[llvm] 521ebc1 - GlobalISel: Move finalizeLowering call later

[Matt Arsenault via llvm-commits]

Author: Matt Arsenault
Date: 2020-07-06T09:19:40-04:00
New Revision: 521ebc168152ab72047e2e7c81c8c6724b3e7623

URL: https://github.com/llvm/llvm-project/commit/521ebc168152ab72047e2e7c81c8c6724b3e7623
DIFF: https://github.com/llvm/llvm-project/commit/521ebc168152ab72047e2e7c81c8c6724b3e7623.diff

LOG: GlobalISel: Move finalizeLowering call later

This matches the DAG behavior where this is called after the loop
checking for calls. The AMDGPU implementation depends on knowing if
there are calls in the function or not, so move this later.

Another problem is finalizeLowering is actually called twice; I was
seeing weird inconsistencies since the first call would produce
unexpected results and the second run would correct them in some
contexts. Since this requires disabling the verifier, and it's useful
to serialize the MIR immediately after selection, FinalizeISel should
probably not be a real pass.

Added: 
    

Modified: 
    llvm/lib/CodeGen/GlobalISel/InstructionSelect.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/GlobalISel/InstructionSelect.cpp b/llvm/lib/CodeGen/GlobalISel/InstructionSelect.cpp
index fc114c6edc0d..f32278d07052 100644
--- a/llvm/lib/CodeGen/GlobalISel/InstructionSelect.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/InstructionSelect.cpp
@@ -223,9 +223,6 @@ bool InstructionSelect::runOnMachineFunction(MachineFunction &MF) {
     return false;
   }
 #endif
-  auto &TLI = *MF.getSubtarget().getTargetLowering();
-  TLI.finalizeLowering(MF);
-
   // Determine if there are any calls in this machine function. Ported from
   // SelectionDAG.
   MachineFrameInfo &MFI = MF.getFrameInfo();
@@ -241,6 +238,9 @@ bool InstructionSelect::runOnMachineFunction(MachineFunction &MF) {
     }
   }
 
+  // FIXME: FinalizeISel pass calls finalizeLowering, so it's called twice.
+  auto &TLI = *MF.getSubtarget().getTargetLowering();
+  TLI.finalizeLowering(MF);
 
   LLVM_DEBUG({
     dbgs() << "Rules covered by selecting function: " << MF.getName() << ":";
@@ -249,11 +249,7 @@ bool InstructionSelect::runOnMachineFunction(MachineFunction &MF) {
     dbgs() << "\n\n";
   });
   CoverageInfo.emit(CoveragePrefix,
-                    MF.getSubtarget()
-                        .getTargetLowering()
-                        ->getTargetMachine()
-                        .getTarget()
-                        .getBackendName());
+                    TLI.getTargetMachine().getTarget().getBackendName());
 
   // If we successfully selected the function nothing is going to use the vreg
   // types after us (otherwise MIRPrinter would need them). Make sure the types