69 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<ast>' -polly-print-ast -polly-ast-detect-parallel -disable-output < %s | FileCheck %s2;3; CHECK: #pragma known-parallel reduction (^ : MemRef_sum)4; void f(int N, int M, int P, int sum[P][M]) {5; for (int i = 0; i < N; i++)6; for (int j = 0; j < P; j++)7; CHECK: #pragma simd8; for (int k = 0; k < M; k++)9; sum[j][k] ^= j;10; }11;12target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-n32-S64"13 14define void @f(i32 %N, i32 %M, i32 %P, ptr %sum) {15entry:16 br label %for.cond17 18for.cond: ; preds = %for.inc11, %entry19 %i.0 = phi i32 [ 0, %entry ], [ %inc12, %for.inc11 ]20 %cmp = icmp slt i32 %i.0, %N21 br i1 %cmp, label %for.body, label %for.end1322 23for.body: ; preds = %for.cond24 br label %for.cond125 26for.cond1: ; preds = %for.inc8, %for.body27 %j.0 = phi i32 [ 0, %for.body ], [ %inc9, %for.inc8 ]28 %cmp2 = icmp slt i32 %j.0, %P29 br i1 %cmp2, label %for.body3, label %for.end1030 31for.body3: ; preds = %for.cond132 br label %for.cond433 34for.cond4: ; preds = %for.inc, %for.body335 %k.0 = phi i32 [ 0, %for.body3 ], [ %inc, %for.inc ]36 %cmp5 = icmp slt i32 %k.0, %M37 br i1 %cmp5, label %for.body6, label %for.end38 39for.body6: ; preds = %for.cond440 %tmp = mul nsw i32 %j.0, %M41 %arrayidx.sum = add i32 %tmp, %k.042 %arrayidx7 = getelementptr inbounds i32, ptr %sum, i32 %arrayidx.sum43 %tmp1 = load i32, ptr %arrayidx7, align 444 %xor = xor i32 %tmp1, %j.045 store i32 %xor, ptr %arrayidx7, align 446 br label %for.inc47 48for.inc: ; preds = %for.body649 %inc = add nsw i32 %k.0, 150 br label %for.cond451 52for.end: ; preds = %for.cond453 br label %for.inc854 55for.inc8: ; preds = %for.end56 %inc9 = add nsw i32 %j.0, 157 br label %for.cond158 59for.end10: ; preds = %for.cond160 br label %for.inc1161 62for.inc11: ; preds = %for.end1063 %inc12 = add nsw i32 %i.0, 164 br label %for.cond65 66for.end13: ; preds = %for.cond67 ret void68}69