85 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<import-jscop;codegen>' -polly-import-jscop-postfix=transformed -verify-dom-info -polly-allow-nonaffine-loops -S < %s | FileCheck %s2 3; This test verifies that partial writes within non-affine loops are code4; generated correctly.5 6; CHECK:polly.stmt.bb3:7; CHECK-NEXT: %polly.subregion.iv = phi i32 [ %polly.subregion.iv.inc, %polly.stmt.bb5.cont ], [ 0, %polly.stmt.bb3.entry ]8; CHECK-NEXT: %polly.j.0 = phi i64 [ %j.0.phiops.reload, %polly.stmt.bb3.entry ], [ %p_tmp10, %polly.stmt.bb5.cont ]9; CHECK-NEXT: %8 = zext nneg i64 %polly.indvar to i6510; CHECK-NEXT: %9 = add nsw i64 %polly.indvar, -111; CHECK-NEXT: %10 = zext i64 %9 to i6512; CHECK-NEXT: %11 = mul i65 %8, %1013; CHECK-NEXT: %12 = lshr i65 %11, 114; CHECK-NEXT: %13 = trunc i65 %12 to i6415; CHECK-NEXT: %14 = shl i64 %13, 116; CHECK-NEXT: %15 = add i64 %polly.indvar, %1417; CHECK-NEXT: %p_tmp4 = icmp slt i64 %polly.j.0, %1518; CHECK-NEXT: %polly.subregion.iv.inc = add i32 %polly.subregion.iv, 119; CHECK-NEXT: br i1 %p_tmp4, label %polly.stmt.bb5, label %polly.stmt.bb11.exit20 21; CHECK:polly.stmt.bb5:22; CHECK-NEXT: %p_tmp6 = getelementptr inbounds float, ptr %B, i64 4223; CHECK-NEXT: %tmp7_p_scalar_ = load float, ptr %p_tmp624; CHECK-NEXT: %p_tmp8 = fadd float %tmp7_p_scalar_, 1.000000e+0025; CHECK-NEXT: %16 = icmp sle i64 %polly.indvar, 926; CHECK-NEXT: %polly.Stmt_bb3__TO__bb11_MayWrite2.cond = icmp ne i1 %16, false27; CHECK-NEXT: br i1 %polly.Stmt_bb3__TO__bb11_MayWrite2.cond, label %polly.stmt.bb5.Stmt_bb3__TO__bb11_MayWrite2.partial, label %polly.stmt.bb5.cont28 29; CHECK:polly.stmt.bb5.Stmt_bb3__TO__bb11_MayWrite2.partial: ; preds = %polly.stmt.bb530; CHECK-NEXT: %polly.access.B3 = getelementptr float, ptr %B, i64 4231; CHECK-NEXT: store float %p_tmp8, ptr %polly.access.B332; CHECK-NEXT: br label %polly.stmt.bb5.cont33 34; CHECK:polly.stmt.bb5.cont:35; CHECK-NEXT: %p_tmp10 = add nuw nsw i64 %polly.j.0, 136; CHECK-NEXT: br label %polly.stmt.bb337 38 39 40; void foo(long A[], float B[], long *x) {41; for (long i = 0; i < 1024; i++)42; for (long j = *x; j < i * i; j++)43; B[42]++;44; }45;46target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"47 48define void @partial_write_in_region_with_loop(ptr %A, ptr %B, ptr %xptr) {49bb:50 br label %bb151 52bb1: ; preds = %bb12, %bb53 %i.0 = phi i64 [ 0, %bb ], [ %tmp13, %bb12 ]54 %exitcond = icmp ne i64 %i.0, 102455 br i1 %exitcond, label %bb2, label %bb1456 57bb2: ; preds = %bb158 %x = load i64, ptr %xptr59 br label %bb360 61bb3: ; preds = %bb9, %bb262 %j.0 = phi i64 [ %x, %bb2 ], [ %tmp10, %bb5 ]63 %tmp = mul nsw i64 %i.0, %i.064 %tmp4 = icmp slt i64 %j.0, %tmp65 br i1 %tmp4, label %bb5, label %bb1166 67bb5: ; preds = %bb368 %tmp6 = getelementptr inbounds float, ptr %B, i64 4269 %tmp7 = load float, ptr %tmp6, align 470 %tmp8 = fadd float %tmp7, 1.000000e+0071 store float %tmp8, ptr %tmp6, align 472 %tmp10 = add nuw nsw i64 %j.0, 173 br label %bb374 75bb11: ; preds = %bb376 br label %bb1277 78bb12: ; preds = %bb1179 %tmp13 = add nuw nsw i64 %i.0, 180 br label %bb181 82bb14: ; preds = %bb183 ret void84}85