AMDGPU/SI: Add support for shrinking v_cndmask_b32_e32 instructions

[arsenm]

LGTM

lib/Target/AMDGPU/SIShrinkInstructions.cpp

static bool canShrink(MachineInstr &MI, const SIInstrInfo *TII,
   const MachineOperand *Src2 = TII->getNamedOperand(MI, AMDGPU::OpName::src2);
   // Can't shrink instruction with three operands.
   // FIXME: v_cndmask_b32 has 3 operands and is shrinkable, but we need to add
   // a special case for it.  It can only be shrunk if the third operand
   // is vcc.  We should handle this the same way we handle vopc, by addding
   // a register allocation hint pre-regalloc and then do the shrining
   // post-regalloc.
   if (Src2) {
-    if (MI.getOpcode() != AMDGPU::V_MAC_F32_e64)
-      return false;
+    switch (MI.getOpcode()) {
+      default: return false;
 
-    const MachineOperand *Src2Mod =
-        TII->getNamedOperand(MI, AMDGPU::OpName::src2_modifiers);
-    if (!isVGPR(Src2, TRI, MRI) || (Src2Mod && Src2Mod->getImm() != 0))
+      case AMDGPU::V_MAC_F32_e64:
+        if (!isVGPR(Src2, TRI, MRI) ||
+            TII->hasModifiersSet(MI, AMDGPU::OpName::src2_modifiers))
           return false;
+        break;
+
+      case AMDGPU::V_CNDMASK_B32_e64:
+        break;
+    }
   }
 
   const MachineOperand *Src1 = TII->getNamedOperand(MI, AMDGPU::OpName::src1);
   const MachineOperand *Src1Mod =
       TII->getNamedOperand(MI, AMDGPU::OpName::src1_modifiers);
 
   if (Src1 && (!isVGPR(Src1, TRI, MRI) || (Src1Mod && Src1Mod->getImm() != 0)))
     return false;
for (I = MBB.begin(); I != MBB.end(); I = Next) {
           // this pass again after RA and shrink it if it outputs to VCC.
           MRI.setRegAllocationHint(MI.getOperand(0).getReg(), 0, AMDGPU::VCC);
           continue;
         }
         if (DstReg != AMDGPU::VCC)
           continue;
       }
 
+      if (Op32 == AMDGPU::V_CNDMASK_B32_e32) {
+        // We shrink V_CNDMASK_B32_e64 using regalloc hints like we do for VOPC
+        // instructions.
+        const MachineOperand *Src2 =
+            TII->getNamedOperand(MI, AMDGPU::OpName::src2);
+        if (!Src2->isReg())
+          continue;
+        unsigned SReg = Src2->getReg();
+        if (TargetRegisterInfo::isVirtualRegister(SReg)) {
+          MRI.setRegAllocationHint(SReg, 0, AMDGPU::VCC);
+          continue;
+        }
+        if (SReg != AMDGPU::VCC)
+          continue;
+      }
+
       // We can shrink this instruction
       DEBUG(dbgs() << "Shrinking "; MI.dump(); dbgs() << '\n';);
 
       MachineInstrBuilder Inst32 =
           BuildMI(MBB, I, MI.getDebugLoc(), TII->get(Op32));
 
       // dst
       Inst32.addOperand(MI.getOperand(0));

test/CodeGen/AMDGPU/llvm.round.ll

 ; RUN: llc -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=R600 -check-prefix=FUNC %s
 
 ; FUNC-LABEL: {{^}}round_f32:
 ; SI-DAG: s_load_dword [[SX:s[0-9]+]]
 ; SI-DAG: s_mov_b32 [[K:s[0-9]+]], 0x7fffffff
 ; SI: v_trunc_f32_e32 [[TRUNC:v[0-9]+]], [[SX]]
 ; SI: v_sub_f32_e32 [[SUB:v[0-9]+]], [[SX]], [[TRUNC]]
 ; SI: v_mov_b32_e32 [[VX:v[0-9]+]], [[SX]]
 ; SI: v_bfi_b32 [[COPYSIGN:v[0-9]+]], [[K]], 1.0, [[VX]]
-; SI: v_cmp_le_f32_e64 [[CMP:s\[[0-9]+:[0-9]+\]]], 0.5, |[[SUB]]|
-; SI: v_cndmask_b32_e64 [[SEL:v[0-9]+]], 0, [[VX]], [[CMP]]
+; SI: v_cmp_le_f32_e64 vcc, 0.5, |[[SUB]]|
+; SI: v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, [[VX]]
 ; SI: v_add_f32_e32 [[RESULT:v[0-9]+]], [[SEL]], [[TRUNC]]
 ; SI: buffer_store_dword [[RESULT]]
 
 ; R600: TRUNC {{.*}}, [[ARG:KC[0-9]\[[0-9]+\]\.[XYZW]]]
 ; R600-DAG: ADD  {{.*}},
 ; R600-DAG: BFI_INT
 ; R600-DAG: SETGE
 ; R600-DAG: CNDE

test/CodeGen/AMDGPU/select-vectors.ll

 ; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 
 ; Test expansion of scalar selects on vectors.
 ; Evergreen not enabled since it seems to be having problems with doubles.
 
 
 ; FUNC-LABEL: {{^}}select_v4i8:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v4i8(<4 x i8> addrspace(1)* %out, <4 x i8> %a, <4 x i8> %b, i8 %c) nounwind {
   %cmp = icmp eq i8 %c, 0
   %select = select i1 %cmp, <4 x i8> %a, <4 x i8> %b
   store <4 x i8> %select, <4 x i8> addrspace(1)* %out, align 4
   ret void
 }
 
 ; FUNC-LABEL: {{^}}select_v4i16:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> %a, <4 x i16> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <4 x i16> %a, <4 x i16> %b
   store <4 x i16> %select, <4 x i16> addrspace(1)* %out, align 4
   ret void
 }
 
 ; FUNC-LABEL: {{^}}select_v2i32:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 ; SI: buffer_store_dwordx2
 define void @select_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a, <2 x i32> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <2 x i32> %a, <2 x i32> %b
   store <2 x i32> %select, <2 x i32> addrspace(1)* %out, align 8
   ret void
 }
 
 ; FUNC-LABEL: {{^}}select_v4i32:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 ; SI: buffer_store_dwordx4
 define void @select_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %a, <4 x i32> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <4 x i32> %a, <4 x i32> %b
   store <4 x i32> %select, <4 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FUNC-LABEL: {{^}}select_v8i32:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v8i32(<8 x i32> addrspace(1)* %out, <8 x i32> %a, <8 x i32> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <8 x i32> %a, <8 x i32> %b
   store <8 x i32> %select, <8 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FUNC-LABEL: {{^}}select_v2f32:

 define void @select_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4 x float> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <4 x float> %a, <4 x float> %b
   store <4 x float> %select, <4 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FUNC-LABEL: {{^}}select_v8f32:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8 x float> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <8 x float> %a, <8 x float> %b
   store <8 x float> %select, <8 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FUNC-LABEL: {{^}}select_v2f64:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %a, <2 x double> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <2 x double> %a, <2 x double> %b
   store <2 x double> %select, <2 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FUNC-LABEL: {{^}}select_v4f64:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %a, <4 x double> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <4 x double> %a, <4 x double> %b
   store <4 x double> %select, <4 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FUNC-LABEL: {{^}}select_v8f64:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %a, <8 x double> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <8 x double> %a, <8 x double> %b
   store <8 x double> %select, <8 x double> addrspace(1)* %out, align 16
   ret void
 }

test/CodeGen/AMDGPU/select64.ll

define void @v_select_trunc_i64_2(i32 addrspace(1)* %out, i32 %cond, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) nounwind {
   ret void
 }
 
 ; CHECK-LABEL: {{^}}v_select_i64_split_imm:
 ; CHECK: s_mov_b32 [[SHI:s[0-9]+]], 63
 ; CHECK: s_mov_b32 [[SLO:s[0-9]+]], 0
 ; CHECK-DAG: v_mov_b32_e32 [[VHI:v[0-9]+]], [[SHI]]
 ; CHECK-DAG: v_mov_b32_e32 [[VLO:v[0-9]+]], [[SLO]]
-; CHECK-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, [[VLO]], {{v[0-9]+}}
-; CHECK-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, [[VHI]], {{v[0-9]+}}
+; CHECK-DAG: v_cndmask_b32_e32 {{v[0-9]+}}, [[VLO]], {{v[0-9]+}}
+; CHECK-DAG: v_cndmask_b32_e32 {{v[0-9]+}}, [[VHI]], {{v[0-9]+}}
 ; CHECK: s_endpgm
 define void @v_select_i64_split_imm(i64 addrspace(1)* %out, i32 %cond, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) nounwind {
   %cmp = icmp ugt i32 %cond, 5
   %a = load i64, i64 addrspace(1)* %aptr, align 8
   %b = load i64, i64 addrspace(1)* %bptr, align 8
   %sel = select i1 %cmp, i64 %a, i64 270582939648 ; 63 << 32
   store i64 %sel, i64 addrspace(1)* %out, align 8
   ret void
 }

test/CodeGen/AMDGPU/sint_to_fp.f64.ll

 ; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
 
 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
 
 ; SI-LABEL: {{^}}sint_to_fp_i32_to_f64
 ; SI: v_cvt_f64_i32_e32
 define void @sint_to_fp_i32_to_f64(double addrspace(1)* %out, i32 %in) {
   %result = sitofp i32 %in to double
   store double %result, double addrspace(1)* %out
   ret void
 }
 
 ; FIXME: select on 0, 0
 ; SI-LABEL: {{^}}sint_to_fp_i1_f64:
-; SI: v_cmp_eq_i32_e64 [[CMP:s\[[0-9]+:[0-9]\]]],
+; SI: v_cmp_eq_i32_e64 vcc,
 ; We can't fold the SGPRs into v_cndmask_b32_e64, because it already
-; uses an SGPR for [[CMP]]
-; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, v{{[0-9]+}}, [[CMP]]
-; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 0, [[CMP]]
+; uses an SGPR (implicit vcc).
+; SI: v_cndmask_b32_e32 v{{[0-9]+}}, 0, v{{[0-9]+}}
+; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 0, vcc
 ; SI: buffer_store_dwordx2
 ; SI: s_endpgm
 define void @sint_to_fp_i1_f64(double addrspace(1)* %out, i32 %in) {
   %cmp = icmp eq i32 %in, 0
   %fp = sitofp i1 %cmp to double
   store double %fp, double addrspace(1)* %out, align 4
   ret void
 }

test/CodeGen/AMDGPU/uint_to_fp.f64.ll

 define void @s_uint_to_fp_v4i32_to_v4f64(<4 x double> addrspace(1)* %out, <4 x i32> %in) {
   %cast = uitofp <4 x i32> %in to <4 x double>
   store <4 x double> %cast, <4 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FIXME: select on 0, 0
 ; SI-LABEL: {{^}}uint_to_fp_i1_to_f64:
-; SI: v_cmp_eq_i32_e64 [[CMP:s\[[0-9]+:[0-9]\]]],
-; We can't fold the SGPRs into v_cndmask_b32_e64, because it already
-; uses an SGPR for [[CMP]]
-; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, v{{[0-9]+}}, [[CMP]]
-; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 0, [[CMP]]
+; SI: v_cmp_eq_i32_e64 vcc
+; We can't fold the SGPRs into v_cndmask_b32_e32, because it already
+; uses an SGPR (implicit vcc).
+; SI: v_cndmask_b32_e32 v{{[0-9]+}}, 0, v{{[0-9]+}}
+; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 0, vcc
 ; SI: buffer_store_dwordx2
 ; SI: s_endpgm
 define void @uint_to_fp_i1_to_f64(double addrspace(1)* %out, i32 %in) {
   %cmp = icmp eq i32 %in, 0
   %fp = uitofp i1 %cmp to double
   store double %fp, double addrspace(1)* %out, align 4
   ret void
 }

test/CodeGen/AMDGPU/vselect.ll

-;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG %s
-;RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI %s
-;RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck --check-prefix=SI %s
+;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG --check-prefix=FUNC %s
+;RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=FUNC %s
+;RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=VI --check-prefix=FUNC %s
+
+;FUNC-LABEL: {{^}}test_select_v2i32:
 
-;EG: {{^}}test_select_v2i32:
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
-;SI: {{^}}test_select_v2i32:
-;SI: v_cndmask_b32_e64
 ;SI: v_cndmask_b32_e64
+;SI: v_cndmask_b32_e32
 
 define void @test_select_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in0, <2 x i32> addrspace(1)* %in1) {
 entry:
   %0 = load <2 x i32>, <2 x i32> addrspace(1)* %in0
   %1 = load <2 x i32>, <2 x i32> addrspace(1)* %in1
   %cmp = icmp ne <2 x i32> %0, %1
   %result = select <2 x i1> %cmp, <2 x i32> %0, <2 x i32> %1
   store <2 x i32> %result, <2 x i32> addrspace(1)* %out
   ret void
 }
 
-;EG: {{^}}test_select_v2f32:
+;FUNC-LABEL: {{^}}test_select_v2f32:
+
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
-;SI: {{^}}test_select_v2f32:
-;SI: v_cndmask_b32_e64
 ;SI: v_cndmask_b32_e64
+;SI: v_cndmask_b32_e32
 
 define void @test_select_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %in0, <2 x float> addrspace(1)* %in1) {
 entry:
   %0 = load <2 x float>, <2 x float> addrspace(1)* %in0
   %1 = load <2 x float>, <2 x float> addrspace(1)* %in1
   %cmp = fcmp une <2 x float> %0, %1
   %result = select <2 x i1> %cmp, <2 x float> %0, <2 x float> %1
   store <2 x float> %result, <2 x float> addrspace(1)* %out
   ret void
 }
 
-;EG: {{^}}test_select_v4i32:
+;FUNC-LABEL: {{^}}test_select_v4i32:
+
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
-;SI: {{^}}test_select_v4i32:
-;SI: v_cndmask_b32_e64
-;SI: v_cndmask_b32_e64
-;SI: v_cndmask_b32_e64
-;SI: v_cndmask_b32_e64
+; FIXME: The shrinking does not happen on tonga
+
+;SI: v_cndmask_b32
+;SI: v_cndmask_b32
+;SI: v_cndmask_b32
+;SI: v_cndmask_b32
 
 define void @test_select_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in0, <4 x i32> addrspace(1)* %in1) {
 entry:
   %0 = load <4 x i32>, <4 x i32> addrspace(1)* %in0
   %1 = load <4 x i32>, <4 x i32> addrspace(1)* %in1
   %cmp = icmp ne <4 x i32> %0, %1
   %result = select <4 x i1> %cmp, <4 x i32> %0, <4 x i32> %1
   store <4 x i32> %result, <4 x i32> addrspace(1)* %out
   ret void
 }
 
-;EG: {{^}}test_select_v4f32:
+;FUNC-LABEL: {{^}}test_select_v4f32:
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
 define void @test_select_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in0, <4 x float> addrspace(1)* %in1) {
 entry:
   %0 = load <4 x float>, <4 x float> addrspace(1)* %in0
   %1 = load <4 x float>, <4 x float> addrspace(1)* %in1
   %cmp = fcmp une <4 x float> %0, %1
   %result = select <4 x i1> %cmp, <4 x float> %0, <4 x float> %1
   store <4 x float> %result, <4 x float> addrspace(1)* %out
   ret void
 }

test/CodeGen/AMDGPU/xor.ll

define void @xor_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in0, <4 x i32> addrspace(1)* %in1) {
   ret void
 }
 
 ; FUNC-LABEL: {{^}}xor_i1:
 ; EG: XOR_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], PS}}
 
 ; SI-DAG: v_cmp_le_f32_e32 [[CMP0:vcc]], 0, {{v[0-9]+}}
 ; SI-DAG: v_cmp_le_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]], 1.0, {{v[0-9]+}}
-; SI: s_xor_b64 [[XOR:s\[[0-9]+:[0-9]+\]]], [[CMP0]], [[CMP1]]
-; SI: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, [[XOR]]
+; SI: s_xor_b64 [[XOR:vcc]], [[CMP0]], [[CMP1]]
+; SI: v_cndmask_b32_e32 [[RESULT:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}
 ; SI: buffer_store_dword [[RESULT]]
 ; SI: s_endpgm
 define void @xor_i1(float addrspace(1)* %out, float addrspace(1)* %in0, float addrspace(1)* %in1) {
   %a = load float, float addrspace(1) * %in0
   %b = load float, float addrspace(1) * %in1
   %acmp = fcmp oge float %a, 0.000000e+00
   %bcmp = fcmp oge float %b, 1.000000e+00
   %xor = xor i1 %acmp, %bcmp