From 632c039aeaf4556f37c02b03a102aaa667fa517c Mon Sep 17 00:00:00 2001
From: Yukun He <23156053+hyukn@users.noreply.github.com>
Date: Thu, 12 Feb 2026 09:54:58 +0800
Subject: [PATCH] [TRTLLM-10793][feat] Add BOLT compatible build flags for
 further experimental usage. (#11297)

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
---
 cpp/CMakeLists.txt                            | 23 +++++++++++++++++++
 .../nixl_utils/CMakeLists.txt                 |  9 +++++++-
 cpp/tensorrt_llm/nanobind/CMakeLists.txt      |  7 +++++-
 scripts/build_wheel.py                        | 12 ++++++++++
 4 files changed, 49 insertions(+), 2 deletions(-)

diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
index 06ddbe0435..ba66a866a6 100644
--- a/cpp/CMakeLists.txt
+++ b/cpp/CMakeLists.txt
@@ -369,6 +369,29 @@ if(CMAKE_SYSTEM_PROCESSOR STREQUAL x86_64)
   set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,--no-relax")
 endif()
 
+# ==== BOLT Compatible Flags ====
+# These flags enable LLVM BOLT binary compatible:
+# -fno-reorder-blocks-and-partition: Prevents compiler from splitting hot/cold
+# blocks -fno-plt: Removes PLT stubs for additional performance
+# -Wl,--emit-relocs (or -Wl,-q): Preserves relocation info for BOLT Note:
+# Binaries should NOT be stripped for BOLT to work Usage: cmake
+# -DENABLE_BOLT_COMPATIBLE=ON ...
+option(ENABLE_BOLT_COMPATIBLE "Enable BOLT-compatible build flags" OFF)
+if(ENABLE_BOLT_COMPATIBLE AND NOT WIN32)
+  message(STATUS "BOLT compatible flags enabled")
+  # Compiler flags for C/C++
+  add_compile_options(-fno-reorder-blocks-and-partition -fno-plt)
+  # Linker flags - applies to shared, module, and executable targets
+  add_link_options(-Wl,--emit-relocs)
+  # Disable stripping - required for BOLT (affects pybind11 POST_BUILD strip)
+  set(CMAKE_STRIP
+      ""
+      CACHE STRING "Disabled for BOLT compatibility" FORCE)
+  message(STATUS "BOLT: Disabled CMAKE_STRIP to prevent binary stripping")
+endif()
+# Note: For nanobind modules, use NOSTRIP option in nanobind_add_module()
+# ==== End BOLT Compatible Flags ====
+
 # Disable deprecated declarations warnings
 if(NOT WIN32)
   set(CMAKE_CXX_FLAGS "-Wno-deprecated-declarations ${CMAKE_CXX_FLAGS}")
diff --git a/cpp/tensorrt_llm/executor/cache_transmission/nixl_utils/CMakeLists.txt b/cpp/tensorrt_llm/executor/cache_transmission/nixl_utils/CMakeLists.txt
index 8b99e698f0..7700c64a90 100644
--- a/cpp/tensorrt_llm/executor/cache_transmission/nixl_utils/CMakeLists.txt
+++ b/cpp/tensorrt_llm/executor/cache_transmission/nixl_utils/CMakeLists.txt
@@ -67,7 +67,14 @@ if(NIXL_ENABLED OR MOONCAKE_ENABLED)
   set(AGENT_BINDING_SOURCES "")
   list(APPEND AGENT_BINDING_SOURCES agentBindings.cpp)
 
-  nanobind_add_module(${TRANSFER_AGENT_BINDING_TARGET} ${AGENT_BINDING_SOURCES})
+  # NOSTRIP: Disable stripping for BOLT compatibility (requires --emit-relocs)
+  if(ENABLE_BOLT_COMPATIBLE)
+    nanobind_add_module(${TRANSFER_AGENT_BINDING_TARGET} NOSTRIP
+                        ${AGENT_BINDING_SOURCES})
+  else()
+    nanobind_add_module(${TRANSFER_AGENT_BINDING_TARGET}
+                        ${AGENT_BINDING_SOURCES})
+  endif()
   message(STATUS "Building tensorrt_llm_transfer_agent_binding with nanobind")
 
   target_compile_options(${TRANSFER_AGENT_BINDING_TARGET} PRIVATE -Wno-error)
diff --git a/cpp/tensorrt_llm/nanobind/CMakeLists.txt b/cpp/tensorrt_llm/nanobind/CMakeLists.txt
index a47f8dd1d4..999c2b736c 100755
--- a/cpp/tensorrt_llm/nanobind/CMakeLists.txt
+++ b/cpp/tensorrt_llm/nanobind/CMakeLists.txt
@@ -29,7 +29,12 @@ set(SRCS
 
 include_directories(${PROJECT_SOURCE_DIR}/include)
 
-nanobind_add_module(${TRTLLM_NB_MODULE} ${SRCS})
+# NOSTRIP: Disable stripping for BOLT compatibility (requires --emit-relocs)
+if(ENABLE_BOLT_COMPATIBLE)
+  nanobind_add_module(${TRTLLM_NB_MODULE} NOSTRIP ${SRCS})
+else()
+  nanobind_add_module(${TRTLLM_NB_MODULE} ${SRCS})
+endif()
 
 set_property(TARGET ${TRTLLM_NB_MODULE} PROPERTY POSITION_INDEPENDENT_CODE ON)
 
diff --git a/scripts/build_wheel.py b/scripts/build_wheel.py
index f61b99466a..6dbfc64740 100755
--- a/scripts/build_wheel.py
+++ b/scripts/build_wheel.py
@@ -598,6 +598,18 @@ def main(*,
     if nvrtc_dynamic_linking:
         cmake_def_args.append(f"-DNVRTC_DYNAMIC_LINKING=ON")
 
+    # BOLT compatibility: Force dynamic linking for NVIDIA libraries
+    # Static NVIDIA libraries (libnvrtc_static.a, etc.) lack --emit-relocs,
+    # which BOLT requires for proper binary optimization.
+    bolt_enabled = any("ENABLE_BOLT_COMPATIBLE=ON" in var
+                       for var in extra_cmake_vars)
+    if bolt_enabled:
+        if not nvrtc_dynamic_linking:
+            cmake_def_args.append("-DNVRTC_DYNAMIC_LINKING=ON")
+            print(
+                "-- BOLT: Forcing NVRTC_DYNAMIC_LINKING=ON (static NVIDIA libs lack relocations)"
+            )
+
     targets = ["tensorrt_llm", "nvinfer_plugin_tensorrt_llm"]
 
     if cpp_only: