Merge pull request #315 from YdrMaster/main

issue/314 修复 nvidia 上编译和测试问题

xmake.lua

@@ -22,7 +22,7 @@ end
 option("cpu")
     set_default(true)
     set_showmenu(true)
-    set_description("Whether to complie implementations for CPU")
+    set_description("Whether to compile implementations for CPU")
 option_end()
 
 option("omp")
@@ -44,32 +44,29 @@ end
 option("nv-gpu")
     set_default(false)
     set_showmenu(true)
-    set_description("Whether to complie implementations for Nvidia GPU")
+    set_description("Whether to compile implementations for Nvidia GPU")
 option_end()
 
 if has_config("nv-gpu") then
-    add_defines("ENABLE_CUDA_API")
+    add_defines("ENABLE_NVIDIA_API")
     includes("xmake/cuda.lua")
 end
 
--- 天数智芯
-option("iluvatar-gpu")
-    set_default(false)
+option("cudnn")
+    set_default(true)
     set_showmenu(true)
-    set_description("Whether to complie implementations for Iluvatar GPU")
+    set_description("Whether to compile cudnn for Nvidia GPU")
 option_end()
 
-if has_config("iluvatar-gpu") then
-    add_defines("ENABLE_CUDA_API")
-    add_defines("ENABLE_ILUVATAR_CUDA_API")
-    includes("xmake/iluvatar.lua")
+if has_config("cudnn") then
+    add_defines("ENABLE_CUDNN_API")
 end
 
 -- 寒武纪
 option("cambricon-mlu")
     set_default(false)
     set_showmenu(true)
-    set_description("Whether to complie implementations for Cambricon MLU")
+    set_description("Whether to compile implementations for Cambricon MLU")
 option_end()
 
 if has_config("cambricon-mlu") then
@@ -81,7 +78,7 @@ end
 option("ascend-npu")
     set_default(false)
     set_showmenu(true)
-    set_description("Whether to complie implementations for Huawei Ascend NPU")
+    set_description("Whether to compile implementations for Huawei Ascend NPU")
 option_end()
 
 if has_config("ascend-npu") then
@@ -89,11 +86,24 @@ if has_config("ascend-npu") then
     includes("xmake/ascend.lua")
 end
 
+-- 天数智芯
+option("iluvatar-gpu")
+    set_default(false)
+    set_showmenu(true)
+    set_description("Whether to compile implementations for Iluvatar GPU")
+option_end()
+
+if has_config("iluvatar-gpu") then
+    add_defines("ENABLE_CUDA_API")
+    add_defines("ENABLE_ILUVATAR_CUDA_API")
+    includes("xmake/iluvatar.lua")
+end
+
 -- 沐曦
 option("metax-gpu")
     set_default(false)
     set_showmenu(true)
-    set_description("Whether to complie implementations for MetaX GPU")
+    set_description("Whether to compile implementations for MetaX GPU")
 option_end()
 
 if has_config("metax-gpu") then
@@ -105,7 +115,7 @@ end
 option("moore-gpu")
     set_default(false)
     set_showmenu(true)
-    set_description("Whether to complie implementations for Moore Threads GPU")
+    set_description("Whether to compile implementations for Moore Threads GPU")
 option_end()
 
 if has_config("moore-gpu") then
@@ -117,11 +127,10 @@ end
 option("sugon-dcu")
     set_default(false)
     set_showmenu(true)
-    set_description("Whether to complie implementations for Sugon DCU")
+    set_description("Whether to compile implementations for Sugon DCU")
 option_end()
 
 if has_config("sugon-dcu") then
-    add_defines("ENABLE_CUDA_API")
     add_defines("ENABLE_SUGON_CUDA_API")
 end
 
@@ -142,7 +151,7 @@ option("ccl")
 set_default(false)
     set_default(false)
     set_showmenu(true)
-    set_description("Wether to complie implementations for InfiniCCL")
+    set_description("Wether to compile implementations for InfiniCCL")
 option_end()
 
 if has_config("ccl") then

xmake/cuda.lua

-
-local CUDA_ROOT = os.getenv("CUDA_ROOT") or os.getenv("CUDA_HOME") or os.getenv("CUDA_PATH")
 local CUDNN_ROOT = os.getenv("CUDNN_ROOT") or os.getenv("CUDNN_HOME") or os.getenv("CUDNN_PATH")
-if CUDA_ROOT ~= nil then
-    add_includedirs(CUDA_ROOT .. "/include")
-end
 if CUDNN_ROOT ~= nil then
     add_includedirs(CUDNN_ROOT .. "/include")
 end
@@ -15,10 +10,20 @@ target("infiniop-cuda")
 
     set_policy("build.cuda.devlink", true)
     set_toolchains("cuda")
-    add_links("cuda", "cublas", "cudnn")
-    add_linkdirs(CUDA_ROOT .. "/lib64/stubs")
+    add_links("cudart", "cublas")
+    if has_config("cudnn") then
+        add_links("cudnn")
+    end
     add_cugencodes("native")
 
+    on_load(function (target)
+        import("lib.detect.find_tool")
+        local nvcc = find_tool("nvcc")
+        if nvcc ~= nil then
+            target:add("linkdirs", path.directory(path.directory(nvcc.program)) .. "/lib64/stubs")
+        end
+    end)
+
     if is_plat("windows") then
         add_cuflags("-Xcompiler=/utf-8", "--expt-relaxed-constexpr", "--allow-unsupported-compiler")
         add_cuflags("-Xcompiler=/W3", "-Xcompiler=/WX")
@@ -38,6 +43,8 @@ target("infiniop-cuda")
         end
     end
 
+    add_cuflags("-Xcompiler=-Wno-error=deprecated-declarations")
+
     set_languages("cxx17")
     add_files("../src/infiniop/devices/cuda/*.cu", "../src/infiniop/ops/*/cuda/*.cu", "../build/ninetoothed/*.c")
 target_end()
@@ -92,5 +99,5 @@ target("infiniccl-cuda")
         end
     end
     set_languages("cxx17")
-    
+
 target_end()