Fix the bug of misaligned address in vectorization (#72106)

baoqiwen · web-flow · commit 018ffdd67f5b · 2025-04-09T14:46:09.000+08:00
diff --git a/paddle/cinn/optim/vectorize_for_trans.cc b/paddle/cinn/optim/vectorize_for_trans.cc
@@ -141,18 +141,40 @@ Expr GetOriginOffsetWithVectorizeAxis(Expr offset, Var var_iter) {
   return origin_offset;
 }
 
-bool CheckoutTensorAddrLegalCastToVectorize(const std::vector<ir::Expr> &shapes,
-                                            const int vectorize_factor) {
-  int64_t nums = 1;
-  for (auto &size : shapes) {
-    auto const_val = size.As<ir::IntImm>();
+bool CheckTensorAddrLegalCastToVectorize(const std::vector<ir::Expr> &indices,
+                                         const std::vector<ir::Expr> &shapes,
+                                         const int vectorize_factor) {
+  auto flattened_value = 1;
+  for (int i = 0; i < indices.size(); ++i) {
+    auto const_val = shapes[i].As<ir::IntImm>();
     PADDLE_ENFORCE_NOT_NULL(const_val,
                             ::common::errors::InvalidArgument(
                                 "vectorize tiling only support static shape"));
-    nums *= const_val->value;
+    ir::Expr index = indices[i];
+    index = optim::ArithSimplify(index);
+    if (index.is_constant() && index.get_constant() == 0) {
+      // If the index is zero (indicating broadcast behavior), reset
+      // flattened_value to 1.
+      flattened_value = 1;
+    } else {
+      flattened_value *= const_val->value;
+    }
   }
-  if (nums % vectorize_factor == 0) return true;
-  return false;
+  return flattened_value % vectorize_factor == 0;
+}
+
+bool CheckTensorOffsetZero(const Expr &offset, const Var &iter_var) {
+  Expr only_vectorize_axis_offset =
+      UpdateOffsetOnlyContainsVectorizeAxis(offset, iter_var);
+  Expr origin_offset =
+      GetOriginOffsetWithVectorizeAxis(only_vectorize_axis_offset, iter_var);
+  Expr compare = CalculateOffsetWithVectorizeAxis(
+      only_vectorize_axis_offset, origin_offset, iter_var, 1);
+  auto const_val = compare.As<ir::IntImm>();
+  if (!const_val || const_val->value != 0) {
+    return false;
+  }
+  return true;
 }
 
 class ForOpWithMultiScheduleBlockSupportVectorize
@@ -331,7 +353,7 @@ class ScheduleBlockTensorVectorizeTeller : public ir::IRMutator<Expr *> {
    *    {
    *      vectorize[4] for (v1, 0, 4)
    *      {
-   *        float a[i, j, v1] =  float b[(i * 64 + j * 4 + v1) / 4]
+   *        float a[i, j, v1] = float b[(i * 64 + j * 4 + v1) / 4]
    *      }
    *    }
    *  }
@@ -345,14 +367,39 @@ class ScheduleBlockTensorVectorizeTeller : public ir::IRMutator<Expr *> {
    *    {
    *      vectorize[4] for (v1, 0, 4)
    *      {
-   *        float a[i, j, v1] =  select(i < 2, float b[i, j, v1], float c[i - 2,
+   *        float a[i, j, v1] = select(i < 2, float b[i, j, v1], float c[i - 2,
    * j, v1])
    *      }
    *    }
    *  }
    * c[i - 2, j, v1] when i = 0, j = 0, v1 = 0, offset = -128
    *
-   * Situation 3 . Tensor offset with vectorize axis is continuous
+   * Situation 3. Do not handle the scenario where there is a % b in the
+   * computation of the index, but b % factor != 0. serial for (i, 0, 4)
+   *  {
+   *    serial for (j, 0, 16)
+   *    {
+   *      vectorize[4] for (v1, 0, 4)
+   *      {
+   *        float a[i, j, v1] = float b[(i * 64 + j * 4 + v1) % 3]
+   *      }
+   *    }
+   *  }
+   *
+   *  misaligned address
+   *
+   * Situation 4. Do not handle the offset 0.
+   *  {
+   *    serial for (j, 0, 16)
+   *    {
+   *      vectorize[4] for (v1, 0, 4)
+   *      {
+   *        float a[i, j, v1] = float b[(i * 64 + j * 4 + v1) / 4]
+   *      }
+   *    }
+   *  }
+   *
+   *  misaligned address
    */
   bool TensorCanVectorize(ir::LoadStoreAddrMnger *node,
                           const std::vector<ir::Expr> &indices) {
@@ -378,40 +425,14 @@ class ScheduleBlockTensorVectorizeTeller : public ir::IRMutator<Expr *> {
       return false;
     }
 
-    Expr only_vectorize_axis_offset =
-        UpdateOffsetOnlyContainsVectorizeAxis(offset, iter_var_);
-
-    Expr origin_offset =
-        GetOriginOffsetWithVectorizeAxis(only_vectorize_axis_offset, iter_var_);
-    bool offset_is_zero = true;
-    bool tensor_is_continuous = true;
-    for (int i = 1; i < factor_; i++) {
-      Expr compare = CalculateOffsetWithVectorizeAxis(
-          only_vectorize_axis_offset, origin_offset, iter_var_, i);
-      auto const_val = compare.As<ir::IntImm>();
-      if (!const_val) return false;
-      if (const_val->value != 0) {
-        offset_is_zero = false;
-      }
-
-      if (const_val->value != i) {
-        tensor_is_continuous = false;
-        break;
-      }
-    }
-
-    if (offset_is_zero) {
-      return false;
-    }
-
-    if (!tensor_is_continuous) {
-      vectorize_tensors_.clear();
-      scalar_tensor_without_vectorize_axis_.clear();
-      schedule_block_can_vectorize_ = false;
+    // situation 3. Do not handle the scenario where there is a % b in the
+    // computation of the index, but b % factor != 0.
+    if (!CheckTensorAddrLegalCastToVectorize(indices, tensor->shape, factor_)) {
       return false;
     }
 
-    if (!CheckoutTensorAddrLegalCastToVectorize(tensor->shape, factor_)) {
+    // situation 4. Do not handle the offset 0.
+    if (CheckTensorOffsetZero(offset, iter_var_)) {
       return false;
     }
 
