CUDA/Clang: Simplify --register-link-binaries logic

Move the logic for appending cubin afterwards, so the check can simply be
empty().
With the Makefile generator the option is now at the front instead of being
intermixed with the actual bins.

Source/cmMakefileTargetGenerator.cxx

@@ -1519,22 +1519,13 @@ void cmMakefileTargetGenerator::WriteDeviceLinkRule(
 
   // Link device code for each architecture.
   for (const std::string& architectureKind : architectures) {
-    // Clang always generates real code, so strip the specifier.
-    const std::string architecture =
-      architectureKind.substr(0, architectureKind.find('-'));
-    const std::string cubin =
-      cmStrCat(objectDir, "sm_", architecture, ".cubin");
-
-    profiles += cmStrCat(" -im=profile=sm_", architecture, ",file=", cubin);
-    fatbinaryDepends.emplace_back(cubin);
-
     std::string registerFileCmd;
 
     // The generated register file contains macros that when expanded
     // register the device routines. Because the routines are the same for
     // all architectures the register file will be the same too. Thus
     // generate it only on the first invocation to reduce overhead.
-    if (fatbinaryDepends.size() == 1) {
+    if (fatbinaryDepends.empty()) {
       std::string const registerFileRel =
         cmStrCat(relPath, relObjectDir, "cmake_cuda_register.h");
       registerFileCmd =
@@ -1542,6 +1533,15 @@ void cmMakefileTargetGenerator::WriteDeviceLinkRule(
       cleanFiles.push_back(registerFileRel);
     }
 
+    // Clang always generates real code, so strip the specifier.
+    const std::string architecture =
+      architectureKind.substr(0, architectureKind.find('-'));
+    const std::string cubin =
+      cmStrCat(objectDir, "sm_", architecture, ".cubin");
+
+    profiles += cmStrCat(" -im=profile=sm_", architecture, ",file=", cubin);
+    fatbinaryDepends.emplace_back(cubin);
+
     std::string command = cmStrCat(
       this->Makefile->GetRequiredDefinition("CMAKE_CUDA_DEVICE_LINKER"),
       " -arch=sm_", architecture, registerFileCmd, " -o=$@ ",

Source/cmNinjaNormalTargetGenerator.cxx

@@ -753,10 +753,6 @@ void cmNinjaNormalTargetGenerator::WriteDeviceLinkStatements(
     const std::string cubin =
       cmStrCat(ninjaOutputDir, "/sm_", architecture, ".cubin");
 
-    fatbinary.Variables["PROFILES"] +=
-      cmStrCat(" -im=profile=sm_", architecture, ",file=", cubin);
-    fatbinary.ExplicitDeps.emplace_back(cubin);
-
     cmNinjaBuild dlink(this->LanguageLinkerCudaDeviceRule(config));
     dlink.ExplicitDeps = explicitDeps;
     dlink.Outputs = { cubin };
@@ -766,11 +762,15 @@ void cmNinjaNormalTargetGenerator::WriteDeviceLinkStatements(
     // the device routines. Because the routines are the same for all
     // architectures the register file will be the same too. Thus generate it
     // only on the first invocation to reduce overhead.
-    if (fatbinary.ExplicitDeps.size() == 1) {
+    if (fatbinary.ExplicitDeps.empty()) {
       dlink.Variables["REGISTER"] = cmStrCat(
         "--register-link-binaries=", ninjaOutputDir, "/cmake_cuda_register.h");
     }
 
+    fatbinary.Variables["PROFILES"] +=
+      cmStrCat(" -im=profile=sm_", architecture, ",file=", cubin);
+    fatbinary.ExplicitDeps.emplace_back(cubin);
+
     this->GetGlobalGenerator()->WriteBuild(this->GetCommonFileStream(), dlink);
   }