使用Mingw cannot get program for cu

[chaseaishang]

使用MSVS可以正常编译cuda.
但是切换为其他就报错,cannot get program for cu

PS <> xmake -D -v
error: @programdir\core\main.lua:302: @programdir\actions\build\main.lua:206: ...mdir\core\sandbox\modules\import\core\base\scheduler.lua:96: ...mdir\core\sandbox\modules\import\core\base\scheduler.lua:56: @programdir\modules\private\async\runjobs.lua:256: ...amdir\core\sandbox\modules\import\core\tool\compiler.lua:37: cannot get program for cu
stack traceback:
[C]: in function 'error'
[@programdir\core\base\os.lua:913]:
[...amdir\core\sandbox\modules\import\core\tool\compiler.lua:37]: in function 'load'
[@programdir\modules\private\action\build\object.lua:40]: in function 'script'
[@programdir\modules\private\action\build\object.lua:104]: in function 'build_object'
[@programdir\modules\private\action\build\object.lua:129]: in function 'jobfunc'
[@programdir\modules\private\async\runjobs.lua:232]:

stack traceback:
[C]: in function 'error'
@programdir\core\base\os.lua:913: in function 'base/os.raiselevel'
(...tail calls...)
@programdir\modules\private\async\runjobs.lua:256: in field 'catch'
@programdir\core\sandbox\modules\try.lua:123: in global 'try'
@programdir\modules\private\async\runjobs.lua:218: in upvalue 'cotask'
@programdir\core\base\scheduler.lua:404: in function <@programdir\core\base\scheduler.lua:397>
stack traceback:
@programdir\core\base\utils.lua:290: in function <@programdir\core\base\utils.lua:280>
[C]: in function 'error'
@programdir\core\base\os.lua:913: in function 'base/os.raiselevel'
(...tail calls...)
...mdir\core\sandbox\modules\import\core\base\scheduler.lua:56: in field 'co_start_withopt'
@programdir\modules\private\async\runjobs.lua:217: in function <@programdir\modules\private\async\runjobs.lua:162>
[C]: in function 'xpcall'
@programdir\core\base\utils.lua:280: in function 'base/utils.trycall'
(...tail calls...)
@programdir\core\base\task.lua:501: in function 'base/task.run'
@programdir\core\main.lua:300: in upvalue 'cotask'
@programdir\core\base\scheduler.lua:404: in function <@programdir\core\base\scheduler.lua:397>
stack traceback:
[C]: in function 'error'
@programdir\core\base\os.lua:913: in function 'base/os.raiselevel'
(...tail calls...)
@programdir\core\main.lua:302: in upvalue 'cotask'
@programdir\core\base\scheduler.lua:404: in function <@programdir\core\base\scheduler.lua:397>
@programdir\core\base\utils.lua:290: in function <@programdir\core\base\utils.lua:280>
[C]: in function 'error'
@programdir\core\base\os.lua:913: in function 'os.raiselevel'
(...tail calls...)
...mdir\core\sandbox\modules\import\core\base\scheduler.lua:56: in field 'co_start_withopt'
@programdir\modules\private\async\runjobs.lua:217: in function <@programdir\modules\private\async\runjobs.lua:162>
[C]: in function 'xpcall'
@programdir\core\base\utils.lua:280: in function 'base/utils.trycall'
@programdir\core\base\scheduler.lua:520: in function 'base/scheduler.co_group_begin'
...mdir\core\sandbox\modules\import\core\base\scheduler.lua:94: in field 'co_group_begin'
@programdir\modules\private\async\runjobs.lua:162: in function <@programdir\modules\private\async\runjobs.lua:56>
(...tail calls...)
@programdir\actions\build\build.lua:259: in function <@programdir\actions\build\build.lua:247>
(...tail calls...)
@programdir\actions\build\main.lua:109: in global '_do_build'
@programdir\actions\build\main.lua:183: in function <@programdir\actions\build\main.lua:173>
[C]: in function 'xpcall'
@programdir\core\base\utils.lua:280: in function 'base/utils.trycall'
@programdir\core\sandbox\modules\try.lua:117: in global 'try'
@programdir\actions\build\main.lua:171: in function <@programdir\actions\build\main.lua:133>
(...tail calls...)
[C]: in function 'xpcall'
@programdir\core\base\utils.lua:280: in function 'base/utils.trycall'
(...tail calls...)
@programdir\core\base\task.lua:501: in function 'base/task.run'
@programdir\core\main.lua:300: in upvalue 'cotask'
@programdir\core\base\scheduler.lua:404: in function <@programdir\core\base\scheduler.lua:397>

[waruqi]

你怎么切的。。这边没对 mingw cuda 做过特别支持。

[chaseaishang]

就用这个xmake f -c -p mingw -vD

[chaseaishang]

看着都找到了。。。。。
checking for architecture ... x86_64
checking for mingw directory ... E:\DevelopEnvironment\MinGW\mingw64
checking for Cuda SDK directory ... C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0
checking for C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\bin\nvcc ... ok
checking for cuda devices

DEVICE #0
name = "NVIDIA GeForce RTX 2060"
totalGlobalMem = 6442123264
sharedMemPerBlock = 49152
regsPerBlock = 65536
warpSize = 32
memPitch = 2147483647
maxThreadsPerBlock = 1024
maxThreadsDim = (1024, 1024, 64)
maxGridSize = (2147483647, 65535, 65535)
clockRate = 1680000
totalConstMem = 65536
major = 7
minor = 5
textureAlignment = 512
texturePitchAlignment = 32
deviceOverlap = true
multiProcessorCount = 30
kernelExecTimeoutEnabled = false
integrated = false
canMapHostMemory = true
computeMode = 0
maxTexture1D = 131072
maxTexture1DMipmap = 32768
maxTexture1DLinear = 268435456
maxTexture2D = (131072, 65536)
maxTexture2DMipmap = (32768, 32768)
maxTexture2DLinear = (131072, 65000, 2097120)
maxTexture2DGather = (32768, 32768)
maxTexture3D = (16384, 16384, 16384)
maxTexture3DAlt = (8192, 8192, 32768)
maxTextureCubemap = 32768
maxTexture1DLayered = (32768, 2048)
maxTexture2DLayered = (32768, 32768, 2048)
maxTextureCubemapLayered = (32768, 2046)
maxSurface1D = 32768
maxSurface2D = (131072, 65536)
maxSurface3D = (16384, 16384, 16384)
maxSurface1DLayered = (32768, 2048)
maxSurface2DLayered = (32768, 32768, 2048)
maxSurfaceCubemap = 32768
pciDomainID = 0
tccDriver = false
asyncEngineCount = 2
unifiedAddressing = true
memoryClockRate = 7001000
memoryBusWidth = 192
l2CacheSize = 3145728
maxThreadsPerMultiProcessor = 1024
streamPrioritiesSupported = true
globalL1CacheSupported = true
localL1CacheSupported = true
sharedMemPerMultiprocessor = 65536
regsPerMultiprocessor = 65536
isMultiGpuBoard = false
multiGpuBoardGroupID = 0
singleToDoublePrecisionPerfRatio = 32
pageableMemoryAccess = false
concurrentManagedAccess = false
managedMemory = true
computePreemptionSupported = true
canUseHostPointerForRegisteredMem = false
cooperativeLaunch = true
cooperativeMultiDeviceLaunch = false
sharedMemPerBlockOptin = 65536
pageableMemoryAccessUsesHostPageTables = false
directManagedMemAccessFromHost = false
uuid = "c8b1cd5e-2c0f-fa19-2e64-c7d89f438936"
luid = "35fd000000000000"
luidDeviceNodeMask = 1
found device #0: NVIDIA GeForce RTX 2060 with compute 7.5 capability
configure
{
arch = x86_64
ccache = true
ndk_stdcxx = true
buildir = build
cuda = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0
host = windows
mingw = E:\DevelopEnvironment\MinGW\mingw64
clean = true
mode = release
plat = mingw
kind = static
}

[chaseaishang]

换成clang也不行
PS E:\workplace\c++\testxmake\test> xmake
checking for Microsoft Visual Studio (x64) version ... 2022
checking for Cuda SDK directory ... C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0
[ 25%]: compiling.release src\main.cu
error: clang: error: unknown argument: '-nologo'
clang: error: unknown argument: '-EHsc'
clang: error: no such file or directory: 'cuda_runtime.h'
clang: error: no such file or directory: 'E:workplacec++testxmaketestsrcmain.cu'
clang: error: no input files

in src\main.cu
warning: add_cuflags("-rdc=true") is ignored, please pass {force = true} or call set_policy("check.auto_ignore_flags", false) if you want to set it.
PS E:\workplace\c++\testxmake\test>

[waruqi]

目前 mingw 平台，没自动绑定 cuda 工具链，只有 windows 平台才绑定了。。

就算 mingw 下，调用 nvcc 编译 .cu，默认应该也是用的 msvc 的 ccbin，确定跟 mingw c++ 混编兼容么？

另外，我只知道 windows 上 nvcc 支持 msvc 的 ccbin ，其他 clang mingw gcc 的 ccbin 也支持？我怎么记得之前 issues 有人说不支持来着。

[chaseaishang]

行吧，我找到的也都说Windows下MinGW不行。都是用MSVC
但是我的clang为啥也不行。。。
用CMake也是只有MSVC行。。。
PS E:\workplace\c++\testxmake\test> xmake -rv
[ 25%]: compiling.release src\main.cu
"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\bin\nvcc" -c -ccbin=clang "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\include" -m64 -gencode arch=compute_75,code=sm_75 -o build.objs\test\windows\x64\release\src\main.cu.obj src\main.cu
error: clang: error: unknown argument: '-nologo'
clang: error: unknown argument: '-EHsc'
clang: error: no such file or directory: 'cuda_runtime.h'
clang: error: no such file or directory: 'E:workplacec++testxmaketestsrcmain.cu'
clang: error: no input files
warning: add_cuflags("-rdc=true") is ignored, please pass {force = true} or call set_policy("check.auto_ignore_flags", false) if you want to set it.
PS E:\workplace\c++\testxmake\test> xmake -rv
checking for C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\bin\nvcc ... ok
checking for the cuda compiler (cu) ... nvcc
checking for cl.exe ... E:\vistualStudio\VC\Tools\MSVC\14.34.31933\bin\HostX64\x64\cl.exe
checking for the c compiler (cc) ... cl.exe
checking for C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\bin\C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\bin\nvcc ... no
checking for flags (-O3) ... no
checking for flags (-rdc=true) ... no
checking for flags (-gencode arch=compute_75,code=sm_75) ... ok
[ 25%]: compiling.release src\main.cu
"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\bin\nvcc" -c -ccbin=clang "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\include" -m64 -gencode arch=compute_75,code=sm_75 -o build.objs\test\windows\x64\release\src\main.cu.obj src\main.cu
checking for flags (-MD -MF) ... no
checking for flags (-M) ... no
error: clang: error: unknown argument: '-nologo'
clang: error: unknown argument: '-EHsc'
clang: error: no such file or directory: 'cuda_runtime.h'
clang: error: no such file or directory: 'E:workplacec++testxmaketestsrcmain.cu'
clang: error: no input fileswarning: .\xmake.lua:6: cannot match add_files("*.cu") in target(test)
warning: add_cuflags("-rdc=true") is ignored, please pass {force = true} or call set_policy("check.auto_ignore_flags", false) if you want to set it.
PS E:\workplace\c++\testxmake\test>

[waruqi]

但是我的clang为啥也不行。。。

我记得之前有人 issues 反馈过，windows 上 ccbin 仅仅支持 cl.exe 。。你切 clang 也不支持。。

只有 linux 下，你可以切 clang

[waruqi]

这跟 xmake 好像也没啥关系。。你可以直接调 nvcc 命令试，不行，那就是 nvcc 的问题。。你得到 nvcc 那边反馈

[chaseaishang]

ok,ok。看来还是不支持。我之前不知道在哪看到xmake可以支持其他cxx编译器编译。。。
我还是老实用MSVC吧。谢谢你了