200 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 52; RUN: llc < %s -mtriple=aarch64 -mattr=-fullfp16 -stop-after=finalize-isel | FileCheck %s --check-prefixes=CHECK-CVT3; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 -stop-after=finalize-isel | FileCheck %s --check-prefixes=CHECK-FP164 5; Check that the output instructions have the same fast math flags as the input6; fadd, even when fadd is promoted to float type.7 8define half @normal_fadd(half %x, half %y) {9 ; CHECK-CVT-LABEL: name: normal_fadd10 ; CHECK-CVT: bb.0.entry:11 ; CHECK-CVT-NEXT: liveins: $h0, $h112 ; CHECK-CVT-NEXT: {{ $}}13 ; CHECK-CVT-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h114 ; CHECK-CVT-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h015 ; CHECK-CVT-NEXT: [[FCVTSHr:%[0-9]+]]:fpr32 = nofpexcept FCVTSHr [[COPY]], implicit $fpcr16 ; CHECK-CVT-NEXT: [[FCVTSHr1:%[0-9]+]]:fpr32 = nofpexcept FCVTSHr [[COPY1]], implicit $fpcr17 ; CHECK-CVT-NEXT: [[FADDSrr:%[0-9]+]]:fpr32 = nofpexcept FADDSrr killed [[FCVTSHr1]], killed [[FCVTSHr]], implicit $fpcr18 ; CHECK-CVT-NEXT: [[FCVTHSr:%[0-9]+]]:fpr16 = nofpexcept FCVTHSr killed [[FADDSrr]], implicit $fpcr19 ; CHECK-CVT-NEXT: $h0 = COPY [[FCVTHSr]]20 ; CHECK-CVT-NEXT: RET_ReallyLR implicit $h021 ;22 ; CHECK-FP16-LABEL: name: normal_fadd23 ; CHECK-FP16: bb.0.entry:24 ; CHECK-FP16-NEXT: liveins: $h0, $h125 ; CHECK-FP16-NEXT: {{ $}}26 ; CHECK-FP16-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h127 ; CHECK-FP16-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h028 ; CHECK-FP16-NEXT: [[FADDHrr:%[0-9]+]]:fpr16 = nofpexcept FADDHrr [[COPY1]], [[COPY]], implicit $fpcr29 ; CHECK-FP16-NEXT: $h0 = COPY [[FADDHrr]]30 ; CHECK-FP16-NEXT: RET_ReallyLR implicit $h031entry:32 %add = fadd half %x, %y33 ret half %add34}35 36define half @fast_fadd(half %x, half %y) {37 ; CHECK-CVT-LABEL: name: fast_fadd38 ; CHECK-CVT: bb.0.entry:39 ; CHECK-CVT-NEXT: liveins: $h0, $h140 ; CHECK-CVT-NEXT: {{ $}}41 ; CHECK-CVT-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h142 ; CHECK-CVT-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h043 ; CHECK-CVT-NEXT: [[FCVTSHr:%[0-9]+]]:fpr32 = nnan ninf nsz arcp contract afn reassoc nofpexcept FCVTSHr [[COPY]], implicit $fpcr44 ; CHECK-CVT-NEXT: [[FCVTSHr1:%[0-9]+]]:fpr32 = nnan ninf nsz arcp contract afn reassoc nofpexcept FCVTSHr [[COPY1]], implicit $fpcr45 ; CHECK-CVT-NEXT: [[FADDSrr:%[0-9]+]]:fpr32 = nnan ninf nsz arcp contract afn reassoc nofpexcept FADDSrr killed [[FCVTSHr1]], killed [[FCVTSHr]], implicit $fpcr46 ; CHECK-CVT-NEXT: [[FCVTHSr:%[0-9]+]]:fpr16 = nnan ninf nsz arcp contract afn reassoc nofpexcept FCVTHSr killed [[FADDSrr]], implicit $fpcr47 ; CHECK-CVT-NEXT: $h0 = COPY [[FCVTHSr]]48 ; CHECK-CVT-NEXT: RET_ReallyLR implicit $h049 ;50 ; CHECK-FP16-LABEL: name: fast_fadd51 ; CHECK-FP16: bb.0.entry:52 ; CHECK-FP16-NEXT: liveins: $h0, $h153 ; CHECK-FP16-NEXT: {{ $}}54 ; CHECK-FP16-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h155 ; CHECK-FP16-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h056 ; CHECK-FP16-NEXT: [[FADDHrr:%[0-9]+]]:fpr16 = nnan ninf nsz arcp contract afn reassoc nofpexcept FADDHrr [[COPY1]], [[COPY]], implicit $fpcr57 ; CHECK-FP16-NEXT: $h0 = COPY [[FADDHrr]]58 ; CHECK-FP16-NEXT: RET_ReallyLR implicit $h059entry:60 %add = fadd fast half %x, %y61 ret half %add62}63 64define half @ninf_fadd(half %x, half %y) {65 ; CHECK-CVT-LABEL: name: ninf_fadd66 ; CHECK-CVT: bb.0.entry:67 ; CHECK-CVT-NEXT: liveins: $h0, $h168 ; CHECK-CVT-NEXT: {{ $}}69 ; CHECK-CVT-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h170 ; CHECK-CVT-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h071 ; CHECK-CVT-NEXT: [[FCVTSHr:%[0-9]+]]:fpr32 = ninf nofpexcept FCVTSHr [[COPY]], implicit $fpcr72 ; CHECK-CVT-NEXT: [[FCVTSHr1:%[0-9]+]]:fpr32 = ninf nofpexcept FCVTSHr [[COPY1]], implicit $fpcr73 ; CHECK-CVT-NEXT: [[FADDSrr:%[0-9]+]]:fpr32 = ninf nofpexcept FADDSrr killed [[FCVTSHr1]], killed [[FCVTSHr]], implicit $fpcr74 ; CHECK-CVT-NEXT: [[FCVTHSr:%[0-9]+]]:fpr16 = ninf nofpexcept FCVTHSr killed [[FADDSrr]], implicit $fpcr75 ; CHECK-CVT-NEXT: $h0 = COPY [[FCVTHSr]]76 ; CHECK-CVT-NEXT: RET_ReallyLR implicit $h077 ;78 ; CHECK-FP16-LABEL: name: ninf_fadd79 ; CHECK-FP16: bb.0.entry:80 ; CHECK-FP16-NEXT: liveins: $h0, $h181 ; CHECK-FP16-NEXT: {{ $}}82 ; CHECK-FP16-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h183 ; CHECK-FP16-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h084 ; CHECK-FP16-NEXT: [[FADDHrr:%[0-9]+]]:fpr16 = ninf nofpexcept FADDHrr [[COPY1]], [[COPY]], implicit $fpcr85 ; CHECK-FP16-NEXT: $h0 = COPY [[FADDHrr]]86 ; CHECK-FP16-NEXT: RET_ReallyLR implicit $h087entry:88 %add = fadd ninf half %x, %y89 ret half %add90}91 92; Check that when we have the right fast math flags the converts in between the93; two fadds are removed.94 95define half @normal_fadd_sequence(half %x, half %y, half %z) {96 ; CHECK-CVT-LABEL: name: normal_fadd_sequence97 ; CHECK-CVT: bb.0.entry:98 ; CHECK-CVT-NEXT: liveins: $h0, $h1, $h299 ; CHECK-CVT-NEXT: {{ $}}100 ; CHECK-CVT-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h2101 ; CHECK-CVT-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h1102 ; CHECK-CVT-NEXT: [[COPY2:%[0-9]+]]:fpr16 = COPY $h0103 ; CHECK-CVT-NEXT: [[FCVTSHr:%[0-9]+]]:fpr32 = nofpexcept FCVTSHr [[COPY1]], implicit $fpcr104 ; CHECK-CVT-NEXT: [[FCVTSHr1:%[0-9]+]]:fpr32 = nofpexcept FCVTSHr [[COPY2]], implicit $fpcr105 ; CHECK-CVT-NEXT: [[FADDSrr:%[0-9]+]]:fpr32 = nofpexcept FADDSrr killed [[FCVTSHr1]], killed [[FCVTSHr]], implicit $fpcr106 ; CHECK-CVT-NEXT: [[FCVTHSr:%[0-9]+]]:fpr16 = nofpexcept FCVTHSr killed [[FADDSrr]], implicit $fpcr107 ; CHECK-CVT-NEXT: [[FCVTSHr2:%[0-9]+]]:fpr32 = nofpexcept FCVTSHr killed [[FCVTHSr]], implicit $fpcr108 ; CHECK-CVT-NEXT: [[FCVTSHr3:%[0-9]+]]:fpr32 = nofpexcept FCVTSHr [[COPY]], implicit $fpcr109 ; CHECK-CVT-NEXT: [[FADDSrr1:%[0-9]+]]:fpr32 = nofpexcept FADDSrr killed [[FCVTSHr2]], killed [[FCVTSHr3]], implicit $fpcr110 ; CHECK-CVT-NEXT: [[FCVTHSr1:%[0-9]+]]:fpr16 = nofpexcept FCVTHSr killed [[FADDSrr1]], implicit $fpcr111 ; CHECK-CVT-NEXT: $h0 = COPY [[FCVTHSr1]]112 ; CHECK-CVT-NEXT: RET_ReallyLR implicit $h0113 ;114 ; CHECK-FP16-LABEL: name: normal_fadd_sequence115 ; CHECK-FP16: bb.0.entry:116 ; CHECK-FP16-NEXT: liveins: $h0, $h1, $h2117 ; CHECK-FP16-NEXT: {{ $}}118 ; CHECK-FP16-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h2119 ; CHECK-FP16-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h1120 ; CHECK-FP16-NEXT: [[COPY2:%[0-9]+]]:fpr16 = COPY $h0121 ; CHECK-FP16-NEXT: [[FADDHrr:%[0-9]+]]:fpr16 = nofpexcept FADDHrr [[COPY2]], [[COPY1]], implicit $fpcr122 ; CHECK-FP16-NEXT: [[FADDHrr1:%[0-9]+]]:fpr16 = nofpexcept FADDHrr killed [[FADDHrr]], [[COPY]], implicit $fpcr123 ; CHECK-FP16-NEXT: $h0 = COPY [[FADDHrr1]]124 ; CHECK-FP16-NEXT: RET_ReallyLR implicit $h0125entry:126 %add1 = fadd half %x, %y127 %add2 = fadd half %add1, %z128 ret half %add2129}130 131define half @nnan_ninf_contract_fadd_sequence(half %x, half %y, half %z) {132 ; CHECK-CVT-LABEL: name: nnan_ninf_contract_fadd_sequence133 ; CHECK-CVT: bb.0.entry:134 ; CHECK-CVT-NEXT: liveins: $h0, $h1, $h2135 ; CHECK-CVT-NEXT: {{ $}}136 ; CHECK-CVT-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h2137 ; CHECK-CVT-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h1138 ; CHECK-CVT-NEXT: [[COPY2:%[0-9]+]]:fpr16 = COPY $h0139 ; CHECK-CVT-NEXT: [[FCVTSHr:%[0-9]+]]:fpr32 = nnan ninf contract nofpexcept FCVTSHr [[COPY1]], implicit $fpcr140 ; CHECK-CVT-NEXT: [[FCVTSHr1:%[0-9]+]]:fpr32 = nnan ninf contract nofpexcept FCVTSHr [[COPY2]], implicit $fpcr141 ; CHECK-CVT-NEXT: [[FADDSrr:%[0-9]+]]:fpr32 = nnan ninf contract nofpexcept FADDSrr killed [[FCVTSHr1]], killed [[FCVTSHr]], implicit $fpcr142 ; CHECK-CVT-NEXT: [[FCVTSHr2:%[0-9]+]]:fpr32 = nnan ninf contract nofpexcept FCVTSHr [[COPY]], implicit $fpcr143 ; CHECK-CVT-NEXT: [[FADDSrr1:%[0-9]+]]:fpr32 = nnan ninf contract nofpexcept FADDSrr killed [[FADDSrr]], killed [[FCVTSHr2]], implicit $fpcr144 ; CHECK-CVT-NEXT: [[FCVTHSr:%[0-9]+]]:fpr16 = nnan ninf contract nofpexcept FCVTHSr killed [[FADDSrr1]], implicit $fpcr145 ; CHECK-CVT-NEXT: $h0 = COPY [[FCVTHSr]]146 ; CHECK-CVT-NEXT: RET_ReallyLR implicit $h0147 ;148 ; CHECK-FP16-LABEL: name: nnan_ninf_contract_fadd_sequence149 ; CHECK-FP16: bb.0.entry:150 ; CHECK-FP16-NEXT: liveins: $h0, $h1, $h2151 ; CHECK-FP16-NEXT: {{ $}}152 ; CHECK-FP16-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h2153 ; CHECK-FP16-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h1154 ; CHECK-FP16-NEXT: [[COPY2:%[0-9]+]]:fpr16 = COPY $h0155 ; CHECK-FP16-NEXT: [[FADDHrr:%[0-9]+]]:fpr16 = nnan ninf contract nofpexcept FADDHrr [[COPY2]], [[COPY1]], implicit $fpcr156 ; CHECK-FP16-NEXT: [[FADDHrr1:%[0-9]+]]:fpr16 = nnan ninf contract nofpexcept FADDHrr killed [[FADDHrr]], [[COPY]], implicit $fpcr157 ; CHECK-FP16-NEXT: $h0 = COPY [[FADDHrr1]]158 ; CHECK-FP16-NEXT: RET_ReallyLR implicit $h0159entry:160 %add1 = fadd nnan ninf contract half %x, %y161 %add2 = fadd nnan ninf contract half %add1, %z162 ret half %add2163}164 165define half @ninf_fadd_sequence(half %x, half %y, half %z) {166 ; CHECK-CVT-LABEL: name: ninf_fadd_sequence167 ; CHECK-CVT: bb.0.entry:168 ; CHECK-CVT-NEXT: liveins: $h0, $h1, $h2169 ; CHECK-CVT-NEXT: {{ $}}170 ; CHECK-CVT-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h2171 ; CHECK-CVT-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h1172 ; CHECK-CVT-NEXT: [[COPY2:%[0-9]+]]:fpr16 = COPY $h0173 ; CHECK-CVT-NEXT: [[FCVTSHr:%[0-9]+]]:fpr32 = ninf nofpexcept FCVTSHr [[COPY1]], implicit $fpcr174 ; CHECK-CVT-NEXT: [[FCVTSHr1:%[0-9]+]]:fpr32 = ninf nofpexcept FCVTSHr [[COPY2]], implicit $fpcr175 ; CHECK-CVT-NEXT: [[FADDSrr:%[0-9]+]]:fpr32 = ninf nofpexcept FADDSrr killed [[FCVTSHr1]], killed [[FCVTSHr]], implicit $fpcr176 ; CHECK-CVT-NEXT: [[FCVTHSr:%[0-9]+]]:fpr16 = ninf nofpexcept FCVTHSr killed [[FADDSrr]], implicit $fpcr177 ; CHECK-CVT-NEXT: [[FCVTSHr2:%[0-9]+]]:fpr32 = ninf nofpexcept FCVTSHr killed [[FCVTHSr]], implicit $fpcr178 ; CHECK-CVT-NEXT: [[FCVTSHr3:%[0-9]+]]:fpr32 = ninf nofpexcept FCVTSHr [[COPY]], implicit $fpcr179 ; CHECK-CVT-NEXT: [[FADDSrr1:%[0-9]+]]:fpr32 = ninf nofpexcept FADDSrr killed [[FCVTSHr2]], killed [[FCVTSHr3]], implicit $fpcr180 ; CHECK-CVT-NEXT: [[FCVTHSr1:%[0-9]+]]:fpr16 = ninf nofpexcept FCVTHSr killed [[FADDSrr1]], implicit $fpcr181 ; CHECK-CVT-NEXT: $h0 = COPY [[FCVTHSr1]]182 ; CHECK-CVT-NEXT: RET_ReallyLR implicit $h0183 ;184 ; CHECK-FP16-LABEL: name: ninf_fadd_sequence185 ; CHECK-FP16: bb.0.entry:186 ; CHECK-FP16-NEXT: liveins: $h0, $h1, $h2187 ; CHECK-FP16-NEXT: {{ $}}188 ; CHECK-FP16-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $h2189 ; CHECK-FP16-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $h1190 ; CHECK-FP16-NEXT: [[COPY2:%[0-9]+]]:fpr16 = COPY $h0191 ; CHECK-FP16-NEXT: [[FADDHrr:%[0-9]+]]:fpr16 = ninf nofpexcept FADDHrr [[COPY2]], [[COPY1]], implicit $fpcr192 ; CHECK-FP16-NEXT: [[FADDHrr1:%[0-9]+]]:fpr16 = ninf nofpexcept FADDHrr killed [[FADDHrr]], [[COPY]], implicit $fpcr193 ; CHECK-FP16-NEXT: $h0 = COPY [[FADDHrr1]]194 ; CHECK-FP16-NEXT: RET_ReallyLR implicit $h0195entry:196 %add1 = fadd ninf half %x, %y197 %add2 = fadd ninf half %add1, %z198 ret half %add2199}200