[llvm] [RISCV] Prefer SP over FP for frame index access when offset fits within compressed immediate range. (PR #193962)

[Craig Topper via llvm-commits]

================
@@ -1518,8 +1518,30 @@ RISCVFrameLowering::getFrameIndexReference(const MachineFunction &MF, int FI,
       assert(!MFI.hasVarSizedObjects());
       FrameReg = SPReg;
     }
+  } else if (!RI->hasStackRealignment(MF)) {
+    // Note: Keeping the following as multiple 'if' statements rather than
+    // merging to a single expression for readability.
+    if (!hasFP(MF)) {
+      // No FP available, must use SP.
+      FrameReg = SPReg;
+    } else {
+      FrameReg = FPReg;
+      // Both FP and SP are candidates.
+      if (RVFI->getRVVStackSize() == 0 && !MFI.hasVarSizedObjects()) {
+        // Prefer SP when the SP-relative offset fits in the compressed
+        // instruction immediate range.
----------------
topperc wrote:

If it's not harmful to uncompressed instructions, I would prefer to not a separate behavior for compressed and uncompressed here. It just adds complexity to the compiler.

https://github.com/llvm/llvm-project/pull/193962