A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 4, 01:15 UTC
Require C++20 for compatibility with PyTorch (#40380)
Signed-off-by: Richard Barnes <rbarnes@meta.com> Co-authored-by: Claude <noreply@anthropic.com>
Richard Barnes committed
May 9, 2026 at 01:04 UTC
d6563d693c817ec762ccd0bb6c44c56b4f3dc4a3
3 files changed
+6
-3
CMakeLists.txt
+5
-1
@@ -13,8 +13,12 @@ cmake_minimum_required(VERSION 3.26)
13
# cmake --install . --component _C
14
project(vllm_extensions LANGUAGES CXX)
15
16
-set(CMAKE_CXX_STANDARD 17)
16
+set(CMAKE_CXX_STANDARD 20)
17
set(CMAKE_CXX_STANDARD_REQUIRED ON)
18
+set(CMAKE_CUDA_STANDARD 20)
19
+set(CMAKE_CUDA_STANDARD_REQUIRED ON)
20
+set(CMAKE_HIP_STANDARD 20)
21
+set(CMAKE_HIP_STANDARD_REQUIRED ON)
22
23
24
# CUDA by default, can be overridden by using -DVLLM_TARGET_DEVICE=... (used by setup.py)
cmake/cpu_extension.cmake
+1
-1
@@ -1,7 +1,7 @@
1
include(FetchContent)
2
3
set(CMAKE_CXX_STANDARD_REQUIRED ON)
4
-set(CMAKE_CXX_STANDARD 17)
4
+set(CMAKE_CXX_STANDARD 20)
5
set(CMAKE_CXX_EXTENSIONS ON)
6
set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
7
cmake/external_projects/deepgemm.cmake
-1
@@ -76,7 +76,6 @@ if(DEEPGEMM_ARCHS)
76
"${deepgemm_SOURCE_DIR}/third-party/fmt/include")
77
78
target_compile_options(_deep_gemm_C PRIVATE
79
- $<$<COMPILE_LANGUAGE:CXX>:-std=c++17>
79
$<$<COMPILE_LANGUAGE:CXX>:-O3>
80
$<$<COMPILE_LANGUAGE:CXX>:-Wno-psabi>
81
$<$<COMPILE_LANGUAGE:CXX>:-Wno-deprecated-declarations>)