lite

build/Linux/Release/amdgpu/onnxruntime/core/providers/rocm/tensor/concat_impl.cu

@@ -13,7 +13,7 @@ namespace rocm {
 namespace {
 #ifdef USE_ROCM
 constexpr int kNumElementsPerThread = 2;
-constexpr int kNumThreadsPerBlock = 512;
+constexpr int kNumThreadsPerBlock = 256;
 #else
 constexpr int kNumElementsPerThread = GridDim::maxElementsPerThread;
 constexpr int kNumThreadsPerBlock = GridDim::maxThreadsPerBlock;