From 4c0c55bba14dd0b0e4197a8dcda5a71b76ee020a Mon Sep 17 00:00:00 2001
From: Zhou Wei <52485244+zhouwei25@users.noreply.github.com>
Date: Wed, 17 Mar 2021 11:48:18 +0800
Subject: [PATCH] support Geforce RTX 30+ GPU (#31529)

---
 CMakeLists.txt   |  6 +++++-
 cmake/cuda.cmake | 12 ++++++++++--
 2 files changed, 15 insertions(+), 3 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 992c3f1c4fa..10b3b0aba4e 100755
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -95,7 +95,7 @@ if(WIN32)
             endif()
         endforeach(flag_var)
     endif()
-
+    
     # NOTE(Avin0323): Less parallel count result in faster compilation.
     math(EXPR PROCESS_MAX "${CPU_CORES} * 2 / 3")
     # windows build turn off warnings, use parallel compiling.
@@ -123,6 +123,10 @@ if(WIN32)
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /wd4068 /wd4129 /wd4244 /wd4267 /wd4297 /wd4530 /wd4577 /wd4819 /wd4838")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /wd4068 /wd4129 /wd4244 /wd4267 /wd4297 /wd4530 /wd4577 /wd4819 /wd4838")
 
+    foreach(flag_var CMAKE_SHARED_LINKER_FLAGS CMAKE_STATIC_LINKER_FLAGS CMAKE_EXE_LINKER_FLAGS CMAKE_LINKER_FLAGS)
+        string(APPEND ${flag_var} "/ignore:4049 /ignore:4217 /ignore:4006 /ignore:4221")
+    endforeach(flag_var)
+
     if (WITH_WIN_DUMP_DBG)
         set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /Zi")
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /Zi")
diff --git a/cmake/cuda.cmake b/cmake/cuda.cmake
index 2f4f5449f48..c4d1384312e 100644
--- a/cmake/cuda.cmake
+++ b/cmake/cuda.cmake
@@ -91,7 +91,7 @@ function(select_nvcc_arch_flags out_variable)
 
   if(${CUDA_ARCH_NAME} STREQUAL "Manual")
     set(CUDA_ARCH_BIN ${paddle_known_gpu_archs} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
-    set(CUDA_ARCH_PTX "50"                     CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
+    set(CUDA_ARCH_PTX ""                        CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
     mark_as_advanced(CUDA_ARCH_BIN CUDA_ARCH_PTX)
   else()
     unset(CUDA_ARCH_BIN CACHE)
@@ -175,14 +175,22 @@ elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 10.0) # CUDA 9.x
   set(paddle_known_gpu_archs ${paddle_known_gpu_archs9})
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
 elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0) # CUDA 10.x
   set(paddle_known_gpu_archs ${paddle_known_gpu_archs10})
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
-elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 12.0) # CUDA 11.x
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
+elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 11.2) # CUDA 11.0/11.1
   set(paddle_known_gpu_archs ${paddle_known_gpu_archs11})
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
+elseif (${CMAKE_CUDA_COMPILER_VERSION} LESS 12.0) # CUDA 11.2+
+  set(paddle_known_gpu_archs "${paddle_known_gpu_archs11} 86")
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D_MWAITXINTRIN_H_INCLUDED")
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -D__STRICT_ANSI__")
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
 endif()
 
 if (NOT ${CMAKE_CUDA_COMPILER_VERSION} LESS 10.0)
-- 
GitLab