[BugFix] fix nvptx not supported by device_enabled error (apache#9585)

* [BugFix] fix nvptx not supported by device_enabled error Signed-off-by: ZQPei <[email protected]> * [BugFix] shared.dyn support for codegen_nvptx Signed-off-by: ZQPei <[email protected]>
mehrdadh · Dec 1, 2021 · 08bd4ca · 08bd4ca
1 parent 261cc4e
commit 08bd4ca
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/src/runtime/module.cc b/src/runtime/module.cc
@@ -148,7 +148,7 @@ bool RuntimeEnabled(const std::string& target) {
   } else if (target == "hexagon") {
     f_name = "device_api.hexagon";
   } else if (target.length() >= 5 && target.substr(0, 5) == "nvptx") {
-    f_name = "device_api.gpu";
+    f_name = "device_api.cuda";
   } else if (target.length() >= 4 && target.substr(0, 4) == "rocm") {
     f_name = "device_api.rocm";
   } else if (target.length() >= 4 && target.substr(0, 4) == "llvm") {

diff --git a/src/target/llvm/codegen_nvptx.cc b/src/target/llvm/codegen_nvptx.cc
@@ -138,7 +138,7 @@ class CodeGenNVPTX : public CodeGenLLVM {
     if (sync == "warp") {
       // TODO(tqchen) warp sync in CUDA9
       return nullptr;
-    } else if (sync == "shared") {
+    } else if (sync == "shared" || sync == "shared.dyn") {
       llvm::Function* f =
           llvm::Intrinsic::getDeclaration(module_.get(), ::llvm::Intrinsic::nvvm_barrier0);
       return builder_->CreateCall(f, {});