fix

apache · May 17, 2022 · 3d2c90d · 3d2c90d
1 parent 403050b
commit 3d2c90d
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/src/target/source/codegen_cuda.cc b/src/target/source/codegen_cuda.cc
@@ -840,7 +840,7 @@ void CodeGenCUDA::VisitExpr_(const CallNode* op, std::ostream& os) {
       os << "for (int i = 0; i < 4; ++i) {\n";
       os << dst << "[(i / 2 * 8 + threadIdx.x / 4) * " << stride
          << " + outer * 8 + (threadIdx.x % 4) * 2 + i % 2]"
-         << " = " << src << "[" << src_offset << " + i * outer * 4];\n";
+         << " = " << src << "[" << src_offset << " + outer * 4 + i];\n";
       os << "}\n";
       os << "}\n";
     }