[codegen,amdgpu] Enhance MIR DIE and re-arrange it for AMDGPU.

[hliao]

Summary

• dead-mi-elimination assumes MIR in the SSA form and cannot be arranged after phi elimination or DeSSA. It's enhanced to handle the dead register definition by skipping use check on it. Once a register def is dead, all its uses, if any, should be undef.
• Re-arrange the DIE in RA phase for AMDGPU by placing it directly after detect-dead-lanes.
• Many relevant tests are refined due to different register assignment.

[hliao]

llvm/test/CodeGen/AMDGPU/dead-machine-elim-after-dead-lane.ll

2

This is the new test exposing the previous issue due to running MIR DIE after DeSSA. It results in illegal MIR.

[rampitec]

You have skipped the dead MO, but was pass reordering really necessary? It seems we have higher register pressure with this change.

llvm/test/CodeGen/AMDGPU/dead-mi-use-same-intr.mir

0

The test became useless. The point was to check def of the same reg as use in the same instruction past SSA.

llvm/test/CodeGen/AMDGPU/spill-vgpr-to-agpr.ll

249

That seem to affect register pressure quite badly. Register numbering in other places suggests other places are negatively affected too.

[arsenm]

I'm not sure this logic works in the presence of an undef subregister def

llvm/lib/CodeGen/DeadMachineInstructionElim.cpp

87

Typo regsiter

[hliao]

llvm/test/CodeGen/AMDGPU/dead-mi-use-same-intr.mir

0

dead-mi-elimination only works for SSA form. If we really need a DIE for non-SSA MIR, we need to check dead def with live range.

llvm/test/CodeGen/AMDGPU/spill-vgpr-to-agpr.ll

249

I have checked what causes the register pressure, yet. The changes in other places show register pressure is increased at the beginning but released earlier than the previous assignment. It's difficult to say which is good or bad. But, in general, except this spilling case, all other cases have no extra register requirements.

[hliao]

In D72709#1820225, @arsenm wrote:

I'm not sure this logic works in the presence of an undef subregister def

In SSA form, there's no subreg def.

[hliao]

In D72709#1820182, @rampitec wrote:

You have skipped the dead MO, but was pass reordering really necessary? It seems we have higher register pressure with this change.

The issue is that DIE on MIR assumes SSA form MIR. It's incorrect to run it after DeSSA. It results in invalid IR. I thought we should fix this first and look into the fix for the higher register pressure issue here.

[rampitec]

llvm/test/CodeGen/AMDGPU/dead-mi-use-same-intr.mir

0

Still the test became useless and needs to be removed with this change.

[hliao]

revise after review comments.

[hliao]

revise after review comments

[arsenm]

dead-mi-elimination should set SSA in getRequiredProperties if it really requires SSA

[rampitec]

LGTM

[uabelho]

Hi,

In my out-of-tree target we are running DeadMachineInstructionElim after SSA, and now of course the new assert fails.

But you're saying that it needs SSA or it can cause miscompiles?

[hliao]

yeah, here one example

//
//    a: X:sub1 := ..
//    b: Y      := use(.., X, ..)
//    c: X:sub0 := ..
//       .. no use of neither X nor Y ..
//

when DIE traverses MIR in the reverse order, it won't remove (c) as X is still used by (b). But, when (b) is checked, as Y is not used, it's removed. Then, (a) will be removed since now there's no more use of X. However, after that, only (c) is left. but it is not full definition of X and there's no other definition dominating this def (in fact, a partial def follow a implicit use.)

Even without sub register, similar cases results in the invalid MIR as well, especially considering the PHI node in loop. If you have to DIE in RA phase, you may consider put it before phi-elimination. In general, after DeSSA, only COPYs should be removed and no other should be removed without carefully live range checking.

>>! In D72709#1823444, @uabelho wrote:
> Hi,
> 
> In my out-of-tree target we are running DeadMachineInstructionElim after SSA, and now of course the new assert fails.
> 
> But you're saying that it needs SSA or it can cause miscompiles?

llvm/lib/CodeGen/DeadMachineInstructionElim.cpp

for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     const MachineOperand &MO = MI->getOperand(i);
     if (MO.isReg() && MO.isDef()) {
       Register Reg = MO.getReg();
       if (Register::isPhysicalRegister(Reg)) {
         // Don't delete live physreg defs, or any reserved register defs.
         if (LivePhysRegs.test(Reg) || MRI->isReserved(Reg))
           return false;
       } else {
+        if (MO.isDead()) {
+#ifndef NDEBUG
+          // Sanity check on uses of this dead register. All of them should be
+          // 'undef'.
+          for (auto &U : MRI->use_nodbg_operands(Reg))
+            assert(U.isUndef() && "'Undef' use on a 'dead' register is found!");
+#endif
+          continue;
+        }
         for (const MachineInstr &Use : MRI->use_nodbg_instructions(Reg)) {
           if (&Use != MI)
             // This def has a non-debug use. Don't delete the instruction!
             return false;
         }
       }
     }
   }

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

void GCNPassConfig::addOptimizedRegAlloc() {
   // TwoAddressInstructions, otherwise the processing of the tied operand of
   // SI_ELSE will introduce a copy of the tied operand source after the else.
   insertPass(&PHIEliminationID, &SILowerControlFlowID, false);
 
   // This must be run just after RegisterCoalescing.
   insertPass(&RegisterCoalescerID, &SIPreAllocateWWMRegsID, false);
 
   if (EnableDCEInRA)
-    insertPass(&RenameIndependentSubregsID, &DeadMachineInstructionElimID);
+    insertPass(&DetectDeadLanesID, &DeadMachineInstructionElimID);
 
   TargetPassConfig::addOptimizedRegAlloc();
 }
 
 bool GCNPassConfig::addPreRewrite() {
   if (EnableRegReassign) {
     addPass(&GCNNSAReassignID);
     addPass(&GCNRegBankReassignID);

llvm/test/CodeGen/AMDGPU/atomic_optimizations_local_pointer.ll

 ; GFX9-NEXT:    s_mov_b32 s2, -1
 ; GFX9-NEXT:    s_nop 2
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX1064-LABEL: add_i64_constant:
 ; GFX1064:       ; %bb.0: ; %entry
-; GFX1064-NEXT:    v_cmp_ne_u32_e64 s[2:3], 1, 0
+; GFX1064-NEXT:    v_cmp_ne_u32_e64 s[4:5], 1, 0
 ; GFX1064-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX1064-NEXT:    ; implicit-def: $vgpr1_vgpr2
-; GFX1064-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s2, 0
-; GFX1064-NEXT:    v_mbcnt_hi_u32_b32_e64 v0, s3, v0
+; GFX1064-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s4, 0
+; GFX1064-NEXT:    v_mbcnt_hi_u32_b32_e64 v0, s5, v0
 ; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
-; GFX1064-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX1064-NEXT:    ; mask branch BB5_2
 ; GFX1064-NEXT:    s_cbranch_execz BB5_2
 ; GFX1064-NEXT:  BB5_1:
-; GFX1064-NEXT:    s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
 ; GFX1064-NEXT:    v_mov_b32_e32 v3, local_var64@abs32@lo
-; GFX1064-NEXT:    v_mul_hi_u32_u24_e64 v2, s2, 5
-; GFX1064-NEXT:    v_mul_u32_u24_e64 v1, s2, 5
+; GFX1064-NEXT:    v_mul_hi_u32_u24_e64 v2, s4, 5
+; GFX1064-NEXT:    v_mul_u32_u24_e64 v1, s4, 5
 ; GFX1064-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX1064-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX1064-NEXT:    ds_add_rtn_u64 v[1:2], v3, v[1:2]
 ; GFX1064-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX1064-NEXT:    buffer_gl0_inv
 ; GFX1064-NEXT:    buffer_gl1_inv
 ; GFX1064-NEXT:  BB5_2:
 ; GFX1064-NEXT:    v_nop
-; GFX1064-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[2:3]
 ; GFX1064-NEXT:    v_readfirstlane_b32 s2, v1
 ; GFX1064-NEXT:    v_readfirstlane_b32 s3, v2
 ; GFX1064-NEXT:    v_mad_u64_u32 v[0:1], s[2:3], v0, 5, s[2:3]
 ; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000
 ; GFX1064-NEXT:    s_mov_b32 s2, -1
 ; GFX1064-NEXT:    s_nop 2
 ; GFX1064-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1064-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0

 ; GFX9-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX9-NEXT:    s_mov_b32 s2, -1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX1064-LABEL: sub_i64_constant:
 ; GFX1064:       ; %bb.0: ; %entry
-; GFX1064-NEXT:    v_cmp_ne_u32_e64 s[2:3], 1, 0
+; GFX1064-NEXT:    v_cmp_ne_u32_e64 s[4:5], 1, 0
 ; GFX1064-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX1064-NEXT:    ; implicit-def: $vgpr1_vgpr2
-; GFX1064-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s2, 0
-; GFX1064-NEXT:    v_mbcnt_hi_u32_b32_e64 v0, s3, v0
+; GFX1064-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s4, 0
+; GFX1064-NEXT:    v_mbcnt_hi_u32_b32_e64 v0, s5, v0
 ; GFX1064-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
-; GFX1064-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; GFX1064-NEXT:    s_and_saveexec_b64 s[2:3], vcc
 ; GFX1064-NEXT:    ; mask branch BB11_2
 ; GFX1064-NEXT:    s_cbranch_execz BB11_2
 ; GFX1064-NEXT:  BB11_1:
-; GFX1064-NEXT:    s_bcnt1_i32_b64 s2, s[2:3]
+; GFX1064-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
 ; GFX1064-NEXT:    v_mov_b32_e32 v3, local_var64@abs32@lo
-; GFX1064-NEXT:    v_mul_hi_u32_u24_e64 v2, s2, 5
-; GFX1064-NEXT:    v_mul_u32_u24_e64 v1, s2, 5
+; GFX1064-NEXT:    v_mul_hi_u32_u24_e64 v2, s4, 5
+; GFX1064-NEXT:    v_mul_u32_u24_e64 v1, s4, 5
 ; GFX1064-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX1064-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX1064-NEXT:    ds_sub_rtn_u64 v[1:2], v3, v[1:2]
 ; GFX1064-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX1064-NEXT:    buffer_gl0_inv
 ; GFX1064-NEXT:    buffer_gl1_inv
 ; GFX1064-NEXT:  BB11_2:
 ; GFX1064-NEXT:    v_nop
-; GFX1064-NEXT:    s_or_b64 exec, exec, s[4:5]
+; GFX1064-NEXT:    s_or_b64 exec, exec, s[2:3]
 ; GFX1064-NEXT:    v_readfirstlane_b32 s2, v1
 ; GFX1064-NEXT:    v_mul_u32_u24_e32 v1, 5, v0
 ; GFX1064-NEXT:    v_readfirstlane_b32 s3, v2
 ; GFX1064-NEXT:    v_mul_hi_u32_u24_e32 v2, 5, v0
 ; GFX1064-NEXT:    v_sub_co_u32_e64 v0, vcc, s2, v1
 ; GFX1064-NEXT:    s_mov_b32 s2, -1
 ; GFX1064-NEXT:    v_sub_co_ci_u32_e32 v1, vcc, s3, v2, vcc
 ; GFX1064-NEXT:    s_mov_b32 s3, 0x31016000

llvm/test/CodeGen/AMDGPU/bswap.ll

 ; SI-NEXT:    v_bfi_b32 v5, s12, v14, v13
 ; SI-NEXT:    v_bfi_b32 v4, s12, v16, v15
 ; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[8:11], 0 offset:16
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: test_bswap_v8i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; VI-NEXT:    s_mov_b32 s12, 0xff00ff
-; VI-NEXT:    s_mov_b32 s11, 0xf000
-; VI-NEXT:    s_mov_b32 s10, -1
+; VI-NEXT:    s_mov_b32 s3, 0xf000
+; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_mov_b32 s8, s0
-; VI-NEXT:    s_mov_b32 s9, s1
-; VI-NEXT:    s_load_dwordx8 s[0:7], s[2:3], 0x0
+; VI-NEXT:    s_mov_b32 s0, s4
+; VI-NEXT:    s_mov_b32 s1, s5
+; VI-NEXT:    s_load_dwordx8 s[4:11], s[6:7], 0x0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_alignbit_b32 v0, s3, s3, 8
-; VI-NEXT:    v_alignbit_b32 v1, s3, s3, 24
+; VI-NEXT:    v_alignbit_b32 v0, s7, s7, 8
+; VI-NEXT:    v_alignbit_b32 v1, s7, s7, 24
 ; VI-NEXT:    v_bfi_b32 v3, s12, v1, v0
-; VI-NEXT:    v_alignbit_b32 v2, s2, s2, 8
-; VI-NEXT:    v_alignbit_b32 v4, s2, s2, 24
-; VI-NEXT:    v_alignbit_b32 v0, s1, s1, 8
-; VI-NEXT:    v_alignbit_b32 v1, s1, s1, 24
+; VI-NEXT:    v_alignbit_b32 v2, s6, s6, 8
+; VI-NEXT:    v_alignbit_b32 v4, s6, s6, 24
+; VI-NEXT:    v_alignbit_b32 v0, s5, s5, 8
+; VI-NEXT:    v_alignbit_b32 v1, s5, s5, 24
 ; VI-NEXT:    v_bfi_b32 v2, s12, v4, v2
 ; VI-NEXT:    v_bfi_b32 v1, s12, v1, v0
-; VI-NEXT:    v_alignbit_b32 v0, s0, s0, 8
-; VI-NEXT:    v_alignbit_b32 v4, s0, s0, 24
+; VI-NEXT:    v_alignbit_b32 v0, s4, s4, 8
+; VI-NEXT:    v_alignbit_b32 v4, s4, s4, 24
 ; VI-NEXT:    v_bfi_b32 v0, s12, v4, v0
-; VI-NEXT:    v_alignbit_b32 v4, s7, s7, 8
-; VI-NEXT:    v_alignbit_b32 v5, s7, s7, 24
+; VI-NEXT:    v_alignbit_b32 v4, s11, s11, 8
+; VI-NEXT:    v_alignbit_b32 v5, s11, s11, 24
 ; VI-NEXT:    v_bfi_b32 v7, s12, v5, v4
-; VI-NEXT:    v_alignbit_b32 v4, s6, s6, 8
-; VI-NEXT:    v_alignbit_b32 v5, s6, s6, 24
+; VI-NEXT:    v_alignbit_b32 v4, s10, s10, 8
+; VI-NEXT:    v_alignbit_b32 v5, s10, s10, 24
 ; VI-NEXT:    v_bfi_b32 v6, s12, v5, v4
-; VI-NEXT:    v_alignbit_b32 v4, s5, s5, 8
-; VI-NEXT:    v_alignbit_b32 v5, s5, s5, 24
+; VI-NEXT:    v_alignbit_b32 v4, s9, s9, 8
+; VI-NEXT:    v_alignbit_b32 v5, s9, s9, 24
 ; VI-NEXT:    v_bfi_b32 v5, s12, v5, v4
-; VI-NEXT:    v_alignbit_b32 v4, s4, s4, 8
-; VI-NEXT:    v_alignbit_b32 v8, s4, s4, 24
+; VI-NEXT:    v_alignbit_b32 v4, s8, s8, 8
+; VI-NEXT:    v_alignbit_b32 v8, s8, s8, 24
 ; VI-NEXT:    v_bfi_b32 v4, s12, v8, v4
-; VI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[8:11], 0 offset:16
-; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; VI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
+; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
   %val = load <8 x i32>, <8 x i32> addrspace(1)* %in, align 32
   %bswap = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %val) nounwind readnone
   store <8 x i32> %bswap, <8 x i32> addrspace(1)* %out, align 32
   ret void
 }
 
 define amdgpu_kernel void @test_bswap_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %in) nounwind {
 ; SI-LABEL: test_bswap_i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s3, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
-; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_load_dwordx2 s[6:7], s[6:7], 0x0
+; SI-NEXT:    s_mov_b32 s2, -1
 ; SI-NEXT:    s_mov_b32 s19, 0xff0000
 ; SI-NEXT:    s_mov_b32 s9, 0
 ; SI-NEXT:    s_mov_b32 s15, 0xff00
 ; SI-NEXT:    s_mov_b32 s11, s9
 ; SI-NEXT:    s_mov_b32 s12, s9
 ; SI-NEXT:    s_mov_b32 s14, s9
 ; SI-NEXT:    s_mov_b32 s16, s9
 ; SI-NEXT:    s_mov_b32 s18, s9
-; SI-NEXT:    s_mov_b32 s4, s0
-; SI-NEXT:    s_mov_b32 s5, s1
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    v_mov_b32_e32 v0, s2
-; SI-NEXT:    v_alignbit_b32 v1, s3, v0, 24
-; SI-NEXT:    v_alignbit_b32 v0, s3, v0, 8
-; SI-NEXT:    s_lshr_b32 s8, s3, 24
-; SI-NEXT:    s_lshr_b32 s10, s3, 8
-; SI-NEXT:    s_lshl_b64 s[0:1], s[2:3], 8
-; SI-NEXT:    s_lshl_b64 s[20:21], s[2:3], 24
-; SI-NEXT:    s_lshl_b32 s17, s2, 24
-; SI-NEXT:    s_lshl_b32 s0, s2, 8
+; SI-NEXT:    v_mov_b32_e32 v0, s6
+; SI-NEXT:    v_alignbit_b32 v1, s7, v0, 24
+; SI-NEXT:    v_alignbit_b32 v0, s7, v0, 8
+; SI-NEXT:    s_lshr_b32 s8, s7, 24
+; SI-NEXT:    s_lshr_b32 s10, s7, 8
+; SI-NEXT:    s_lshl_b64 s[4:5], s[6:7], 8
+; SI-NEXT:    s_lshl_b64 s[20:21], s[6:7], 24
+; SI-NEXT:    s_lshl_b32 s17, s6, 24
+; SI-NEXT:    s_lshl_b32 s4, s6, 8
 ; SI-NEXT:    v_and_b32_e32 v1, s19, v1
 ; SI-NEXT:    v_and_b32_e32 v0, 0xff000000, v0
 ; SI-NEXT:    s_and_b32 s10, s10, s15
-; SI-NEXT:    s_and_b32 s13, s1, 0xff
+; SI-NEXT:    s_and_b32 s13, s5, 0xff
 ; SI-NEXT:    s_and_b32 s15, s21, s15
-; SI-NEXT:    s_and_b32 s19, s0, s19
+; SI-NEXT:    s_and_b32 s19, s4, s19
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v1
-; SI-NEXT:    s_or_b64 s[0:1], s[10:11], s[8:9]
-; SI-NEXT:    s_or_b64 s[2:3], s[14:15], s[12:13]
+; SI-NEXT:    s_or_b64 s[4:5], s[10:11], s[8:9]
+; SI-NEXT:    s_or_b64 s[6:7], s[14:15], s[12:13]
 ; SI-NEXT:    s_or_b64 s[8:9], s[16:17], s[18:19]
-; SI-NEXT:    v_or_b32_e32 v0, s0, v0
-; SI-NEXT:    v_mov_b32_e32 v1, s1
-; SI-NEXT:    s_or_b64 s[0:1], s[8:9], s[2:3]
-; SI-NEXT:    v_or_b32_e32 v0, s0, v0
-; SI-NEXT:    v_or_b32_e32 v1, s1, v1
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; SI-NEXT:    v_or_b32_e32 v0, s4, v0
+; SI-NEXT:    v_mov_b32_e32 v1, s5
+; SI-NEXT:    s_or_b64 s[4:5], s[8:9], s[6:7]
+; SI-NEXT:    v_or_b32_e32 v0, s4, v0
+; SI-NEXT:    v_or_b32_e32 v1, s5, v1
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: test_bswap_i64:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; VI-NEXT:    s_mov_b32 s12, 0xff0000
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_endpgm
   %bswap = call i64 @llvm.bswap.i64(i64 %val) nounwind readnone
   store i64 %bswap, i64 addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @test_bswap_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %in) nounwind {
 ; SI-LABEL: test_bswap_v2i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s3, 0xf000
 ; SI-NEXT:    s_mov_b32 s2, -1
 ; SI-NEXT:    s_mov_b32 s31, 0xff0000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_load_dwordx4 s[4:7], s[10:11], 0x0
-; SI-NEXT:    s_mov_b32 s11, 0
+; SI-NEXT:    s_load_dwordx4 s[8:11], s[6:7], 0x0
+; SI-NEXT:    s_mov_b32 s7, 0
 ; SI-NEXT:    s_mov_b32 s22, 0xff000000
 ; SI-NEXT:    s_mov_b32 s27, 0xff00
 ; SI-NEXT:    s_movk_i32 s25, 0xff
-; SI-NEXT:    s_mov_b32 s13, s11
-; SI-NEXT:    s_mov_b32 s14, s11
-; SI-NEXT:    s_mov_b32 s16, s11
-; SI-NEXT:    s_mov_b32 s18, s11
-; SI-NEXT:    s_mov_b32 s20, s11
-; SI-NEXT:    s_mov_b32 s23, s11
-; SI-NEXT:    s_mov_b32 s24, s11
-; SI-NEXT:    s_mov_b32 s26, s11
-; SI-NEXT:    s_mov_b32 s28, s11
-; SI-NEXT:    s_mov_b32 s30, s11
-; SI-NEXT:    s_mov_b32 s0, s8
-; SI-NEXT:    s_mov_b32 s1, s9
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    v_mov_b32_e32 v0, s6
-; SI-NEXT:    v_alignbit_b32 v1, s7, v0, 24
-; SI-NEXT:    v_alignbit_b32 v0, s7, v0, 8
-; SI-NEXT:    s_lshr_b32 s10, s7, 24
-; SI-NEXT:    s_lshr_b32 s12, s7, 8
-; SI-NEXT:    s_lshl_b64 s[8:9], s[6:7], 8
-; SI-NEXT:    s_lshl_b64 s[32:33], s[6:7], 24
-; SI-NEXT:    s_lshl_b32 s19, s6, 24
-; SI-NEXT:    s_lshl_b32 s21, s6, 8
-; SI-NEXT:    v_mov_b32_e32 v2, s4
-; SI-NEXT:    v_alignbit_b32 v3, s5, v2, 24
-; SI-NEXT:    v_alignbit_b32 v2, s5, v2, 8
-; SI-NEXT:    s_lshr_b32 s32, s5, 8
-; SI-NEXT:    s_lshl_b64 s[6:7], s[4:5], 8
-; SI-NEXT:    s_and_b32 s15, s9, s25
-; SI-NEXT:    s_lshl_b64 s[8:9], s[4:5], 24
-; SI-NEXT:    s_lshl_b32 s29, s4, 24
-; SI-NEXT:    s_lshl_b32 s4, s4, 8
+; SI-NEXT:    s_mov_b32 s13, s7
+; SI-NEXT:    s_mov_b32 s14, s7
+; SI-NEXT:    s_mov_b32 s16, s7
+; SI-NEXT:    s_mov_b32 s18, s7
+; SI-NEXT:    s_mov_b32 s20, s7
+; SI-NEXT:    s_mov_b32 s23, s7
+; SI-NEXT:    s_mov_b32 s24, s7
+; SI-NEXT:    s_mov_b32 s26, s7
+; SI-NEXT:    s_mov_b32 s28, s7
+; SI-NEXT:    s_mov_b32 s30, s7
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    v_mov_b32_e32 v0, s10
+; SI-NEXT:    v_alignbit_b32 v1, s11, v0, 24
+; SI-NEXT:    v_alignbit_b32 v0, s11, v0, 8
+; SI-NEXT:    s_lshr_b32 s6, s11, 24
+; SI-NEXT:    s_lshr_b32 s12, s11, 8
+; SI-NEXT:    s_lshl_b64 s[4:5], s[10:11], 8
+; SI-NEXT:    s_lshl_b64 s[32:33], s[10:11], 24
+; SI-NEXT:    s_lshl_b32 s19, s10, 24
+; SI-NEXT:    s_lshl_b32 s21, s10, 8
+; SI-NEXT:    v_mov_b32_e32 v2, s8
+; SI-NEXT:    v_alignbit_b32 v3, s9, v2, 24
+; SI-NEXT:    v_alignbit_b32 v2, s9, v2, 8
+; SI-NEXT:    s_lshr_b32 s32, s9, 8
+; SI-NEXT:    s_lshl_b64 s[10:11], s[8:9], 8
+; SI-NEXT:    s_and_b32 s15, s5, s25
+; SI-NEXT:    s_lshl_b64 s[4:5], s[8:9], 24
+; SI-NEXT:    s_lshl_b32 s29, s8, 24
+; SI-NEXT:    s_lshl_b32 s4, s8, 8
 ; SI-NEXT:    v_and_b32_e32 v1, s31, v1
 ; SI-NEXT:    v_and_b32_e32 v0, s22, v0
 ; SI-NEXT:    s_and_b32 s12, s12, s27
 ; SI-NEXT:    s_and_b32 s17, s33, s27
 ; SI-NEXT:    s_and_b32 s21, s21, s31
 ; SI-NEXT:    v_and_b32_e32 v3, s31, v3
 ; SI-NEXT:    v_and_b32_e32 v2, s22, v2
 ; SI-NEXT:    s_and_b32 s22, s32, s27
-; SI-NEXT:    s_and_b32 s25, s7, s25
-; SI-NEXT:    s_and_b32 s27, s9, s27
+; SI-NEXT:    s_and_b32 s25, s11, s25
+; SI-NEXT:    s_and_b32 s27, s5, s27
 ; SI-NEXT:    s_and_b32 s31, s4, s31
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v1
-; SI-NEXT:    s_or_b64 s[6:7], s[12:13], s[10:11]
-; SI-NEXT:    s_or_b64 s[8:9], s[16:17], s[14:15]
+; SI-NEXT:    s_or_b64 s[4:5], s[12:13], s[6:7]
+; SI-NEXT:    s_or_b64 s[10:11], s[16:17], s[14:15]
 ; SI-NEXT:    s_or_b64 s[12:13], s[18:19], s[20:21]
 ; SI-NEXT:    v_or_b32_e32 v1, v2, v3
-; SI-NEXT:    s_lshr_b32 s10, s5, 24
-; SI-NEXT:    s_or_b64 s[4:5], s[26:27], s[24:25]
+; SI-NEXT:    s_lshr_b32 s6, s9, 24
+; SI-NEXT:    s_or_b64 s[8:9], s[26:27], s[24:25]
 ; SI-NEXT:    s_or_b64 s[14:15], s[28:29], s[30:31]
-; SI-NEXT:    v_or_b32_e32 v0, s6, v0
-; SI-NEXT:    v_mov_b32_e32 v3, s7
-; SI-NEXT:    s_or_b64 s[6:7], s[12:13], s[8:9]
-; SI-NEXT:    s_or_b64 s[8:9], s[22:23], s[10:11]
-; SI-NEXT:    s_or_b64 s[4:5], s[14:15], s[4:5]
-; SI-NEXT:    v_or_b32_e32 v2, s6, v0
-; SI-NEXT:    v_or_b32_e32 v3, s7, v3
-; SI-NEXT:    v_or_b32_e32 v0, s8, v1
-; SI-NEXT:    v_mov_b32_e32 v1, s9
 ; SI-NEXT:    v_or_b32_e32 v0, s4, v0
-; SI-NEXT:    v_or_b32_e32 v1, s5, v1
+; SI-NEXT:    v_mov_b32_e32 v3, s5
+; SI-NEXT:    s_or_b64 s[4:5], s[12:13], s[10:11]
+; SI-NEXT:    s_or_b64 s[6:7], s[22:23], s[6:7]
+; SI-NEXT:    s_or_b64 s[8:9], s[14:15], s[8:9]
+; SI-NEXT:    v_or_b32_e32 v2, s4, v0
+; SI-NEXT:    v_or_b32_e32 v3, s5, v3
+; SI-NEXT:    v_or_b32_e32 v0, s6, v1
+; SI-NEXT:    v_mov_b32_e32 v1, s7
+; SI-NEXT:    v_or_b32_e32 v0, s8, v0
+; SI-NEXT:    v_or_b32_e32 v1, s9, v1
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: test_bswap_v2i64:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; VI-NEXT:    s_mov_b32 s9, 0
 ; VI-NEXT:    s_mov_b32 s14, 0xff0000

llvm/test/CodeGen/AMDGPU/copy-illegal-type.ll

; VI-NEXT:    s_endpgm
   store <4 x i8> %add, <4 x i8> addrspace(1)* %out1, align 4
   ret void
 }
 
 ; FIXME: Need to handle non-uniform case for function below (load without gep).
 define amdgpu_kernel void @test_copy_v4i8_x2_extra_use(<4 x i8> addrspace(1)* %out0, <4 x i8> addrspace(1)* %out1, <4 x i8> addrspace(1)* %out2, <4 x i8> addrspace(1)* %in) nounwind {
 ; SI-LABEL: test_copy_v4i8_x2_extra_use:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s11, 0xf000
+; SI-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s3, 0xf000
 ; SI-NEXT:    s_mov_b32 s14, 0
-; SI-NEXT:    s_mov_b32 s15, s11
+; SI-NEXT:    s_mov_b32 s15, s3
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_mov_b64 s[12:13], s[6:7]
+; SI-NEXT:    s_mov_b64 s[12:13], s[10:11]
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    buffer_load_dword v0, v[0:1], s[12:15], 0 addr64
 ; SI-NEXT:    s_mov_b32 s16, 0xff00
 ; SI-NEXT:    s_movk_i32 s17, 0xff
-; SI-NEXT:    s_mov_b32 s10, -1
-; SI-NEXT:    s_mov_b32 s8, s4
-; SI-NEXT:    s_mov_b32 s9, s5
-; SI-NEXT:    s_mov_b32 s4, s2
-; SI-NEXT:    s_mov_b32 s5, s3
-; SI-NEXT:    s_mov_b32 s6, s10
-; SI-NEXT:    s_mov_b32 s7, s11
-; SI-NEXT:    s_mov_b32 s2, s10
-; SI-NEXT:    s_mov_b32 s3, s11
+; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s0, s8
+; SI-NEXT:    s_mov_b32 s1, s9
+; SI-NEXT:    s_mov_b32 s8, s6
+; SI-NEXT:    s_mov_b32 s9, s7
+; SI-NEXT:    s_mov_b32 s10, s2
+; SI-NEXT:    s_mov_b32 s11, s3
+; SI-NEXT:    s_mov_b32 s6, s2
+; SI-NEXT:    s_mov_b32 s7, s3
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_add_i32_e32 v3, vcc, 9, v0
 ; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; SI-NEXT:    v_and_b32_e32 v4, s16, v1
 ; SI-NEXT:    v_add_i32_e32 v1, vcc, 9, v1
 ; SI-NEXT:    v_and_b32_e32 v2, s16, v0
 ; SI-NEXT:    v_and_b32_e32 v3, s17, v3
 ; SI-NEXT:    v_or_b32_e32 v2, v2, v3
 ; SI-NEXT:    v_and_b32_e32 v1, s17, v1
 ; SI-NEXT:    v_add_i32_e32 v2, vcc, 0x900, v2
 ; SI-NEXT:    v_or_b32_e32 v1, v4, v1
 ; SI-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_or_b32_e32 v1, v1, v2
 ; SI-NEXT:    v_add_i32_e32 v1, vcc, 0x9000000, v1
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_dword v1, off, s[8:11], 0
 ; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
-; SI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
-; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: test_copy_v4i8_x2_extra_use:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; VI-NEXT:    s_movk_i32 s12, 0xff00
 ; VI-NEXT:    s_movk_i32 s13, 0xff

llvm/test/CodeGen/AMDGPU/dead-machine-elim-after-dead-lane.ll

+; RUN: llc -march=amdgcn -verify-machineinstrs %s -o - | FileCheck %s
+
+; CHECK-LABEL: foo
+; CHECK-NOT: BUFFER_LOAD_DWORDX2_OFFSET
+; After dead code elimination, that buffer load should be eliminated finally
+; after dead lane detection.
+define amdgpu_kernel void @foo() {
+entry:
+  switch i8 undef, label %foo.exit [
+    i8 4, label %sw.bb4
+    i8 10, label %sw.bb10
+  ]
+
+sw.bb4:
+  %x = load i64, i64 addrspace(1)* undef, align 8
+  %c = sitofp i64 %x to float
+  %v = insertelement <2 x float> <float undef, float 0.000000e+00>, float %c, i32 0
+  br label %foo.exit
+
+sw.bb10:
+  unreachable
+
+foo.exit:
+  %agg = phi <2 x float> [ %v, %sw.bb4 ], [ zeroinitializer, %entry ]
+  %s = extractelement <2 x float> %agg, i32 1
+  store float %s, float addrspace(1)* undef, align 4
+  ret void
+}

llvm/test/CodeGen/AMDGPU/dead-mi-use-same-intr.mir

-# RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs -run-pass dead-mi-elimination -o - %s | FileCheck -check-prefix=GCN %s
-
-# GCN-LABEL: name: dead_undef
-# GCN:      bb.0:
-# GCN-NEXT: S_ENDPGM 0
----
-name:            dead_undef
-tracksRegLiveness: true
-registers:
-  - { id: 0, class: vgpr_32 }
-  - { id: 1, class: vgpr_32 }
-  - { id: 2, class: vgpr_32 }
-body: |
-  bb.0:
-  %0 = IMPLICIT_DEF
-  %1 = IMPLICIT_DEF
-  dead %2:vgpr_32 = V_MAC_F32_e32 %0:vgpr_32, %1:vgpr_32, undef %2:vgpr_32, implicit $exec
-  S_ENDPGM 0
-
-# GCN-LABEL: name: dead_defined
-# GCN:      bb.0:
-# GCN-NEXT: S_ENDPGM 0
----
-name:            dead_defined
-tracksRegLiveness: true
-registers:
-  - { id: 0, class: vgpr_32 }
-  - { id: 1, class: vgpr_32 }
-  - { id: 2, class: vgpr_32 }
-body: |
-  bb.0:
-  %0 = IMPLICIT_DEF
-  %1 = IMPLICIT_DEF
-  %2 = IMPLICIT_DEF
-  dead %2:vgpr_32 = V_MAC_F32_e32 %0:vgpr_32, %1:vgpr_32, %2:vgpr_32, implicit $exec
-  S_ENDPGM 0
-
-# Probably this dead mac can be removed anyway.
-# GCN-LABEL: name: dead_def_live_use
-# GCN:      dead %2:vgpr_32 = V_MAC_F32_e32 %0, %1, %2, implicit $exec
----
-name:            dead_def_live_use
-tracksRegLiveness: true
-registers:
-  - { id: 0, class: vgpr_32 }
-  - { id: 1, class: vgpr_32 }
-  - { id: 2, class: vgpr_32 }
-body: |
-  bb.0:
-  %0 = IMPLICIT_DEF
-  %1 = IMPLICIT_DEF
-  %2 = IMPLICIT_DEF
-  GLOBAL_STORE_DWORD undef $vgpr0_vgpr1, %2, 0, 0, 0, 0, implicit $exec
-  dead %2:vgpr_32 = V_MAC_F32_e32 %0:vgpr_32, %1:vgpr_32, %2:vgpr_32, implicit $exec
-  S_ENDPGM 0

llvm/test/CodeGen/AMDGPU/idot8u.ll

entry:
   ret void
 }
 
 ; TODO: Remove the unnecessary instruction(that is zero-extending the
 ; 2nd MAD) to have the pattern-recognizer to kick in.
 define amdgpu_kernel void @udot8_acc16(<8 x i4> addrspace(1)* %src1,
 ; GFX7-LABEL: udot8_acc16:
 ; GFX7:       ; %bb.0: ; %entry
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
-; GFX7-NEXT:    s_mov_b32 s11, 0xf000
-; GFX7-NEXT:    s_mov_b32 s10, -1
-; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX7-NEXT:    s_load_dword s0, s[4:5], 0x0
-; GFX7-NEXT:    buffer_load_ushort v0, off, s[8:11], 0
-; GFX7-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[8:9], 0x0
+; GFX7-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
+; GFX7-NEXT:    s_load_dword s1, s[10:11], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_lshr_b32 s2, s0, 28
-; GFX7-NEXT:    s_bfe_u32 s4, s0, 0x40018
+; GFX7-NEXT:    s_bfe_u32 s8, s0, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s15, s1, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s16, s1, 0x40014
 ; GFX7-NEXT:    s_bfe_u32 s17, s1, 0x40010
 ; GFX7-NEXT:    s_bfe_u32 s18, s1, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s19, s1, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s20, s1, 0x40004
 ; GFX7-NEXT:    s_lshr_b32 s14, s1, 28
 ; GFX7-NEXT:    s_and_b32 s1, s1, 15
-; GFX7-NEXT:    s_bfe_u32 s5, s0, 0x40014
-; GFX7-NEXT:    s_bfe_u32 s6, s0, 0x40010
-; GFX7-NEXT:    s_bfe_u32 s7, s0, 0x4000c
+; GFX7-NEXT:    s_bfe_u32 s9, s0, 0x40014
+; GFX7-NEXT:    s_bfe_u32 s10, s0, 0x40010
+; GFX7-NEXT:    s_bfe_u32 s11, s0, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s12, s0, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s13, s0, 0x40004
 ; GFX7-NEXT:    s_and_b32 s0, s0, 15
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s20
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s19
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s18
 ; GFX7-NEXT:    v_mov_b32_e32 v5, s17
 ; GFX7-NEXT:    v_mov_b32_e32 v6, s16
 ; GFX7-NEXT:    v_mov_b32_e32 v7, s15
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s0, v1, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s13, v2, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s12, v3, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v4, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s6, v5, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s5, v6, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v7, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v4, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v5, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v6, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v7, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s14
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s2, v1, v0
-; GFX7-NEXT:    buffer_store_short v0, off, s[8:11], 0
+; GFX7-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: udot8_acc16:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
entry:
   ret void
 }
 
 ; TODO: Remove the unnecessary instruction(that is zero-extending the
 ; 2nd MAD) to have the pattern-recognizer to kick in.
 define amdgpu_kernel void @udot8_acc8(<8 x i4> addrspace(1)* %src1,
 ; GFX7-LABEL: udot8_acc8:
 ; GFX7:       ; %bb.0: ; %entry
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
-; GFX7-NEXT:    s_mov_b32 s11, 0xf000
-; GFX7-NEXT:    s_mov_b32 s10, -1
-; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX7-NEXT:    s_load_dword s0, s[4:5], 0x0
-; GFX7-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
-; GFX7-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[8:9], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
+; GFX7-NEXT:    s_load_dword s1, s[10:11], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_lshr_b32 s2, s0, 28
-; GFX7-NEXT:    s_bfe_u32 s4, s0, 0x40018
+; GFX7-NEXT:    s_bfe_u32 s8, s0, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s15, s1, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s16, s1, 0x40014
 ; GFX7-NEXT:    s_bfe_u32 s17, s1, 0x40010
 ; GFX7-NEXT:    s_bfe_u32 s18, s1, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s19, s1, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s20, s1, 0x40004
 ; GFX7-NEXT:    s_lshr_b32 s14, s1, 28
 ; GFX7-NEXT:    s_and_b32 s1, s1, 15
-; GFX7-NEXT:    s_bfe_u32 s5, s0, 0x40014
-; GFX7-NEXT:    s_bfe_u32 s6, s0, 0x40010
-; GFX7-NEXT:    s_bfe_u32 s7, s0, 0x4000c
+; GFX7-NEXT:    s_bfe_u32 s9, s0, 0x40014
+; GFX7-NEXT:    s_bfe_u32 s10, s0, 0x40010
+; GFX7-NEXT:    s_bfe_u32 s11, s0, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s12, s0, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s13, s0, 0x40004
 ; GFX7-NEXT:    s_and_b32 s0, s0, 15
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s20
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s19
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s18
 ; GFX7-NEXT:    v_mov_b32_e32 v5, s17
 ; GFX7-NEXT:    v_mov_b32_e32 v6, s16
 ; GFX7-NEXT:    v_mov_b32_e32 v7, s15
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s0, v1, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s13, v2, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s12, v3, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v4, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s6, v5, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s5, v6, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v7, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v4, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v5, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v6, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v7, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s14
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s2, v1, v0
-; GFX7-NEXT:    buffer_store_byte v0, off, s[8:11], 0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: udot8_acc8:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
entry:
   ret void
 }
 
 ; TODO: Remove the two unnecessary instructions(and+add after 2nd MAD)
 ; to have the pattern-recognizer to kick in.
 define amdgpu_kernel void @udot8_acc4(<8 x i4> addrspace(1)* %src1,
 ; GFX7-LABEL: udot8_acc4:
 ; GFX7:       ; %bb.0: ; %entry
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
-; GFX7-NEXT:    s_mov_b32 s11, 0xf000
-; GFX7-NEXT:    s_mov_b32 s10, -1
-; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX7-NEXT:    s_load_dword s0, s[4:5], 0x0
-; GFX7-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
-; GFX7-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[8:9], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
+; GFX7-NEXT:    s_load_dword s1, s[10:11], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_lshr_b32 s2, s0, 28
-; GFX7-NEXT:    s_bfe_u32 s4, s0, 0x40018
+; GFX7-NEXT:    s_bfe_u32 s8, s0, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s15, s1, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s16, s1, 0x40014
 ; GFX7-NEXT:    s_bfe_u32 s17, s1, 0x40010
 ; GFX7-NEXT:    s_bfe_u32 s18, s1, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s19, s1, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s20, s1, 0x40004
 ; GFX7-NEXT:    s_lshr_b32 s14, s1, 28
 ; GFX7-NEXT:    s_and_b32 s1, s1, 15
-; GFX7-NEXT:    s_bfe_u32 s5, s0, 0x40014
-; GFX7-NEXT:    s_bfe_u32 s6, s0, 0x40010
-; GFX7-NEXT:    s_bfe_u32 s7, s0, 0x4000c
+; GFX7-NEXT:    s_bfe_u32 s9, s0, 0x40014
+; GFX7-NEXT:    s_bfe_u32 s10, s0, 0x40010
+; GFX7-NEXT:    s_bfe_u32 s11, s0, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s12, s0, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s13, s0, 0x40004
 ; GFX7-NEXT:    s_and_b32 s0, s0, 15
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s20
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s19
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s18
 ; GFX7-NEXT:    v_mov_b32_e32 v5, s17
 ; GFX7-NEXT:    v_mov_b32_e32 v6, s16
 ; GFX7-NEXT:    v_mov_b32_e32 v7, s15
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s0, v1, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s13, v2, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s12, v3, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v4, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s6, v5, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s5, v6, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v7, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v4, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v5, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v6, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v7, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s14
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s2, v1, v0
 ; GFX7-NEXT:    v_and_b32_e32 v0, 15, v0
-; GFX7-NEXT:    buffer_store_byte v0, off, s[8:11], 0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: udot8_acc4:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
entry:
   ret void
 }
 
 ; TODO: Currently, permutation of udot8 is turned off due to a huge increase
 ; in the compile time.
 define amdgpu_kernel void @udot8_CommutationInsideMAD(<8 x i4> addrspace(1)* %src1,
 ; GFX7-LABEL: udot8_CommutationInsideMAD:
 ; GFX7:       ; %bb.0: ; %entry
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
-; GFX7-NEXT:    s_mov_b32 s11, 0xf000
-; GFX7-NEXT:    s_mov_b32 s10, -1
-; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX7-NEXT:    s_load_dword s0, s[4:5], 0x0
-; GFX7-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
-; GFX7-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[8:9], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
+; GFX7-NEXT:    s_load_dword s1, s[10:11], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_lshr_b32 s2, s0, 28
-; GFX7-NEXT:    s_bfe_u32 s4, s0, 0x40018
+; GFX7-NEXT:    s_bfe_u32 s8, s0, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s15, s1, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s16, s1, 0x40014
 ; GFX7-NEXT:    s_bfe_u32 s17, s1, 0x40010
 ; GFX7-NEXT:    s_bfe_u32 s18, s1, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s19, s1, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s20, s1, 0x40004
 ; GFX7-NEXT:    s_lshr_b32 s14, s1, 28
 ; GFX7-NEXT:    s_and_b32 s1, s1, 15
-; GFX7-NEXT:    s_bfe_u32 s5, s0, 0x40014
-; GFX7-NEXT:    s_bfe_u32 s6, s0, 0x40010
-; GFX7-NEXT:    s_bfe_u32 s7, s0, 0x4000c
+; GFX7-NEXT:    s_bfe_u32 s9, s0, 0x40014
+; GFX7-NEXT:    s_bfe_u32 s10, s0, 0x40010
+; GFX7-NEXT:    s_bfe_u32 s11, s0, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s12, s0, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s13, s0, 0x40004
 ; GFX7-NEXT:    s_and_b32 s0, s0, 15
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s20
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s19
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s18
 ; GFX7-NEXT:    v_mov_b32_e32 v5, s17
 ; GFX7-NEXT:    v_mov_b32_e32 v6, s16
 ; GFX7-NEXT:    v_mov_b32_e32 v7, s15
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s0, v1, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s13, v2, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s12, v3, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v4, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s6, v5, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s5, v6, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v7, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v4, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v5, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v6, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v7, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s14
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s2, v1, v0
 ; GFX7-NEXT:    v_and_b32_e32 v0, 15, v0
-; GFX7-NEXT:    buffer_store_byte v0, off, s[8:11], 0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: udot8_CommutationInsideMAD:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
entry:
   store i8 %add8, i8 addrspace(1)* %dst, align 4
   ret void
 }
 
 ; TODO: Once the adictional "and+add" are removed, the pattern will be recognized.
 define amdgpu_kernel void @udot8_acc4_vecMul(<8 x i4> addrspace(1)* %src1,
 ; GFX7-LABEL: udot8_acc4_vecMul:
 ; GFX7:       ; %bb.0: ; %entry
-; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
-; GFX7-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
-; GFX7-NEXT:    s_mov_b32 s11, 0xf000
-; GFX7-NEXT:    s_mov_b32 s10, -1
-; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX7-NEXT:    s_load_dword s0, s[4:5], 0x0
-; GFX7-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
-; GFX7-NEXT:    s_load_dword s1, s[6:7], 0x0
+; GFX7-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
+; GFX7-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
+; GFX7-NEXT:    s_mov_b32 s7, 0xf000
+; GFX7-NEXT:    s_mov_b32 s6, -1
+; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX7-NEXT:    s_load_dword s0, s[8:9], 0x0
+; GFX7-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
+; GFX7-NEXT:    s_load_dword s1, s[10:11], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    s_lshr_b32 s2, s0, 28
-; GFX7-NEXT:    s_bfe_u32 s4, s0, 0x40018
+; GFX7-NEXT:    s_bfe_u32 s8, s0, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s15, s1, 0x40018
 ; GFX7-NEXT:    s_bfe_u32 s16, s1, 0x40014
 ; GFX7-NEXT:    s_bfe_u32 s17, s1, 0x40010
 ; GFX7-NEXT:    s_bfe_u32 s18, s1, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s19, s1, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s20, s1, 0x40004
 ; GFX7-NEXT:    s_lshr_b32 s14, s1, 28
 ; GFX7-NEXT:    s_and_b32 s1, s1, 15
-; GFX7-NEXT:    s_bfe_u32 s5, s0, 0x40014
-; GFX7-NEXT:    s_bfe_u32 s6, s0, 0x40010
-; GFX7-NEXT:    s_bfe_u32 s7, s0, 0x4000c
+; GFX7-NEXT:    s_bfe_u32 s9, s0, 0x40014
+; GFX7-NEXT:    s_bfe_u32 s10, s0, 0x40010
+; GFX7-NEXT:    s_bfe_u32 s11, s0, 0x4000c
 ; GFX7-NEXT:    s_bfe_u32 s12, s0, 0x40008
 ; GFX7-NEXT:    s_bfe_u32 s13, s0, 0x40004
 ; GFX7-NEXT:    s_and_b32 s0, s0, 15
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX7-NEXT:    v_mov_b32_e32 v2, s20
 ; GFX7-NEXT:    v_mov_b32_e32 v3, s19
 ; GFX7-NEXT:    v_mov_b32_e32 v4, s18
 ; GFX7-NEXT:    v_mov_b32_e32 v5, s17
 ; GFX7-NEXT:    v_mov_b32_e32 v6, s16
 ; GFX7-NEXT:    v_mov_b32_e32 v7, s15
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s0, v1, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s13, v2, v0
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s12, v3, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v4, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s6, v5, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s5, v6, v0
-; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v7, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v4, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v5, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v6, v0
+; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v7, v0
 ; GFX7-NEXT:    v_mov_b32_e32 v1, s14
 ; GFX7-NEXT:    v_mad_u32_u24 v0, s2, v1, v0
 ; GFX7-NEXT:    v_and_b32_e32 v0, 15, v0
-; GFX7-NEXT:    buffer_store_byte v0, off, s[8:11], 0
+; GFX7-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; GFX7-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: udot8_acc4_vecMul:
 ; GFX8:       ; %bb.0: ; %entry
 ; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0

llvm/test/CodeGen/AMDGPU/insert_vector_elt.ll

; VI-NEXT:    s_endpgm
   %vecins = insertelement <4 x double> %a, double 8.0, i32 %b
   store <4 x double> %vecins, <4 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @dynamic_insertelement_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %a, i32 %b) #0 {
 ; SI-LABEL: dynamic_insertelement_v8f64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[24:25], s[4:5], 0x0
-; SI-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x10
+; SI-NEXT:    s_load_dwordx2 s[8:9], s[4:5], 0x0
+; SI-NEXT:    s_load_dwordx16 s[12:27], s[4:5], 0x10
 ; SI-NEXT:    s_load_dword s4, s[4:5], 0x20
 ; SI-NEXT:    v_mov_b32_e32 v16, 64
-; SI-NEXT:    s_mov_b32 s27, 0x100f000
-; SI-NEXT:    s_mov_b32 s26, -1
+; SI-NEXT:    s_mov_b32 s11, 0x100f000
+; SI-NEXT:    s_mov_b32 s10, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    v_mov_b32_e32 v0, s8
+; SI-NEXT:    v_mov_b32_e32 v0, s12
 ; SI-NEXT:    s_and_b32 s4, s4, 7
 ; SI-NEXT:    s_lshl_b32 s4, s4, 3
-; SI-NEXT:    v_mov_b32_e32 v1, s9
-; SI-NEXT:    v_mov_b32_e32 v2, s10
-; SI-NEXT:    v_mov_b32_e32 v3, s11
-; SI-NEXT:    v_mov_b32_e32 v4, s12
-; SI-NEXT:    v_mov_b32_e32 v5, s13
-; SI-NEXT:    v_mov_b32_e32 v6, s14
-; SI-NEXT:    v_mov_b32_e32 v7, s15
-; SI-NEXT:    v_mov_b32_e32 v8, s16
-; SI-NEXT:    v_mov_b32_e32 v9, s17
-; SI-NEXT:    v_mov_b32_e32 v10, s18
-; SI-NEXT:    v_mov_b32_e32 v11, s19
-; SI-NEXT:    v_mov_b32_e32 v12, s20
-; SI-NEXT:    v_mov_b32_e32 v13, s21
-; SI-NEXT:    v_mov_b32_e32 v14, s22
-; SI-NEXT:    v_mov_b32_e32 v15, s23
+; SI-NEXT:    v_mov_b32_e32 v1, s13
+; SI-NEXT:    v_mov_b32_e32 v2, s14
+; SI-NEXT:    v_mov_b32_e32 v3, s15
+; SI-NEXT:    v_mov_b32_e32 v4, s16
+; SI-NEXT:    v_mov_b32_e32 v5, s17
+; SI-NEXT:    v_mov_b32_e32 v6, s18
+; SI-NEXT:    v_mov_b32_e32 v7, s19
+; SI-NEXT:    v_mov_b32_e32 v8, s20
+; SI-NEXT:    v_mov_b32_e32 v9, s21
+; SI-NEXT:    v_mov_b32_e32 v10, s22
+; SI-NEXT:    v_mov_b32_e32 v11, s23
+; SI-NEXT:    v_mov_b32_e32 v12, s24
+; SI-NEXT:    v_mov_b32_e32 v13, s25
+; SI-NEXT:    v_mov_b32_e32 v14, s26
+; SI-NEXT:    v_mov_b32_e32 v15, s27
 ; SI-NEXT:    buffer_store_dwordx4 v[12:15], off, s[0:3], s7 offset:112
 ; SI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[0:3], s7 offset:96
 ; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], s7 offset:80
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], s7 offset:64
 ; SI-NEXT:    v_or_b32_e32 v16, s4, v16
 ; SI-NEXT:    v_mov_b32_e32 v0, 0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0x40200000
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], v16, s[0:3], s7 offen
 ; SI-NEXT:    buffer_load_dwordx4 v[0:3], off, s[0:3], s7 offset:64
 ; SI-NEXT:    buffer_load_dwordx4 v[4:7], off, s[0:3], s7 offset:80
 ; SI-NEXT:    buffer_load_dwordx4 v[8:11], off, s[0:3], s7 offset:96
 ; SI-NEXT:    buffer_load_dwordx4 v[12:15], off, s[0:3], s7 offset:112
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[12:15], off, s[24:27], 0 offset:48
-; SI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[24:27], 0 offset:32
-; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[24:27], 0 offset:16
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[24:27], 0
+; SI-NEXT:    buffer_store_dwordx4 v[12:15], off, s[8:11], 0 offset:48
+; SI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[8:11], 0 offset:32
+; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[8:11], 0 offset:16
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: dynamic_insertelement_v8f64:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[24:25], s[4:5], 0x0
-; VI-NEXT:    s_load_dwordx16 s[8:23], s[4:5], 0x40
+; VI-NEXT:    s_load_dwordx2 s[8:9], s[4:5], 0x0
+; VI-NEXT:    s_load_dwordx16 s[12:27], s[4:5], 0x40
 ; VI-NEXT:    s_load_dword s4, s[4:5], 0x80
 ; VI-NEXT:    v_mov_b32_e32 v16, 64
-; VI-NEXT:    s_mov_b32 s27, 0x1100f000
-; VI-NEXT:    s_mov_b32 s26, -1
+; VI-NEXT:    s_mov_b32 s11, 0x1100f000
+; VI-NEXT:    s_mov_b32 s10, -1
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v0, s8
+; VI-NEXT:    v_mov_b32_e32 v0, s12
 ; VI-NEXT:    s_and_b32 s4, s4, 7
 ; VI-NEXT:    s_lshl_b32 s4, s4, 3
-; VI-NEXT:    v_mov_b32_e32 v1, s9
-; VI-NEXT:    v_mov_b32_e32 v2, s10
-; VI-NEXT:    v_mov_b32_e32 v3, s11
-; VI-NEXT:    v_mov_b32_e32 v4, s12
-; VI-NEXT:    v_mov_b32_e32 v5, s13
-; VI-NEXT:    v_mov_b32_e32 v6, s14
-; VI-NEXT:    v_mov_b32_e32 v7, s15
-; VI-NEXT:    v_mov_b32_e32 v8, s16
-; VI-NEXT:    v_mov_b32_e32 v9, s17
-; VI-NEXT:    v_mov_b32_e32 v10, s18
-; VI-NEXT:    v_mov_b32_e32 v11, s19
-; VI-NEXT:    v_mov_b32_e32 v12, s20
-; VI-NEXT:    v_mov_b32_e32 v13, s21
-; VI-NEXT:    v_mov_b32_e32 v14, s22
-; VI-NEXT:    v_mov_b32_e32 v15, s23
+; VI-NEXT:    v_mov_b32_e32 v1, s13
+; VI-NEXT:    v_mov_b32_e32 v2, s14
+; VI-NEXT:    v_mov_b32_e32 v3, s15
+; VI-NEXT:    v_mov_b32_e32 v4, s16
+; VI-NEXT:    v_mov_b32_e32 v5, s17
+; VI-NEXT:    v_mov_b32_e32 v6, s18
+; VI-NEXT:    v_mov_b32_e32 v7, s19
+; VI-NEXT:    v_mov_b32_e32 v8, s20
+; VI-NEXT:    v_mov_b32_e32 v9, s21
+; VI-NEXT:    v_mov_b32_e32 v10, s22
+; VI-NEXT:    v_mov_b32_e32 v11, s23
+; VI-NEXT:    v_mov_b32_e32 v12, s24
+; VI-NEXT:    v_mov_b32_e32 v13, s25
+; VI-NEXT:    v_mov_b32_e32 v14, s26
+; VI-NEXT:    v_mov_b32_e32 v15, s27
+; VI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[0:3], s7 offset:96
 ; VI-NEXT:    buffer_store_dwordx4 v[12:15], off, s[0:3], s7 offset:112
 ; VI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[0:3], s7 offset:96
 ; VI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], s7 offset:80
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], s7 offset:64
 ; VI-NEXT:    v_or_b32_e32 v16, s4, v16
 ; VI-NEXT:    v_mov_b32_e32 v0, 0
 ; VI-NEXT:    v_mov_b32_e32 v1, 0x40200000
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], v16, s[0:3], s7 offen
 ; VI-NEXT:    buffer_load_dwordx4 v[0:3], off, s[0:3], s7 offset:64
 ; VI-NEXT:    buffer_load_dwordx4 v[4:7], off, s[0:3], s7 offset:80
 ; VI-NEXT:    buffer_load_dwordx4 v[8:11], off, s[0:3], s7 offset:96
 ; VI-NEXT:    buffer_load_dwordx4 v[12:15], off, s[0:3], s7 offset:112
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    buffer_store_dwordx4 v[12:15], off, s[24:27], 0 offset:48
-; VI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[24:27], 0 offset:32
-; VI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[24:27], 0 offset:16
-; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[24:27], 0
+; VI-NEXT:    buffer_store_dwordx4 v[12:15], off, s[8:11], 0 offset:48
+; VI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[8:11], 0 offset:32
+; VI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[8:11], 0 offset:16
+; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
 ; VI-NEXT:    s_endpgm
   %vecins = insertelement <8 x double> %a, double 8.0, i32 %b
   store <8 x double> %vecins, <8 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 declare <4 x float> @llvm.amdgcn.image.gather4.lz.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readnone }

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.ds.ordered.swap.ll

 ; GCN-NEXT: s_waitcnt expcnt(0) lgkmcnt(0)
 define amdgpu_cs float @ds_ordered_swap(i32 addrspace(2)* inreg %gds, i32 %value) {
   %val = call [e-mail](i32 addrspace(2)* %gds, i32 %value, i32 0, i32 0, i1 false, i32 1, i1 true, i1 true)
   %r = bitcast i32 %val to float
   ret float %r
 }
 
 ; FUNC-LABEL: {{^}}ds_ordered_swap_conditional:
-; GCN: v_cmp_ne_u32_e32 vcc, 0, v0
+; GCN: v_cmp_ne_u32_e32 vcc, 0, v[[VALUE:[0-9]+]]
 ; GCN: s_and_saveexec_b64 s[[SAVED:\[[0-9]+:[0-9]+\]]], vcc
 ; // We have to use s_cbranch, because ds_ordered_count has side effects with EXEC=0
 ; GCN: s_cbranch_execz [[BB:BB._.]]
 ; GCN: s_mov_b32 m0, s0
 ; VIGFX9-NEXT: s_nop 0
-; GCN-NEXT: ds_ordered_count v{{[0-9]+}}, v0 offset:4868 gds
+; GCN-NEXT: ds_ordered_count v{{[0-9]+}}, v[[VALUE]] offset:4868 gds
 ; GCN-NEXT: [[BB]]:
 ; // Wait for expcnt(0) before modifying EXEC
 ; GCN-NEXT: s_waitcnt expcnt(0)
 ; GCN-NEXT: s_or_b64 exec, exec, s[[SAVED]]
 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
 define amdgpu_cs float @ds_ordered_swap_conditional(i32 addrspace(2)* inreg %gds, i32 %value) {
 entry:
   %c = icmp ne i32 %value, 0

llvm/test/CodeGen/AMDGPU/llvm.round.f64.ll

 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_cndmask_b32_e64 v4, v4, v8, s[0:1]
 ; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[4:7], 0 offset:16
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; CI-LABEL: round_v4f64:
 ; CI:       ; %bb.0:
-; CI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x9
-; CI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x11
-; CI-NEXT:    s_brev_b32 s12, -2
+; CI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; CI-NEXT:    s_load_dwordx8 s[8:15], s[0:1], 0x11
+; CI-NEXT:    s_brev_b32 s2, -2
 ; CI-NEXT:    v_mov_b32_e32 v12, 0x3ff00000
-; CI-NEXT:    s_mov_b32 s11, 0xf000
-; CI-NEXT:    s_mov_b32 s10, -1
+; CI-NEXT:    s_mov_b32 s7, 0xf000
+; CI-NEXT:    s_mov_b32 s6, -1
 ; CI-NEXT:    s_waitcnt lgkmcnt(0)
-; CI-NEXT:    v_trunc_f64_e32 v[0:1], s[2:3]
-; CI-NEXT:    v_mov_b32_e32 v4, s3
-; CI-NEXT:    v_add_f64 v[2:3], s[2:3], -v[0:1]
-; CI-NEXT:    v_bfi_b32 v4, s12, v12, v4
+; CI-NEXT:    v_trunc_f64_e32 v[0:1], s[10:11]
+; CI-NEXT:    v_mov_b32_e32 v4, s11
+; CI-NEXT:    v_add_f64 v[2:3], s[10:11], -v[0:1]
+; CI-NEXT:    v_bfi_b32 v4, s2, v12, v4
 ; CI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[2:3]|, 0.5
-; CI-NEXT:    v_trunc_f64_e32 v[8:9], s[0:1]
+; CI-NEXT:    v_trunc_f64_e32 v[8:9], s[8:9]
 ; CI-NEXT:    v_cndmask_b32_e32 v3, 0, v4, vcc
 ; CI-NEXT:    v_mov_b32_e32 v2, 0
 ; CI-NEXT:    v_add_f64 v[2:3], v[0:1], v[2:3]
-; CI-NEXT:    v_add_f64 v[0:1], s[0:1], -v[8:9]
-; CI-NEXT:    v_mov_b32_e32 v4, s1
+; CI-NEXT:    v_add_f64 v[0:1], s[8:9], -v[8:9]
+; CI-NEXT:    v_mov_b32_e32 v4, s9
 ; CI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[0:1]|, 0.5
-; CI-NEXT:    v_bfi_b32 v4, s12, v12, v4
+; CI-NEXT:    v_bfi_b32 v4, s2, v12, v4
 ; CI-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
-; CI-NEXT:    v_trunc_f64_e32 v[4:5], s[6:7]
-; CI-NEXT:    v_mov_b32_e32 v10, s7
-; CI-NEXT:    v_add_f64 v[6:7], s[6:7], -v[4:5]
-; CI-NEXT:    v_bfi_b32 v10, s12, v12, v10
+; CI-NEXT:    v_trunc_f64_e32 v[4:5], s[14:15]
+; CI-NEXT:    v_mov_b32_e32 v10, s15
+; CI-NEXT:    v_add_f64 v[6:7], s[14:15], -v[4:5]
+; CI-NEXT:    v_bfi_b32 v10, s2, v12, v10
 ; CI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[6:7]|, 0.5
 ; CI-NEXT:    v_mov_b32_e32 v6, 0
 ; CI-NEXT:    v_cndmask_b32_e32 v7, 0, v10, vcc
-; CI-NEXT:    v_trunc_f64_e32 v[10:11], s[4:5]
+; CI-NEXT:    v_trunc_f64_e32 v[10:11], s[12:13]
 ; CI-NEXT:    v_add_f64 v[6:7], v[4:5], v[6:7]
-; CI-NEXT:    v_add_f64 v[4:5], s[4:5], -v[10:11]
-; CI-NEXT:    v_mov_b32_e32 v13, s5
+; CI-NEXT:    v_add_f64 v[4:5], s[12:13], -v[10:11]
+; CI-NEXT:    v_mov_b32_e32 v13, s13
 ; CI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[4:5]|, 0.5
-; CI-NEXT:    v_bfi_b32 v12, s12, v12, v13
+; CI-NEXT:    v_bfi_b32 v12, s2, v12, v13
 ; CI-NEXT:    v_mov_b32_e32 v0, 0
 ; CI-NEXT:    v_cndmask_b32_e32 v5, 0, v12, vcc
 ; CI-NEXT:    v_mov_b32_e32 v4, 0
 ; CI-NEXT:    v_add_f64 v[4:5], v[10:11], v[4:5]
 ; CI-NEXT:    v_add_f64 v[0:1], v[8:9], v[0:1]
-; CI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[8:11], 0 offset:16
-; CI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; CI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[4:7], 0 offset:16
+; CI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; CI-NEXT:    s_endpgm
   %result = call <4 x double> @llvm.round.v4f64(<4 x double> %in) #1
   store <4 x double> %result, <4 x double> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @round_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %in) #0 {
 ; SI-LABEL: round_v8f64:

llvm/test/CodeGen/AMDGPU/loop_break.ll

 
 ; OPT: bb9:
 ; OPT: call void @llvm.amdgcn.end.cf.i64(i64
 
 ; GCN-LABEL: {{^}}break_loop:
 ; GCN:      s_mov_b64         [[ACCUM_MASK:s\[[0-9]+:[0-9]+\]]], 0{{$}}
 
 ; GCN: [[LOOP_ENTRY:BB[0-9]+_[0-9]+]]: ; %bb1
-; GCN:     s_add_i32 s4, s4, 1
+; GCN:     s_add_i32 s6, s6, 1
 ; GCN:     s_or_b64 [[INNER_MASK:s\[[0-9]+:[0-9]+\]]], [[INNER_MASK]], exec
-; GCN:     s_cmp_gt_i32 s4, -1
+; GCN:     s_cmp_gt_i32 s6, -1
 ; GCN:     s_cbranch_scc1   [[FLOW:BB[0-9]+_[0-9]+]]
 
 ; GCN: ; %bb4
 ; GCN:      buffer_load_dword
 ; GCN:      v_cmp_ge_i32_e32  vcc
 ; GCN:      s_andn2_b64 [[INNER_MASK]], [[INNER_MASK]], exec
 ; GCN:      s_and_b64 [[BROKEN_MASK:s\[[0-9]+:[0-9]+\]]], vcc, exec
 ; GCN:      s_or_b64  [[INNER_MASK]], [[INNER_MASK]], [[BROKEN_MASK]]

llvm/test/CodeGen/AMDGPU/select.f16.ll

entry:
   store half %r.val, half addrspace(1)* %r
   ret void
 }
 
 define amdgpu_kernel void @select_v2f16(
 ; SI-LABEL: select_v2f16:
 ; SI:       ; %bb.0: ; %entry
 ; SI-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x9
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x11
-; SI-NEXT:    s_mov_b32 s15, 0xf000
-; SI-NEXT:    s_mov_b32 s14, -1
-; SI-NEXT:    s_mov_b32 s22, s14
+; SI-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x11
+; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s22, s2
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    s_mov_b32 s16, s10
 ; SI-NEXT:    s_mov_b32 s17, s11
-; SI-NEXT:    s_mov_b32 s10, s14
-; SI-NEXT:    s_mov_b32 s11, s15
+; SI-NEXT:    s_mov_b32 s10, s2
+; SI-NEXT:    s_mov_b32 s11, s3
 ; SI-NEXT:    s_mov_b32 s20, s6
 ; SI-NEXT:    s_mov_b32 s21, s7
-; SI-NEXT:    s_mov_b32 s23, s15
-; SI-NEXT:    s_mov_b32 s2, s14
-; SI-NEXT:    s_mov_b32 s3, s15
+; SI-NEXT:    s_mov_b32 s23, s3
+; SI-NEXT:    s_mov_b32 s14, s2
+; SI-NEXT:    s_mov_b32 s15, s3
 ; SI-NEXT:    buffer_load_dword v0, off, s[20:23], 0
-; SI-NEXT:    s_mov_b32 s18, s14
-; SI-NEXT:    s_mov_b32 s19, s15
+; SI-NEXT:    s_mov_b32 s18, s2
+; SI-NEXT:    s_mov_b32 s19, s3
 ; SI-NEXT:    buffer_load_dword v1, off, s[8:11], 0
-; SI-NEXT:    buffer_load_dword v2, off, s[0:3], 0
+; SI-NEXT:    buffer_load_dword v2, off, s[12:15], 0
 ; SI-NEXT:    buffer_load_dword v3, off, s[16:19], 0
-; SI-NEXT:    s_mov_b32 s12, s4
-; SI-NEXT:    s_mov_b32 s13, s5
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
 ; SI-NEXT:    v_cvt_f32_f16_e32 v5, v5
 ; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_cvt_f32_f16_e32 v4, v2
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_lshrrev_b32_e32 v7, 16, v3
 ; SI-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
 ; SI-NEXT:    v_cvt_f32_f16_e32 v6, v6
 ; SI-NEXT:    v_cvt_f32_f16_e32 v2, v2
 ; SI-NEXT:    v_cvt_f32_f16_e32 v7, v7
 ; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
 ; SI-NEXT:    v_cvt_f32_f16_e32 v3, v3
 ; SI-NEXT:    v_cmp_lt_f32_e32 vcc, v5, v6
 ; SI-NEXT:    v_cndmask_b32_e32 v2, v2, v7, vcc
 ; SI-NEXT:    v_cmp_lt_f32_e32 vcc, v0, v1
 ; SI-NEXT:    v_cndmask_b32_e32 v0, v4, v3, vcc
 ; SI-NEXT:    v_cvt_f16_f32_e32 v2, v2
 ; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v2
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v1
-; SI-NEXT:    buffer_store_dword v0, off, s[12:15], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: select_v2f16:
 ; VI:       ; %bb.0: ; %entry
 ; VI-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x24
 ; VI-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x44
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1

llvm/test/CodeGen/AMDGPU/spill-vgpr-to-agpr.ll

 ; GFX908-FIXME-NOT: SCRATCH_RSRC
 ; GFX908-DAG: v_accvgpr_write_b32 a0, v
 ; GFX900:     buffer_store_dword v
 ; GFX900:     buffer_load_dword v
 ; GFX908-FIXME-NOT: buffer_
 ; GFX908-DAG  v_accvgpr_read_b32
 
 ; GCN:    NumVgprs: 256
-; GFX900: ScratchSize: 580
+; GFX900: ScratchSize: 644
 ; GFX908-FIXME: ScratchSize: 0
 ; GCN:    VGPRBlocks: 63
 ; GCN:    NumVGPRsForWavesPerEU: 256
 define amdgpu_kernel void @max_256_vgprs_spill_9x32_2bb(<32 x float> addrspace(1)* %p) {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %p1 = getelementptr inbounds <32 x float>, <32 x float> addrspace(1)* %p, i32 %tid
   %p2 = getelementptr inbounds <32 x float>, <32 x float> addrspace(1)* %p1, i32 %tid
   %p3 = getelementptr inbounds <32 x float>, <32 x float> addrspace(1)* %p2, i32 %tid