replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

AMDGPU/GlobalISel: Remove manual store select code 

This regresses the weird types that are newly treated as legal load
types, but fixes incorrectly using flat instrucions on SI.

llvm-svn: 367512
This commit is contained in:
Matt Arsenault 2019-08-01 03:52:40 +00:00
parent ae87b9f2c2
commit 57495268ac
10 changed files with 395 additions and 447 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
View File

@ -842,56 +842,8 @@ bool AMDGPUInstructionSelector::selectG_SELECT(MachineInstr &I) const {
bool AMDGPUInstructionSelector::selectG_STORE(
MachineInstr &I, CodeGenCoverage &CoverageInfo) const {
MachineBasicBlock *BB = I.getParent();
MachineFunction *MF = BB->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
const DebugLoc &DL = I.getDebugLoc();
LLT PtrTy = MRI.getType(I.getOperand(1).getReg());
if (PtrTy.getSizeInBits() != 64) {
initM0(I);
return selectImpl(I, CoverageInfo);
}
if (selectImpl(I, CoverageInfo))
return true;
unsigned StoreSize = RBI.getSizeInBits(I.getOperand(0).getReg(), MRI, TRI);
unsigned Opcode;
// FIXME: Remove this when integers > s32 naturally selected.
switch (StoreSize) {
default:
return false;
case 32:
Opcode = AMDGPU::FLAT_STORE_DWORD;
break;
case 64:
Opcode = AMDGPU::FLAT_STORE_DWORDX2;
break;
case 96:
Opcode = AMDGPU::FLAT_STORE_DWORDX3;
break;
case 128:
Opcode = AMDGPU::FLAT_STORE_DWORDX4;
break;
}
MachineInstr *Flat = BuildMI(*BB, &I, DL, TII.get(Opcode))
.add(I.getOperand(1))
.add(I.getOperand(0))
.addImm(0) // offset
.addImm(0) // glc
.addImm(0) // slc
.addImm(0); // dlc
// Now that we selected an opcode, we need to constrain the register
// operands to use appropriate classes.
bool Ret = constrainSelectedInstRegOperands(*Flat, TII, TRI, RBI);
I.eraseFromParent();
return Ret;
initM0(I);
return selectImpl(I, CoverageInfo);
}
static int sizeToSubRegIndex(unsigned Size) {

29
llvm/lib/Target/AMDGPU/FLATInstructions.td
View File

@ -777,8 +777,6 @@ def : FlatLoadPat <FLAT_LOAD_USHORT, extloadi16_flat, i32>;
def : FlatLoadPat <FLAT_LOAD_USHORT, zextloadi16_flat, i32>;
def : FlatLoadPat <FLAT_LOAD_USHORT, load_flat, i16>;
def : FlatLoadPat <FLAT_LOAD_SSHORT, sextloadi16_flat, i32>;
def : FlatLoadPat <FLAT_LOAD_DWORD, load_flat, i32>;
def : FlatLoadPat <FLAT_LOAD_DWORDX2, load_flat, v2i32>;
def : FlatLoadPat <FLAT_LOAD_DWORDX3, load_flat, v3i32>;
def : FlatLoadPat <FLAT_LOAD_DWORDX4, load_flat, v4i32>;
@ -787,8 +785,17 @@ def : FlatLoadAtomicPat <FLAT_LOAD_DWORDX2, atomic_load_64_flat, i64>;
def : FlatStorePat <FLAT_STORE_BYTE, truncstorei8_flat, i32>;
def : FlatStorePat <FLAT_STORE_SHORT, truncstorei16_flat, i32>;
def : FlatStorePat <FLAT_STORE_DWORD, store_flat, i32>;
def : FlatStorePat <FLAT_STORE_DWORDX2, store_flat, v2i32, VReg_64>;
foreach vt = [i32, f32, v2i16, v2f16] in {
def : FlatLoadPat <FLAT_LOAD_DWORD, load_flat, vt>;
def : FlatStorePat <FLAT_STORE_DWORD, store_flat, vt>;
}
foreach vt = VReg_64.RegTypes in {
def : FlatStorePat <FLAT_STORE_DWORDX2, store_flat, vt, VReg_64>;
def : FlatLoadPat <FLAT_LOAD_DWORDX2, load_flat, vt>;
}
def : FlatStorePat <FLAT_STORE_DWORDX3, store_flat, v3i32, VReg_96>;
def : FlatStorePat <FLAT_STORE_DWORDX4, store_flat, v4i32, VReg_128>;
@ -860,8 +867,16 @@ def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, zextloadi16_global, i32>;
def : FlatLoadSignedPat <GLOBAL_LOAD_SSHORT, sextloadi16_global, i32>;
def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, load_global, i16>;
def : FlatLoadSignedPat <GLOBAL_LOAD_DWORD, load_global, i32>;
def : FlatLoadSignedPat <GLOBAL_LOAD_DWORDX2, load_global, v2i32>;
foreach vt = [i32, f32, v2i16, v2f16] in {
def : FlatLoadSignedPat <GLOBAL_LOAD_DWORD, load_global, vt>;
def : FlatStoreSignedPat <GLOBAL_STORE_DWORD, store_global, vt, VGPR_32>;
}
foreach vt = VReg_64.RegTypes in {
def : FlatLoadSignedPat <GLOBAL_LOAD_DWORDX2, load_global, vt>;
def : FlatStoreSignedPat <GLOBAL_STORE_DWORDX2, store_global, vt, VReg_64>;
}
def : FlatLoadSignedPat <GLOBAL_LOAD_DWORDX3, load_global, v3i32>;
def : FlatLoadSignedPat <GLOBAL_LOAD_DWORDX4, load_global, v4i32>;
@ -872,8 +887,6 @@ def : FlatStoreSignedPat <GLOBAL_STORE_BYTE, truncstorei8_global, i32, VGPR_32>;
def : FlatStoreSignedPat <GLOBAL_STORE_BYTE, truncstorei8_global, i16, VGPR_32>;
def : FlatStoreSignedPat <GLOBAL_STORE_SHORT, truncstorei16_global, i32, VGPR_32>;
def : FlatStoreSignedPat <GLOBAL_STORE_SHORT, store_global, i16, VGPR_32>;
def : FlatStoreSignedPat <GLOBAL_STORE_DWORD, store_global, i32, VGPR_32>;
def : FlatStoreSignedPat <GLOBAL_STORE_DWORDX2, store_global, v2i32, VReg_64>;
def : FlatStoreSignedPat <GLOBAL_STORE_DWORDX3, store_global, v3i32, VReg_96>;
def : FlatStoreSignedPat <GLOBAL_STORE_DWORDX4, store_global, v4i32, VReg_128>;

75
llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir
View File

@ -242,24 +242,25 @@ body: |
; GFX7-LABEL: name: load_flat_s64
; GFX7: liveins: $vgpr0_vgpr1
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[LOAD:%[0-9]+]]:vreg_64(s64) = G_LOAD [[COPY]](p1) :: (load 8)
; GFX7: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8)
; GFX7: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX8-LABEL: name: load_flat_s64
; GFX8: liveins: $vgpr0_vgpr1
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[LOAD:%[0-9]+]]:vreg_64(s64) = G_LOAD [[COPY]](p1) :: (load 8)
; GFX8: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8)
; GFX8: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX9-LABEL: name: load_flat_s64
; GFX9: liveins: $vgpr0_vgpr1
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[LOAD:%[0-9]+]]:vreg_64(s64) = G_LOAD [[COPY]](p1) :: (load 8)
; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8)
; GFX9: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX10-LABEL: name: load_flat_s64
; GFX10: liveins: $vgpr0_vgpr1
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[LOAD:%[0-9]+]]:vreg_64(s64) = G_LOAD [[COPY]](p1) :: (load 8)
; GFX10: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8)
; GFX10: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = G_LOAD %0 :: (load 8, align 8, addrspace 0)
$vgpr0_vgpr1 = COPY %1
@ -571,24 +572,25 @@ body: |
; GFX7-LABEL: name: load_flat_v2s16
; GFX7: liveins: $vgpr0_vgpr1
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[LOAD:%[0-9]+]]:vgpr_32(<2 x s16>) = G_LOAD [[COPY]](p1) :: (load 4)
; GFX7: $vgpr0 = COPY [[LOAD]](<2 x s16>)
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4)
; GFX7: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
; GFX8-LABEL: name: load_flat_v2s16
; GFX8: liveins: $vgpr0_vgpr1
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[LOAD:%[0-9]+]]:vgpr_32(<2 x s16>) = G_LOAD [[COPY]](p1) :: (load 4)
; GFX8: $vgpr0 = COPY [[LOAD]](<2 x s16>)
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4)
; GFX8: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
; GFX9-LABEL: name: load_flat_v2s16
; GFX9: liveins: $vgpr0_vgpr1
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[LOAD:%[0-9]+]]:vgpr_32(<2 x s16>) = G_LOAD [[COPY]](p1) :: (load 4)
; GFX9: $vgpr0 = COPY [[LOAD]](<2 x s16>)
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4)
; GFX9: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
; GFX10-LABEL: name: load_flat_v2s16
; GFX10: liveins: $vgpr0_vgpr1
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[LOAD:%[0-9]+]]:vgpr_32(<2 x s16>) = G_LOAD [[COPY]](p1) :: (load 4)
; GFX10: $vgpr0 = COPY [[LOAD]](<2 x s16>)
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4)
; GFX10: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x s16>) = G_LOAD %0 :: (load 4, align 4, addrspace 0)
$vgpr0 = COPY %1
@ -608,24 +610,25 @@ body: |
; GFX7-LABEL: name: load_flat_v4s16
; GFX7: liveins: $vgpr0_vgpr1
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[LOAD:%[0-9]+]]:vreg_64(<4 x s16>) = G_LOAD [[COPY]](p1) :: (load 8)
; GFX7: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8)
; GFX7: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX8-LABEL: name: load_flat_v4s16
; GFX8: liveins: $vgpr0_vgpr1
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[LOAD:%[0-9]+]]:vreg_64(<4 x s16>) = G_LOAD [[COPY]](p1) :: (load 8)
; GFX8: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8)
; GFX8: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX9-LABEL: name: load_flat_v4s16
; GFX9: liveins: $vgpr0_vgpr1
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[LOAD:%[0-9]+]]:vreg_64(<4 x s16>) = G_LOAD [[COPY]](p1) :: (load 8)
; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8)
; GFX9: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX10-LABEL: name: load_flat_v4s16
; GFX10: liveins: $vgpr0_vgpr1
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[LOAD:%[0-9]+]]:vreg_64(<4 x s16>) = G_LOAD [[COPY]](p1) :: (load 8)
; GFX10: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8)
; GFX10: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<4 x s16>) = G_LOAD %0 :: (load 8, align 8, addrspace 0)
$vgpr0_vgpr1 = COPY %1

75
llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-global.mir
View File

@ -248,24 +248,25 @@ body: |
; GFX7-LABEL: name: load_global_s64
; GFX7: liveins: $vgpr0_vgpr1
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[LOAD:%[0-9]+]]:vreg_64(s64) = G_LOAD [[COPY]](p1) :: (load 8, addrspace 1)
; GFX7: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8, addrspace 1)
; GFX7: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX8-LABEL: name: load_global_s64
; GFX8: liveins: $vgpr0_vgpr1
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[LOAD:%[0-9]+]]:vreg_64(s64) = G_LOAD [[COPY]](p1) :: (load 8, addrspace 1)
; GFX8: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8, addrspace 1)
; GFX8: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX9-LABEL: name: load_global_s64
; GFX9: liveins: $vgpr0_vgpr1
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[LOAD:%[0-9]+]]:vreg_64(s64) = G_LOAD [[COPY]](p1) :: (load 8, addrspace 1)
; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec :: (load 8, addrspace 1)
; GFX9: $vgpr0_vgpr1 = COPY [[GLOBAL_LOAD_DWORDX2_]]
; GFX10-LABEL: name: load_global_s64
; GFX10: liveins: $vgpr0_vgpr1
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[LOAD:%[0-9]+]]:vreg_64(s64) = G_LOAD [[COPY]](p1) :: (load 8, addrspace 1)
; GFX10: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec :: (load 8, addrspace 1)
; GFX10: $vgpr0_vgpr1 = COPY [[GLOBAL_LOAD_DWORDX2_]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = G_LOAD %0 :: (load 8, align 8, addrspace 1)
$vgpr0_vgpr1 = COPY %1
@ -581,24 +582,25 @@ body: |
; GFX7-LABEL: name: load_global_v2s16
; GFX7: liveins: $vgpr0_vgpr1
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[LOAD:%[0-9]+]]:vgpr_32(<2 x s16>) = G_LOAD [[COPY]](p1) :: (load 4, addrspace 1)
; GFX7: $vgpr0 = COPY [[LOAD]](<2 x s16>)
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4, addrspace 1)
; GFX7: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
; GFX8-LABEL: name: load_global_v2s16
; GFX8: liveins: $vgpr0_vgpr1
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[LOAD:%[0-9]+]]:vgpr_32(<2 x s16>) = G_LOAD [[COPY]](p1) :: (load 4, addrspace 1)
; GFX8: $vgpr0 = COPY [[LOAD]](<2 x s16>)
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 4, addrspace 1)
; GFX8: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
; GFX9-LABEL: name: load_global_v2s16
; GFX9: liveins: $vgpr0_vgpr1
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[LOAD:%[0-9]+]]:vgpr_32(<2 x s16>) = G_LOAD [[COPY]](p1) :: (load 4, addrspace 1)
; GFX9: $vgpr0 = COPY [[LOAD]](<2 x s16>)
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
; GFX10-LABEL: name: load_global_v2s16
; GFX10: liveins: $vgpr0_vgpr1
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[LOAD:%[0-9]+]]:vgpr_32(<2 x s16>) = G_LOAD [[COPY]](p1) :: (load 4, addrspace 1)
; GFX10: $vgpr0 = COPY [[LOAD]](<2 x s16>)
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x s16>) = G_LOAD %0 :: (load 4, align 4, addrspace 1)
$vgpr0 = COPY %1
@ -618,24 +620,25 @@ body: |
; GFX7-LABEL: name: load_global_v4s16
; GFX7: liveins: $vgpr0_vgpr1
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[LOAD:%[0-9]+]]:vreg_64(<4 x s16>) = G_LOAD [[COPY]](p1) :: (load 8, addrspace 1)
; GFX7: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8, addrspace 1)
; GFX7: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX8-LABEL: name: load_global_v4s16
; GFX8: liveins: $vgpr0_vgpr1
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[LOAD:%[0-9]+]]:vreg_64(<4 x s16>) = G_LOAD [[COPY]](p1) :: (load 8, addrspace 1)
; GFX8: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (load 8, addrspace 1)
; GFX8: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
; GFX9-LABEL: name: load_global_v4s16
; GFX9: liveins: $vgpr0_vgpr1
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[LOAD:%[0-9]+]]:vreg_64(<4 x s16>) = G_LOAD [[COPY]](p1) :: (load 8, addrspace 1)
; GFX9: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec :: (load 8, addrspace 1)
; GFX9: $vgpr0_vgpr1 = COPY [[GLOBAL_LOAD_DWORDX2_]]
; GFX10-LABEL: name: load_global_v4s16
; GFX10: liveins: $vgpr0_vgpr1
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[LOAD:%[0-9]+]]:vreg_64(<4 x s16>) = G_LOAD [[COPY]](p1) :: (load 8, addrspace 1)
; GFX10: $vgpr0_vgpr1 = COPY [[LOAD]](<4 x s16>)
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2 [[COPY]], 0, 0, 0, 0, implicit $exec :: (load 8, addrspace 1)
; GFX10: $vgpr0_vgpr1 = COPY [[GLOBAL_LOAD_DWORDX2_]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<4 x s16>) = G_LOAD %0 :: (load 8, align 8, addrspace 1)
$vgpr0_vgpr1 = COPY %1

299
llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-store-flat.mir
View File

@ -131,23 +131,23 @@ body: |
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8)
; GFX8-LABEL: name: store_flat_s64
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8)
; GFX9-LABEL: name: store_flat_s64
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8)
; GFX10-LABEL: name: store_flat_s64
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store 8, align 8, addrspace 0)
@ -166,25 +166,24 @@ body: |
; GFX7-LABEL: name: store_flat_s96
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX7: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
; GFX7: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store 12, align 16)
; GFX8-LABEL: name: store_flat_s96
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX8: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
; GFX8: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store 12, align 16)
; GFX9-LABEL: name: store_flat_s96
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX9: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
; GFX9: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store 12, align 16)
; GFX10-LABEL: name: store_flat_s96
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX10: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
; GFX10: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store 12, align 16)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
G_STORE %1, %0 :: (store 12, align 16, addrspace 0)
@ -203,25 +202,24 @@ body: |
; GFX7-LABEL: name: store_flat_s128
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: G_STORE [[COPY1]](s128), [[COPY]](p1) :: (store 16)
; GFX8-LABEL: name: store_flat_s128
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: G_STORE [[COPY1]](s128), [[COPY]](p1) :: (store 16)
; GFX9-LABEL: name: store_flat_s128
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: G_STORE [[COPY1]](s128), [[COPY]](p1) :: (store 16)
; GFX10-LABEL: name: store_flat_s128
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: G_STORE [[COPY1]](s128), [[COPY]](p1) :: (store 16)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
G_STORE %1, %0 :: (store 16, align 16, addrspace 0)
@ -355,23 +353,23 @@ body: |
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 4)
; GFX8-LABEL: name: store_flat_v2s16
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 4)
; GFX9-LABEL: name: store_flat_v2s16
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 4)
; GFX10-LABEL: name: store_flat_v2s16
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 4)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x s16>) = COPY $vgpr2
G_STORE %1, %0 :: (store 4, align 4, addrspace 0)
@ -393,23 +391,23 @@ body: |
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8)
; GFX8-LABEL: name: store_flat_v4s16
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8)
; GFX9-LABEL: name: store_flat_v4s16
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8)
; GFX10-LABEL: name: store_flat_v4s16
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<4 x s16>) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store 8, align 8, addrspace 0)
@ -429,25 +427,24 @@ body: |
; GFX7-LABEL: name: store_flat_v6s16
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX7: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
; GFX7: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store 12, align 16)
; GFX8-LABEL: name: store_flat_v6s16
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX8: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
; GFX8: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store 12, align 16)
; GFX9-LABEL: name: store_flat_v6s16
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX9: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
; GFX9: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store 12, align 16)
; GFX10-LABEL: name: store_flat_v6s16
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX10: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
; GFX10: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store 12, align 16)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
G_STORE %1, %0 :: (store 12, align 16, addrspace 0)
@ -466,25 +463,24 @@ body: |
; GFX7-LABEL: name: store_flat_v8s16
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: G_STORE [[COPY1]](<8 x s16>), [[COPY]](p1) :: (store 16)
; GFX8-LABEL: name: store_flat_v8s16
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: G_STORE [[COPY1]](<8 x s16>), [[COPY]](p1) :: (store 16)
; GFX9-LABEL: name: store_flat_v8s16
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: G_STORE [[COPY1]](<8 x s16>), [[COPY]](p1) :: (store 16)
; GFX10-LABEL: name: store_flat_v8s16
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: G_STORE [[COPY1]](<8 x s16>), [[COPY]](p1) :: (store 16)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
G_STORE %1, %0 :: (store 16, align 16, addrspace 0)
@ -504,25 +500,24 @@ body: |
; GFX7-LABEL: name: store_flat_v2s64
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: G_STORE [[COPY1]](<2 x s64>), [[COPY]](p1) :: (store 16)
; GFX8-LABEL: name: store_flat_v2s64
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: G_STORE [[COPY1]](<2 x s64>), [[COPY]](p1) :: (store 16)
; GFX9-LABEL: name: store_flat_v2s64
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: G_STORE [[COPY1]](<2 x s64>), [[COPY]](p1) :: (store 16)
; GFX10-LABEL: name: store_flat_v2s64
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: G_STORE [[COPY1]](<2 x s64>), [[COPY]](p1) :: (store 16)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
G_STORE %1, %0 :: (store 16, align 16, addrspace 0)
@ -542,25 +537,24 @@ body: |
; GFX7-LABEL: name: store_flat_p1
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(p1) = COPY $vgpr2_vgpr3
; GFX7: G_STORE [[COPY1]](p1), [[COPY]](p1) :: (store 8)
; GFX8-LABEL: name: store_flat_p1
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(p1) = COPY $vgpr2_vgpr3
; GFX8: G_STORE [[COPY1]](p1), [[COPY]](p1) :: (store 8)
; GFX9-LABEL: name: store_flat_p1
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(p1) = COPY $vgpr2_vgpr3
; GFX9: G_STORE [[COPY1]](p1), [[COPY]](p1) :: (store 8)
; GFX10-LABEL: name: store_flat_p1
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(p1) = COPY $vgpr2_vgpr3
; GFX10: G_STORE [[COPY1]](p1), [[COPY]](p1) :: (store 8)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(p1) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store 8, align 8, addrspace 0)
@ -580,25 +574,24 @@ body: |
; GFX7-LABEL: name: store_flat_v2p1
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: G_STORE [[COPY1]](<2 x p1>), [[COPY]](p1) :: (store 16)
; GFX8-LABEL: name: store_flat_v2p1
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: G_STORE [[COPY1]](<2 x p1>), [[COPY]](p1) :: (store 16)
; GFX9-LABEL: name: store_flat_v2p1
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: G_STORE [[COPY1]](<2 x p1>), [[COPY]](p1) :: (store 16)
; GFX10-LABEL: name: store_flat_v2p1
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: G_STORE [[COPY1]](<2 x p1>), [[COPY]](p1) :: (store 16)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
G_STORE %1, %0 :: (store 16, align 16, addrspace 0)
@ -618,25 +611,24 @@ body: |
; GFX7-LABEL: name: store_flat_p3
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY $vgpr2
; GFX7: G_STORE [[COPY1]](p3), [[COPY]](p1) :: (store 4)
; GFX8-LABEL: name: store_flat_p3
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY $vgpr2
; GFX8: G_STORE [[COPY1]](p3), [[COPY]](p1) :: (store 4)
; GFX9-LABEL: name: store_flat_p3
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY $vgpr2
; GFX9: G_STORE [[COPY1]](p3), [[COPY]](p1) :: (store 4)
; GFX10-LABEL: name: store_flat_p3
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY $vgpr2
; GFX10: G_STORE [[COPY1]](p3), [[COPY]](p1) :: (store 4)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(p3) = COPY $vgpr2
G_STORE %1, %0 :: (store 4, align 4, addrspace 0)
@ -656,25 +648,24 @@ body: |
; GFX7-LABEL: name: store_flat_v2p3
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
; GFX7: G_STORE [[COPY1]](<2 x p3>), [[COPY]](p1) :: (store 8)
; GFX8-LABEL: name: store_flat_v2p3
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
; GFX8: G_STORE [[COPY1]](<2 x p3>), [[COPY]](p1) :: (store 8)
; GFX9-LABEL: name: store_flat_v2p3
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
; GFX9: G_STORE [[COPY1]](<2 x p3>), [[COPY]](p1) :: (store 8)
; GFX10-LABEL: name: store_flat_v2p3
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
; GFX10: G_STORE [[COPY1]](<2 x p3>), [[COPY]](p1) :: (store 8)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store 8, align 8, addrspace 0)
@ -693,25 +684,24 @@ body: |
; GFX7-LABEL: name: store_atomic_flat_s32
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; GFX7: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store monotonic 4)
; GFX8-LABEL: name: store_atomic_flat_s32
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; GFX8: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store monotonic 4)
; GFX9-LABEL: name: store_atomic_flat_s32
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; GFX9: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store monotonic 4)
; GFX10-LABEL: name: store_atomic_flat_s32
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; GFX10: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store monotonic 4)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s32) = COPY $vgpr2
G_STORE %1, %0 :: (store monotonic 4, align 4, addrspace 0)
@ -731,25 +721,24 @@ body: |
; GFX7-LABEL: name: store_atomic_flat_s64
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
; GFX7: G_STORE [[COPY1]](s64), [[COPY]](p1) :: (store monotonic 8)
; GFX8-LABEL: name: store_atomic_flat_s64
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
; GFX8: G_STORE [[COPY1]](s64), [[COPY]](p1) :: (store monotonic 8)
; GFX9-LABEL: name: store_atomic_flat_s64
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
; GFX9: G_STORE [[COPY1]](s64), [[COPY]](p1) :: (store monotonic 8)
; GFX10-LABEL: name: store_atomic_flat_s64
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
; GFX10: G_STORE [[COPY1]](s64), [[COPY]](p1) :: (store monotonic 8)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store monotonic 8, align 8, addrspace 0)

299
llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-store-global.mir
View File

@ -131,23 +131,23 @@ body: |
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8, addrspace 1)
; GFX8-LABEL: name: store_global_s64
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8, addrspace 1)
; GFX9-LABEL: name: store_global_s64
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: GLOBAL_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (store 8, addrspace 1)
; GFX10-LABEL: name: store_global_s64
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: GLOBAL_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (store 8, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store 8, align 8, addrspace 1)
@ -166,25 +166,24 @@ body: |
; GFX7-LABEL: name: store_global_s96
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX7: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
; GFX7: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store 12, align 16, addrspace 1)
; GFX8-LABEL: name: store_global_s96
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX8: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
; GFX8: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store 12, align 16, addrspace 1)
; GFX9-LABEL: name: store_global_s96
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX9: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
; GFX9: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store 12, align 16, addrspace 1)
; GFX10-LABEL: name: store_global_s96
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX10: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
; GFX10: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store 12, align 16, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
G_STORE %1, %0 :: (store 12, align 16, addrspace 1)
@ -203,25 +202,24 @@ body: |
; GFX7-LABEL: name: store_global_s128
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: G_STORE [[COPY1]](s128), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX8-LABEL: name: store_global_s128
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: G_STORE [[COPY1]](s128), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX9-LABEL: name: store_global_s128
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: G_STORE [[COPY1]](s128), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX10-LABEL: name: store_global_s128
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: G_STORE [[COPY1]](s128), [[COPY]](p1) :: (store 16, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s128) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
G_STORE %1, %0 :: (store 16, align 16, addrspace 1)
@ -355,23 +353,23 @@ body: |
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 4, addrspace 1)
; GFX8-LABEL: name: store_global_v2s16
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 4, addrspace 1)
; GFX9-LABEL: name: store_global_v2s16
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: GLOBAL_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (store 4, addrspace 1)
; GFX10-LABEL: name: store_global_v2s16
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: GLOBAL_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (store 4, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x s16>) = COPY $vgpr2
G_STORE %1, %0 :: (store 4, align 4, addrspace 1)
@ -393,23 +391,23 @@ body: |
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8, addrspace 1)
; GFX8-LABEL: name: store_global_v4s16
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr :: (store 8, addrspace 1)
; GFX9-LABEL: name: store_global_v4s16
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: GLOBAL_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (store 8, addrspace 1)
; GFX10-LABEL: name: store_global_v4s16
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: GLOBAL_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (store 8, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<4 x s16>) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store 8, align 8, addrspace 1)
@ -429,25 +427,24 @@ body: |
; GFX7-LABEL: name: store_global_v6s16
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX7: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
; GFX7: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store 12, align 16, addrspace 1)
; GFX8-LABEL: name: store_global_v6s16
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX8: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
; GFX8: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store 12, align 16, addrspace 1)
; GFX9-LABEL: name: store_global_v6s16
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX9: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
; GFX9: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store 12, align 16, addrspace 1)
; GFX10-LABEL: name: store_global_v6s16
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
; GFX10: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
; GFX10: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store 12, align 16, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
G_STORE %1, %0 :: (store 12, align 16, addrspace 1)
@ -466,25 +463,24 @@ body: |
; GFX7-LABEL: name: store_global_v8s16
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: G_STORE [[COPY1]](<8 x s16>), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX8-LABEL: name: store_global_v8s16
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: G_STORE [[COPY1]](<8 x s16>), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX9-LABEL: name: store_global_v8s16
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: G_STORE [[COPY1]](<8 x s16>), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX10-LABEL: name: store_global_v8s16
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: G_STORE [[COPY1]](<8 x s16>), [[COPY]](p1) :: (store 16, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<8 x s16>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
G_STORE %1, %0 :: (store 16, align 16, addrspace 1)
@ -504,25 +500,24 @@ body: |
; GFX7-LABEL: name: store_global_v2s64
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: G_STORE [[COPY1]](<2 x s64>), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX8-LABEL: name: store_global_v2s64
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: G_STORE [[COPY1]](<2 x s64>), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX9-LABEL: name: store_global_v2s64
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: G_STORE [[COPY1]](<2 x s64>), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX10-LABEL: name: store_global_v2s64
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: G_STORE [[COPY1]](<2 x s64>), [[COPY]](p1) :: (store 16, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x s64>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
G_STORE %1, %0 :: (store 16, align 16, addrspace 1)
@ -542,25 +537,24 @@ body: |
; GFX7-LABEL: name: store_global_p1
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(p1) = COPY $vgpr2_vgpr3
; GFX7: G_STORE [[COPY1]](p1), [[COPY]](p1) :: (store 8, addrspace 1)
; GFX8-LABEL: name: store_global_p1
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(p1) = COPY $vgpr2_vgpr3
; GFX8: G_STORE [[COPY1]](p1), [[COPY]](p1) :: (store 8, addrspace 1)
; GFX9-LABEL: name: store_global_p1
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(p1) = COPY $vgpr2_vgpr3
; GFX9: G_STORE [[COPY1]](p1), [[COPY]](p1) :: (store 8, addrspace 1)
; GFX10-LABEL: name: store_global_p1
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(p1) = COPY $vgpr2_vgpr3
; GFX10: G_STORE [[COPY1]](p1), [[COPY]](p1) :: (store 8, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(p1) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store 8, align 8, addrspace 1)
@ -580,25 +574,24 @@ body: |
; GFX7-LABEL: name: store_global_v2p1
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX7: G_STORE [[COPY1]](<2 x p1>), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX8-LABEL: name: store_global_v2p1
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX8: G_STORE [[COPY1]](<2 x p1>), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX9-LABEL: name: store_global_v2p1
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX9: G_STORE [[COPY1]](<2 x p1>), [[COPY]](p1) :: (store 16, addrspace 1)
; GFX10-LABEL: name: store_global_v2p1
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_128 = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: FLAT_STORE_DWORDX4 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
; GFX10: G_STORE [[COPY1]](<2 x p1>), [[COPY]](p1) :: (store 16, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x p1>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
G_STORE %1, %0 :: (store 16, align 16, addrspace 1)
@ -618,25 +611,24 @@ body: |
; GFX7-LABEL: name: store_global_p3
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY $vgpr2
; GFX7: G_STORE [[COPY1]](p3), [[COPY]](p1) :: (store 4, addrspace 1)
; GFX8-LABEL: name: store_global_p3
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY $vgpr2
; GFX8: G_STORE [[COPY1]](p3), [[COPY]](p1) :: (store 4, addrspace 1)
; GFX9-LABEL: name: store_global_p3
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY $vgpr2
; GFX9: G_STORE [[COPY1]](p3), [[COPY]](p1) :: (store 4, addrspace 1)
; GFX10-LABEL: name: store_global_p3
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY $vgpr2
; GFX10: G_STORE [[COPY1]](p3), [[COPY]](p1) :: (store 4, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(p3) = COPY $vgpr2
G_STORE %1, %0 :: (store 4, align 4, addrspace 1)
@ -656,25 +648,24 @@ body: |
; GFX7-LABEL: name: store_global_v2p3
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
; GFX7: G_STORE [[COPY1]](<2 x p3>), [[COPY]](p1) :: (store 8, addrspace 1)
; GFX8-LABEL: name: store_global_v2p3
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
; GFX8: G_STORE [[COPY1]](<2 x p3>), [[COPY]](p1) :: (store 8, addrspace 1)
; GFX9-LABEL: name: store_global_v2p3
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
; GFX9: G_STORE [[COPY1]](<2 x p3>), [[COPY]](p1) :: (store 8, addrspace 1)
; GFX10-LABEL: name: store_global_v2p3
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
; GFX10: G_STORE [[COPY1]](<2 x p3>), [[COPY]](p1) :: (store 8, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(<2 x p3>) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store 8, align 8, addrspace 1)
@ -693,25 +684,24 @@ body: |
; GFX7-LABEL: name: store_atomic_global_s32
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; GFX7: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store monotonic 4, addrspace 1)
; GFX8-LABEL: name: store_atomic_global_s32
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; GFX8: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store monotonic 4, addrspace 1)
; GFX9-LABEL: name: store_atomic_global_s32
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; GFX9: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store monotonic 4, addrspace 1)
; GFX10-LABEL: name: store_atomic_global_s32
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: FLAT_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
; GFX10: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store monotonic 4, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s32) = COPY $vgpr2
G_STORE %1, %0 :: (store monotonic 4, align 4, addrspace 1)
@ -731,25 +721,24 @@ body: |
; GFX7-LABEL: name: store_atomic_global_s64
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX7: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
; GFX7: G_STORE [[COPY1]](s64), [[COPY]](p1) :: (store monotonic 8, addrspace 1)
; GFX8-LABEL: name: store_atomic_global_s64
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX8: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
; GFX8: G_STORE [[COPY1]](s64), [[COPY]](p1) :: (store monotonic 8, addrspace 1)
; GFX9-LABEL: name: store_atomic_global_s64
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX9: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
; GFX9: G_STORE [[COPY1]](s64), [[COPY]](p1) :: (store monotonic 8, addrspace 1)
; GFX10-LABEL: name: store_atomic_global_s64
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: FLAT_STORE_DWORDX2 [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec, implicit $flat_scr
; GFX10: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr(s64) = COPY $vgpr2_vgpr3
; GFX10: G_STORE [[COPY1]](s64), [[COPY]](p1) :: (store monotonic 8, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = COPY $vgpr2_vgpr3
G_STORE %1, %0 :: (store monotonic 8, align 8, addrspace 1)

3
llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.if.break.i64.ll
View File

@ -1,5 +1,5 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel -mtriple=amdgcn--amdhsa -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
; RUN: llc -global-isel -mtriple=amdgcn--amdhsa -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
define amdgpu_kernel void @test_wave64(i32 %arg0, [8 x i32], i64 %saved) {
; GCN-LABEL: test_wave64:
@ -13,7 +13,6 @@ define amdgpu_kernel void @test_wave64(i32 %arg0, [8 x i32], i64 %saved) {
; GCN-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1]
; GCN-NEXT: v_mov_b32_e32 v0, s0
; GCN-NEXT: v_mov_b32_e32 v1, s1
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN-NEXT: flat_store_dwordx2 v[0:1], v[0:1]
; GCN-NEXT: s_endpgm
entry:

2
llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.kernarg.segment.ptr.ll
View File

@ -1,6 +1,6 @@
; RUN: llc -global-isel -mtriple=amdgcn--amdhsa -mattr=-code-object-v3 -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefixes=CO-V2,HSA,ALL %s
; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=hawaii -mattr=+flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=CO-V2,OS-MESA3D,MESA,ALL %s
; RUN: llc -global-isel -mtriple=amdgcn-mesa-unknown -verify-machineinstrs < %s | FileCheck -check-prefixes=OS-UNKNOWN,MESA,ALL %s
; RUN: llc -global-isel -mtriple=amdgcn-mesa-unknown -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefixes=OS-UNKNOWN,MESA,ALL %s
; ALL-LABEL: {{^}}test:
; CO-V2: enable_sgpr_kernarg_segment_ptr = 1

4
llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.workgroup.id.ll
View File

@ -1,8 +1,8 @@
; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa -mattr=-code-object-v3 -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=CO-V2 -check-prefix=CI-HSA %s
; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa -mattr=-code-object-v3 -mcpu=carrizo -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=CO-V2 -check-prefix=VI-HSA %s
; RUN: llc -global-isel -mtriple=amdgcn-- -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=UNKNOWN-OS -check-prefix=SI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-- -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=UNKNOWN-OS -check-prefix=SI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-- -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=UNKNOWN-OS -check-prefix=VI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-unknown-mesa3d -mattr=-code-object-v3 -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefixes=ALL,CO-V2,SI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-unknown-mesa3d -mattr=-code-object-v3 -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefixes=ALL,CO-V2,SI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-unknown-mesa3d -mattr=-code-object-v3 -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=ALL,CO-V2,VI-MESA %s
declare i32 @llvm.amdgcn.workgroup.id.x() #0

4
llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.workitem.id.ll
View File

@ -1,8 +1,8 @@
; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa -mattr=-code-object-v3 -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=CO-V2 -check-prefix=CI-HSA %s
; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa -mattr=-code-object-v3 -mcpu=carrizo -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=CO-V2 -check-prefix=VI-HSA %s
; RUN: llc -global-isel -mtriple=amdgcn-- -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=MESA -check-prefix=SI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-- -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=MESA -check-prefix=SI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-- -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=ALL -check-prefix=MESA -check-prefix=VI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-unknown-mesa3d -mattr=-code-object-v3 -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefixes=ALL,CO-V2,SI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-unknown-mesa3d -mattr=-code-object-v3 -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefixes=ALL,CO-V2,SI-MESA %s
; RUN: llc -global-isel -mtriple=amdgcn-unknown-mesa3d -mattr=-code-object-v3 -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=ALL,CO-V2,VI-MESA %s
declare i32 @llvm.amdgcn.workitem.id.x() #0