[llvm] [SDAG] Add missing float type legalizations for FMODF (PR #128055)

[Benjamin Maxwell via llvm-commits]

================
@@ -2616,6 +2624,18 @@ bool SelectionDAG::expandMultipleResultFPLibCall(
 
   auto [Call, CallChain] = TLI->LowerCallTo(CLI);
 
+  if (CallRetResNo && !Node->hasAnyUseOfValue(*CallRetResNo)) {
+    // This is needed for x87, which uses a floating-point stack. If (for
----------------
MacDue wrote:

> The "generic" issue is that libcalls inserted after SelectionDAGBuilder are weird. Normal calls have a unique ordering: the end of one call is attached to the beginning of the next. 

Yes, I realize this pain (they only kinda work). Ideally, we would just do what `SelectionDAGBuilder` does, but that depends on IR order -- which we've lost at this point.

> The x86-specific issue is that the "CopyFromReg" has a side-effect. Well, not really a side-effect, but the COPY needs to be present so that later code can correctly compute the depth of the x87 stack. 

Yep, it is a hidden side-effect, which is not really modeled in the DAG, which does make this awkward and bug-prone case -- since most targets (maybe all but x87 FP) don't work like this. 

> My primary issue with the chain barrier is it doesn't generalize properly. For example, say you have a libcall that returns a complex number as an fp0/fp1 pair, and subsequent code throws away fp1. You run into basically the same issue,

Could you give an example in pseudo-SDAG syntax? The fix in this PR does not depend on the value from the copy-from-reg being used, only that the chain has some later use. 




https://github.com/llvm/llvm-project/pull/128055