[llvm] Add remaining patterns for floating-point flag matches (PR #173912)

[Mikołaj Piróg via llvm-commits]

================
@@ -81,19 +82,76 @@ template <typename T> inline OneUse_match<T> m_OneUse(const T &SubPattern) {
   return SubPattern;
 }
 
-template <typename SubPattern_t> struct AllowReassoc_match {
+template <typename SubPattern_t, int Flag> struct AllowFmf_match {
   SubPattern_t SubPattern;
-
-  AllowReassoc_match(const SubPattern_t &SP) : SubPattern(SP) {}
+  FastMathFlags FMF;
+
+  AllowFmf_match(const SubPattern_t &SP) : SubPattern(SP) {
+    if (Flag == FastMathFlags::AllowReassoc)
+      FMF.setAllowReassoc();
+    if (Flag == FastMathFlags::AllowReciprocal)
+      FMF.setAllowReciprocal();
+    if (Flag == FastMathFlags::AllowContract)
+      FMF.setAllowContract();
+    if (Flag == FastMathFlags::ApproxFunc)
+      FMF.setApproxFunc();
+    if (Flag == FastMathFlags::NoNaNs)
+      FMF.setNoNaNs();
+    if (Flag == FastMathFlags::NoInfs)
+      FMF.setNoInfs();
+    if (Flag == FastMathFlags::NoSignedZeros)
+      FMF.setNoSignedZeros();
----------------
mikolaj-pirog wrote:

I'm not aware of particular reasons why is that -- the ctor that takes int is private. Maybe because we can't control if the value of the int makes sense for FastMathFlags? I think it should be public regardless + have an assert to check if the flag value makes sense. What do you think?

https://github.com/llvm/llvm-project/pull/173912