[AMDGPU] Check if type transforms to i16 (VI+) when getting AMDGPUISD::FFBH_U32

[kzhuravl]

Summary

This will prevent following regression when enabling i16 support (D18049):

test/CodeGen/AMDGPU/ctlz.ll
test/CodeGen/AMDGPU/ctlz_zero_undef.ll

[• tstellarAMD]

lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp

482 ↗

(On Diff #75252)

Shouldn't we only be doing this for uniform ops? Otherwise the SelectionDAG should be able to do the promotion.

[kzhuravl]

We do not need to promote it. Fixed during combining.

[arsenm]

lib/Target/AMDGPU/AMDGPUISelLowering.cpp

2461 ↗

(On Diff #75500)

Subtarget->hasI16

2468 ↗

(On Diff #75500)

Variable name probably shouldn't match the node type name

lib/Target/AMDGPU/AMDGPUISelLowering.h

33 ↗

(On Diff #75500)

I think SDLoc is supposed to be passed by const reference now

[kzhuravl]

lib/Target/AMDGPU/AMDGPUISelLowering.cpp

2461 ↗

(On Diff #75500)

This will involve moving has16BitInsts() from SISubtarget to AMDGPUSubtarget. Are there any objections?

[kzhuravl]

Address review feedback

[kzhuravl]

Ping

[• tstellarAMD]

LGTM.

llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h

 
 namespace llvm {
 
 class AMDGPUMachineFunction;
 class AMDGPUSubtarget;
 class MachineRegisterInfo;
 
 class AMDGPUTargetLowering : public TargetLowering {
+private:
+  /// \returns AMDGPUISD::FFBH_U32 node if the incoming \p Op may have been
+  /// legalized from a smaller type VT. Need to match pre-legalized type because
+  /// the generic legalization inserts the add/sub between the select and
+  /// compare.
+  SDValue getFFBH_U32(SelectionDAG &DAG, SDValue Op, const SDLoc &DL) const;
+
 protected:
   const AMDGPUSubtarget *Subtarget;
 
   SDValue LowerEXTRACT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
   /// \brief Split a vector store into multiple scalar stores.
   /// \returns The resulting chain.

llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp

if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Val))
     return C->isAllOnesValue();
   return false;
 }
 
 static bool isCtlzOpc(unsigned Opc) {
   return Opc == ISD::CTLZ || Opc == ISD::CTLZ_ZERO_UNDEF;
 }
 
-// Get FFBH node if the incoming op may have been type legalized from a smaller
-// type VT.
-// Need to match pre-legalized type because the generic legalization inserts the
-// add/sub between the select and compare.
-static SDValue getFFBH_U32(const TargetLowering &TLI, SelectionDAG &DAG,
-                           const SDLoc &SL, SDValue Op) {
+SDValue AMDGPUTargetLowering::getFFBH_U32(SelectionDAG &DAG,
+                                          SDValue Op,
+                                          const SDLoc &DL) const {
   EVT VT = Op.getValueType();
-  EVT LegalVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
-  if (LegalVT != MVT::i32)
+  EVT LegalVT = getTypeToTransformTo(*DAG.getContext(), VT);
+  if (LegalVT != MVT::i32 && (Subtarget->has16BitInsts() &&
+                              LegalVT != MVT::i16))
     return SDValue();
 
   if (VT != MVT::i32)
-    Op = DAG.getNode(ISD::ZERO_EXTEND, SL, MVT::i32, Op);
+    Op = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i32, Op);
 
-  SDValue FFBH = DAG.getNode(AMDGPUISD::FFBH_U32, SL, MVT::i32, Op);
+  SDValue FFBH = DAG.getNode(AMDGPUISD::FFBH_U32, DL, MVT::i32, Op);
   if (VT != MVT::i32)
-    FFBH = DAG.getNode(ISD::TRUNCATE, SL, VT, FFBH);
+    FFBH = DAG.getNode(ISD::TRUNCATE, DL, VT, FFBH);
 
   return FFBH;
 }
 
 // The native instructions return -1 on 0 input. Optimize out a select that
 // produces -1 on 0.
 //
 // TODO: If zero is not undef, we could also do this if the output is compared
SDValue AMDGPUTargetLowering::performCtlzCombine(const SDLoc &SL, SDValue Cond,
   ISD::CondCode CCOpcode = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
   SDValue CmpLHS = Cond.getOperand(0);
 
   // select (setcc x, 0, eq), -1, (ctlz_zero_undef x) -> ffbh_u32 x
   if (CCOpcode == ISD::SETEQ &&
       isCtlzOpc(RHS.getOpcode()) &&
       RHS.getOperand(0) == CmpLHS &&
       isNegativeOne(LHS)) {
-    return getFFBH_U32(*this, DAG, SL, CmpLHS);
+    return getFFBH_U32(DAG, CmpLHS, SL);
   }
 
   // select (setcc x, 0, ne), (ctlz_zero_undef x), -1 -> ffbh_u32 x
   if (CCOpcode == ISD::SETNE &&
       isCtlzOpc(LHS.getOpcode()) &&
       LHS.getOperand(0) == CmpLHS &&
       isNegativeOne(RHS)) {
-    return getFFBH_U32(*this, DAG, SL, CmpLHS);
+    return getFFBH_U32(DAG, CmpLHS, SL);
   }
 
   return SDValue();
 }
 
 SDValue AMDGPUTargetLowering::performSelectCombine(SDNode *N,
                                                    DAGCombinerInfo &DCI) const {
   SDValue Cond = N->getOperand(0);

llvm/trunk/lib/Target/AMDGPU/AMDGPUSubtarget.h

public:
   int getLDSBankCount() const {
     return LDSBankCount;
   }
 
   unsigned getMaxPrivateElementSize() const {
     return MaxPrivateElementSize;
   }
 
+  bool has16BitInsts() const {
+    return Has16BitInsts;
+  }
+
   bool hasHWFP64() const {
     return FP64;
   }
 
   bool hasFastFMAF32() const {
     return FastFMAF32;
   }
 
public:
   bool hasFlatAddressSpace() const {
     return FlatAddressSpace;
   }
 
   bool hasSMemRealTime() const {
     return HasSMemRealTime;
   }
 
-  bool has16BitInsts() const {
-    return Has16BitInsts;
-  }
-
   bool hasMovrel() const {
     return HasMovrel;
   }
 
   bool hasVGPRIndexMode() const {
     return HasVGPRIndexMode;
   }
 

llvm/trunk/test/CodeGen/AMDGPU/ctlz.ll

define void @v_ctlz_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrspace(1)* noalias %valptr) nounwind {
   %val = load <4 x i32>, <4 x i32> addrspace(1)* %valptr, align 16
   %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %val, i1 false) nounwind readnone
   store <4 x i32> %ctlz, <4 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_i8:
 ; GCN: buffer_load_ubyte [[VAL:v[0-9]+]],
-; GCN-DAG: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
-; GCN-DAG: v_cmp_eq_u32_e32 vcc, 0, [[CTLZ]]
-; GCN-DAG: v_cndmask_b32_e64 [[CORRECTED_FFBH:v[0-9]+]], [[FFBH]], 32, vcc
-; GCN: v_add_i32_e32 [[RESULT:v[0-9]+]], vcc, 0xffffffe8, [[CORRECTED_FFBH]]
+; GCN-DAG: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
 ; GCN: buffer_store_byte [[RESULT]],
 define void @v_ctlz_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
   %val = load i8, i8 addrspace(1)* %valptr
   %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 false) nounwind readnone
   store i8 %ctlz, i8 addrspace(1)* %out
   ret void
 }
 
define void @v_ctlz_i32_sel_ne_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
   %cmp = icmp ne i32 %ctlz, 32
   %sel = select i1 %cmp, i32 %ctlz, i32 -1
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
-; FIXME: Need to handle non-uniform case for function below (load without gep).
 ; FUNC-LABEL: {{^}}v_ctlz_i8_sel_eq_neg1:
 ; GCN: {{buffer|flat}}_load_ubyte [[VAL:v[0-9]+]],
 ; GCN: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
 ; GCN: {{buffer|flat}}_store_byte [[FFBH]],
  define void @v_ctlz_i8_sel_eq_neg1(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
   %tid = call i32 @llvm.r600.read.tidig.x()
   %valptr.gep = getelementptr i8, i8 addrspace(1)* %valptr, i32 %tid
   %val = load i8, i8 addrspace(1)* %valptr.gep

llvm/trunk/test/CodeGen/AMDGPU/ctlz_zero_undef.ll

-; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
 ; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
 
 declare i8 @llvm.ctlz.i8(i8, i1) nounwind readnone
 
 declare i32 @llvm.ctlz.i32(i32, i1) nounwind readnone
 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1) nounwind readnone
 declare <4 x i32> @llvm.ctlz.v4i32(<4 x i32>, i1) nounwind readnone
 
 declare i64 @llvm.ctlz.i64(i64, i1) nounwind readnone
 declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1) nounwind readnone
 declare <4 x i64> @llvm.ctlz.v4i64(<4 x i64>, i1) nounwind readnone
 
 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
 
 ; FUNC-LABEL: {{^}}s_ctlz_zero_undef_i32:
-; SI: s_load_dword [[VAL:s[0-9]+]],
-; SI: s_flbit_i32_b32 [[SRESULT:s[0-9]+]], [[VAL]]
-; SI: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]]
-; SI: buffer_store_dword [[VRESULT]],
-; SI: s_endpgm
+; GCN: s_load_dword [[VAL:s[0-9]+]],
+; GCN: s_flbit_i32_b32 [[SRESULT:s[0-9]+]], [[VAL]]
+; GCN: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]]
+; GCN: buffer_store_dword [[VRESULT]],
+; GCN: s_endpgm
 ; EG: MEM_RAT_CACHELESS STORE_RAW [[RESULT:T[0-9]+\.[XYZW]]]
 ; EG: FFBH_UINT {{\*? *}}[[RESULT]]
 define void @s_ctlz_zero_undef_i32(i32 addrspace(1)* noalias %out, i32 %val) nounwind {
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   store i32 %ctlz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i32:
-; SI: buffer_load_dword [[VAL:v[0-9]+]],
-; SI: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; SI: buffer_store_dword [[RESULT]],
-; SI: s_endpgm
+; GCN: buffer_load_dword [[VAL:v[0-9]+]],
+; GCN: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GCN: buffer_store_dword [[RESULT]],
+; GCN: s_endpgm
 ; EG: MEM_RAT_CACHELESS STORE_RAW [[RESULT:T[0-9]+\.[XYZW]]]
 ; EG: FFBH_UINT {{\*? *}}[[RESULT]]
 define void @v_ctlz_zero_undef_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr, align 4
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   store i32 %ctlz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_v2i32:
-; SI: buffer_load_dwordx2
-; SI: v_ffbh_u32_e32
-; SI: v_ffbh_u32_e32
-; SI: buffer_store_dwordx2
-; SI: s_endpgm
+; GCN: buffer_load_dwordx2
+; GCN: v_ffbh_u32_e32
+; GCN: v_ffbh_u32_e32
+; GCN: buffer_store_dwordx2
+; GCN: s_endpgm
 ; EG: MEM_RAT_CACHELESS STORE_RAW [[RESULT:T[0-9]+]]{{\.[XYZW]}}
 ; EG: FFBH_UINT {{\*? *}}[[RESULT]]
 ; EG: FFBH_UINT {{\*? *}}[[RESULT]]
 define void @v_ctlz_zero_undef_v2i32(<2 x i32> addrspace(1)* noalias %out, <2 x i32> addrspace(1)* noalias %valptr) nounwind {
   %val = load <2 x i32>, <2 x i32> addrspace(1)* %valptr, align 8
   %ctlz = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %val, i1 true) nounwind readnone
   store <2 x i32> %ctlz, <2 x i32> addrspace(1)* %out, align 8
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_v4i32:
-; SI: buffer_load_dwordx4
-; SI: v_ffbh_u32_e32
-; SI: v_ffbh_u32_e32
-; SI: v_ffbh_u32_e32
-; SI: v_ffbh_u32_e32
-; SI: buffer_store_dwordx4
-; SI: s_endpgm
+; GCN: buffer_load_dwordx4
+; GCN: v_ffbh_u32_e32
+; GCN: v_ffbh_u32_e32
+; GCN: v_ffbh_u32_e32
+; GCN: v_ffbh_u32_e32
+; GCN: buffer_store_dwordx4
+; GCN: s_endpgm
 ; EG: MEM_RAT_CACHELESS STORE_RAW [[RESULT:T[0-9]+]]{{\.[XYZW]}}
 ; EG: FFBH_UINT {{\*? *}}[[RESULT]]
 ; EG: FFBH_UINT {{\*? *}}[[RESULT]]
 ; EG: FFBH_UINT {{\*? *}}[[RESULT]]
 ; EG: FFBH_UINT {{\*? *}}[[RESULT]]
 define void @v_ctlz_zero_undef_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrspace(1)* noalias %valptr) nounwind {
   %val = load <4 x i32>, <4 x i32> addrspace(1)* %valptr, align 16
   %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %val, i1 true) nounwind readnone
   store <4 x i32> %ctlz, <4 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i8:
-; SI: buffer_load_ubyte [[VAL:v[0-9]+]],
-; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
-; SI: v_add_i32_e32 [[RESULT:v[0-9]+]], vcc, 0xffffffe8, [[FFBH]]
-; SI: buffer_store_byte [[RESULT]],
+; GCN: buffer_load_ubyte [[VAL:v[0-9]+]],
+; GCN: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GCN: buffer_store_byte [[RESULT]],
 define void @v_ctlz_zero_undef_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
   %val = load i8, i8 addrspace(1)* %valptr
   %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 true) nounwind readnone
   store i8 %ctlz, i8 addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}s_ctlz_zero_undef_i64:
-; SI: s_load_dwordx2 s{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
-; SI-DAG: v_cmp_eq_u32_e64 vcc, s[[HI]], 0{{$}}
-; SI-DAG: s_flbit_i32_b32 [[FFBH_LO:s[0-9]+]], s[[LO]]
-; SI-DAG: s_add_i32 [[ADD:s[0-9]+]], [[FFBH_LO]], 32
-; SI-DAG: s_flbit_i32_b32 [[FFBH_HI:s[0-9]+]], s[[HI]]
-; SI-DAG: v_mov_b32_e32 [[VFFBH_LO:v[0-9]+]], [[FFBH_LO]]
-; SI-DAG: v_mov_b32_e32 [[VFFBH_HI:v[0-9]+]], [[FFBH_HI]]
-; SI-DAG: v_cndmask_b32_e32 v[[CTLZ:[0-9]+]], [[VFFBH_HI]], [[VFFBH_LO]]
-; SI-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
-; SI: {{buffer|flat}}_store_dwordx2 v{{\[}}[[CTLZ]]:[[CTLZ_HI]]{{\]}}
+; GCN: s_load_dwordx2 s{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
+; GCN-DAG: v_cmp_eq_u32_e64 vcc, s[[HI]], 0{{$}}
+; GCN-DAG: s_flbit_i32_b32 [[FFBH_LO:s[0-9]+]], s[[LO]]
+; GCN-DAG: s_add_i32 [[ADD:s[0-9]+]], [[FFBH_LO]], 32
+; GCN-DAG: s_flbit_i32_b32 [[FFBH_HI:s[0-9]+]], s[[HI]]
+; GCN-DAG: v_mov_b32_e32 [[VFFBH_LO:v[0-9]+]], [[FFBH_LO]]
+; GCN-DAG: v_mov_b32_e32 [[VFFBH_HI:v[0-9]+]], [[FFBH_HI]]
+; GCN-DAG: v_cndmask_b32_e32 v[[CTLZ:[0-9]+]], [[VFFBH_HI]], [[VFFBH_LO]]
+; GCN-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
+; GCN: {{buffer|flat}}_store_dwordx2 v{{\[}}[[CTLZ]]:[[CTLZ_HI]]{{\]}}
 define void @s_ctlz_zero_undef_i64(i64 addrspace(1)* noalias %out, i64 %val) nounwind {
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 true)
   store i64 %ctlz, i64 addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}s_ctlz_zero_undef_i64_trunc:
 define void @s_ctlz_zero_undef_i64_trunc(i32 addrspace(1)* noalias %out, i64 %val) nounwind {
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 true)
   %trunc = trunc i64 %ctlz to i32
   store i32 %trunc, i32 addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i64:
-; SI-DAG: {{buffer|flat}}_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
-; SI-DAG: v_cmp_eq_u32_e64 [[CMPHI:s\[[0-9]+:[0-9]+\]]], 0, v[[HI]]
-; SI-DAG: v_ffbh_u32_e32 [[FFBH_LO:v[0-9]+]], v[[LO]]
-; SI-DAG: v_add_i32_e32 [[ADD:v[0-9]+]], vcc, 32, [[FFBH_LO]]
-; SI-DAG: v_ffbh_u32_e32 [[FFBH_HI:v[0-9]+]], v[[HI]]
-; SI-DAG: v_cndmask_b32_e64 v[[CTLZ:[0-9]+]], [[FFBH_HI]], [[FFBH_LO]]
-; SI-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
-; SI: {{buffer|flat}}_store_dwordx2 {{.*}}v{{\[}}[[CTLZ]]:[[CTLZ_HI]]{{\]}}
+; GCN-DAG: {{buffer|flat}}_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
+; GCN-DAG: v_cmp_eq_u32_e64 [[CMPHI:s\[[0-9]+:[0-9]+\]]], 0, v[[HI]]
+; GCN-DAG: v_ffbh_u32_e32 [[FFBH_LO:v[0-9]+]], v[[LO]]
+; GCN-DAG: v_add_i32_e32 [[ADD:v[0-9]+]], vcc, 32, [[FFBH_LO]]
+; GCN-DAG: v_ffbh_u32_e32 [[FFBH_HI:v[0-9]+]], v[[HI]]
+; GCN-DAG: v_cndmask_b32_e64 v[[CTLZ:[0-9]+]], [[FFBH_HI]], [[FFBH_LO]]
+; GCN-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
+; GCN: {{buffer|flat}}_store_dwordx2 {{.*}}v{{\[}}[[CTLZ]]:[[CTLZ_HI]]{{\]}}
 define void @v_ctlz_zero_undef_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
   %tid = call i32 @llvm.r600.read.tidig.x()
   %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
   %out.gep = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
   %val = load i64, i64 addrspace(1)* %in.gep
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 true)
   store i64 %ctlz, i64 addrspace(1)* %out.gep
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i64_trunc:
 define void @v_ctlz_zero_undef_i64_trunc(i32 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
   %tid = call i32 @llvm.r600.read.tidig.x()
   %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
   %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
   %val = load i64, i64 addrspace(1)* %in.gep
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 true)
   %trunc = trunc i64 %ctlz to i32
   store i32 %trunc, i32 addrspace(1)* %out.gep
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i32_sel_eq_neg1:
-; SI: buffer_load_dword [[VAL:v[0-9]+]],
-; SI: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; SI: buffer_store_dword [[RESULT]],
+; GCN: buffer_load_dword [[VAL:v[0-9]+]],
+; GCN: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GCN: buffer_store_dword [[RESULT]],
  define void @v_ctlz_zero_undef_i32_sel_eq_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp eq i32 %val, 0
   %sel = select i1 %cmp, i32 -1, i32 %ctlz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i32_sel_ne_neg1:
-; SI: buffer_load_dword [[VAL:v[0-9]+]],
-; SI: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
-; SI: buffer_store_dword [[RESULT]],
+; GCN: buffer_load_dword [[VAL:v[0-9]+]],
+; GCN: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; GCN: buffer_store_dword [[RESULT]],
 define void @v_ctlz_zero_undef_i32_sel_ne_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp ne i32 %val, 0
   %sel = select i1 %cmp, i32 %ctlz, i32 -1
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
-; FIXME: Need to handle non-uniform case for function below (load without gep).
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i8_sel_eq_neg1:
-; SI: {{buffer|flat}}_load_ubyte [[VAL:v[0-9]+]],
-; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
-; SI: {{buffer|flat}}_store_byte [[FFBH]],
+; GCN: {{buffer|flat}}_load_ubyte [[VAL:v[0-9]+]],
+; GCN: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
+; GCN: {{buffer|flat}}_store_byte [[FFBH]],
 define void @v_ctlz_zero_undef_i8_sel_eq_neg1(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
   %tid = call i32 @llvm.r600.read.tidig.x()
   %valptr.gep = getelementptr i8, i8 addrspace(1)* %valptr, i32 %tid
   %val = load i8, i8 addrspace(1)* %valptr.gep
   %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 true) nounwind readnone
   %cmp = icmp eq i8 %val, 0
   %sel = select i1 %cmp, i8 -1, i8 %ctlz
   store i8 %sel, i8 addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i32_sel_eq_neg1_two_use:
-; SI: buffer_load_dword [[VAL:v[0-9]+]],
-; SI-DAG: v_ffbh_u32_e32 [[RESULT0:v[0-9]+]], [[VAL]]
-; SI-DAG: v_cmp_eq_u32_e32 vcc, 0, [[VAL]]
-; SI-DAG: v_cndmask_b32_e64 [[RESULT1:v[0-9]+]], 0, 1, vcc
-; SI-DAG: buffer_store_dword [[RESULT0]]
-; SI-DAG: buffer_store_byte [[RESULT1]]
-; SI: s_endpgm
+; GCN: buffer_load_dword [[VAL:v[0-9]+]],
+; GCN-DAG: v_ffbh_u32_e32 [[RESULT0:v[0-9]+]], [[VAL]]
+; GCN-DAG: v_cmp_eq_u32_e32 vcc, 0, [[VAL]]
+; GCN-DAG: v_cndmask_b32_e64 [[RESULT1:v[0-9]+]], 0, 1, vcc
+; GCN-DAG: buffer_store_dword [[RESULT0]]
+; GCN-DAG: buffer_store_byte [[RESULT1]]
+; GCN: s_endpgm
  define void @v_ctlz_zero_undef_i32_sel_eq_neg1_two_use(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp eq i32 %val, 0
   %sel = select i1 %cmp, i32 -1, i32 %ctlz
   store volatile i32 %sel, i32 addrspace(1)* %out
   store volatile i1 %cmp, i1 addrspace(1)* undef
   ret void
 }
 
 ; Selected on wrong constant
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i32_sel_eq_0:
-; SI: buffer_load_dword
-; SI: v_ffbh_u32_e32
-; SI: v_cmp
-; SI: v_cndmask
-; SI: buffer_store_dword
+; GCN: buffer_load_dword
+; GCN: v_ffbh_u32_e32
+; GCN: v_cmp
+; GCN: v_cndmask
+; GCN: buffer_store_dword
  define void @v_ctlz_zero_undef_i32_sel_eq_0(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp eq i32 %val, 0
   %sel = select i1 %cmp, i32 0, i32 %ctlz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; Selected on wrong constant
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i32_sel_ne_0:
-; SI: buffer_load_dword
-; SI: v_ffbh_u32_e32
-; SI: v_cmp
-; SI: v_cndmask
-; SI: buffer_store_dword
+; GCN: buffer_load_dword
+; GCN: v_ffbh_u32_e32
+; GCN: v_cmp
+; GCN: v_cndmask
+; GCN: buffer_store_dword
 define void @v_ctlz_zero_undef_i32_sel_ne_0(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp ne i32 %val, 0
   %sel = select i1 %cmp, i32 %ctlz, i32 0
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; Compare on wrong constant
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i32_sel_eq_cmp_non0:
-; SI: buffer_load_dword
-; SI: v_ffbh_u32_e32
-; SI: v_cmp
-; SI: v_cndmask
-; SI: buffer_store_dword
+; GCN: buffer_load_dword
+; GCN: v_ffbh_u32_e32
+; GCN: v_cmp
+; GCN: v_cndmask
+; GCN: buffer_store_dword
  define void @v_ctlz_zero_undef_i32_sel_eq_cmp_non0(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp eq i32 %val, 1
   %sel = select i1 %cmp, i32 0, i32 %ctlz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; Selected on wrong constant
 ; FUNC-LABEL: {{^}}v_ctlz_zero_undef_i32_sel_ne_cmp_non0:
-; SI: buffer_load_dword
-; SI: v_ffbh_u32_e32
-; SI: v_cmp
-; SI: v_cndmask
-; SI: buffer_store_dword
+; GCN: buffer_load_dword
+; GCN: v_ffbh_u32_e32
+; GCN: v_cmp
+; GCN: v_cndmask
+; GCN: buffer_store_dword
 define void @v_ctlz_zero_undef_i32_sel_ne_cmp_non0(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
   %val = load i32, i32 addrspace(1)* %valptr
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp ne i32 %val, 1
   %sel = select i1 %cmp, i32 %ctlz, i32 0
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }