1; RUN: opt %loadNPMPolly -aa-pipeline=basic-aa -polly-allow-nonaffine-branches -polly-allow-nonaffine-loops=false '-passes=print<polly-detect>' -disable-output < %s 2>&1 | FileCheck %s --check-prefix=REJECTNONAFFINELOOPS 2; RUN: opt %loadNPMPolly -aa-pipeline=basic-aa -polly-allow-nonaffine-branches -polly-allow-nonaffine-loops=true '-passes=print<polly-detect>' -disable-output < %s 2>&1 | FileCheck %s --check-prefix=ALLOWNONAFFINELOOPS 3; RUN: opt %loadNPMPolly -aa-pipeline=basic-aa -polly-allow-nonaffine-branches -polly-allow-nonaffine-loops=true -polly-allow-nonaffine '-passes=print<polly-detect>' -disable-output < %s 2>&1 | FileCheck %s --check-prefix=ALLOWNONAFFINELOOPSANDACCESSES 4; 5; Here we have a non-affine loop (in the context of the loop nest) 6; and also a non-affine access (A[k]). While we can always detect the 7; innermost loop as a SCoP of depth 1, we have to reject the loop nest if not 8; both, non-affine loops as well as non-affine accesses are allowed. 9; 10; REJECTNONAFFINELOOPS: Valid Region for Scop: bb15 => bb13 11; REJECTNONAFFINELOOPS-NOT: Valid 12; ALLOWNONAFFINELOOPS: Valid Region for Scop: bb15 => bb13 13; ALLOWNONAFFINELOOPS-NOT: Valid 14; ALLOWNONAFFINELOOPSANDACCESSES: Valid Region for Scop: bb11 => bb29 15; 16; void f(int *A) { 17; for (int i = 0; i < 1024; i++) 18; for (int j = 0; j < 1024; j++) 19; for (int k = 0; k < i * j; k++) 20; A[k] += A[i] + A[j]; 21; } 22; 23target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 24 25define void @f(ptr %A) { 26bb: 27 br label %bb11 28 29bb11: ; preds = %bb28, %bb 30 %indvars.iv8 = phi i64 [ %indvars.iv.next9, %bb28 ], [ 0, %bb ] 31 %indvars.iv1 = phi i32 [ %indvars.iv.next2, %bb28 ], [ 0, %bb ] 32 %exitcond10 = icmp ne i64 %indvars.iv8, 1024 33 br i1 %exitcond10, label %bb12, label %bb29 34 35bb12: ; preds = %bb11 36 br label %bb13 37 38bb13: ; preds = %bb26, %bb12 39 %indvars.iv5 = phi i64 [ %indvars.iv.next6, %bb26 ], [ 0, %bb12 ] 40 %indvars.iv3 = phi i32 [ %indvars.iv.next4, %bb26 ], [ 0, %bb12 ] 41 %exitcond7 = icmp ne i64 %indvars.iv5, 1024 42 br i1 %exitcond7, label %bb14, label %bb27 43 44bb14: ; preds = %bb13 45 br label %bb15 46 47bb15: ; preds = %bb24, %bb14 48 %indvars.iv = phi i64 [ %indvars.iv.next, %bb24 ], [ 0, %bb14 ] 49 %lftr.wideiv = trunc i64 %indvars.iv to i32 50 %exitcond = icmp ne i32 %lftr.wideiv, %indvars.iv3 51 br i1 %exitcond, label %bb16, label %bb25 52 53bb16: ; preds = %bb15 54 %tmp = getelementptr inbounds i32, ptr %A, i64 %indvars.iv8 55 %tmp17 = load i32, ptr %tmp, align 4 56 %tmp18 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv5 57 %tmp19 = load i32, ptr %tmp18, align 4 58 %tmp20 = add nsw i32 %tmp17, %tmp19 59 %tmp21 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv 60 %tmp22 = load i32, ptr %tmp21, align 4 61 %tmp23 = add nsw i32 %tmp22, %tmp20 62 store i32 %tmp23, ptr %tmp21, align 4 63 br label %bb24 64 65bb24: ; preds = %bb16 66 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 67 br label %bb15 68 69bb25: ; preds = %bb15 70 br label %bb26 71 72bb26: ; preds = %bb25 73 %indvars.iv.next6 = add nuw nsw i64 %indvars.iv5, 1 74 %indvars.iv.next4 = add nuw nsw i32 %indvars.iv3, %indvars.iv1 75 br label %bb13 76 77bb27: ; preds = %bb13 78 br label %bb28 79 80bb28: ; preds = %bb27 81 %indvars.iv.next9 = add nuw nsw i64 %indvars.iv8, 1 82 %indvars.iv.next2 = add nuw nsw i32 %indvars.iv1, 1 83 br label %bb11 84 85bb29: ; preds = %bb11 86 ret void 87} 88