[llvm] [ProfileData] Clean up validateRecord (PR #95488)

[Kazu Hirata via llvm-commits]

================
@@ -1035,16 +1035,13 @@ static const char *ValueProfKindStr[] = {
 
 Error InstrProfWriter::validateRecord(const InstrProfRecord &Func) {
   for (uint32_t VK = 0; VK <= IPVK_Last; VK++) {
-    uint32_t NS = Func.getNumValueSites(VK);
-    if (!NS)
+    if (VK == IPVK_IndirectCallTarget || VK == IPVK_VTableTarget)
       continue;
+    uint32_t NS = Func.getNumValueSites(VK);
     for (uint32_t S = 0; S < NS; S++) {
-      uint32_t ND = Func.getNumValueDataForSite(VK, S);
-      std::unique_ptr<InstrProfValueData[]> VD = Func.getValueForSite(VK, S);
       DenseSet<uint64_t> SeenValues;
-      for (uint32_t I = 0; I < ND; I++)
-        if ((VK != IPVK_IndirectCallTarget && VK != IPVK_VTableTarget) &&
----------------
kazutakahirata wrote:

I happen to have a raw instrumentation profile lying around for clang (that is, clang as an application).  When I generate an indexed profile with `validateRecord` modified to check all value profile kinds, including `IPVK_IndirectCallTarget` and `IPVK_VTableTarget`, then I get an `invalid_prof` error.  So, duplicate values for the two kinds seem to be a real thing.  I have yet to investigate where they come from though.

https://github.com/llvm/llvm-project/pull/95488