130 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<scops>' -polly-print-detect -polly-print-scops -disable-output < %s 2>&1 | FileCheck %s2;3; A valid Scop would print the list of it's statements, we check that we do not4; see that list.5;6; CHECK-NOT: Statements7;8; FIXME: Handling this is an open problem, at the moment we just bail out.9;10; void foo(float *A, float *B,11; long p1,12; long p2,13; long p3,14; long p4,15; long p5,16; long p6,17; long p7,18; long p8,19; long p9,20; long p10,21; long p11,22; long p12) {23; for (long i = 0; i < 100; i++) {24; A[i] =25; B[i + p1] +26; B[i + p2] +27; B[i + p3] +28; B[i + p4] +29; B[i + p5] +30; B[i + p6] +31; B[i + p7] +32; B[i + p8] +33; B[i + p9] +34; B[i + p10] +35; B[i + p11] +36; B[i + p12];37; }38; }39;40; Computing the minimal and maximal element accessed in B is very expensive.41; Expressing the minimal element itself yields a rather complex isl_pw_aff which42; looks as follows:43; { ...44; MemRef_B[(100 + p11)] : p2 <= -1 + p1 and p3 <= -1 + p1 and p4 <= -1 + p145; and p5 <= -1 + p1 and p6 <= -1 + p1 and46; p7 <= -1 + p1 and p8 <= -1 + p1 and p9 <= -1 + p147; and p10 <= -1 + p1 and p11 >= p1 and48; p12 <= -1 + p11;49; MemRef_B[(100 + p12)] : p2 <= -1 + p1 and p3 <= -1 + p1 and p4 <= -1 + p150; and p5 <= -1 + p1 and p6 <= -1 + p1 and51; p7 <= -1 + p1 and p8 <= -1 + p1 and p9 <= -1 + p152; and p10 <= -1 + p1 and p11 <= -1 + p1 and p12 >= p1;53;54; and this isl_pw_aff is then 1:1 translated into a isl ast expression.55;56; In the best case, we would create a run-time check such as:57;58; if (B[99 + max(p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12)] < A[0]59; || A[99] B[min(p0, p1, p2, p3, p4, p5, p6, p7, p8, p9]))60;61 62target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"63 64; Function Attrs: nounwind uwtable65define void @foo(ptr %A, ptr %B, i64 %p1, i64 %p2, i64 %p3, i64 %p4, i64 %p5, i64 %p6, i64 %p7, i64 %p8, i64 %p9, i64 %p10, i64 %p11, i64 %p12) #0 {66entry:67 br label %entry.split68 69entry.split: ; preds = %entry70 br label %for.body71 72for.body: ; preds = %entry.split, %for.body73 %i.01 = phi i64 [ 0, %entry.split ], [ %tmp25, %for.body ]74 %tmp = add i64 %p1, %i.0175 %arrayidx = getelementptr float, ptr %B, i64 %tmp76 %tmp2 = add i64 %p2, %i.0177 %arrayidx2 = getelementptr float, ptr %B, i64 %tmp278 %tmp3 = add i64 %p3, %i.0179 %arrayidx5 = getelementptr float, ptr %B, i64 %tmp380 %tmp4 = add i64 %p4, %i.0181 %arrayidx8 = getelementptr float, ptr %B, i64 %tmp482 %tmp5 = add i64 %p5, %i.0183 %arrayidx11 = getelementptr float, ptr %B, i64 %tmp584 %tmp6 = add i64 %p6, %i.0185 %arrayidx14 = getelementptr float, ptr %B, i64 %tmp686 %tmp7 = add i64 %p7, %i.0187 %arrayidx17 = getelementptr float, ptr %B, i64 %tmp788 %tmp8 = add i64 %p8, %i.0189 %arrayidx20 = getelementptr float, ptr %B, i64 %tmp890 %tmp9 = add i64 %p9, %i.0191 %arrayidx23 = getelementptr float, ptr %B, i64 %tmp992 %tmp10 = add i64 %p10, %i.0193 %arrayidx26 = getelementptr float, ptr %B, i64 %tmp1094 %tmp11 = add i64 %p11, %i.0195 %arrayidx29 = getelementptr float, ptr %B, i64 %tmp1196 %tmp12 = add i64 %p12, %i.0197 %arrayidx32 = getelementptr float, ptr %B, i64 %tmp1298 %arrayidx34 = getelementptr float, ptr %A, i64 %i.0199 %tmp13 = load float, ptr %arrayidx, align 4100 %tmp14 = load float, ptr %arrayidx2, align 4101 %add3 = fadd float %tmp13, %tmp14102 %tmp15 = load float, ptr %arrayidx5, align 4103 %add6 = fadd float %add3, %tmp15104 %tmp16 = load float, ptr %arrayidx8, align 4105 %add9 = fadd float %add6, %tmp16106 %tmp17 = load float, ptr %arrayidx11, align 4107 %add12 = fadd float %add9, %tmp17108 %tmp18 = load float, ptr %arrayidx14, align 4109 %add15 = fadd float %add12, %tmp18110 %tmp19 = load float, ptr %arrayidx17, align 4111 %add18 = fadd float %add15, %tmp19112 %tmp20 = load float, ptr %arrayidx20, align 4113 %add21 = fadd float %add18, %tmp20114 %tmp21 = load float, ptr %arrayidx23, align 4115 %add24 = fadd float %add21, %tmp21116 %tmp22 = load float, ptr %arrayidx26, align 4117 %add27 = fadd float %add24, %tmp22118 %tmp23 = load float, ptr %arrayidx29, align 4119 %add30 = fadd float %add27, %tmp23120 %tmp24 = load float, ptr %arrayidx32, align 4121 %add33 = fadd float %add30, %tmp24122 store float %add33, ptr %arrayidx34, align 4123 %tmp25 = add nsw i64 %i.01, 1124 %exitcond = icmp ne i64 %tmp25, 100125 br i1 %exitcond, label %for.body, label %for.end126 127for.end: ; preds = %for.body128 ret void129}130