[AArch64][SVE] Implement floating-point comparison & reduction intrinsics

[kmclaughlin]

Summary

Adds intrinsics for the following:

• fadda & faddv
• fminv, fmaxv, fminnmv & fmaxnmv
• facge & facgt
• fcmp[eq|ge|gt|ne|uo]

[sdesmalen]

Thanks, LGTM!

llvm/include/llvm/IR/IntrinsicsAArch64.td

let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
   class AdvSIMD_3VectorArgIndexed_Intrinsic
     : Intrinsic<[llvm_anyvector_ty],
                 [LLVMMatchType<0>,
                  LLVMMatchType<0>,
                  LLVMMatchType<0>,
                  llvm_i32_ty],
                 [IntrNoMem]>;
 
+  class AdvSIMD_SVE_Compare_Intrinsic
+    : Intrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>],
+                [LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                 llvm_anyvector_ty,
+                 LLVMMatchType<0>],
+                [IntrNoMem]>;
+
   class AdvSIMD_SVE_CNT_Intrinsic
     : Intrinsic<[LLVMVectorOfBitcastsToInt<0>],
                 [LLVMVectorOfBitcastsToInt<0>,
                  LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
                  llvm_anyvector_ty],
                 [IntrNoMem]>;
 
+  class AdvSIMD_SVE_Reduce_Intrinsic
+    : Intrinsic<[llvm_anyfloat_ty],
+                [LLVMScalarOrSameVectorWidth<1, llvm_i1_ty>,
+                 llvm_anyvector_ty],
+                [IntrNoMem]>;
+
+  class AdvSIMD_SVE_ReduceWithInit_Intrinsic
+    : Intrinsic<[llvm_anyfloat_ty],
+                [LLVMScalarOrSameVectorWidth<1, llvm_i1_ty>,
+                 LLVMMatchType<0>,
+                 llvm_anyvector_ty],
+                [IntrNoMem]>;
+
   class AdvSIMD_SVE_Unpack_Intrinsic
     : Intrinsic<[llvm_anyvector_ty],
                [LLVMSubdivide2VectorType<0>],
                [IntrNoMem]>;
 
   class AdvSIMD_SVE_CADD_Intrinsic
     : Intrinsic<[llvm_anyvector_ty],
                 [LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
                  LLVMMatchType<0>,
                  LLVMMatchType<0>,

 def int_aarch64_sve_fsqrt      : AdvSIMD_Merged1VectorArg_Intrinsic;
 def int_aarch64_sve_fsub       : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_fsubr      : AdvSIMD_Pred2VectorArg_Intrinsic;
 def int_aarch64_sve_ftmad_x    : AdvSIMD_2VectorArgIndexed_Intrinsic;
 def int_aarch64_sve_ftsmul_x   : AdvSIMD_SVE_TSMUL_Intrinsic;
 def int_aarch64_sve_ftssel_x   : AdvSIMD_SVE_TSMUL_Intrinsic;
 
 //
+// Floating-point reductions
+//
+
+def int_aarch64_sve_fadda   : AdvSIMD_SVE_ReduceWithInit_Intrinsic;
+def int_aarch64_sve_faddv   : AdvSIMD_SVE_Reduce_Intrinsic;
+def int_aarch64_sve_fmaxv   : AdvSIMD_SVE_Reduce_Intrinsic;
+def int_aarch64_sve_fmaxnmv : AdvSIMD_SVE_Reduce_Intrinsic;
+def int_aarch64_sve_fminv   : AdvSIMD_SVE_Reduce_Intrinsic;
+def int_aarch64_sve_fminnmv : AdvSIMD_SVE_Reduce_Intrinsic;
+
+//
 // Floating-point comparisons
 //
 
+def int_aarch64_sve_facge : AdvSIMD_SVE_Compare_Intrinsic;
+def int_aarch64_sve_facgt : AdvSIMD_SVE_Compare_Intrinsic;
+
+def int_aarch64_sve_fcmpeq : AdvSIMD_SVE_Compare_Intrinsic;
+def int_aarch64_sve_fcmpge : AdvSIMD_SVE_Compare_Intrinsic;
+def int_aarch64_sve_fcmpgt : AdvSIMD_SVE_Compare_Intrinsic;
+def int_aarch64_sve_fcmpne : AdvSIMD_SVE_Compare_Intrinsic;
+def int_aarch64_sve_fcmpuo : AdvSIMD_SVE_Compare_Intrinsic;
+
 def int_aarch64_sve_fcvtzs_i32f16 : Builtin_SVCVT<"svcvt_s32_f16_m", llvm_nxv4i32_ty, llvm_nxv8f16_ty>;
 
 //
 // Predicate operations
 //
 
 def int_aarch64_sve_punpkhi : AdvSIMD_SVE_PUNPKHI_Intrinsic;
 def int_aarch64_sve_punpklo : AdvSIMD_SVE_PUNPKHI_Intrinsic;
 }

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

let Predicates = [HasSVE] in {
 
   defm FMLA_ZZZI : sve_fp_fma_by_indexed_elem<0b0, "fmla", int_aarch64_sve_fmla_lane>;
   defm FMLS_ZZZI : sve_fp_fma_by_indexed_elem<0b1, "fmls", int_aarch64_sve_fmls_lane>;
 
   defm FCMLA_ZZZI : sve_fp_fcmla_by_indexed_elem<"fcmla", int_aarch64_sve_fcmla_lane>;
   defm FMUL_ZZZI   : sve_fp_fmul_by_indexed_elem<"fmul", int_aarch64_sve_fmul_lane>;
 
   // SVE floating point reductions.
-  defm FADDA_VPZ   : sve_fp_2op_p_vd<0b000, "fadda">;
-  defm FADDV_VPZ   : sve_fp_fast_red<0b000, "faddv">;
-  defm FMAXNMV_VPZ : sve_fp_fast_red<0b100, "fmaxnmv">;
-  defm FMINNMV_VPZ : sve_fp_fast_red<0b101, "fminnmv">;
-  defm FMAXV_VPZ   : sve_fp_fast_red<0b110, "fmaxv">;
-  defm FMINV_VPZ   : sve_fp_fast_red<0b111, "fminv">;
+  defm FADDA_VPZ   : sve_fp_2op_p_vd<0b000, "fadda", int_aarch64_sve_fadda>;
+  defm FADDV_VPZ   : sve_fp_fast_red<0b000, "faddv", int_aarch64_sve_faddv>;
+  defm FMAXNMV_VPZ : sve_fp_fast_red<0b100, "fmaxnmv", int_aarch64_sve_fmaxnmv>;
+  defm FMINNMV_VPZ : sve_fp_fast_red<0b101, "fminnmv", int_aarch64_sve_fminnmv>;
+  defm FMAXV_VPZ   : sve_fp_fast_red<0b110, "fmaxv", int_aarch64_sve_fmaxv>;
+  defm FMINV_VPZ   : sve_fp_fast_red<0b111, "fminv", int_aarch64_sve_fminv>;
 
   // Splat immediate (unpredicated)
   defm DUP_ZI   : sve_int_dup_imm<"dup">;
   defm FDUP_ZI  : sve_int_dup_fpimm<"fdup">;
   defm DUPM_ZI : sve_int_dup_mask_imm<"dupm">;
 
   // Splat immediate (predicated)
   defm CPY_ZPmI  : sve_int_dup_imm_pred_merge<"cpy">;
let Predicates = [HasSVE] in {
   defm CMPLE_PPzZI : sve_int_scmp_vi<0b011, "cmple">;
   defm CMPEQ_PPzZI : sve_int_scmp_vi<0b100, "cmpeq">;
   defm CMPNE_PPzZI : sve_int_scmp_vi<0b101, "cmpne">;
   defm CMPHS_PPzZI : sve_int_ucmp_vi<0b00, "cmphs">;
   defm CMPHI_PPzZI : sve_int_ucmp_vi<0b01, "cmphi">;
   defm CMPLO_PPzZI : sve_int_ucmp_vi<0b10, "cmplo">;
   defm CMPLS_PPzZI : sve_int_ucmp_vi<0b11, "cmpls">;
 
-  defm FCMGE_PPzZZ : sve_fp_3op_p_pd<0b000, "fcmge">;
-  defm FCMGT_PPzZZ : sve_fp_3op_p_pd<0b001, "fcmgt">;
-  defm FCMEQ_PPzZZ : sve_fp_3op_p_pd<0b010, "fcmeq">;
-  defm FCMNE_PPzZZ : sve_fp_3op_p_pd<0b011, "fcmne">;
-  defm FCMUO_PPzZZ : sve_fp_3op_p_pd<0b100, "fcmuo">;
-  defm FACGE_PPzZZ : sve_fp_3op_p_pd<0b101, "facge">;
-  defm FACGT_PPzZZ : sve_fp_3op_p_pd<0b111, "facgt">;
+  defm FCMGE_PPzZZ : sve_fp_3op_p_pd<0b000, "fcmge", int_aarch64_sve_fcmpge>;
+  defm FCMGT_PPzZZ : sve_fp_3op_p_pd<0b001, "fcmgt", int_aarch64_sve_fcmpgt>;
+  defm FCMEQ_PPzZZ : sve_fp_3op_p_pd<0b010, "fcmeq", int_aarch64_sve_fcmpeq>;
+  defm FCMNE_PPzZZ : sve_fp_3op_p_pd<0b011, "fcmne", int_aarch64_sve_fcmpne>;
+  defm FCMUO_PPzZZ : sve_fp_3op_p_pd<0b100, "fcmuo", int_aarch64_sve_fcmpuo>;
+  defm FACGE_PPzZZ : sve_fp_3op_p_pd<0b101, "facge", int_aarch64_sve_facge>;
+  defm FACGT_PPzZZ : sve_fp_3op_p_pd<0b111, "facgt", int_aarch64_sve_facgt>;
 
   defm FCMGE_PPzZ0 : sve_fp_2op_p_pd<0b000, "fcmge">;
   defm FCMGT_PPzZ0 : sve_fp_2op_p_pd<0b001, "fcmgt">;
   defm FCMLT_PPzZ0 : sve_fp_2op_p_pd<0b010, "fcmlt">;
   defm FCMLE_PPzZ0 : sve_fp_2op_p_pd<0b011, "fcmle">;
   defm FCMEQ_PPzZ0 : sve_fp_2op_p_pd<0b100, "fcmeq">;
   defm FCMNE_PPzZ0 : sve_fp_2op_p_pd<0b110, "fcmne">;
 

llvm/lib/Target/AArch64/SVEInstrFormats.td

: I<(outs dstRegClass:$Vd), (ins PPR3bAny:$Pg, zprty:$Zn),
   let Inst{21-19} = 0b000;
   let Inst{18-16} = opc;
   let Inst{15-13} = 0b001;
   let Inst{12-10} = Pg;
   let Inst{9-5}   = Zn;
   let Inst{4-0}   = Vd;
 }
 
-multiclass sve_fp_fast_red<bits<3> opc, string asm> {
+multiclass sve_fp_fast_red<bits<3> opc, string asm, SDPatternOperator op> {
   def _H : sve_fp_fast_red<0b01, opc, asm, ZPR16, FPR16>;
   def _S : sve_fp_fast_red<0b10, opc, asm, ZPR32, FPR32>;
   def _D : sve_fp_fast_red<0b11, opc, asm, ZPR64, FPR64>;
+
+  def : SVE_2_Op_Pat<f16, op, nxv8i1, nxv8f16, !cast<Instruction>(NAME # _H)>;
+  def : SVE_2_Op_Pat<f32, op, nxv4i1, nxv4f32, !cast<Instruction>(NAME # _S)>;
+  def : SVE_2_Op_Pat<f64, op, nxv2i1, nxv2f64, !cast<Instruction>(NAME # _D)>;
 }
 
 
 //===----------------------------------------------------------------------===//
 // SVE Floating Point Accumulating Reduction Group
 //===----------------------------------------------------------------------===//
 
 class sve_fp_2op_p_vd<bits<2> sz, bits<3> opc, string asm,
: I<(outs dstRegClass:$Vdn), (ins PPR3bAny:$Pg, dstRegClass:$_Vdn, zprty:$Zm),
   let Inst{15-13} = 0b001;
   let Inst{12-10} = Pg;
   let Inst{9-5}   = Zm;
   let Inst{4-0}   = Vdn;
 
   let Constraints = "$Vdn = $_Vdn";
 }
 
-multiclass sve_fp_2op_p_vd<bits<3> opc, string asm> {
+multiclass sve_fp_2op_p_vd<bits<3> opc, string asm, SDPatternOperator op> {
   def _H : sve_fp_2op_p_vd<0b01, opc, asm, ZPR16, FPR16>;
   def _S : sve_fp_2op_p_vd<0b10, opc, asm, ZPR32, FPR32>;
   def _D : sve_fp_2op_p_vd<0b11, opc, asm, ZPR64, FPR64>;
+
+  def : SVE_3_Op_Pat<f16, op, nxv8i1, f16, nxv8f16, !cast<Instruction>(NAME # _H)>;
+  def : SVE_3_Op_Pat<f32, op, nxv4i1, f32, nxv4f32, !cast<Instruction>(NAME # _S)>;
+  def : SVE_3_Op_Pat<f64, op, nxv2i1, f64, nxv2f64, !cast<Instruction>(NAME # _D)>;
 }
 
 //===----------------------------------------------------------------------===//
 // SVE Floating Point Compare - Vectors Group
 //===----------------------------------------------------------------------===//
 
 class sve_fp_3op_p_pd<bits<2> sz, bits<3> opc, string asm, PPRRegOp pprty,
                       ZPRRegOp zprty>
: I<(outs pprty:$Pd), (ins PPR3bAny:$Pg, zprty:$Zn, zprty:$Zm),
   let Inst{14}    = 0b1;
   let Inst{13}    = opc{1};
   let Inst{12-10} = Pg;
   let Inst{9-5}   = Zn;
   let Inst{4}     = opc{0};
   let Inst{3-0}   = Pd;
 }
 
-multiclass sve_fp_3op_p_pd<bits<3> opc, string asm> {
+multiclass sve_fp_3op_p_pd<bits<3> opc, string asm, SDPatternOperator op> {
   def _H : sve_fp_3op_p_pd<0b01, opc, asm, PPR16, ZPR16>;
   def _S : sve_fp_3op_p_pd<0b10, opc, asm, PPR32, ZPR32>;
   def _D : sve_fp_3op_p_pd<0b11, opc, asm, PPR64, ZPR64>;
+
+  def : SVE_3_Op_Pat<nxv8i1, op, nxv8i1, nxv8f16, nxv8f16, !cast<Instruction>(NAME # _H)>;
+  def : SVE_3_Op_Pat<nxv4i1, op, nxv4i1, nxv4f32, nxv4f32, !cast<Instruction>(NAME # _S)>;
+  def : SVE_3_Op_Pat<nxv2i1, op, nxv2i1, nxv2f64, nxv2f64, !cast<Instruction>(NAME # _D)>;
 }
 
 
 //===----------------------------------------------------------------------===//
 // SVE Floating Point Compare - with Zero Group
 //===----------------------------------------------------------------------===//
 
 class sve_fp_2op_p_pd<bits<2> sz, bits<3> opc, string asm, PPRRegOp pprty,

llvm/test/CodeGen/AArch64/sve-intrinsics-fp-compares.ll

+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+;
+; FACGE
+;
+
+define <vscale x 8 x i1> @facge_h(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %b) {
+; CHECK-LABEL: facge_h:
+; CHECK: facge p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT: ret
+  %out = call <vscale x 8 x i1> @llvm.aarch64.sve.facge.nxv8f16(<vscale x 8 x i1> %pg,
+                                                                <vscale x 8 x half> %a,
+                                                                <vscale x 8 x half> %b)
+  ret <vscale x 8 x i1> %out
+}
+
+define <vscale x 4 x i1> @facge_s(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: facge_s:
+; CHECK: facge p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT: ret
+  %out = call <vscale x 4 x i1> @llvm.aarch64.sve.facge.nxv4f32(<vscale x 4 x i1> %pg,
+                                                                <vscale x 4 x float> %a,
+                                                                <vscale x 4 x float> %b)
+  ret <vscale x 4 x i1> %out
+}
+
+define <vscale x 2 x i1> @facge_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: facge_d:
+; CHECK: facge p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT: ret
+  %out = call <vscale x 2 x i1> @llvm.aarch64.sve.facge.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                <vscale x 2 x double> %a,
+                                                                <vscale x 2 x double> %b)
+  ret <vscale x 2 x i1> %out
+}
+
+;
+; FACGT
+;
+
+define <vscale x 8 x i1> @facgt_h(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %b) {
+; CHECK-LABEL: facgt_h:
+; CHECK: facgt p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT: ret
+  %out = call <vscale x 8 x i1> @llvm.aarch64.sve.facgt.nxv8f16(<vscale x 8 x i1> %pg,
+                                                                <vscale x 8 x half> %a,
+                                                                <vscale x 8 x half> %b)
+  ret <vscale x 8 x i1> %out
+}
+
+define <vscale x 4 x i1> @facgt_s(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: facgt_s:
+; CHECK: facgt p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT: ret
+  %out = call <vscale x 4 x i1> @llvm.aarch64.sve.facgt.nxv4f32(<vscale x 4 x i1> %pg,
+                                                                <vscale x 4 x float> %a,
+                                                                <vscale x 4 x float> %b)
+  ret <vscale x 4 x i1> %out
+}
+
+define <vscale x 2 x i1> @facgt_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: facgt_d:
+; CHECK: facgt p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT: ret
+  %out = call <vscale x 2 x i1> @llvm.aarch64.sve.facgt.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                <vscale x 2 x double> %a,
+                                                                <vscale x 2 x double> %b)
+  ret <vscale x 2 x i1> %out
+}
+
+;
+; FCMEQ
+;
+
+define <vscale x 8 x i1> @fcmeq_h(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %b) {
+; CHECK-LABEL: fcmeq_h:
+; CHECK: fcmeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT: ret
+  %out = call <vscale x 8 x i1> @llvm.aarch64.sve.fcmpeq.nxv8f16(<vscale x 8 x i1> %pg,
+                                                                 <vscale x 8 x half> %a,
+                                                                 <vscale x 8 x half> %b)
+  ret <vscale x 8 x i1> %out
+}
+
+define <vscale x 4 x i1> @fcmeq_s(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: fcmeq_s:
+; CHECK: fcmeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT: ret
+  %out = call <vscale x 4 x i1> @llvm.aarch64.sve.fcmpeq.nxv4f32(<vscale x 4 x i1> %pg,
+                                                                 <vscale x 4 x float> %a,
+                                                                 <vscale x 4 x float> %b)
+  ret <vscale x 4 x i1> %out
+}
+
+define <vscale x 2 x i1> @fcmeq_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: fcmeq_d:
+; CHECK: fcmeq p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT: ret
+  %out = call <vscale x 2 x i1> @llvm.aarch64.sve.fcmpeq.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                 <vscale x 2 x double> %a,
+                                                                 <vscale x 2 x double> %b)
+  ret <vscale x 2 x i1> %out
+}
+
+;
+; FCMGE
+;
+
+define <vscale x 8 x i1> @fcmge_h(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %b) {
+; CHECK-LABEL: fcmge_h:
+; CHECK: fcmge p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT: ret
+  %out = call <vscale x 8 x i1> @llvm.aarch64.sve.fcmpge.nxv8f16(<vscale x 8 x i1> %pg,
+                                                                 <vscale x 8 x half> %a,
+                                                                 <vscale x 8 x half> %b)
+  ret <vscale x 8 x i1> %out
+}
+
+define <vscale x 4 x i1> @fcmge_s(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: fcmge_s:
+; CHECK: fcmge p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT: ret
+  %out = call <vscale x 4 x i1> @llvm.aarch64.sve.fcmpge.nxv4f32(<vscale x 4 x i1> %pg,
+                                                                 <vscale x 4 x float> %a,
+                                                                 <vscale x 4 x float> %b)
+  ret <vscale x 4 x i1> %out
+}
+
+define <vscale x 2 x i1> @fcmge_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: fcmge_d:
+; CHECK: fcmge p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT: ret
+  %out = call <vscale x 2 x i1> @llvm.aarch64.sve.fcmpge.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                 <vscale x 2 x double> %a,
+                                                                 <vscale x 2 x double> %b)
+  ret <vscale x 2 x i1> %out
+}
+
+;
+; FCMGT
+;
+
+define <vscale x 8 x i1> @fcmgt_h(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %b) {
+; CHECK-LABEL: fcmgt_h:
+; CHECK: fcmgt p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT: ret
+  %out = call <vscale x 8 x i1> @llvm.aarch64.sve.fcmpgt.nxv8f16(<vscale x 8 x i1> %pg,
+                                                                 <vscale x 8 x half> %a,
+                                                                 <vscale x 8 x half> %b)
+  ret <vscale x 8 x i1> %out
+}
+
+define <vscale x 4 x i1> @fcmgt_s(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: fcmgt_s:
+; CHECK: fcmgt p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT: ret
+  %out = call <vscale x 4 x i1> @llvm.aarch64.sve.fcmpgt.nxv4f32(<vscale x 4 x i1> %pg,
+                                                                 <vscale x 4 x float> %a,
+                                                                 <vscale x 4 x float> %b)
+  ret <vscale x 4 x i1> %out
+}
+
+define <vscale x 2 x i1> @fcmgt_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: fcmgt_d:
+; CHECK: fcmgt p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT: ret
+  %out = call <vscale x 2 x i1> @llvm.aarch64.sve.fcmpgt.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                 <vscale x 2 x double> %a,
+                                                                 <vscale x 2 x double> %b)
+  ret <vscale x 2 x i1> %out
+}
+
+;
+; FCMNE
+;
+
+define <vscale x 8 x i1> @fcmne_h(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %b) {
+; CHECK-LABEL: fcmne_h:
+; CHECK: fcmne p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT: ret
+  %out = call <vscale x 8 x i1> @llvm.aarch64.sve.fcmpne.nxv8f16(<vscale x 8 x i1> %pg,
+                                                                 <vscale x 8 x half> %a,
+                                                                 <vscale x 8 x half> %b)
+  ret <vscale x 8 x i1> %out
+}
+
+define <vscale x 4 x i1> @fcmne_s(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: fcmne_s:
+; CHECK: fcmne p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT: ret
+  %out = call <vscale x 4 x i1> @llvm.aarch64.sve.fcmpne.nxv4f32(<vscale x 4 x i1> %pg,
+                                                                 <vscale x 4 x float> %a,
+                                                                 <vscale x 4 x float> %b)
+  ret <vscale x 4 x i1> %out
+}
+
+define <vscale x 2 x i1> @fcmne_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: fcmne_d:
+; CHECK: fcmne p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT: ret
+  %out = call <vscale x 2 x i1> @llvm.aarch64.sve.fcmpne.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                 <vscale x 2 x double> %a,
+                                                                 <vscale x 2 x double> %b)
+  ret <vscale x 2 x i1> %out
+}
+
+;
+; FCMPUO
+;
+
+define <vscale x 8 x i1> @fcmuo_h(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %b) {
+; CHECK-LABEL: fcmuo_h:
+; CHECK: fcmuo p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT: ret
+  %out = call <vscale x 8 x i1> @llvm.aarch64.sve.fcmpuo.nxv8f16(<vscale x 8 x i1> %pg,
+                                                                 <vscale x 8 x half> %a,
+                                                                 <vscale x 8 x half> %b)
+  ret <vscale x 8 x i1> %out
+}
+
+define <vscale x 4 x i1> @fcmuo_s(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a, <vscale x 4 x float> %b) {
+; CHECK-LABEL: fcmuo_s:
+; CHECK: fcmuo p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT: ret
+  %out = call <vscale x 4 x i1> @llvm.aarch64.sve.fcmpuo.nxv4f32(<vscale x 4 x i1> %pg,
+                                                                 <vscale x 4 x float> %a,
+                                                                 <vscale x 4 x float> %b)
+  ret <vscale x 4 x i1> %out
+}
+
+define <vscale x 2 x i1> @fcmuo_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: fcmuo_d:
+; CHECK: fcmuo p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT: ret
+  %out = call <vscale x 2 x i1> @llvm.aarch64.sve.fcmpuo.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                 <vscale x 2 x double> %a,
+                                                                 <vscale x 2 x double> %b)
+  ret <vscale x 2 x i1> %out
+}
+
+declare <vscale x 8 x i1> @llvm.aarch64.sve.facge.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>, <vscale x 8 x half>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.facge.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>, <vscale x 4 x float>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.facge.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+
+declare <vscale x 8 x i1> @llvm.aarch64.sve.facgt.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>, <vscale x 8 x half>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.facgt.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>, <vscale x 4 x float>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.facgt.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+
+declare <vscale x 8 x i1> @llvm.aarch64.sve.fcmpeq.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>, <vscale x 8 x half>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.fcmpeq.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>, <vscale x 4 x float>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpeq.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+
+declare <vscale x 8 x i1> @llvm.aarch64.sve.fcmpge.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>, <vscale x 8 x half>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.fcmpge.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>, <vscale x 4 x float>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpge.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+
+declare <vscale x 8 x i1> @llvm.aarch64.sve.fcmpgt.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>, <vscale x 8 x half>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.fcmpgt.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>, <vscale x 4 x float>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpgt.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+
+declare <vscale x 8 x i1> @llvm.aarch64.sve.fcmpne.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>, <vscale x 8 x half>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.fcmpne.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>, <vscale x 4 x float>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpne.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+
+declare <vscale x 8 x i1> @llvm.aarch64.sve.fcmpuo.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>, <vscale x 8 x half>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.fcmpuo.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>, <vscale x 4 x float>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpuo.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)

llvm/test/CodeGen/AArch64/sve-intrinsics-fp-reduce.ll

+; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve < %s | FileCheck %s
+
+;
+; FADDA
+;
+
+define half @fadda_f16(<vscale x 8 x i1> %pg, half %init, <vscale x 8 x half> %a) {
+; CHECK-LABEL: fadda_f16:
+; CHECK: fadda h0, p0, h0, z1.h
+; CHECK-NEXT: ret
+  %res = call half @llvm.aarch64.sve.fadda.nxv8f16(<vscale x 8 x i1> %pg,
+                                                   half %init,
+                                                   <vscale x 8 x half> %a)
+  ret half %res
+}
+
+define float @fadda_f32(<vscale x 4 x i1> %pg, float %init, <vscale x 4 x float> %a) {
+; CHECK-LABEL: fadda_f32:
+; CHECK: fadda s0, p0, s0, z1.s
+; CHECK-NEXT: ret
+  %res = call float @llvm.aarch64.sve.fadda.nxv4f32(<vscale x 4 x i1> %pg,
+                                                    float %init,
+                                                    <vscale x 4 x float> %a)
+  ret float %res
+}
+
+define double @fadda_f64(<vscale x 2 x i1> %pg, double %init, <vscale x 2 x double> %a) {
+; CHECK-LABEL: fadda_f64:
+; CHECK: fadda d0, p0, d0, z1.d
+; CHECK-NEXT: ret
+  %res = call double @llvm.aarch64.sve.fadda.nxv2f64(<vscale x 2 x i1> %pg,
+                                                     double %init,
+                                                     <vscale x 2 x double> %a)
+  ret double %res
+}
+
+;
+; FADDV
+;
+
+define half @faddv_f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) {
+; CHECK-LABEL: faddv_f16:
+; CHECK: faddv h0, p0, z0.h
+; CHECK-NEXT: ret
+  %res = call half @llvm.aarch64.sve.faddv.nxv8f16(<vscale x 8 x i1> %pg,
+                                                   <vscale x 8 x half> %a)
+  ret half %res
+}
+
+define float @faddv_f32(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a) {
+; CHECK-LABEL: faddv_f32:
+; CHECK: faddv s0, p0, z0.s
+; CHECK-NEXT: ret
+  %res = call float @llvm.aarch64.sve.faddv.nxv4f32(<vscale x 4 x i1> %pg,
+                                                    <vscale x 4 x float> %a)
+  ret float %res
+}
+
+define double @faddv_f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a) {
+; CHECK-LABEL: faddv_f64:
+; CHECK: faddv d0, p0, z0.d
+; CHECK-NEXT: ret
+  %res = call double @llvm.aarch64.sve.faddv.nxv2f64(<vscale x 2 x i1> %pg,
+                                                     <vscale x 2 x double> %a)
+  ret double %res
+}
+
+;
+; FMAXNMV
+;
+
+define half @fmaxnmv_f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) {
+; CHECK-LABEL: fmaxnmv_f16:
+; CHECK: fmaxnmv h0, p0, z0.h
+; CHECK-NEXT: ret
+  %res = call half @llvm.aarch64.sve.fmaxnmv.nxv8f16(<vscale x 8 x i1> %pg,
+                                                     <vscale x 8 x half> %a)
+  ret half %res
+}
+
+define float @fmaxnmv_f32(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a) {
+; CHECK-LABEL: fmaxnmv_f32:
+; CHECK: fmaxnmv s0, p0, z0.s
+; CHECK-NEXT: ret
+  %res = call float @llvm.aarch64.sve.fmaxnmv.nxv4f32(<vscale x 4 x i1> %pg,
+                                                      <vscale x 4 x float> %a)
+  ret float %res
+}
+
+define double @fmaxnmv_f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a) {
+; CHECK-LABEL: fmaxnmv_f64:
+; CHECK: fmaxnmv d0, p0, z0.d
+; CHECK-NEXT: ret
+  %res = call double @llvm.aarch64.sve.fmaxnmv.nxv2f64(<vscale x 2 x i1> %pg,
+                                                       <vscale x 2 x double> %a)
+  ret double %res
+}
+
+;
+; FMAXV
+;
+
+define half @fmaxv_f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) {
+; CHECK-LABEL: fmaxv_f16:
+; CHECK: fmaxv h0, p0, z0.h
+; CHECK-NEXT: ret
+  %res = call half @llvm.aarch64.sve.fmaxv.nxv8f16(<vscale x 8 x i1> %pg,
+                                                   <vscale x 8 x half> %a)
+  ret half %res
+}
+
+define float @fmaxv_f32(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a) {
+; CHECK-LABEL: fmaxv_f32:
+; CHECK: fmaxv s0, p0, z0.s
+; CHECK-NEXT: ret
+  %res = call float @llvm.aarch64.sve.fmaxv.nxv4f32(<vscale x 4 x i1> %pg,
+                                                    <vscale x 4 x float> %a)
+  ret float %res
+}
+
+define double @fmaxv_f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a) {
+; CHECK-LABEL: fmaxv_f64:
+; CHECK: fmaxv d0, p0, z0.d
+; CHECK-NEXT: ret
+  %res = call double @llvm.aarch64.sve.fmaxv.nxv2f64(<vscale x 2 x i1> %pg,
+                                                     <vscale x 2 x double> %a)
+  ret double %res
+}
+
+;
+; FMINNMV
+;
+
+define half @fminnmv_f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) {
+; CHECK-LABEL: fminnmv_f16:
+; CHECK: fminnmv h0, p0, z0.h
+; CHECK-NEXT: ret
+  %res = call half @llvm.aarch64.sve.fminnmv.nxv8f16(<vscale x 8 x i1> %pg,
+                                                     <vscale x 8 x half> %a)
+  ret half %res
+}
+
+define float @fminnmv_f32(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a) {
+; CHECK-LABEL: fminnmv_f32:
+; CHECK: fminnmv s0, p0, z0.s
+; CHECK-NEXT: ret
+  %res = call float @llvm.aarch64.sve.fminnmv.nxv4f32(<vscale x 4 x i1> %pg,
+                                                      <vscale x 4 x float> %a)
+  ret float %res
+}
+
+define double @fminnmv_f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a) {
+; CHECK-LABEL: fminnmv_f64:
+; CHECK: fminnmv d0, p0, z0.d
+; CHECK-NEXT: ret
+  %res = call double @llvm.aarch64.sve.fminnmv.nxv2f64(<vscale x 2 x i1> %pg,
+                                                       <vscale x 2 x double> %a)
+  ret double %res
+}
+
+;
+; FMINV
+;
+
+define half @fminv_f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) {
+; CHECK-LABEL: fminv_f16:
+; CHECK: fminv h0, p0, z0.h
+; CHECK-NEXT: ret
+  %res = call half @llvm.aarch64.sve.fminv.nxv8f16(<vscale x 8 x i1> %pg,
+                                                   <vscale x 8 x half> %a)
+  ret half %res
+}
+
+define float @fminv_f32(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a) {
+; CHECK-LABEL: fminv_f32:
+; CHECK: fminv s0, p0, z0.s
+; CHECK-NEXT: ret
+  %res = call float @llvm.aarch64.sve.fminv.nxv4f32(<vscale x 4 x i1> %pg,
+                                                    <vscale x 4 x float> %a)
+  ret float %res
+}
+
+define double @fminv_f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a) {
+; CHECK-LABEL: fminv_f64:
+; CHECK: fminv d0, p0, z0.d
+; CHECK-NEXT: ret
+  %res = call double @llvm.aarch64.sve.fminv.nxv2f64(<vscale x 2 x i1> %pg,
+                                                     <vscale x 2 x double> %a)
+  ret double %res
+}
+
+declare half @llvm.aarch64.sve.fadda.nxv8f16(<vscale x 8 x i1>, half, <vscale x 8 x half>)
+declare float @llvm.aarch64.sve.fadda.nxv4f32(<vscale x 4 x i1>, float, <vscale x 4 x float>)
+declare double @llvm.aarch64.sve.fadda.nxv2f64(<vscale x 2 x i1>, double, <vscale x 2 x double>)
+
+declare half @llvm.aarch64.sve.faddv.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>)
+declare float @llvm.aarch64.sve.faddv.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>)
+declare double @llvm.aarch64.sve.faddv.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>)
+
+declare half @llvm.aarch64.sve.fmaxnmv.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>)
+declare float @llvm.aarch64.sve.fmaxnmv.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>)
+declare double @llvm.aarch64.sve.fmaxnmv.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>)
+
+declare half @llvm.aarch64.sve.fmaxv.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>)
+declare float @llvm.aarch64.sve.fmaxv.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>)
+declare double @llvm.aarch64.sve.fmaxv.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>)
+
+declare half @llvm.aarch64.sve.fminnmv.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>)
+declare float @llvm.aarch64.sve.fminnmv.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>)
+declare double @llvm.aarch64.sve.fminnmv.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>)
+
+declare half @llvm.aarch64.sve.fminv.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>)
+declare float @llvm.aarch64.sve.fminv.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>)
+declare double @llvm.aarch64.sve.fminv.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>)