[SYCL] Improve sycl-post-link performance with -split=kernel (#6689)

Right now we are computing a new callgraph in every call to
extractCallGraph. extractCallGraph is called every time we do a module
split, so for -split=kernel, that would be once per kernel. For modules
with many kernels, this can take a very long time. We only need to
compute this once because the input IR doesn't seem to change between
splits.

This improves performance of sycl-post-link from ~45min to ~7min for an
example with 13k kernels

Signed-off-by: Sarnie, Nick <[email protected]>

Author: sarnex

llvm/tools/sycl-post-link/ModuleSplitter.cpp

@@ -391,9 +391,10 @@ ModuleDesc extractSubModule(const ModuleDesc &MD,
 // The function produces a copy of input LLVM IR module M with only those entry
 // points that are specified in ModuleEntryPoints vector.
 ModuleDesc extractCallGraph(const ModuleDesc &MD,
-                            EntryPointGroup &&ModuleEntryPoints) {
+                            EntryPointGroup &&ModuleEntryPoints,
+                            const CallGraph &CG) {
   SetVector<const GlobalValue *> GVs;
-  collectFunctionsToExtract(GVs, ModuleEntryPoints, CallGraph{MD.getModule()});
+  collectFunctionsToExtract(GVs, ModuleEntryPoints, CG);
   collectGlobalVarsToExtract(GVs, MD.getModule());
 
   ModuleDesc SplitM = extractSubModule(MD, GVs, std::move(ModuleEntryPoints));
@@ -414,11 +415,15 @@ class ModuleCopier : public ModuleSplitterBase {
 class ModuleSplitter : public ModuleSplitterBase {
 public:
   ModuleSplitter(ModuleDesc &&MD, EntryPointGroupVec &&GroupVec)
-      : ModuleSplitterBase(std::move(MD), std::move(GroupVec)) {}
+      : ModuleSplitterBase(std::move(MD), std::move(GroupVec)),
+        CG(Input.getModule()) {}
 
   ModuleDesc nextSplit() override {
-    return extractCallGraph(Input, nextGroup());
+    return extractCallGraph(Input, nextGroup(), CG);
   }
+
+private:
+  CallGraph CG;
 };
 
 } // namespace