Cuda update

README.md

@@ -278,7 +278,7 @@ target("test")
     set_kind("binary")
     add_files("src/*.cu")
     add_cugencodes("native")
-    add_cugencodes("compute_30")
+    add_cugencodes("compute_35")
 ```
 
 WDK/UMDF Driver Program:

README_zh.md

@@ -280,7 +280,7 @@ target("test")
     set_kind("binary")
     add_files("src/*.cu")
     add_cugencodes("native")
-    add_cugencodes("compute_30")
+    add_cugencodes("compute_35")
 ```
 
 WDK/UMDF驱动程序:

tests/projects/cuda/console/xmake.lua

@@ -14,9 +14,9 @@ target("cuda_console")
     add_files("src/*.cu")
 
     -- generate SASS code for each SM architecture
-    add_cugencodes("sm_30", "sm_35", "sm_37", "sm_50", "sm_52", "sm_60", "sm_61", "sm_70")
+    add_cugencodes("sm_35", "sm_37", "sm_50", "sm_52", "sm_60", "sm_61", "sm_70", "sm_75")
 
     -- generate PTX code from the highest SM architecture to guarantee forward-compatibility
-    add_cugencodes("compute_70")
+    add_cugencodes("compute_75")
 
 

tests/projects/cuda/console_2/xmake.lua

@@ -3,7 +3,7 @@
 add_rules("mode.debug", "mode.release")
 
 -- generate PTX code for the virtual architecture to guarantee compatibility
-add_cugencodes("compute_30")
+add_cugencodes("compute_35")
 
 -- define target
 target("bin")

xmake/modules/core/tools/nvcc.lua

@@ -33,6 +33,7 @@ function init(self)
     -- init cuflags
     if not is_plat("windows", "mingw") then
         self:set("shared.cuflags", "-Xcompiler -fPIC")
+        self:set("binary.cuflags", "-Xcompiler -fPIE")
     end
 
     -- add -ccbin

xmake/modules/detect/tools/nvcc/has_flags.lua

@@ -112,12 +112,24 @@ function _check_try_running(flags, opt, islinker)
         io.writefile(sourcefile, "int main(int argc, char** argv)\n{return 0;}")
     end
 
-    -- check flags
-    if islinker then
-        return _try_running(opt.program, table.join(flags, "-o", os.nuldev(), sourcefile))
-    else
-        return _try_running(opt.program, table.join(flags, "-c", "-o", os.nuldev(), sourcefile))
+    local args = table.join("-o", os.nuldev(), sourcefile)
+
+    if not islinker then
+        table.insert(args, 1, "-c")
+    end
+
+    -- avoid recursion
+    if flags[1] ~= "-allow-unsupported-compiler" then
+        -- add -allow-unsupported-compiler if supported to suppress error of unsupported compiler,
+        -- which caused all checks failed.
+        local allow_unsupported_compiler = _has_flags({"-allow-unsupported-compiler"}, opt)
+        if allow_unsupported_compiler then
+            table.insert(args, 1, "-allow-unsupported-compiler")
+        end
     end
+
+    -- check flags
+    return _try_running(opt.program, table.join(flags, args))
 end
 
 -- has_flags(flags)?
@@ -126,7 +138,7 @@ end
 --
 -- @return      true or false
 --
-function main(flags, opt)
+function _has_flags(flags, opt)
 
     -- is linker?
     local islinker = _islinker(flags, opt)
@@ -140,3 +152,7 @@ function main(flags, opt)
     return _check_try_running(flags, opt, islinker)
 end
 
+function main(...)
+    return _has_flags(...)
+end
+

xmake/modules/lib/detect/find_cudadevices.lua

@@ -238,6 +238,7 @@ function _order_by_flops(devices)
     ,   [70] =     64
     ,   [72] =     64
     ,   [75] =     64
+    ,   [80] =     64
     }
 
     for _, dev in ipairs(devices) do

xmake/rules/cuda/gencodes/xmake.lua

@@ -42,8 +42,9 @@ rule("cuda.gencodes")
         import("core.base.hashset")
 
         -- sm_20 and compute_20 is supported until CUDA 8
-        local known_v_archs = hashset.of(20, 30, 32, 35, 37, 50, 52, 53, 60, 61, 62, 70, 72, 75)
-        local known_r_archs = hashset.of(20, 30, 32, 35, 37, 50, 52, 53, 60, 61, 62, 70, 72, 75)
+        -- sm_30 and compute_30 is supported until CUDA 10
+        local known_v_archs = hashset.of(20, 30, 32, 35, 37, 50, 52, 53, 60, 61, 62, 70, 72, 75, 80)
+        local known_r_archs = hashset.of(20, 30, 32, 35, 37, 50, 52, 53, 60, 61, 62, 70, 72, 75, 80)
 
         local function nf_cugencode(archs)
             if type(archs) ~= 'string' then

xmake/templates/cuda/console/project/xmake.lua

@@ -21,10 +21,10 @@ target("${TARGETNAME}")
     add_cugencodes("native")
 
     -- generate PTX code for the virtual architecture to guarantee compatibility
-    add_cugencodes("compute_30")
+    add_cugencodes("compute_35")
 
     -- -- generate SASS code for each SM architecture
-    -- add_cugencodes("sm_30", "sm_35", "sm_37", "sm_50", "sm_52", "sm_60", "sm_61", "sm_70", "sm_75")
+    -- add_cugencodes("sm_35", "sm_37", "sm_50", "sm_52", "sm_60", "sm_61", "sm_70", "sm_75")
 
     -- -- generate PTX code from the highest SM architecture to guarantee forward-compatibility
     -- add_cugencodes("compute_75")