AMDGPU: Make hazard recognizer aware of maximum clause sizes

[arsenm]

Summary

Supposedly clauses can be formed up until vmcnt/lgkmcnt are
saturated.

The limit for vmcnt on gfx9 is pretty big, so this requires
looking ahead 64 instructions. I don't know how much of an
impact this has on compile time. Reducing it should be
conservatively correct at the cost of extra nops inserted
into unreasonably large clauses.

Another issue is this doesn't account for how some
instructions increase the counter by more than 1,
but this should be conservatively correct.

[rampitec]

LGTM

[t-tye]

lib/Target/AMDGPU/GCNHazardRecognizer.cpp

366–368

We must also not put stores in the same clause if they may write the same address. Putting stores in their own clause will trivially satisfy this requirement.

How does this code ensure multiple consecutive stores are not put in the same clause? Seems it will only break the clause if a read is before a store. How is a store before a read broken into a separate clause?

This code seems to be per single basic block? What ensures that a clause cannot straddle across basic blocks (due to fall through)? Since no checking is being done across basic blocks the conservative action is to break clauses that end a basic block.

374

Would it be better to explicitly break the clause, as a context save restore may bring a wave back in mid clause execution and so the counters may saturate at a different instruction. By breaking the clause explicitly this will not be an issue.

[arsenm]

lib/Target/AMDGPU/GCNHazardRecognizer.cpp

366–368

This was already handled, the clause is broken if it's a store

374

I don't understand what the issue is here. The counters stop working correctly?

[t-tye]

lib/Target/AMDGPU/GCNHazardRecognizer.cpp

374

When a context save happens the wave could be in the middle of a clause. When the context restore happens it behaves as if the clause started from the point of the context save (since all previous memory operations are waited to complete before the context save, and so the outstanding waitcnt counters start at 0 on the context restore). So the counters may saturate further down the clause than this code predicted.

XNACK replay can cause a similar situation since during replay the clause is being effectively started further in.

So my suggestion is that if the code is only scanning over a certain list of instructions to ensure the clause rules, it should explicitly break the clause at the point it scans up to. Otherwise a context save/restore or XNACK replay could effectively execute a clause that spans over instructions that have not been checked for hazards.

In practice it seems unlikely that clauses will reach the hardware counter limit, so this will likely not result in reduced performance. However, it ensures correctness in the unlikely case that they do.

So the counters are still working correctly, they just may resume from 0 mid clause if the hardware does a context save/restore or XNACK replay which implicitly waits for the outstanding memory operations to complete.

[t-tye]

lib/Target/AMDGPU/GCNHazardRecognizer.cpp

366–368

Does that address all the items in the previous comment?

Also curious which line of code achieves that as it does not appear to be this line of code.

[arsenm]

Rebase

[kerbowa]

llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp

60 ↗

(On Diff #255163)

gfx10 has 6 bits for LGKMCNT.

lib/Target/AMDGPU/AMDGPUSubtarget.h

public:
 
   bool enableSubRegLiveness() const override {
     return true;
   }
 
   void setScalarizeGlobalBehavior(bool b) { ScalarizeGlobal = b;}
   bool getScalarizeGlobalBehavior() const { return ScalarizeGlobal;}
 
+  unsigned getMaxLgkmcnt() const {
+    return AMDGPU::getLgkmcntBitMask(
+      AMDGPU::IsaInfo::getIsaVersion(getFeatureBits()));
+  }
+
+  unsigned getMaxVmcnt() const {
+    return AMDGPU::getVmcntBitMask(
+      AMDGPU::IsaInfo::getIsaVersion(getFeatureBits()));
+  }
+
   /// \returns Number of execution units per compute unit supported by the
   /// subtarget.
   unsigned getEUsPerCU() const {
     return AMDGPU::IsaInfo::getEUsPerCU(getFeatureBits());
   }
 
   /// \returns Maximum number of work groups per compute unit supported by the
   /// subtarget and limited by given \p FlatWorkGroupSize.

lib/Target/AMDGPU/GCNHazardRecognizer.h

class GCNHazardRecognizer final : public ScheduleHazardRecognizer {
   // will be added to EmittedInstrs, when AdvanceCycle() or RecedeCycle() is
   // called.
   MachineInstr *CurrCycleInstr;
   std::list<MachineInstr*> EmittedInstrs;
   const MachineFunction &MF;
   const SISubtarget &ST;
   const SIInstrInfo &TII;
   const SIRegisterInfo &TRI;
+  unsigned MaxSMEMClauseSize;
+  unsigned MaxVMEMClauseSize;
 
   /// RegUnits of uses in the current soft memory clause.
   BitVector ClauseUses;
 
   /// RegUnits of defs in the current soft memory clause.
   BitVector ClauseDefs;
 
   void resetClause() {

lib/Target/AMDGPU/GCNHazardRecognizer.cpp

 #include <vector>
 
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
 // Hazard Recoginizer Implementation
 //===----------------------------------------------------------------------===//
 
+static cl::opt<unsigned>
+HazardMaxLookAhead("amdgpu-hazard-max-lookahead", cl::Hidden, cl::init(16),
+                   cl::desc("Maximum instruction count to look for hazards"));
+
 GCNHazardRecognizer::GCNHazardRecognizer(const MachineFunction &MF) :
   CurrCycleInstr(nullptr),
   MF(MF),
   ST(MF.getSubtarget<SISubtarget>()),
   TII(*ST.getInstrInfo()),
   TRI(TII.getRegisterInfo()),
+  MaxSMEMClauseSize(ST.getMaxLgkmcnt() + 1),
+  MaxVMEMClauseSize(ST.getMaxVmcnt() + 1),
   ClauseUses(TRI.getNumRegUnits()),
   ClauseDefs(TRI.getNumRegUnits()) {
-  MaxLookAhead = 5;
+
+  // VMCNT is 4 bits, except on GFX9 where it is 6 bits, therefore the maximum
+  // size of a VMEM clause is 16 or 64.
+  //
+  // LGKMCNT is always 4 bits for an SMEM clause.
+
+  // Setting it so high only allows for skipping a nop for extremely large
+  // clauses, so it should be fine to reduce this if it's a compile time issue.
+  MaxLookAhead = std::max(MaxVMEMClauseSize, MaxSMEMClauseSize);
+
+  if (HazardMaxLookAhead.getNumOccurrencesFlag())
+    MaxLookAhead = HazardMaxLookAhead;
+
+  // The maximum number of wait states for any normal hazard type is 5.
+  MaxLookAhead = std::max(MaxLookAhead, 5u);
 }
 
 void GCNHazardRecognizer::EmitInstruction(SUnit *SU) {
   EmitInstruction(SU->getInstr());
 }
 
 void GCNHazardRecognizer::EmitInstruction(MachineInstr *MI) {
   CurrCycleInstr = MI;

 int GCNHazardRecognizer::checkSoftClauseHazards(MachineInstr *MEM) {
   // SMEM soft clause are only present on VI+, and only matter if xnack is
   // enabled.
   if (!ST.isXNACKEnabled())
     return 0;
 
   bool IsSMRD = TII.isSMRD(*MEM);
 
+  const unsigned MaxClauseSize = IsSMRD ? MaxSMEMClauseSize : MaxVMEMClauseSize;
+
   resetClause();
 
-  // A soft-clause is any group of consecutive SMEM instructions.  The
-  // instructions in this group may return out of order and/or may be
-  // replayed (i.e. the same instruction issued more than once).
+  // A soft-clause is any group of consecutive SMEM or VMEM instructions. The
+  // instructions in this group may return out of order and/or may be replayed
+  // (i.e. the same instruction issued more than once).
   //
   // In order to handle these situations correctly we need to make sure
   // that when a clause has more than one instruction, no instruction in the
   // clause writes to a register that is read another instruction in the clause
   // (including itself). If we encounter this situaion, we need to break the
   // clause by inserting a non SMEM instruction.
 
+  unsigned ClauseSize = 1;
+
   for (MachineInstr *MI : EmittedInstrs) {
     // When we hit a non-SMEM instruction then we have passed the start of the
     // clause and we can stop.
     if (!MI)
       break;
 
     if (IsSMRD != SIInstrInfo::isSMRD(*MI))
       break;
 
     addClauseInst(*MI);
+
+    // FIXME: The maximum clause size is actually determined by saturating the
+    // counter type. This is > 1 for some SMEM operations.
+    ++ClauseSize;
   }
 
   if (ClauseDefs.none())
     return 0;
 
   // We need to make sure not to put loads and stores in the same clause if they
   // use the same address. For now, just start a new clause whenever we see a
   // store.
   if (MEM->mayStore())
     return 1;
 
+  // The clause is implicitly broken if it exceeds the maximum.
+  if (ClauseSize > MaxClauseSize)
+    return 0;
+
   addClauseInst(*MEM);
 
   // If the set of defs and uses intersect then we cannot add this instruction
   // to the clause, so we have a hazard.
   return ClauseDefs.anyCommon(ClauseUses) ? 1 : 0;
 }
 
 int GCNHazardRecognizer::checkSMRDHazards(MachineInstr *SMRD) {

test/CodeGen/AMDGPU/break-smem-soft-clauses.mir

 # RUN: llc -march=amdgcn -mcpu=carrizo -verify-machineinstrs -run-pass  post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,XNACK %s
 # RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs -run-pass  post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,NOXNACK %s
+# RUN: llc -march=amdgcn -mcpu=gfx901 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,XNACK %s
+# RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,NOXNACK %s
 
 ---
 # Trivial clause at beginning of program
 name: trivial_smem_clause_load_smrd4_x1
 
 body: |
   bb.0:
     ; GCN-LABEL: name: trivial_smem_clause_load_smrd4_x1
body: |
   bb.0:
     ; GCN-LABEL: name: smrd_load8_overwrite_ptr
     ; GCN: %sgpr10_sgpr11 = S_LOAD_DWORDX2_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: S_ENDPGM
     %sgpr10_sgpr11 = S_LOAD_DWORDX2_IMM %sgpr10_sgpr11, 0, 0
     S_ENDPGM
 ...
 ---
-# vmcnt has 4 bits, so maximum 16 outstanding loads. The waitcnt
-# breaks the clause.
+# lgmcnt has 4 bits pre-gfx9, so maximum 16 outstanding loads.
 
 name: break_smem_clause_at_max_smem_clause_size_smrd_load4
 
 body: |
   bb.0:
     ; GCN-LABEL: name: break_smem_clause_at_max_smem_clause_size_smrd_load4
+
     ; GCN: %sgpr13 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr14 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr15 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr16 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
     ; GCN-NEXT: %sgpr17 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr18 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr19 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr20 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
     ; GCN-NEXT: %sgpr21 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr22 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr23 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr24 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
     ; GCN-NEXT: %sgpr25 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr26 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr27 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     ; GCN-NEXT: %sgpr28 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
-    ; GCN-NEXT: %sgpr0 = S_LOAD_DWORD_IMM %sgpr30_sgpr31, 0, 0
-    ; GCN-NEXT: %sgpr0 = S_MOV_B32 %sgpr0, implicit %sgpr13, implicit %sgpr14, implicit %sgpr15, implicit %sgpr16, implicit %sgpr17, implicit %sgpr18, implicit %sgpr19, implicit %sgpr20, implicit %sgpr21, implicit %sgpr22, implicit %sgpr23, implicit %sgpr24, implicit %sgpr25, implicit %sgpr26, implicit %sgpr27, implicit %sgpr28
+
+    ; GCN-NEXT: %sgpr29 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr30 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr31 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr32 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    ; GCN-NEXT: %sgpr33 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr34 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr35 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr36 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    ; GCN-NEXT: %sgpr37 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr38 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr39 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr40 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    ; GCN-NEXT: %sgpr41 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr42 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr43 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr44 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    ; GCN-NEXT: %sgpr10 = S_LOAD_DWORD_IMM %sgpr30_sgpr31, 0, 0
+    ; GCN-NEXT: %sgpr0 = S_MOV_B32 %sgpr0
     ; GCN-NEXT: S_ENDPGM
+
     %sgpr13 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr14 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr15 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr16 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
 
     %sgpr17 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr18 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr19 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr20 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
 
     %sgpr21 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr22 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr23 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr24 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
 
     %sgpr25 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr26 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr27 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
     %sgpr28 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
 
-    %sgpr0 = S_LOAD_DWORD_IMM %sgpr30_sgpr31, 0, 0
-    %sgpr0 = S_MOV_B32 %sgpr0, implicit %sgpr13, implicit %sgpr14, implicit %sgpr15, implicit %sgpr16, implicit %sgpr17, implicit %sgpr18, implicit %sgpr19, implicit %sgpr20, implicit %sgpr21, implicit %sgpr22, implicit %sgpr23, implicit %sgpr24, implicit %sgpr25, implicit %sgpr26, implicit %sgpr27, implicit %sgpr28
+    %sgpr29 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr30 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr31 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr32 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    %sgpr33 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr34 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr35 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr36 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    %sgpr37 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr38 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr39 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr40 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    %sgpr41 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr42 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr43 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr44 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    %sgpr10 = S_LOAD_DWORD_IMM %sgpr30_sgpr31, 0, 0
+    %sgpr0 = S_MOV_B32 %sgpr0
+    S_ENDPGM
+...
+---
+
+name: break_smem_clause_at_max_smem_clause_size_m1_smrd_load4
+
+body: |
+  bb.0:
+    ; GCN-LABEL: name: break_smem_clause_at_max_smem_clause_size_m1_smrd_load4
+
+    ; GCN: %sgpr13 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr14 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr15 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr16 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    ; GCN-NEXT: %sgpr17 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr18 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr19 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr20 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    ; GCN-NEXT: %sgpr21 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr22 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr23 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr24 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    ; GCN-NEXT: %sgpr25 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr26 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    ; GCN-NEXT: %sgpr27 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    ; XNACK-NEXT: S_NOP 0
+    ; GCN-NEXT: %sgpr10 = S_LOAD_DWORD_IMM %sgpr30_sgpr31, 0, 0
+    ; GCN-NEXT: %sgpr0 = S_MOV_B32 %sgpr0
+    ; GCN-NEXT: S_ENDPGM
+
+    %sgpr13 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr14 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr15 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr16 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    %sgpr17 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr18 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr19 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr20 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    %sgpr21 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr22 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr23 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr24 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    %sgpr25 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr26 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+    %sgpr27 = S_LOAD_DWORD_IMM %sgpr10_sgpr11, 0, 0
+
+    %sgpr10 = S_LOAD_DWORD_IMM %sgpr30_sgpr31, 0, 0
+    %sgpr0 = S_MOV_B32 %sgpr0
     S_ENDPGM
 ...
 ---
 
 name: break_smem_clause_simple_load_smrd4_lo_ptr
 
 body: |
   bb.0:

test/CodeGen/AMDGPU/break-vmem-soft-clauses.mir

-# RUN: llc -march=amdgcn -mcpu=carrizo -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,XNACK %s
+# RUN: llc -march=amdgcn -mcpu=carrizo -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,XNACK,XNACK-VI %s
 # RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,NOXNACK %s
+# RUN: llc -march=amdgcn -mcpu=gfx901 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,XNACK,XNACK-GFX9 %s
+# RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefixes=GCN,NOXNACK %s
+
 ---
 # Trivial clause at beginning of program
 name: trivial_clause_load_flat4_x1
 
 body: |
   bb.0:
     ; GCN-LABEL: name: trivial_clause_load_flat4_x1
     ; GCN: %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
bb.0:
     ; GCN-LABEL: name: flat_load8_overwrite_ptr
     ; GCN: %vgpr2_vgpr3 = FLAT_LOAD_DWORDX2 %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: S_ENDPGM
 
     %vgpr2_vgpr3 = FLAT_LOAD_DWORDX2 %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
     S_ENDPGM
 ...
 ---
-# vmcnt has 4 bits, so maximum 16 outstanding loads. The waitcnt
-# breaks the clause.
+# vmcnt has 4 bits pre-gfx9, so maximum 16 outstanding loads can be in a
+# clause, so no breaking is needed.
 
 
-name: break_clause_at_max_clause_size_flat_load4
+name: break_clause_at_max_clause16_size_flat_load4
 
 body: |
   bb.0:
-    ; GCN-LABEL: name: break_clause_at_max_clause_size_flat_load4
+    ; GCN-LABEL: name: break_clause_at_max_clause16_size_flat_load4
     ; GCN: %vgpr2 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr3 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr4 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr5 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr6 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr7 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr8 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr9 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr10 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr11 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr12 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr13 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr14 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr15 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr16 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %vgpr17 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
-    ; XNACK-NEXT: S_NOP 0
+    ; XNACK-GFX9-NEXT: S_NOP 0
     ; GCN-NEXT: %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: %sgpr0 = S_MOV_B32 %sgpr0, implicit %vgpr2, implicit %vgpr3, implicit %vgpr4, implicit %vgpr5, implicit %vgpr6, implicit %vgpr7, implicit %vgpr8, implicit %vgpr9, implicit %vgpr10, implicit %vgpr11, implicit %vgpr12, implicit %vgpr13, implicit %vgpr14, implicit %vgpr15, implicit %vgpr16, implicit %vgpr17, implicit %vgpr18
     ; GCN-NEXT: S_ENDPGM
 
     %vgpr2 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr3 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr4 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr5 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
bb.0:
     %vgpr16 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr17 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
 
     %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
     %sgpr0 = S_MOV_B32 %sgpr0, implicit %vgpr2, implicit %vgpr3, implicit %vgpr4, implicit %vgpr5, implicit %vgpr6, implicit %vgpr7, implicit %vgpr8, implicit %vgpr9, implicit %vgpr10, implicit %vgpr11, implicit %vgpr12, implicit %vgpr13, implicit %vgpr14, implicit %vgpr15, implicit %vgpr16, implicit %vgpr17, implicit %vgpr18
     S_ENDPGM
 ...
 ---
+# vmcnt has 4 bits pre-gfx9, so maximum 16 outstanding loads can be in a
+# clause, so no breaking is needed.
+
+
+name: break_clause_at_max_m1_clause16_size_flat_load4
+
+body: |
+  bb.0:
+    ; GCN-LABEL: name: break_clause_at_max_m1_clause16_size_flat_load4
+    ; GCN: %vgpr2 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr3 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr4 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr5 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr6 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr7 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr8 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr9 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr10 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr11 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr12 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr13 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr14 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr15 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr16 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; XNACK-NEXT: S_NOP 0
+    ; GCN-NEXT: %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %sgpr0 = S_MOV_B32 %sgpr0, implicit %vgpr2, implicit %vgpr3, implicit %vgpr4, implicit %vgpr5, implicit %vgpr6, implicit %vgpr7, implicit %vgpr8, implicit %vgpr9, implicit %vgpr10, implicit %vgpr11, implicit %vgpr12, implicit %vgpr13, implicit %vgpr14, implicit %vgpr15, implicit %vgpr16, implicit %vgpr17, implicit %vgpr18
+    ; GCN-NEXT: S_ENDPGM
+
+    %vgpr2 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr3 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr4 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr5 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr6 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr7 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr8 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr9 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr10 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr11 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr12 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr13 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr14 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr15 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr16 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %sgpr0 = S_MOV_B32 %sgpr0, implicit %vgpr2, implicit %vgpr3, implicit %vgpr4, implicit %vgpr5, implicit %vgpr6, implicit %vgpr7, implicit %vgpr8, implicit %vgpr9, implicit %vgpr10, implicit %vgpr11, implicit %vgpr12, implicit %vgpr13, implicit %vgpr14, implicit %vgpr15, implicit %vgpr16, implicit %vgpr17, implicit %vgpr18
+    S_ENDPGM
+...
+---
+# vmcnt was increased to 6 bits in gfx9, so maximum 64 outstanding loads can be in a
+# clause, so no breaking is needed.
+
+name: break_clause_at_max_clause64_size_flat_load4
+
+body: |
+  bb.0:
+    ; GCN-LABEL: name: break_clause_at_max_clause64_size_flat_load4
+
+    ; GCN: %vgpr2 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr3 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr4 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr5 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr6 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr7 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr8 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr9 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr10 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr11 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr12 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr13 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr14 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr15 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr16 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr17 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr18 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr19 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr20 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr21 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr22 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr23 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr24 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr25 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr26 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr27 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr28 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr29 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr30 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr31 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr32 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr33 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr34 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr35 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr36 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr37 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr38 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr39 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr40 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr41 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr42 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr43 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr44 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr45 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr46 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr47 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr48 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr49 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr50 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr51 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr52 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr53 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr54 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr55 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr56 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr57 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr58 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr59 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr60 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr61 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr62 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr63 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr64 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr65 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %sgpr0 = S_MOV_B32 %sgpr0
+    ; GCN-NEXT: S_ENDPGM
+
+    %vgpr2 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr3 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr4 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr5 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr6 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr7 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr8 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr9 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr10 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr11 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr12 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr13 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr14 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr15 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr16 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr17 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr18 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr19 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr20 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr21 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr22 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr23 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr24 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr25 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr26 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr27 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr28 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr29 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr30 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr31 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr32 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr33 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr34 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr35 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr36 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr37 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr38 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr39 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr40 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr41 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr42 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr43 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr44 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr45 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr46 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr47 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr48 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr49 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr50 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr51 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr52 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr53 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr54 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr55 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr56 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr57 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr58 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr59 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr60 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr61 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr62 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr63 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr64 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr65 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %sgpr0 = S_MOV_B32 %sgpr0
+    S_ENDPGM
+...
+---
+# vmcnt was increased to 6 bits in gfx9, so maximum 64 outstanding loads can be in a
+# clause, so no breaking is needed.
+
+name: break_clause_at_max_m1_clause64_size_flat_load4
+
+body: |
+  bb.0:
+    ; GCN-LABEL: name: break_clause_at_max_m1_clause64_size_flat_load4
+
+    ; GCN: %vgpr2 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr3 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr4 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr5 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr6 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr7 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr8 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr9 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr10 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr11 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr12 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr13 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr14 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr15 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr16 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr17 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr18 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr19 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr20 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr21 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr22 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr23 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr24 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr25 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr26 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr27 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr28 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr29 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr30 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr31 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr32 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr33 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr34 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr35 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr36 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr37 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr38 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr39 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr40 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr41 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT: %vgpr42 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr43 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr44 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT: %vgpr45 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr46 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr47 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr48 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr49 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr50 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr51 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr52 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr53 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr54 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr55 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr56 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr57 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr58 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr59 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr60 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr61 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; GCN-NEXT:  %vgpr62 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr63 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %vgpr64 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    ; XNACK-GFX9-NEXT: S_NOP 0
+    ; GCN-NEXT:  %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
+    ; GCN-NEXT:  %sgpr0 = S_MOV_B32 %sgpr0
+    ; GCN-NEXT: S_ENDPGM
+
+    %vgpr2 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr3 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr4 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr5 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr6 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr7 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr8 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr9 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr10 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr11 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr12 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr13 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr14 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr15 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr16 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr17 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr18 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr19 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr20 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr21 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr22 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr23 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr24 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr25 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr26 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr27 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr28 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr29 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr30 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr31 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr32 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr33 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr34 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr35 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr36 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr37 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr38 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr39 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr40 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr41 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr42 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr43 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr44 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr45 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr46 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr47 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr48 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr49 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr50 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr51 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr52 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr53 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr54 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr55 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr56 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr57 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr58 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr59 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr60 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr61 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr62 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr63 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %vgpr64 = FLAT_LOAD_DWORD %vgpr0_vgpr1, 0, 0, 0, implicit %exec, implicit %flat_scr
+
+    %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
+    %sgpr0 = S_MOV_B32 %sgpr0
+    S_ENDPGM
+...
+---
 
 name: break_clause_simple_load_flat4_lo_ptr
 
 body: |
   bb.0:
     ; GCN-LABEL: name: break_clause_simple_load_flat4_lo_ptr
     ; GCN: %vgpr0 = FLAT_LOAD_DWORD %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; XNACK-NEXT: S_NOP 0
bb.0:
     ; GCN-NEXT: %vgpr2_vgpr3_vgpr4_vgpr5 = FLAT_LOAD_DWORDX4 %vgpr6_vgpr7, 0, 0, 0, implicit %exec, implicit %flat_scr
     ; GCN-NEXT: S_ENDPGM
     %vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr2_vgpr3_vgpr4_vgpr5 = FLAT_LOAD_DWORDX4 %vgpr6_vgpr7, 0, 0, 0, implicit %exec, implicit %flat_scr
     S_ENDPGM
 ...
 ---
 
-# The clause is broken by the waitcnt inserted at the end of the
-# block, so no nop is needed.
-
+# Make sure a clause break is inserted even if the next member is in a following basic block.
 
 name: break_clause_block_boundary_load_flat8_ptr
 
 body: |
   ; GCN-LABEL: name: break_clause_block_boundary_load_flat8_ptr
   ; GCN: bb.0:
   ; GCN-NEXT:   successors: %bb.1(0x80000000)
   ; GCN:   %vgpr0_vgpr1 = FLAT_LOAD_DWORDX2 %vgpr2_vgpr3, 0, 0, 0, implicit %exec, implicit %flat_scr