[IR] fix assert op pd to kernel pass bug (#56314)

* fix assert op pd to kernel pass bug * refine * refine

[IR] fix assert op pd to kernel pass bug (#56314)
* fix assert op pd to kernel pass bug * refine * refine
58708a00 · wanghuancoder · GitHub · 035d185f · 58708a00
隐藏空白更改
内联并排

Showing with 6 addition and 2 deletion

paddle/fluid/ir/transforms/pd_op_to_kernel_pass.cc paddle/fluid/ir/transforms/pd_op_to_kernel_pass.cc +6 -2

未找到文件。
--- a/paddle/fluid/ir/transforms/pd_op_to_kernel_pass.cc
+++ b/paddle/fluid/ir/transforms/pd_op_to_kernel_pass.cc
@@ -305,8 +305,12 @@ phi::KernelKey GetKernelKey(
      if (input_type.isa<dialect::AllocatedDenseTensorType>()) {
        type = input_type.dyn_cast<dialect::AllocatedDenseTensorType>();
      } else if (input_type.isa<ir::VectorType>()) {
-        type = input_type.dyn_cast<ir::VectorType>()[0]
-                   .dyn_cast<dialect::AllocatedDenseTensorType>();
+        if (!input_type.dyn_cast<ir::VectorType>().empty()) {
+          type = input_type.dyn_cast<ir::VectorType>()[0]
+                     .dyn_cast<dialect::AllocatedDenseTensorType>();
+        } else {
+          continue;
+        }
      }

      // fake tensor here