[PATCH] D111433: [PowerPC] Respect rounding mode in the back end

[Qiu Chaofan via Phabricator via llvm-commits]

qiucf added a comment.

Thanks for fixing this!

I guess the reason to restrict them only undef strict-fp is performance? Since FPSCR is not modeled yet, if so, we may need to re-define all float instructions like this patch when implementing FPSCR model?



================
Comment at: llvm/lib/Target/PowerPC/PPCISelLowering.cpp:5228
+    RetOpc = PPCISD::CALL;
+  if (IsStrictFPCall) {
+    switch (RetOpc) {
----------------
Better to move this into a static function?


================
Comment at: llvm/lib/Target/PowerPC/PPCInstr64Bit.td:181
 
+let isCall = 1, PPC970_Unit = 7, Defs = [LR8, RM], hasSideEffects = 0,
+    isCodeGenOnly = 1, Uses = [RM] in {
----------------
Can they be implemented through `multiclass/defm`?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D111433/new/

https://reviews.llvm.org/D111433