[llvm] [PartiallyInlineLibCalls] Emit missed- and passed-optimization remarks when partially inlining sqrt (PR #123966)

[via llvm-commits]

================
@@ -56,6 +56,20 @@ static bool optimizeSQRT(CallInst *Call, Function *CalledFunc,
   // dst = phi(v0, v1)
   //
 
+  ORE->emit([&]() {
+    return OptimizationRemark(DEBUG_TYPE, "SqrtPartiallyInlined",
+                              Call->getDebugLoc(), &CurrBB)
+           << "Partially inlined call to sqrt function despite having to use "
+              "errno for error handling: target has fast sqrt instruction";
+  });
+  ORE->emit([&]() {
+    return OptimizationRemarkMissed(DEBUG_TYPE, "BranchInserted",
+                                    Call->getDebugLoc(), &CurrBB)
+           << "Branch to library sqrt fn had to be inserted to satisfy the "
+              "current target's requirement for math functions to set errno on "
+              "invalid inputs";
----------------
TiborGY wrote:

Hmm, I can see how this could be a matter of interpretation. I have felt the need to separate the two messages as they convey two different aspects of the same event.
First, the `OptimizationRemark` signals that a transformation which is expected to be profitable has been done.
Second, the `OptimizationRemarkMissed` signals that that there is a condition inhibiting this optimization from being even better.
Splitting these enables users to have more relevant information with filtering, eg. if they are only interested in spots where LLVM is struggling to optimize something, they can look at `OptimizationRemarkMissed` only.
Combining the two messages or turning the second one into a second `OptimizationRemark` would defeat this.

Perhaps my mental model is wrong, but to me `OptimizationRemark` means "LLVM did something it believes to be useful" and `OptimizationRemarkMissed` means "LLVM cannot or will not do something that (hypothetically) could have been useful". Hence the separation.

https://github.com/llvm/llvm-project/pull/123966