diff --git a/tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc b/tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc index d3592b6da15..adeb14eef14 100644 --- a/tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc +++ b/tensorflow/compiler/mlir/tools/kernel_gen/transforms/gpu_kernel_to_blob_pass.cc @@ -99,9 +99,7 @@ class GpuKernelToBlobPass llvmModule->setModuleIdentifier("acme"); xla::HloModuleConfig config; - xla::DebugOptions options = xla::GetDebugOptionsFromFlags(); - options.set_xla_gpu_ftz(true); - config.set_debug_options(options); + config.set_debug_options(xla::GetDebugOptionsFromFlags()); using AmdGpuHsaco = std::vector; std::vector images; @@ -150,9 +148,7 @@ class GpuKernelToBlobPass llvmModule->setDataLayout(xla::gpu::nvptx::kDataLayout); xla::HloModuleConfig config; - xla::DebugOptions options = xla::GetDebugOptionsFromFlags(); - options.set_xla_gpu_ftz(true); - config.set_debug_options(options); + config.set_debug_options(xla::GetDebugOptionsFromFlags()); auto enable_fusion = [](llvm::TargetMachine* target) { target->Options.AllowFPOpFusion = llvm::FPOpFusion::FPOpFusionMode::Fast;