1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2 ; RUN: opt < %s -indvars -S | FileCheck %s 3 4 target datalayout = "e-m:e-i64:64-p:64:64:64-n8:16:32:64-S128" 5 6 ; When widening IV and its users, trunc and zext/sext are not needed 7 ; if the original 32-bit user is known to be non-negative, whether 8 ; the IV is considered signed or unsigned. 9 define void @foo(i32* %A, i32* %B, i32* %C, i32 %N) { 10 ; CHECK-LABEL: @foo( 11 ; CHECK-NEXT: entry: 12 ; CHECK-NEXT: [[CMP1:%.*]] = icmp slt i32 0, %N 13 ; CHECK-NEXT: br i1 [[CMP1]], label %for.body.lr.ph, label %for.end 14 ; CHECK: for.body.lr.ph: 15 ; CHECK-NEXT: br label %for.body 16 ; CHECK: for.body: 17 ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV:%.*]].next, %for.inc ], [ 0, %for.body.lr.ph ] 18 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* %B, i64 [[INDVARS_IV]] 19 ; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 20 ; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2 21 ; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* %C, i64 [[TMP1]] 22 ; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4 23 ; CHECK-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP0]], [[TMP2]] 24 ; CHECK-NEXT: [[TRUNC0:%.*]] = trunc i64 [[TMP1]] to i32 25 ; CHECK-NEXT: [[DIV0:%.*]] = udiv i32 5, [[TRUNC0]] 26 ; CHECK-NEXT: [[ADD4:%.*]] = add nsw i32 [[ADD3]], [[DIV0]] 27 ; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* %A, i64 [[INDVARS_IV]] 28 ; CHECK-NEXT: store i32 [[ADD4]], i32* [[ARRAYIDX5]], align 4 29 ; CHECK-NEXT: br label %for.inc 30 ; CHECK: for.inc: 31 ; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1 32 ; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 %N to i64 33 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]] 34 ; CHECK-NEXT: br i1 [[EXITCOND]], label %for.body, label %for.cond.for.end_crit_edge 35 ; CHECK: for.cond.for.end_crit_edge: 36 ; CHECK-NEXT: br label %for.end 37 ; CHECK: for.end: 38 ; CHECK-NEXT: ret void 39 ; 40 entry: 41 %cmp1 = icmp slt i32 0, %N 42 br i1 %cmp1, label %for.body.lr.ph, label %for.end 43 44 for.body.lr.ph: ; preds = %entry 45 br label %for.body 46 47 for.body: ; preds = %for.body.lr.ph, %for.inc 48 %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.inc ] 49 %idxprom = sext i32 %i.02 to i64 50 %arrayidx = getelementptr inbounds i32, i32* %B, i64 %idxprom 51 %0 = load i32, i32* %arrayidx, align 4 52 %add = add nsw i32 %i.02, 2 53 %idxprom1 = zext i32 %add to i64 54 %arrayidx2 = getelementptr inbounds i32, i32* %C, i64 %idxprom1 55 %1 = load i32, i32* %arrayidx2, align 4 56 %add3 = add nsw i32 %0, %1 57 %div0 = udiv i32 5, %add 58 %add4 = add nsw i32 %add3, %div0 59 %idxprom4 = zext i32 %i.02 to i64 60 %arrayidx5 = getelementptr inbounds i32, i32* %A, i64 %idxprom4 61 store i32 %add4, i32* %arrayidx5, align 4 62 br label %for.inc 63 64 for.inc: ; preds = %for.body 65 %inc = add nsw i32 %i.02, 1 66 %cmp = icmp slt i32 %inc, %N 67 br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge 68 69 for.cond.for.end_crit_edge: ; preds = %for.inc 70 br label %for.end 71 72 for.end: ; preds = %for.cond.for.end_crit_edge, %entry 73 ret void 74 } 75 76 define void @foo1(i32* %A, i32* %B, i32* %C, i32 %N) { 77 ; CHECK-LABEL: @foo1( 78 ; CHECK-NEXT: entry: 79 ; CHECK-NEXT: [[CMP1:%.*]] = icmp slt i32 0, %N 80 ; CHECK-NEXT: br i1 [[CMP1]], label %for.body.lr.ph, label %for.end 81 ; CHECK: for.body.lr.ph: 82 ; CHECK-NEXT: br label %for.body 83 ; CHECK: for.body: 84 ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV:%.*]].next, %for.inc ], [ 0, %for.body.lr.ph ] 85 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* %B, i64 [[INDVARS_IV]] 86 ; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 87 ; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2 88 ; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* %C, i64 [[TMP1]] 89 ; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4 90 ; CHECK-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP0]], [[TMP2]] 91 ; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* %A, i64 [[INDVARS_IV]] 92 ; CHECK-NEXT: store i32 [[ADD3]], i32* [[ARRAYIDX5]], align 4 93 ; CHECK-NEXT: br label %for.inc 94 ; CHECK: for.inc: 95 ; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1 96 ; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 %N to i64 97 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]] 98 ; CHECK-NEXT: br i1 [[EXITCOND]], label %for.body, label %for.cond.for.end_crit_edge 99 ; CHECK: for.cond.for.end_crit_edge: 100 ; CHECK-NEXT: br label %for.end 101 ; CHECK: for.end: 102 ; CHECK-NEXT: ret void 103 ; 104 entry: 105 %cmp1 = icmp slt i32 0, %N 106 br i1 %cmp1, label %for.body.lr.ph, label %for.end 107 108 for.body.lr.ph: ; preds = %entry 109 br label %for.body 110 111 for.body: ; preds = %for.body.lr.ph, %for.inc 112 %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.inc ] 113 %idxprom = zext i32 %i.02 to i64 114 %arrayidx = getelementptr inbounds i32, i32* %B, i64 %idxprom 115 %0 = load i32, i32* %arrayidx, align 4 116 %add = add nsw i32 %i.02, 2 117 %idxprom1 = sext i32 %add to i64 118 %arrayidx2 = getelementptr inbounds i32, i32* %C, i64 %idxprom1 119 %1 = load i32, i32* %arrayidx2, align 4 120 %add3 = add nsw i32 %0, %1 121 %idxprom4 = sext i32 %i.02 to i64 122 %arrayidx5 = getelementptr inbounds i32, i32* %A, i64 %idxprom4 123 store i32 %add3, i32* %arrayidx5, align 4 124 br label %for.inc 125 126 for.inc: ; preds = %for.body 127 %inc = add nsw i32 %i.02, 1 128 %cmp = icmp slt i32 %inc, %N 129 br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge 130 131 for.cond.for.end_crit_edge: ; preds = %for.inc 132 br label %for.end 133 134 for.end: ; preds = %for.cond.for.end_crit_edge, %entry 135 ret void 136 } 137 138 139 @a = common global [100 x i32] zeroinitializer, align 16 140 @b = common global [100 x i32] zeroinitializer, align 16 141 142 define i32 @foo2(i32 %M) { 143 ; CHECK-LABEL: @foo2( 144 ; CHECK-NEXT: entry: 145 ; CHECK-NEXT: [[CMP1:%.*]] = icmp slt i32 0, %M 146 ; CHECK-NEXT: br i1 [[CMP1]], label %for.body.lr.ph, label %for.end 147 ; CHECK: for.body.lr.ph: 148 ; CHECK-NEXT: [[TMP0:%.*]] = sext i32 %M to i64 149 ; CHECK-NEXT: br label %for.body 150 ; CHECK: for.body: 151 ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV:%.*]].next, %for.inc ], [ 0, %for.body.lr.ph ] 152 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [100 x i32], [100 x i32]* @a, i64 0, i64 [[INDVARS_IV]] 153 ; CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 154 ; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds [100 x i32], [100 x i32]* @b, i64 0, i64 [[INDVARS_IV]] 155 ; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4 156 ; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[TMP2]] 157 ; CHECK-NEXT: [[TMP3:%.*]] = add nsw i64 [[INDVARS_IV]], [[TMP0]] 158 ; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds [100 x i32], [100 x i32]* @a, i64 0, i64 [[TMP3]] 159 ; CHECK-NEXT: store i32 [[ADD]], i32* [[ARRAYIDX5]], align 4 160 ; CHECK-NEXT: br label %for.inc 161 ; CHECK: for.inc: 162 ; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1 163 ; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 %M to i64 164 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]] 165 ; CHECK-NEXT: br i1 [[EXITCOND]], label %for.body, label %for.cond.for.end_crit_edge 166 ; CHECK: for.cond.for.end_crit_edge: 167 ; CHECK-NEXT: br label %for.end 168 ; CHECK: for.end: 169 ; CHECK-NEXT: [[CALL:%.*]] = call i32 @dummy(i32* getelementptr inbounds ([100 x i32], [100 x i32]* @a, i32 0, i32 0), i32* getelementptr inbounds ([100 x i32], [100 x i32]* @b, i32 0, i32 0)) 170 ; CHECK-NEXT: ret i32 0 171 ; 172 entry: 173 %cmp1 = icmp slt i32 0, %M 174 br i1 %cmp1, label %for.body.lr.ph, label %for.end 175 176 for.body.lr.ph: ; preds = %entry 177 br label %for.body 178 179 for.body: ; preds = %for.body.lr.ph, %for.inc 180 %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.inc ] 181 %idxprom = zext i32 %i.02 to i64 182 %arrayidx = getelementptr inbounds [100 x i32], [100 x i32]* @a, i64 0, i64 %idxprom 183 %0 = load i32, i32* %arrayidx, align 4 184 %idxprom1 = sext i32 %i.02 to i64 185 %arrayidx2 = getelementptr inbounds [100 x i32], [100 x i32]* @b, i64 0, i64 %idxprom1 186 %1 = load i32, i32* %arrayidx2, align 4 187 %add = add nsw i32 %0, %1 188 %add3 = add nsw i32 %i.02, %M 189 %idxprom4 = sext i32 %add3 to i64 190 %arrayidx5 = getelementptr inbounds [100 x i32], [100 x i32]* @a, i64 0, i64 %idxprom4 191 store i32 %add, i32* %arrayidx5, align 4 192 br label %for.inc 193 194 for.inc: ; preds = %for.body 195 %inc = add nsw i32 %i.02, 1 196 %cmp = icmp slt i32 %inc, %M 197 br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge 198 199 for.cond.for.end_crit_edge: ; preds = %for.inc 200 br label %for.end 201 202 for.end: ; preds = %for.cond.for.end_crit_edge, %entry 203 %call = call i32 @dummy(i32* getelementptr inbounds ([100 x i32], [100 x i32]* @a, i32 0, i32 0), i32* getelementptr inbounds ([100 x i32], [100 x i32]* @b, i32 0, i32 0)) 204 ret i32 0 205 } 206 207 declare i32 @dummy(i32*, i32*) 208 209 ; A case where zext should not be eliminated when its operands could only be extended by sext. 210 define i32 @foo3(i32 %M) { 211 ; CHECK-LABEL: @foo3( 212 ; CHECK-NEXT: entry: 213 ; CHECK-NEXT: [[CMP1:%.*]] = icmp slt i32 0, %M 214 ; CHECK-NEXT: br i1 [[CMP1]], label %for.body.lr.ph, label %for.end 215 ; CHECK: for.body.lr.ph: 216 ; CHECK-NEXT: [[TMP0:%.*]] = sext i32 %M to i64 217 ; CHECK-NEXT: br label %for.body 218 ; CHECK: for.body: 219 ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV:%.*]].next, %for.inc ], [ 0, %for.body.lr.ph ] 220 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [100 x i32], [100 x i32]* @a, i64 0, i64 [[INDVARS_IV]] 221 ; CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 222 ; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds [100 x i32], [100 x i32]* @b, i64 0, i64 [[INDVARS_IV]] 223 ; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4 224 ; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[TMP2]] 225 ; CHECK-NEXT: [[TMP3:%.*]] = add nsw i64 [[INDVARS_IV]], [[TMP0]] 226 ; CHECK-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32 227 ; CHECK-NEXT: [[IDXPROM4:%.*]] = zext i32 [[TMP4]] to i64 228 ; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds [100 x i32], [100 x i32]* @a, i64 0, i64 [[IDXPROM4]] 229 ; CHECK-NEXT: store i32 [[ADD]], i32* [[ARRAYIDX5]], align 4 230 ; CHECK-NEXT: br label %for.inc 231 ; CHECK: for.inc: 232 ; CHECK-NEXT: [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1 233 ; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 %M to i64 234 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]] 235 ; CHECK-NEXT: br i1 [[EXITCOND]], label %for.body, label %for.cond.for.end_crit_edge 236 ; CHECK: for.cond.for.end_crit_edge: 237 ; CHECK-NEXT: br label %for.end 238 ; CHECK: for.end: 239 ; CHECK-NEXT: [[CALL:%.*]] = call i32 @dummy(i32* getelementptr inbounds ([100 x i32], [100 x i32]* @a, i32 0, i32 0), i32* getelementptr inbounds ([100 x i32], [100 x i32]* @b, i32 0, i32 0)) 240 ; CHECK-NEXT: ret i32 0 241 ; 242 entry: 243 %cmp1 = icmp slt i32 0, %M 244 br i1 %cmp1, label %for.body.lr.ph, label %for.end 245 246 for.body.lr.ph: ; preds = %entry 247 br label %for.body 248 249 for.body: ; preds = %for.body.lr.ph, %for.inc 250 %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.inc ] 251 %idxprom = sext i32 %i.02 to i64 252 %arrayidx = getelementptr inbounds [100 x i32], [100 x i32]* @a, i64 0, i64 %idxprom 253 %0 = load i32, i32* %arrayidx, align 4 254 %idxprom1 = sext i32 %i.02 to i64 255 %arrayidx2 = getelementptr inbounds [100 x i32], [100 x i32]* @b, i64 0, i64 %idxprom1 256 %1 = load i32, i32* %arrayidx2, align 4 257 %add = add nsw i32 %0, %1 258 %add3 = add nsw i32 %i.02, %M 259 %idxprom4 = zext i32 %add3 to i64 260 %arrayidx5 = getelementptr inbounds [100 x i32], [100 x i32]* @a, i64 0, i64 %idxprom4 261 store i32 %add, i32* %arrayidx5, align 4 262 br label %for.inc 263 264 for.inc: ; preds = %for.body 265 %inc = add nsw i32 %i.02, 1 266 %cmp = icmp slt i32 %inc, %M 267 br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge 268 269 for.cond.for.end_crit_edge: ; preds = %for.inc 270 br label %for.end 271 272 for.end: ; preds = %for.cond.for.end_crit_edge, %entry 273 %call = call i32 @dummy(i32* getelementptr inbounds ([100 x i32], [100 x i32]* @a, i32 0, i32 0), i32* getelementptr inbounds ([100 x i32], [100 x i32]* @b, i32 0, i32 0)) 274 ret i32 0 275 } 276
