147 lines · plain
1; REQUIRES: asserts2; RUN: opt -mattr=+neon -passes=loop-vectorize -debug-only=loop-vectorize -disable-output -S < %s 2>&1 | FileCheck %s --check-prefix=NEON-NO-VECLIB3; RUN: opt -mattr=+sve -passes=loop-vectorize -debug-only=loop-vectorize -disable-output -S < %s 2>&1 | FileCheck %s --check-prefix=SVE-NO-VECLIB4; RUN: opt -mattr=+neon -vector-library=ArmPL -passes=loop-vectorize -debug-only=loop-vectorize -disable-output -S < %s 2>&1 | FileCheck %s --check-prefix=NEON-ARMPL5; RUN: opt -mattr=+neon -vector-library=sleefgnuabi -passes=loop-vectorize -debug-only=loop-vectorize -disable-output -S < %s 2>&1 | FileCheck %s --check-prefix=NEON-SLEEF6; RUN: opt -mattr=+sve -vector-library=ArmPL -passes=loop-vectorize -debug-only=loop-vectorize -disable-output -S < %s 2>&1 | FileCheck %s --check-prefix=SVE-ARMPL7; RUN: opt -mattr=+sve -vector-library=sleefgnuabi -passes=loop-vectorize -debug-only=loop-vectorize -disable-output -S < %s 2>&1 | FileCheck %s --check-prefix=SVE-SLEEF8; RUN: opt -mattr=+sve -vector-library=ArmPL -passes=loop-vectorize -prefer-predicate-over-epilogue=predicate-dont-vectorize -debug-only=loop-vectorize -disable-output -S < %s 2>&1 | FileCheck %s --check-prefix=SVE-ARMPL-TAILFOLD9; RUN: opt -mattr=+sve -vector-library=sleefgnuabi -passes=loop-vectorize -prefer-predicate-over-epilogue=predicate-dont-vectorize -debug-only=loop-vectorize -disable-output -S < %s 2>&1 | FileCheck %s --check-prefix=SVE-SLEEF-TAILFOLD10 11target triple = "aarch64-unknown-linux-gnu"12 13define void @frem_f64(ptr noalias %in.ptr, ptr noalias %out.ptr) {14; NEON-NO-VECLIB-LABEL: 'frem_f64'15; NEON-NO-VECLIB: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem double %in, %in16; NEON-NO-VECLIB: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>17;18; SVE-NO-VECLIB-LABEL: 'frem_f64'19; SVE-NO-VECLIB: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem double %in, %in20; SVE-NO-VECLIB: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>21; SVE-NO-VECLIB: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>22; SVE-NO-VECLIB: Cost of Invalid for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>23;24; NEON-ARMPL-LABEL: 'frem_f64'25; NEON-ARMPL: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem double %in, %in26; NEON-ARMPL: Cost of 10 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>27;28; NEON-SLEEF-LABEL: 'frem_f64'29; NEON-SLEEF: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem double %in, %in30; NEON-SLEEF: Cost of 10 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>31;32; SVE-ARMPL-LABEL: 'frem_f64'33; SVE-ARMPL: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem double %in, %in34; SVE-ARMPL: Cost of 10 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>35; SVE-ARMPL: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>36; SVE-ARMPL: Cost of 10 for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>37;38; SVE-SLEEF-LABEL: 'frem_f64'39; SVE-SLEEF: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem double %in, %in40; SVE-SLEEF: Cost of 10 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>41; SVE-SLEEF: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>42; SVE-SLEEF: Cost of 10 for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>43;44; SVE-ARMPL-TAILFOLD-LABEL: 'frem_f64'45; SVE-ARMPL-TAILFOLD: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem double %in, %in46; SVE-ARMPL-TAILFOLD: Cost of 10 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>47; SVE-ARMPL-TAILFOLD: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>48; SVE-ARMPL-TAILFOLD: Cost of 10 for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>49;50; SVE-SLEEF-TAILFOLD-LABEL: 'frem_f64'51; SVE-SLEEF-TAILFOLD: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem double %in, %in52; SVE-SLEEF-TAILFOLD: Cost of 10 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>53; SVE-SLEEF-TAILFOLD: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>54; SVE-SLEEF-TAILFOLD: Cost of 10 for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>55;56 entry:57 br label %for.body58 59 for.body:60 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]61 %in.gep = getelementptr inbounds double, ptr %in.ptr, i64 %iv62 %in = load double, ptr %in.gep, align 863 %res = frem double %in, %in64 %out.gep = getelementptr inbounds double, ptr %out.ptr, i64 %iv65 store double %res, ptr %out.gep, align 866 %iv.next = add nuw nsw i64 %iv, 167 %exitcond = icmp eq i64 %iv.next, 100068 br i1 %exitcond, label %for.end, label %for.body69 70 for.end:71 ret void72}73 74define void @frem_f32(ptr noalias %in.ptr, ptr noalias %out.ptr) {75; NEON-NO-VECLIB-LABEL: 'frem_f32'76; NEON-NO-VECLIB: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem float %in, %in77; NEON-NO-VECLIB: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>78; NEON-NO-VECLIB: Cost of 52 for VF 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>79;80; SVE-NO-VECLIB-LABEL: 'frem_f32'81; SVE-NO-VECLIB: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem float %in, %in82; SVE-NO-VECLIB: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>83; SVE-NO-VECLIB: Cost of 52 for VF 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>84; SVE-NO-VECLIB: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>85; SVE-NO-VECLIB: Cost of Invalid for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>86; SVE-NO-VECLIB: Cost of Invalid for VF vscale x 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>87;88; NEON-ARMPL-LABEL: 'frem_f32'89; NEON-ARMPL: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem float %in, %in90; NEON-ARMPL: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>91; NEON-ARMPL: Cost of 10 for VF 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>92;93; NEON-SLEEF-LABEL: 'frem_f32'94; NEON-SLEEF: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem float %in, %in95; NEON-SLEEF: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>96; NEON-SLEEF: Cost of 10 for VF 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>97;98; SVE-ARMPL-LABEL: 'frem_f32'99; SVE-ARMPL: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem float %in, %in100; SVE-ARMPL: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>101; SVE-ARMPL: Cost of 10 for VF 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>102; SVE-ARMPL: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>103; SVE-ARMPL: Cost of Invalid for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>104; SVE-ARMPL: Cost of 10 for VF vscale x 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>105;106; SVE-SLEEF-LABEL: 'frem_f32'107; SVE-SLEEF: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem float %in, %in108; SVE-SLEEF: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>109; SVE-SLEEF: Cost of 10 for VF 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>110; SVE-SLEEF: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>111; SVE-SLEEF: Cost of Invalid for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>112; SVE-SLEEF: Cost of 10 for VF vscale x 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>113;114; SVE-ARMPL-TAILFOLD-LABEL: 'frem_f32'115; SVE-ARMPL-TAILFOLD: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem float %in, %in116; SVE-ARMPL-TAILFOLD: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>117; SVE-ARMPL-TAILFOLD: Cost of 10 for VF 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>118; SVE-ARMPL-TAILFOLD: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>119; SVE-ARMPL-TAILFOLD: Cost of Invalid for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>120; SVE-ARMPL-TAILFOLD: Cost of 10 for VF vscale x 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>121;122; SVE-SLEEF-TAILFOLD-LABEL: 'frem_f32'123; SVE-SLEEF-TAILFOLD: LV: Found an estimated cost of 10 for VF 1 For instruction: %res = frem float %in, %in124; SVE-SLEEF-TAILFOLD: Cost of 24 for VF 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>125; SVE-SLEEF-TAILFOLD: Cost of 10 for VF 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>126; SVE-SLEEF-TAILFOLD: Cost of Invalid for VF vscale x 1: WIDEN ir<%res> = frem ir<%in>, ir<%in>127; SVE-SLEEF-TAILFOLD: Cost of Invalid for VF vscale x 2: WIDEN ir<%res> = frem ir<%in>, ir<%in>128; SVE-SLEEF-TAILFOLD: Cost of 10 for VF vscale x 4: WIDEN ir<%res> = frem ir<%in>, ir<%in>129;130 entry:131 br label %for.body132 133 for.body:134 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]135 %in.gep = getelementptr inbounds float, ptr %in.ptr, i64 %iv136 %in = load float, ptr %in.gep, align 8137 %res = frem float %in, %in138 %out.gep = getelementptr inbounds float, ptr %out.ptr, i64 %iv139 store float %res, ptr %out.gep, align 4140 %iv.next = add nuw nsw i64 %iv, 1141 %exitcond = icmp eq i64 %iv.next, 1000142 br i1 %exitcond, label %for.end, label %for.body143 144 for.end:145 ret void146}147