="both", is_ascend=False, dtype="int64"):
             end.append(dshape[i])
     if ret_type == "both":
         values_out, indices_out = output
-        print("end:", end, k)
         values_out = strided_slice(values_out, beg, end, strides)
         indices_out = strided_slice(indices_out, beg, end, strides)
         output = [values_out, indices_out]