Opencl doesn't do data rearrangement

apache · Jun 25, 2020 · fa183ce · fa183ce
1 parent 819f0bc
commit fa183ce
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 5 deletions.
diff --git a/src/target/source/codegen_opencl.cc b/src/target/source/codegen_opencl.cc
@@ -230,7 +230,7 @@ std::string CodeGenOpenCL::CastFromTo(std::string value, DataType from, DataType
 }
 
 void CodeGenOpenCL::VisitExpr_(const CallNode* op, std::ostream& os) {
-  if (op->is_intrinsic(intrinsic::tvm_address_of)) {
+  if (op->op.same_as(builtin::address_of())) {
     // Overload tvm_address_of to add storage scope (e.g. __global).
     const LoadNode* load = op->args[0].as<LoadNode>();
     CHECK(op->args.size() == 1 && load);
@@ -243,9 +243,10 @@ void CodeGenOpenCL::VisitExpr_(const CallNode* op, std::ostream& os) {
     os << " *)" << this->GetVarID(load->buffer_var.get()) << " + ";
     this->PrintExpr(load->index, os);
     os << ')';
-  } else if (op->call_type == CallNode::Extern || op->call_type == CallNode::PureExtern) {
+  } else if (op->op.same_as(builtin_call_extern_)) {
+    auto func = Downcast<StringImm>(op->args[0]);
     // Enable atomics extension if used.
-    if (op->name == "atomic_add") {
+    if (func->value == "atomic_add") {
       enable_atomics_ = true;
     }
     CodeGenC::VisitExpr_(op, os);

diff --git a/tests/python/relay/test_op_level5.py b/tests/python/relay/test_op_level5.py
@@ -270,8 +270,8 @@ def verify_get_valid_counts(dshape, score_threshold, id_index, score_index):
             intrp = relay.create_executor("debug", ctx=ctx, target=target)
             out = intrp.evaluate(func)(np_data)
             tvm.testing.assert_allclose(out[0].asnumpy(), np_out1, rtol=1e-3, atol=1e-04)
-            # get_valid_count for cuda doesn't do data rearrangement
-            if target == 'cuda':
+            # get_valid_count for cuda, opencl doesn't do data rearrangement
+            if target in ['cuda', 'opencl']:
                 return
             tvm.testing.assert_allclose(out[1].asnumpy(), np_out2, rtol=1e-3, atol=1e-04)
             tvm.testing.assert_allclose(out[2].asnumpy(), np_out3, rtol=1e-3, atol=1e-04)