return true;
}
+// This function assumes that \p A and \p B have are identical except for
+// size and offset, and they referecne adjacent memory.
+static MachineMemOperand *combineKnownAdjacentMMOs(MachineFunction &MF,
+ const MachineMemOperand *A,
+ const MachineMemOperand *B) {
+ unsigned MinOffset = std::min(A->getOffset(), B->getOffset());
+ unsigned Size = A->getSize() + B->getSize();
+ return MF.getMachineMemOperand(A, MinOffset, Size);
+}
+
bool SILoadStoreOptimizer::offsetsCanBeCombined(CombineInfo &CI) {
// XXX - Would the same offset be OK? Is there any reason this would happen or
// be useful?
unsigned DestReg = MRI->createVirtualRegister(SuperRC);
unsigned MergedOffset = std::min(CI.Offset0, CI.Offset1);
+ // It shouldn't be possible to get this far if the two instructions
+ // don't have a single memoperand, because MachineInstr::mayAlias()
+ // will return true if this is the case.
+ assert(CI.I->hasOneMemOperand() && CI.Paired->hasOneMemOperand());
+
+ const MachineMemOperand *MMOa = *CI.I->memoperands_begin();
+ const MachineMemOperand *MMOb = *CI.Paired->memoperands_begin();
+
BuildMI(*MBB, CI.Paired, DL, TII->get(Opcode), DestReg)
.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::sbase))
.addImm(MergedOffset) // offset
.addImm(CI.GLC0) // glc
.addImm(CI.DLC0) // dlc
- .cloneMergedMemRefs({&*CI.I, &*CI.Paired});
+ .addMemOperand(combineKnownAdjacentMMOs(*MBB->getParent(), MMOa, MMOb));
std::pair<unsigned, unsigned> SubRegIdx = getSubRegIdxs(CI);
const unsigned SubRegIdx0 = std::get<0>(SubRegIdx);
if (Regs & VADDR)
MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::vaddr));
+ // It shouldn't be possible to get this far if the two instructions
+ // don't have a single memoperand, because MachineInstr::mayAlias()
+ // will return true if this is the case.
+ assert(CI.I->hasOneMemOperand() && CI.Paired->hasOneMemOperand());
+
+ const MachineMemOperand *MMOa = *CI.I->memoperands_begin();
+ const MachineMemOperand *MMOb = *CI.Paired->memoperands_begin();
+
MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::srsrc))
.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::soffset))
.addImm(MergedOffset) // offset
.addImm(CI.SLC0) // slc
.addImm(0) // tfe
.addImm(CI.DLC0) // dlc
- .cloneMergedMemRefs({&*CI.I, &*CI.Paired});
+ .addMemOperand(combineKnownAdjacentMMOs(*MBB->getParent(), MMOa, MMOb));
std::pair<unsigned, unsigned> SubRegIdx = getSubRegIdxs(CI);
const unsigned SubRegIdx0 = std::get<0>(SubRegIdx);
if (Regs & VADDR)
MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::vaddr));
+
+ // It shouldn't be possible to get this far if the two instructions
+ // don't have a single memoperand, because MachineInstr::mayAlias()
+ // will return true if this is the case.
+ assert(CI.I->hasOneMemOperand() && CI.Paired->hasOneMemOperand());
+
+ const MachineMemOperand *MMOa = *CI.I->memoperands_begin();
+ const MachineMemOperand *MMOb = *CI.Paired->memoperands_begin();
+
MIB.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::srsrc))
.add(*TII->getNamedOperand(*CI.I, AMDGPU::OpName::soffset))
.addImm(std::min(CI.Offset0, CI.Offset1)) // offset
.addImm(CI.SLC0) // slc
.addImm(0) // tfe
.addImm(CI.DLC0) // dlc
- .cloneMergedMemRefs({&*CI.I, &*CI.Paired});
+ .addMemOperand(combineKnownAdjacentMMOs(*MBB->getParent(), MMOa, MMOb));
moveInstsAfter(MIB, CI.InstsToMove);
}
attributes #0 = { convergent nounwind }
+
+ define amdgpu_kernel void @merge_mmos() { ret void }
...
---
name: mem_dependency
%18:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM %9, 3, 0, 0 :: (dereferenceable invariant load 4)
S_ENDPGM 0
...
+---
+# CHECK-LABEL: merge_mmos
+# CHECK: S_BUFFER_LOAD_DWORDX2_IMM %0, 0, 0, 0 :: (dereferenceable invariant load 8, align 4)
+# CHECK: BUFFER_LOAD_DWORDX2_OFFSET %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8, align 4)
+# CHECK: BUFFER_STORE_DWORDX2_OFFSET_exact killed %{{[0-9]+}}, %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 8, align 4)
+name: merge_mmos
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $sgpr0_sgpr1_sgpr2_sgpr3
+
+ %0:sreg_128 = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+ %1:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM %0, 0, 0, 0 :: (dereferenceable invariant load 4)
+ %2:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM %0, 1, 0, 0 :: (dereferenceable invariant load 4)
+ %3:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4)
+ %4:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %0, 0, 4, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4)
+ BUFFER_STORE_DWORD_OFFSET_exact %3, %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4)
+ BUFFER_STORE_DWORD_OFFSET_exact %4, %0, 0, 4, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4)
+ S_ENDPGM 0
+
+...
}
...
-# CHECK: BUFFER_STORE_DWORDX2_OFFSET killed %{{[0-9]+}}, %{{[0-9]+}}, 0, 4, 0, 0, 0, 0, implicit $exec :: (store 4 into %ir.out.gep.1, addrspace 1)
+# CHECK: BUFFER_STORE_DWORDX2_OFFSET killed %{{[0-9]+}}, %{{[0-9]+}}, 0, 4, 0, 0, 0, 0, implicit $exec :: (store 8 into %ir.out.gep.1, align 4, addrspace 1)
---
name: test1
liveins:
S_ENDPGM 0
...
-# CHECK: BUFFER_STORE_DWORDX2_OFFSET killed %{{[0-9]+}}, %{{[0-9]+}}, 0, 4, 0, 0, 0, 1, implicit $exec :: (store 4 into %ir.out.gep.1, addrspace 1)
+# CHECK: BUFFER_STORE_DWORDX2_OFFSET killed %{{[0-9]+}}, %{{[0-9]+}}, 0, 4, 0, 0, 0, 1, implicit $exec :: (store 8 into %ir.out.gep.1, align 4, addrspace 1)
---
name: test4
liveins: