1; RUN: opt %loadNPMPolly -S -passes=polly-codegen < %s | FileCheck %s 2; 3; Check that we calculate the maximal access into array A correctly and track the overflow state. 4; 5; CHECK: %[[TMP0:[._0-9a-zA-Z]*]] = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 99, i64 %m) 6; CHECK: %[[TMP0O:[._0-9a-zA-Z]*]] = extractvalue { i64, i1 } %[[TMP0]], 1 7; CHECK: %[[OS0:[._0-9a-zA-Z]*]] = or i1 {{[^,]*}}, %[[TMP0O]] 8; CHECK: %[[TMP0R:[._0-9a-zA-Z]*]] = extractvalue { i64, i1 } %[[TMP0]], 0 9; CHECK: %[[TMP1:[._0-9a-zA-Z]*]] = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %[[TMP0R]], i64 149) 10; CHECK: %[[TMP1O:[._0-9a-zA-Z]*]] = extractvalue { i64, i1 } %[[TMP1]], 1 11; CHECK: %[[OS1:[._0-9a-zA-Z]*]] = or i1 %[[OS0]], %[[TMP1O]] 12; CHECK: %[[TMP1R:[._0-9a-zA-Z]*]] = extractvalue { i64, i1 } %[[TMP1]], 0 13; CHECK: %[[TMP2:[._0-9a-zA-Z]*]] = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %[[TMP1R]], i64 %p) 14; CHECK: %[[TMP2O:[._0-9a-zA-Z]*]] = extractvalue { i64, i1 } %[[TMP2]], 1 15; CHECK: %[[OS2:[._0-9a-zA-Z]*]] = or i1 %[[OS1]], %[[TMP2O]] 16; CHECK: %[[TMP2R:[._0-9a-zA-Z]*]] = extractvalue { i64, i1 } %[[TMP2]], 0 17; CHECK: %[[TMP3:[._0-9a-zA-Z]*]] = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %[[TMP2R]], i64 150) 18; CHECK: %[[TMP3O:[._0-9a-zA-Z]*]] = extractvalue { i64, i1 } %[[TMP3]], 1 19; CHECK: %[[OS3:[._0-9a-zA-Z]*]] = or i1 %[[OS2]], %[[TMP3O]] 20; CHECK: %[[TMP3R:[._0-9a-zA-Z]*]] = extractvalue { i64, i1 } %[[TMP3]], 0 21; CHECK: %polly.access.A{{[0-9]*}} = getelementptr double, ptr %A, i64 %[[TMP3R]] 22; 23; CHECK: %polly.rtc.overflown = xor i1 %[[OS3]], true 24; CHECK: %polly.rtc.result = and i1 %{{[^,]*}}, %polly.rtc.overflown 25; CHECK: br i1 %polly.rtc.result, 26; 27; void foo(long n, long m, long p, double A[n][m][p], int *B) { 28; for (long i = 0; i < 100; i++) 29; for (long j = 0; j < 150; j++) 30; for (long k = 0; k < 150; k++) 31; A[i][j][k] = B[k]; 32; } 33; 34target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 35 36define void @foo(i64 %n, i64 %m, i64 %p, ptr %A, ptr %B) { 37entry: 38 br label %for.cond 39 40for.cond: ; preds = %for.inc03, %entry 41 %i.0 = phi i64 [ 0, %entry ], [ %inc04, %for.inc03 ] 42 %exitcond2 = icmp ne i64 %i.0, 100 43 br i1 %exitcond2, label %for.body, label %for.end15 44 45for.body: ; preds = %for.cond 46 br label %for.cond1 47 48for.cond1: ; preds = %for.inc00, %for.body 49 %j.0 = phi i64 [ 0, %for.body ], [ %inc01, %for.inc00 ] 50 %exitcond1 = icmp ne i64 %j.0, 150 51 br i1 %exitcond1, label %for.body3, label %for.end12 52 53for.body3: ; preds = %for.cond1 54 br label %for.cond4 55 56for.cond4: ; preds = %for.inc, %for.body3 57 %k.0 = phi i64 [ 0, %for.body3 ], [ %inc, %for.inc ] 58 %exitcond = icmp ne i64 %k.0, 150 59 br i1 %exitcond, label %for.body6, label %for.end 60 61for.body6: ; preds = %for.cond4 62 %arrayidx = getelementptr inbounds i32, ptr %B, i64 %k.0 63 %tmp3 = load i32, ptr %arrayidx, align 2 64 %conv = sitofp i32 %tmp3 to double 65 %tmp4 = mul nuw i64 %m, %p 66 %tmp5 = mul nsw i64 %i.0, %tmp4 67 %tmp6 = mul nsw i64 %j.0, %p 68 %arrayidx7.sum = add i64 %tmp5, %tmp6 69 %arrayidx8.sum = add i64 %arrayidx7.sum, %k.0 70 %arrayidx9 = getelementptr inbounds double, ptr %A, i64 %arrayidx8.sum 71 store double %conv, ptr %arrayidx9, align 8 72 br label %for.inc 73 74for.inc: ; preds = %for.body6 75 %inc = add nsw i64 %k.0, 1 76 br label %for.cond4 77 78for.end: ; preds = %for.cond4 79 br label %for.inc00 80 81for.inc00: ; preds = %for.end 82 %inc01 = add nsw i64 %j.0, 1 83 br label %for.cond1 84 85for.end12: ; preds = %for.cond1 86 br label %for.inc03 87 88for.inc03: ; preds = %for.end12 89 %inc04 = add nsw i64 %i.0, 1 90 br label %for.cond 91 92for.end15: ; preds = %for.cond 93 ret void 94} 95