brintos

brintos / llvm-project-archived public Read only

0
0
Text · 6.5 KiB · 2d56fb3 Raw
161 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-unroll -mtriple=arm64-apple-iphoneos -S %s | FileCheck %s3 4; Check we unroll even with optsize, if the result is smaller, either because5; we have single iteration loops or bodies with constant folding opportunities6; after fully unrolling.7 8declare i32 @get()9 10define void @fully_unrolled_single_iteration(ptr %src) #0 {11; CHECK-LABEL: @fully_unrolled_single_iteration(12; CHECK-NEXT:  entry:13; CHECK-NEXT:    [[ARR:%.*]] = alloca [4 x i32], align 414; CHECK-NEXT:    br label [[FOR_BODY:%.*]]15; CHECK:       for.body:16; CHECK-NEXT:    [[V:%.*]] = load i32, ptr [[SRC:%.*]]17; CHECK-NEXT:    store i32 [[V]], ptr [[ARR]], align 418; CHECK-NEXT:    call void @use(ptr nonnull [[ARR]])19; CHECK-NEXT:    ret void20;21entry:22  %arr = alloca [4 x i32], align 423  br label %for.body24 25for.body:                                         ; preds = %for.body, %entry26  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]27  %src.idx = getelementptr inbounds i32, ptr %src, i64 %indvars.iv28  %v = load i32, ptr %src.idx29  %arrayidx = getelementptr inbounds [4 x i32], ptr %arr, i64 0, i64 %indvars.iv30  store i32 %v, ptr %arrayidx, align 431  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 132  %exitcond = icmp eq i64 %indvars.iv.next, 133  br i1 %exitcond, label %for.cond.cleanup, label %for.body34 35for.cond.cleanup:                                 ; preds = %for.cond36  call void @use(ptr nonnull %arr) #437  ret void38}39 40 41define void @fully_unrolled_smaller() #0 {42; CHECK-LABEL: @fully_unrolled_smaller(43; CHECK-NEXT:  entry:44; CHECK-NEXT:    [[ARR:%.*]] = alloca [4 x i32], align 445; CHECK-NEXT:    br label [[FOR_BODY:%.*]]46; CHECK:       for.body:47; CHECK-NEXT:    store i32 16, ptr [[ARR]], align 448; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds [4 x i32], ptr [[ARR]], i64 0, i64 149; CHECK-NEXT:    store i32 4104, ptr [[ARRAYIDX_1]], align 450; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds [4 x i32], ptr [[ARR]], i64 0, i64 251; CHECK-NEXT:    store i32 1048592, ptr [[ARRAYIDX_2]], align 452; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds [4 x i32], ptr [[ARR]], i64 0, i64 353; CHECK-NEXT:    store i32 268435480, ptr [[ARRAYIDX_3]], align 454; CHECK-NEXT:    call void @use(ptr nonnull [[ARR]])55; CHECK-NEXT:    ret void56;57entry:58  %arr = alloca [4 x i32], align 459  br label %for.body60 61for.body:                                         ; preds = %for.body, %entry62  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]63  %indvars.iv.tr = trunc i64 %indvars.iv to i3264  %shl.0 = shl i32 %indvars.iv.tr, 365  %shl.1 = shl i32 16, %shl.066  %or = or i32 %shl.1, %shl.067  %arrayidx = getelementptr inbounds [4 x i32], ptr %arr, i64 0, i64 %indvars.iv68  store i32 %or, ptr %arrayidx, align 469  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 170  %exitcond = icmp eq i64 %indvars.iv, 371  br i1 %exitcond, label %for.cond.cleanup, label %for.body72 73for.cond.cleanup:                                 ; preds = %for.cond74  call void @use(ptr nonnull %arr) #475  ret void76}77 78define void @fully_unrolled_smaller_Oz() #1 {79; CHECK-LABEL: @fully_unrolled_smaller_Oz(80; CHECK-NEXT:  entry:81; CHECK-NEXT:    [[ARR:%.*]] = alloca [4 x i32], align 482; CHECK-NEXT:    br label [[FOR_BODY:%.*]]83; CHECK:       for.body:84; CHECK-NEXT:    store i32 16, ptr [[ARR]], align 485; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds [4 x i32], ptr [[ARR]], i64 0, i64 186; CHECK-NEXT:    store i32 4104, ptr [[ARRAYIDX_1]], align 487; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds [4 x i32], ptr [[ARR]], i64 0, i64 288; CHECK-NEXT:    store i32 1048592, ptr [[ARRAYIDX_2]], align 489; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds [4 x i32], ptr [[ARR]], i64 0, i64 390; CHECK-NEXT:    store i32 268435480, ptr [[ARRAYIDX_3]], align 491; CHECK-NEXT:    call void @use(ptr nonnull [[ARR]])92; CHECK-NEXT:    ret void93;94entry:95  %arr = alloca [4 x i32], align 496  br label %for.body97 98for.body:                                         ; preds = %for.body, %entry99  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]100  %indvars.iv.tr = trunc i64 %indvars.iv to i32101  %shl.0 = shl i32 %indvars.iv.tr, 3102  %shl.1 = shl i32 16, %shl.0103  %or = or i32 %shl.1, %shl.0104  %arrayidx = getelementptr inbounds [4 x i32], ptr %arr, i64 0, i64 %indvars.iv105  store i32 %or, ptr %arrayidx, align 4106  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1107  %exitcond = icmp eq i64 %indvars.iv, 3108  br i1 %exitcond, label %for.cond.cleanup, label %for.body109 110for.cond.cleanup:                                 ; preds = %for.cond111  call void @use(ptr nonnull %arr) #4112  ret void113}114 115 116define void @fully_unrolled_bigger() #0 {117; CHECK-LABEL: @fully_unrolled_bigger(118; CHECK-NEXT:  entry:119; CHECK-NEXT:    [[ARR:%.*]] = alloca [4 x i32], align 4120; CHECK-NEXT:    br label [[FOR_BODY:%.*]]121; CHECK:       for.body:122; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]123; CHECK-NEXT:    [[INDVARS_IV_TR:%.*]] = trunc i64 [[INDVARS_IV]] to i32124; CHECK-NEXT:    [[SHL_0:%.*]] = shl i32 [[INDVARS_IV_TR]], 3125; CHECK-NEXT:    [[SHL_1:%.*]] = shl i32 16, [[SHL_0]]126; CHECK-NEXT:    [[OR:%.*]] = or i32 [[SHL_1]], [[SHL_0]]127; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [4 x i32], ptr [[ARR]], i64 0, i64 [[INDVARS_IV]]128; CHECK-NEXT:    store i32 [[OR]], ptr [[ARRAYIDX]], align 4129; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1130; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV]], 7131; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY]]132; CHECK:       for.cond.cleanup:133; CHECK-NEXT:    call void @use(ptr nonnull [[ARR]])134; CHECK-NEXT:    ret void135;136entry:137  %arr = alloca [4 x i32], align 4138  br label %for.body139 140for.body:                                         ; preds = %for.body, %entry141  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]142  %indvars.iv.tr = trunc i64 %indvars.iv to i32143  %shl.0 = shl i32 %indvars.iv.tr, 3144  %shl.1 = shl i32 16, %shl.0145  %or = or i32 %shl.1, %shl.0146  %arrayidx = getelementptr inbounds [4 x i32], ptr %arr, i64 0, i64 %indvars.iv147  store i32 %or, ptr %arrayidx, align 4148  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1149  %exitcond = icmp eq i64 %indvars.iv, 7150  br i1 %exitcond, label %for.cond.cleanup, label %for.body151 152for.cond.cleanup:                                 ; preds = %for.cond153  call void @use(ptr nonnull %arr) #4154  ret void155}156 157declare void @use(ptr)158 159attributes #0 = { optsize }160attributes #1 = { minsize optsize }161