[AMDGPU] Fix shouldClusterMemOps to process flat loads

[rampitec]

Summary

Flat loads do not have vdata operand but have vdst instead.

[vpykhtin]

LGTM. However it looks like we should fix TD to follow single naming convention.

[rampitec]

In D37502#862146, @vpykhtin wrote:

LGTM. However it looks like we should fix TD to follow single naming convention.

vdata is for store, vdst is for load, so convention seems fine.

[arsenm]

lib/Target/AMDGPU/SIInstrInfo.cpp

368–369

I think falling back like this is more hazardous than checking if it's flat first

lib/Target/AMDGPU/SIInstrInfo.cpp

bool SIInstrInfo::shouldClusterMemOps(MachineInstr &FirstLdSt,
                                       MachineInstr &SecondLdSt,
                                       unsigned NumLoads) const {
   const MachineOperand *FirstDst = nullptr;
   const MachineOperand *SecondDst = nullptr;
 
   if ((isMUBUF(FirstLdSt) && isMUBUF(SecondLdSt)) ||
       (isMTBUF(FirstLdSt) && isMTBUF(SecondLdSt)) ||
       (isFLAT(FirstLdSt) && isFLAT(SecondLdSt))) {
     FirstDst = getNamedOperand(FirstLdSt, AMDGPU::OpName::vdata);
+    if (!FirstDst)
+      FirstDst = getNamedOperand(FirstLdSt, AMDGPU::OpName::vdst);
     SecondDst = getNamedOperand(SecondLdSt, AMDGPU::OpName::vdata);
+    if (!SecondDst)
+      SecondDst = getNamedOperand(SecondLdSt, AMDGPU::OpName::vdst);
   } else if (isSMRD(FirstLdSt) && isSMRD(SecondLdSt)) {
     FirstDst = getNamedOperand(FirstLdSt, AMDGPU::OpName::sdst);
     SecondDst = getNamedOperand(SecondLdSt, AMDGPU::OpName::sdst);
   } else if (isDS(FirstLdSt) && isDS(SecondLdSt)) {
     FirstDst = getNamedOperand(FirstLdSt, AMDGPU::OpName::vdst);
     SecondDst = getNamedOperand(SecondLdSt, AMDGPU::OpName::vdst);
   }
 

test/CodeGen/AMDGPU/cluster-flat-loads.mir

+# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass machine-scheduler %s -o - | FileCheck -check-prefix=GCN %s
+
+# GCN-LABEL: name: cluster_flat_loads
+# GCN:      FLAT_LOAD_DWORD %0, 0
+# GCN-NEXT: FLAT_LOAD_DWORD %0, 4
+# GCN-NEXT: V_ADD_F32_e64
+name: cluster_flat_loads
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: vreg_64 }
+  - { id: 1, class: vgpr_32 }
+  - { id: 2, class: vgpr_32 }
+  - { id: 3, class: vgpr_32 }
+body:             |
+  bb.0:
+    %0 = IMPLICIT_DEF
+    %1 = FLAT_LOAD_DWORD %0, 0, 0, 0, implicit %exec, implicit %flat_scr :: (load 4)
+    %2 = V_ADD_F32_e64 0, killed %1, 0, 1, 0, 0, implicit %exec
+    %3 = FLAT_LOAD_DWORD %0, 4, 0, 0, implicit %exec, implicit %flat_scr :: (load 4)
+...