88 lines · plain
1; RUN: opt %loadNPMPolly -aa-pipeline=basic-aa -polly-allow-nonaffine-branches -polly-allow-nonaffine-loops=false '-passes=polly-custom<detect>' -polly-print-detect -disable-output < %s 2>&1 | FileCheck %s --check-prefix=REJECTNONAFFINELOOPS2; RUN: opt %loadNPMPolly -aa-pipeline=basic-aa -polly-allow-nonaffine-branches -polly-allow-nonaffine-loops=true '-passes=polly-custom<detect>' -polly-print-detect -disable-output < %s 2>&1 | FileCheck %s --check-prefix=ALLOWNONAFFINELOOPS3; RUN: opt %loadNPMPolly -aa-pipeline=basic-aa -polly-allow-nonaffine-branches -polly-allow-nonaffine-loops=true -polly-allow-nonaffine '-passes=polly-custom<detect>' -polly-print-detect -disable-output < %s 2>&1 | FileCheck %s --check-prefix=ALLOWNONAFFINELOOPSANDACCESSES4;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 the7; innermost loop as a SCoP of depth 1, we have to reject the loop nest if not8; both, non-affine loops as well as non-affine accesses are allowed.9;10; REJECTNONAFFINELOOPS: Valid Region for Scop: bb15 => bb1311; REJECTNONAFFINELOOPS-NOT: Valid12; ALLOWNONAFFINELOOPS: Valid Region for Scop: bb15 => bb1313; ALLOWNONAFFINELOOPS-NOT: Valid14; ALLOWNONAFFINELOOPSANDACCESSES: Valid Region for Scop: bb11 => bb2915;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 %bb1128 29bb11: ; preds = %bb28, %bb30 %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, 102433 br i1 %exitcond10, label %bb12, label %bb2934 35bb12: ; preds = %bb1136 br label %bb1337 38bb13: ; preds = %bb26, %bb1239 %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, 102442 br i1 %exitcond7, label %bb14, label %bb2743 44bb14: ; preds = %bb1345 br label %bb1546 47bb15: ; preds = %bb24, %bb1448 %indvars.iv = phi i64 [ %indvars.iv.next, %bb24 ], [ 0, %bb14 ]49 %lftr.wideiv = trunc i64 %indvars.iv to i3250 %exitcond = icmp ne i32 %lftr.wideiv, %indvars.iv351 br i1 %exitcond, label %bb16, label %bb2552 53bb16: ; preds = %bb1554 %tmp = getelementptr inbounds i32, ptr %A, i64 %indvars.iv855 %tmp17 = load i32, ptr %tmp, align 456 %tmp18 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv557 %tmp19 = load i32, ptr %tmp18, align 458 %tmp20 = add nsw i32 %tmp17, %tmp1959 %tmp21 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv60 %tmp22 = load i32, ptr %tmp21, align 461 %tmp23 = add nsw i32 %tmp22, %tmp2062 store i32 %tmp23, ptr %tmp21, align 463 br label %bb2464 65bb24: ; preds = %bb1666 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 167 br label %bb1568 69bb25: ; preds = %bb1570 br label %bb2671 72bb26: ; preds = %bb2573 %indvars.iv.next6 = add nuw nsw i64 %indvars.iv5, 174 %indvars.iv.next4 = add nuw nsw i32 %indvars.iv3, %indvars.iv175 br label %bb1376 77bb27: ; preds = %bb1378 br label %bb2879 80bb28: ; preds = %bb2781 %indvars.iv.next9 = add nuw nsw i64 %indvars.iv8, 182 %indvars.iv.next2 = add nuw nsw i32 %indvars.iv1, 183 br label %bb1184 85bb29: ; preds = %bb1186 ret void87}88