AMDGPU/GlobalISel: Select llvm.amdgcn.class

[arsenm]

Summary

Also fixes missing SubtargetPredicate on f16 class instructions.

[ronlieb]

LGTM

[arsenm]

r371436

lib/Target/AMDGPU/AMDGPUGISel.td

 def gi_vop3mods :
     GIComplexOperandMatcher<s32, "selectVOP3Mods">,
     GIComplexPatternEquiv<VOP3Mods>;
 
 def gi_vop3omods :
     GIComplexOperandMatcher<s32, "selectVOP3OMods">,
     GIComplexPatternEquiv<VOP3OMods>;
 
+def gi_vop3omods0clamp0omod :
+    GIComplexOperandMatcher<s32, "selectVOP3Mods0Clamp0OMod">,
+    GIComplexPatternEquiv<VOP3Mods0Clamp0OMod>;
+
 def gi_vop3opselmods0 :
     GIComplexOperandMatcher<s32, "selectVOP3OpSelMods0">,
     GIComplexPatternEquiv<VOP3OpSelMods0>;
 
 def gi_vop3opselmods :
     GIComplexOperandMatcher<s32, "selectVOP3OpSelMods">,
     GIComplexPatternEquiv<VOP3OpSelMods>;
 

lib/Target/AMDGPU/AMDGPUInstrInfo.td

 def AMDGPUpknorm_i16_f32 : SDNode<"AMDGPUISD::CVT_PKNORM_I16_F32", AMDGPUFPPackOp>;
 def AMDGPUpknorm_u16_f32 : SDNode<"AMDGPUISD::CVT_PKNORM_U16_F32", AMDGPUFPPackOp>;
 def AMDGPUpk_i16_i32 : SDNode<"AMDGPUISD::CVT_PK_I16_I32", AMDGPUIntPackOp>;
 def AMDGPUpk_u16_u32 : SDNode<"AMDGPUISD::CVT_PK_U16_U32", AMDGPUIntPackOp>;
 def AMDGPUfp_to_f16 : SDNode<"AMDGPUISD::FP_TO_FP16" , SDTFPToIntOp>;
 def AMDGPUfp16_zext : SDNode<"AMDGPUISD::FP16_ZEXT" , SDTFPToIntOp>;
 
 
-def AMDGPUfp_class : SDNode<"AMDGPUISD::FP_CLASS", AMDGPUFPClassOp>;
+def AMDGPUfp_class_impl : SDNode<"AMDGPUISD::FP_CLASS", AMDGPUFPClassOp>;
 
 // out = max(a, b) a and b are floats, where a nan comparison fails.
 // This is not commutative because this gives the second operand:
 //   x < nan ? x : nan -> nan
 //   nan < x ? nan : x -> x
 def AMDGPUfmax_legacy : SDNode<"AMDGPUISD::FMAX_LEGACY", SDTFPBinOp,
   []
 >;
def AMDGPUcos : PatFrags<(ops node:$src), [(int_amdgcn_cos node:$src),
                                            (AMDGPUcos_impl node:$src)]>;
 def AMDGPUfract : PatFrags<(ops node:$src), [(int_amdgcn_fract node:$src),
                                              (AMDGPUfract_impl node:$src)]>;
 
 def AMDGPUldexp : PatFrags<(ops node:$src0, node:$src1),
   [(int_amdgcn_ldexp node:$src0, node:$src1),
    (AMDGPUldexp_impl node:$src0, node:$src1)]>;
 
+def AMDGPUfp_class : PatFrags<(ops node:$src0, node:$src1),
+  [(int_amdgcn_class node:$src0, node:$src1),
+  (AMDGPUfp_class_impl node:$src0, node:$src1)]>;
+
 def AMDGPUfmed3 : PatFrags<(ops node:$src0, node:$src1, node:$src2),
   [(int_amdgcn_fmed3 node:$src0, node:$src1, node:$src2),
    (AMDGPUfmed3_impl node:$src0, node:$src1, node:$src2)]>;

lib/Target/AMDGPU/AMDGPUInstructionSelector.h

private:
   selectVCSRC(MachineOperand &Root) const;
 
   InstructionSelector::ComplexRendererFns
   selectVSRC0(MachineOperand &Root) const;
 
   InstructionSelector::ComplexRendererFns
   selectVOP3Mods0(MachineOperand &Root) const;
   InstructionSelector::ComplexRendererFns
+  selectVOP3Mods0Clamp0OMod(MachineOperand &Root) const;
+  InstructionSelector::ComplexRendererFns
   selectVOP3OMods(MachineOperand &Root) const;
   InstructionSelector::ComplexRendererFns
   selectVOP3Mods(MachineOperand &Root) const;
 
   InstructionSelector::ComplexRendererFns
   selectVOP3OpSelMods0(MachineOperand &Root) const;
   InstructionSelector::ComplexRendererFns
   selectVOP3OpSelMods(MachineOperand &Root) const;

lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

AMDGPUInstructionSelector::selectVOP3Mods0(MachineOperand &Root) const {
 
   return {{
       [=](MachineInstrBuilder &MIB) { MIB.addReg(Src); },
       [=](MachineInstrBuilder &MIB) { MIB.addImm(Mods); }, // src0_mods
       [=](MachineInstrBuilder &MIB) { MIB.addImm(0); },    // clamp
       [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }     // omod
   }};
 }
+
+InstructionSelector::ComplexRendererFns
+AMDGPUInstructionSelector::selectVOP3Mods0Clamp0OMod(MachineOperand &Root) const {
+  MachineRegisterInfo &MRI
+    = Root.getParent()->getParent()->getParent()->getRegInfo();
+
+  Register Src;
+  unsigned Mods;
+  std::tie(Src, Mods) = selectVOP3ModsImpl(Root.getReg(), MRI);
+
+  return {{
+      [=](MachineInstrBuilder &MIB) { MIB.addReg(Src); },
+      [=](MachineInstrBuilder &MIB) { MIB.addImm(Mods); }, // src0_mods
+      [=](MachineInstrBuilder &MIB) { MIB.addImm(0); },    // clamp
+      [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }     // omod
+  }};
+}
+
 InstructionSelector::ComplexRendererFns
 AMDGPUInstructionSelector::selectVOP3OMods(MachineOperand &Root) const {
   return {{
       [=](MachineInstrBuilder &MIB) { MIB.add(Root); },
       [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }, // clamp
       [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }  // omod
   }};
 }

lib/Target/AMDGPU/VOPCInstructions.td

 
 multiclass VOPCX_CLASS_F64 <string opName> :
   VOPCX_Class_Pseudos <opName, VOPC_I1_F64_I32, VOPC_F64_I32>;
 
 defm V_CMP_CLASS_F32 : VOPC_CLASS_F32 <"v_cmp_class_f32">;
 defm V_CMPX_CLASS_F32 : VOPCX_CLASS_F32 <"v_cmpx_class_f32">;
 defm V_CMP_CLASS_F64 : VOPC_CLASS_F64 <"v_cmp_class_f64">;
 defm V_CMPX_CLASS_F64 : VOPCX_CLASS_F64 <"v_cmpx_class_f64">;
+
+let SubtargetPredicate = Has16BitInsts in {
 defm V_CMP_CLASS_F16  : VOPC_CLASS_F16 <"v_cmp_class_f16">;
 defm V_CMPX_CLASS_F16 : VOPCX_CLASS_F16 <"v_cmpx_class_f16">;
+}
 
 //===----------------------------------------------------------------------===//
 // V_ICMPIntrinsic Pattern.
 //===----------------------------------------------------------------------===//
 
 // We need to use COPY_TO_REGCLASS to w/a the problem when ReplaceAllUsesWith()
 // complaints it cannot replace i1 <-> i64/i32 if node was not morphed in place.
 multiclass ICMP_Pattern <PatLeaf cond, Instruction inst, ValueType vt> {

test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.class.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs %s -o -  | FileCheck -check-prefix=WAVE64 %s
+# RUN: llc -march=amdgcn -mcpu=gfx1010 -run-pass=instruction-select -verify-machineinstrs %s -o -  | FileCheck -check-prefix=WAVE32 %s
+
+---
+name: class_s32_vcc_sv
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; WAVE64-LABEL: name: class_s32_vcc_sv
+    ; WAVE64: liveins: $sgpr0, $vgpr0
+    ; WAVE64: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE64: [[V_CMP_CLASS_F32_e64_:%[0-9]+]]:sreg_64 = V_CMP_CLASS_F32_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE64: S_ENDPGM 0, implicit [[V_CMP_CLASS_F32_e64_]]
+    ; WAVE32-LABEL: name: class_s32_vcc_sv
+    ; WAVE32: liveins: $sgpr0, $vgpr0
+    ; WAVE32: $vcc_hi = IMPLICIT_DEF
+    ; WAVE32: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE32: [[V_CMP_CLASS_F32_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_CLASS_F32_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE32: S_ENDPGM 0, implicit [[V_CMP_CLASS_F32_e64_]]
+    %0:sgpr(s32) = COPY $sgpr0
+    %1:vgpr(s32) = COPY $vgpr0
+    %2:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %0, %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: class_s32_vcc_vs
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; WAVE64-LABEL: name: class_s32_vcc_vs
+    ; WAVE64: liveins: $sgpr0, $vgpr0
+    ; WAVE64: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE64: [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE64: [[V_CMP_CLASS_F32_e64_:%[0-9]+]]:sreg_64 = V_CMP_CLASS_F32_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE64: S_ENDPGM 0, implicit [[V_CMP_CLASS_F32_e64_]]
+    ; WAVE32-LABEL: name: class_s32_vcc_vs
+    ; WAVE32: liveins: $sgpr0, $vgpr0
+    ; WAVE32: $vcc_hi = IMPLICIT_DEF
+    ; WAVE32: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE32: [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE32: [[V_CMP_CLASS_F32_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_CLASS_F32_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE32: S_ENDPGM 0, implicit [[V_CMP_CLASS_F32_e64_]]
+    %0:vgpr(s32) = COPY $vgpr0
+    %1:sgpr(s32) = COPY $sgpr0
+    %2:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %0, %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: class_s32_vcc_vv
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; WAVE64-LABEL: name: class_s32_vcc_vv
+    ; WAVE64: liveins: $vgpr0, $vgpr1
+    ; WAVE64: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; WAVE64: [[V_CMP_CLASS_F32_e64_:%[0-9]+]]:sreg_64 = V_CMP_CLASS_F32_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE64: S_ENDPGM 0, implicit [[V_CMP_CLASS_F32_e64_]]
+    ; WAVE32-LABEL: name: class_s32_vcc_vv
+    ; WAVE32: liveins: $vgpr0, $vgpr1
+    ; WAVE32: $vcc_hi = IMPLICIT_DEF
+    ; WAVE32: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; WAVE32: [[V_CMP_CLASS_F32_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_CLASS_F32_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE32: S_ENDPGM 0, implicit [[V_CMP_CLASS_F32_e64_]]
+    %0:vgpr(s32) = COPY $vgpr0
+    %1:vgpr(s32) = COPY $vgpr1
+    %2:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %0, %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: class_s64_vcc_sv
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1, $vgpr0
+    ; WAVE64-LABEL: name: class_s64_vcc_sv
+    ; WAVE64: liveins: $sgpr0_sgpr1, $vgpr0
+    ; WAVE64: [[COPY:%[0-9]+]]:sreg_64_xexec = COPY $sgpr0_sgpr1
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE64: [[V_CMP_CLASS_F64_e64_:%[0-9]+]]:sreg_64 = V_CMP_CLASS_F64_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE64: S_ENDPGM 0, implicit [[V_CMP_CLASS_F64_e64_]]
+    ; WAVE32-LABEL: name: class_s64_vcc_sv
+    ; WAVE32: liveins: $sgpr0_sgpr1, $vgpr0
+    ; WAVE32: $vcc_hi = IMPLICIT_DEF
+    ; WAVE32: [[COPY:%[0-9]+]]:sreg_64_xexec = COPY $sgpr0_sgpr1
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE32: [[V_CMP_CLASS_F64_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_CLASS_F64_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE32: S_ENDPGM 0, implicit [[V_CMP_CLASS_F64_e64_]]
+    %0:sgpr(s64) = COPY $sgpr0_sgpr1
+    %1:vgpr(s32) = COPY $vgpr0
+    %2:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %0, %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: class_s64_vcc_vs
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1, $vgpr0
+
+    ; WAVE64-LABEL: name: class_s64_vcc_vs
+    ; WAVE64: liveins: $sgpr0_sgpr1, $vgpr0
+    ; WAVE64: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
+    ; WAVE64: [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE64: [[V_CMP_CLASS_F64_e64_:%[0-9]+]]:sreg_64 = V_CMP_CLASS_F64_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE64: S_ENDPGM 0, implicit [[V_CMP_CLASS_F64_e64_]]
+    ; WAVE32-LABEL: name: class_s64_vcc_vs
+    ; WAVE32: liveins: $sgpr0_sgpr1, $vgpr0
+    ; WAVE32: $vcc_hi = IMPLICIT_DEF
+    ; WAVE32: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
+    ; WAVE32: [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE32: [[V_CMP_CLASS_F64_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_CLASS_F64_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE32: S_ENDPGM 0, implicit [[V_CMP_CLASS_F64_e64_]]
+    %0:vgpr(s64) = COPY $vgpr0_vgpr1
+    %1:sgpr(s32) = COPY $sgpr0
+    %2:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %0, %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: class_s64_vcc_vv
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2
+
+    ; WAVE64-LABEL: name: class_s64_vcc_vv
+    ; WAVE64: liveins: $vgpr0_vgpr1, $vgpr2
+    ; WAVE64: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+    ; WAVE64: [[V_CMP_CLASS_F64_e64_:%[0-9]+]]:sreg_64 = V_CMP_CLASS_F64_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE64: S_ENDPGM 0, implicit [[V_CMP_CLASS_F64_e64_]]
+    ; WAVE32-LABEL: name: class_s64_vcc_vv
+    ; WAVE32: liveins: $vgpr0_vgpr1, $vgpr2
+    ; WAVE32: $vcc_hi = IMPLICIT_DEF
+    ; WAVE32: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
+    ; WAVE32: [[V_CMP_CLASS_F64_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_CLASS_F64_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE32: S_ENDPGM 0, implicit [[V_CMP_CLASS_F64_e64_]]
+    %0:vgpr(s64) = COPY $vgpr0_vgpr1
+    %1:vgpr(s32) = COPY $vgpr2
+    %2:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %0, %1
+    S_ENDPGM 0, implicit %2
+...

test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.class.s16.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck -check-prefix=WAVE32 %s
+# RUN: llc -march=amdgcn -mcpu=gfx1010 -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck -check-prefix=WAVE64 %s
+
+# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=2 -pass-remarks-missed='gisel*'  %s -o /dev/null 2>&1 | FileCheck -check-prefix=SI-ERR %s
+
+# SI-ERR-NOT: remark
+# SI-ERR: remark: <unknown>:0:0: cannot select: %3:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %2:sgpr(s16), %1:vgpr(s32) (in function: class_s16_vcc_sv)
+# SI-ERR-NEXT: remark: <unknown>:0:0: cannot select: %3:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %2:vgpr(s16), %1:sgpr(s32) (in function: class_s16_vcc_vs)
+# SI-ERR-NEXT: remark: <unknown>:0:0: cannot select: %3:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %2:vgpr(s16), %1:vgpr(s32) (in function: class_s16_vcc_vv)
+# SI-ERR-NOT: remark
+
+---
+name: class_s16_vcc_sv
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; WAVE32-LABEL: name: class_s16_vcc_sv
+    ; WAVE32: liveins: $sgpr0, $vgpr0
+    ; WAVE32: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE32: [[V_CMP_CLASS_F16_e64_:%[0-9]+]]:sreg_64 = V_CMP_CLASS_F16_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE32: S_ENDPGM 0, implicit [[V_CMP_CLASS_F16_e64_]]
+    ; WAVE64-LABEL: name: class_s16_vcc_sv
+    ; WAVE64: liveins: $sgpr0, $vgpr0
+    ; WAVE64: $vcc_hi = IMPLICIT_DEF
+    ; WAVE64: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE64: [[V_CMP_CLASS_F16_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_CLASS_F16_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE64: S_ENDPGM 0, implicit [[V_CMP_CLASS_F16_e64_]]
+    %0:sgpr(s32) = COPY $sgpr0
+    %1:vgpr(s32) = COPY $vgpr0
+    %2:sgpr(s16) = G_TRUNC %0
+    %4:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %2, %1
+    S_ENDPGM 0, implicit %4
+...
+
+---
+name: class_s16_vcc_vs
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; WAVE32-LABEL: name: class_s16_vcc_vs
+    ; WAVE32: liveins: $sgpr0, $vgpr0
+    ; WAVE32: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE32: [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE32: [[V_CMP_CLASS_F16_e64_:%[0-9]+]]:sreg_64 = V_CMP_CLASS_F16_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE32: S_ENDPGM 0, implicit [[V_CMP_CLASS_F16_e64_]]
+    ; WAVE64-LABEL: name: class_s16_vcc_vs
+    ; WAVE64: liveins: $sgpr0, $vgpr0
+    ; WAVE64: $vcc_hi = IMPLICIT_DEF
+    ; WAVE64: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE64: [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; WAVE64: [[V_CMP_CLASS_F16_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_CLASS_F16_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE64: S_ENDPGM 0, implicit [[V_CMP_CLASS_F16_e64_]]
+    %0:vgpr(s32) = COPY $vgpr0
+    %1:sgpr(s32) = COPY $sgpr0
+    %2:vgpr(s16) = G_TRUNC %0
+    %4:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %2, %1
+    S_ENDPGM 0, implicit %4
+...
+
+---
+name: class_s16_vcc_vv
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; WAVE32-LABEL: name: class_s16_vcc_vv
+    ; WAVE32: liveins: $vgpr0, $vgpr1
+    ; WAVE32: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE32: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; WAVE32: [[V_CMP_CLASS_F16_e64_:%[0-9]+]]:sreg_64 = V_CMP_CLASS_F16_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE32: S_ENDPGM 0, implicit [[V_CMP_CLASS_F16_e64_]]
+    ; WAVE64-LABEL: name: class_s16_vcc_vv
+    ; WAVE64: liveins: $vgpr0, $vgpr1
+    ; WAVE64: $vcc_hi = IMPLICIT_DEF
+    ; WAVE64: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; WAVE64: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; WAVE64: [[V_CMP_CLASS_F16_e64_:%[0-9]+]]:sreg_32_xm0 = V_CMP_CLASS_F16_e64 0, [[COPY]], [[COPY1]], implicit $exec
+    ; WAVE64: S_ENDPGM 0, implicit [[V_CMP_CLASS_F16_e64_]]
+    %0:vgpr(s32) = COPY $vgpr0
+    %1:vgpr(s32) = COPY $vgpr1
+    %2:vgpr(s16) = G_TRUNC %0
+    %4:vcc(s1) = G_INTRINSIC intrinsic(@llvm.amdgcn.class), %2, %1
+    S_ENDPGM 0, implicit %4
+...