[PATCH] D20295: When looking for a spill slot in reg scavenger, find one that matches RC

[Quentin Colombet via llvm-commits]

qcolombet accepted this revision.
This revision is now accepted and ready to land.

================
Comment at: lib/CodeGen/RegisterScavenging.cpp:414
@@ +413,3 @@
+    // that is the best fit for this register class (in street metric).
+    // Picking a larger slot than necessary could happen if a slot for a
+    // larger register is reserved before a slot for a smaller one. When
----------------
What worries me is that we have a quadratic complexity here (#scavengedRegister x Scavenged.size()).
That being said, I guess it is pretty small in practice. Therefore, let us move on and we will adapt if it turns out to be a problem.

================
Comment at: lib/CodeGen/RegisterScavenging.cpp:438-443
@@ -413,4 +437,8 @@
     // Spill the scavenged register before I.
-    assert(Scavenged[SI].FrameIndex >= 0 &&
-           "Cannot scavenge register without an emergency spill slot!");
+    if (Scavenged[SI].FrameIndex < 0) {
+      Twine Msg = Twine("Error while trying to spill ") + TRI->getName(SReg) +
+           " from class " + TRI->getRegClassName(RC) +
+           ": Cannot scavenge register without an emergency spill slot!";
+      report_fatal_error(Msg);
+    }
     TII->storeRegToStackSlot(*MBB, I, SReg, true, Scavenged[SI].FrameIndex,
----------------
kparzysz wrote:
> At the second thought---wouldn't it be better to keep the "unreachable"?  report_fatal_error does not print the call stack.
We could do both maybe.
The problem is that llvm_unreachable expands to nothing in release mode. Thus we may silently generate broken code or a crash somewhere else in the compiler.


Repository:
  rL LLVM

http://reviews.llvm.org/D20295