Merge pull request #990 from InfiniTensor/demo131

Demo-131 Cuda graph with optimized paged attention

xmake/qy.lua

@@ -88,6 +88,7 @@ target("infiniop-qy")
         add_cuflags("-Xcompiler=-fPIC")
         add_cuflags("--extended-lambda")
         add_culdflags("-Xcompiler=-fPIC")
+        add_cxflags("-fPIC")
         add_cxxflags("-fPIC")
         add_cuflags("--expt-relaxed-constexpr")
         if CUDNN_ROOT ~= nil then
@@ -98,10 +99,10 @@ target("infiniop-qy")
     add_cuflags("-Xcompiler=-Wno-error=deprecated-declarations")
 
     set_languages("cxx17")
-    add_files("../src/infiniop/devices/nvidia/*.cu", "../src/infiniop/ops/*/nvidia/*.cu")
+    add_files("../src/infiniop/devices/nvidia/*.cu", "../src/infiniop/ops/*/nvidia/*.cu", "../src/infiniop/ops/*/*/nvidia/*.cu")
 
     if has_config("ninetoothed") then
-        add_files("../build/ninetoothed/*.c")
+        add_files("../build/ninetoothed/*.c", "../build/ninetoothed/*.cpp")
     end
 target_end()
 
@@ -117,6 +118,7 @@ target("infinirt-qy")
         add_cuflags("-Xcompiler=-fPIC")
         add_culdflags("-Xcompiler=-fPIC")
         add_cxflags("-fPIC")
+        add_cxxflags("-fPIC")
     end
 
     set_languages("cxx17")
@@ -133,6 +135,7 @@ target("infiniccl-qy")
             add_cuflags("-Xcompiler=-fPIC")
             add_culdflags("-Xcompiler=-fPIC")
             add_cxflags("-fPIC")
+            add_cxxflags("-fPIC")
 
             local nccl_root = os.getenv("NCCL_ROOT")
             if nccl_root then

xmake/test.lua

@@ -24,7 +24,7 @@ target("infiniop-test")
     add_links("infiniop", "infinirt")
 
     if has_config("omp") then
-        add_cxflags("-fopenmp")
+        add_cxxflags("-fopenmp")
         add_ldflags("-fopenmp")
     end