[llvm] [AMDGPU] Shrink S_MOV_B64 to S_MOV_B32 during rematerialization (PR #184333)

[Matt Arsenault via llvm-commits]

================
@@ -2609,14 +2609,50 @@ bool SIInstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
 
 void SIInstrInfo::reMaterialize(MachineBasicBlock &MBB,
                                 MachineBasicBlock::iterator I, Register DestReg,
-                                unsigned SubIdx,
-                                const MachineInstr &Orig) const {
+                                unsigned SubIdx, const MachineInstr &Orig,
+                                LaneBitmask UsedLanes) const {
 
   // Try shrinking the instruction to remat only the part needed for current
   // context.
   // TODO: Handle more cases.
   unsigned Opcode = Orig.getOpcode();
   switch (Opcode) {
+  case AMDGPU::S_MOV_B64:
+  case AMDGPU::S_MOV_B64_IMM_PSEUDO: {
+    if (SubIdx != 0)
+      break;
+
+    if (!Orig.getOperand(1).isImm())
----------------
arsenm wrote:

It should be possible to split other pseudo constants and relocations, but that's best left for later and less likely to be useful 

https://github.com/llvm/llvm-project/pull/184333