110 lines · plain
1# RUN: llc -mtriple=hexagon -run-pass pipeliner -debug-only=pipeliner %s -o /dev/null 2>&1 -pipeliner-experimental-cg=true | FileCheck %s2# REQUIRES: asserts3 4# Test that loop carried memory dependencies are correctly added when two5# arrays may point to the same memory location.6#7# ```8# void f(int *a, int *b, int n) {9# for (int i = 0; i < n; i++) {10# a[i] += b[i];11# b[i] += a[i];12# }13# }14# ```15#16# Here is what each instruction does.17# SU(2): Load b[i]18# SU(3): Load a[i]19# SU(5): Store a[i]20# SU(6): Load b[i]21# SU(8): Store b[i]22#23# Note that if there is already a dependency between two instructions, we don't24# add loop-carried on between them since non-loop-carried one imposes stronger25# constraint than loop-carried one.26 27# CHECK: ===== Loop Carried Edges Begin =====28# CHECK-NEXT: Loop carried edges from SU(5)29# CHECK-NEXT: Order30# CHECK-NEXT: SU(2)31# CHECK-NEXT: Loop carried edges from SU(6)32# CHECK-NEXT: Order33# CHECK-NEXT: SU(5)34# CHECK-NEXT: Loop carried edges from SU(8)35# CHECK-NEXT: Order36# CHECK-NEXT: SU(3)37# CHECK-NEXT: SU(5)38# CHECK-NEXT: ===== Loop Carried Edges End =====39 40--- |41 define dso_local void @f(ptr nocapture noundef %a, ptr nocapture noundef %b, i32 noundef %n) local_unnamed_addr {42 entry:43 %cmp12 = icmp sgt i32 %n, 044 br i1 %cmp12, label %for.body, label %for.cond.cleanup45 46 for.cond.cleanup:47 ret void48 49 for.body:50 %lsr.iv15 = phi ptr [ %cgep17, %for.body ], [ %b, %entry ]51 %lsr.iv14 = phi ptr [ %cgep, %for.body ], [ %a, %entry ]52 %lsr.iv = phi i32 [ %lsr.iv.next, %for.body ], [ %n, %entry ]53 %0 = load i32, ptr %lsr.iv15, align 4, !tbaa !554 %1 = load i32, ptr %lsr.iv14, align 4, !tbaa !555 %add = add nsw i32 %1, %056 store i32 %add, ptr %lsr.iv14, align 4, !tbaa !557 %2 = load i32, ptr %lsr.iv15, align 4, !tbaa !558 %add4 = add nsw i32 %2, %add59 store i32 %add4, ptr %lsr.iv15, align 4, !tbaa !560 %lsr.iv.next = add i32 %lsr.iv, -161 %exitcond.not = icmp eq i32 %lsr.iv.next, 062 %cgep = getelementptr i8, ptr %lsr.iv14, i32 463 %cgep17 = getelementptr i8, ptr %lsr.iv15, i32 464 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body65 }66 67 !5 = !{!6, !6, i64 0}68 !6 = !{!"int", !7, i64 0}69 !7 = !{!"omnipotent char", !8, i64 0}70 !8 = !{!"Simple C/C++ TBAA"}71 72...73---74name: f75tracksRegLiveness: true76body: |77 bb.0.entry:78 successors: %bb.3, %bb.179 liveins: $r0, $r1, $r280 81 %8:intregs = COPY $r282 %7:intregs = COPY $r183 %6:intregs = COPY $r084 %9:predregs = C2_cmpgti %8, 085 J2_jumpf %9, %bb.1, implicit-def $pc86 87 bb.3:88 %16:intregs = COPY %889 J2_loop0r %bb.2, %16, implicit-def $lc0, implicit-def $sa0, implicit-def $usr90 J2_jump %bb.2, implicit-def $pc91 92 bb.1.for.cond.cleanup:93 PS_jmpret $r31, implicit-def dead $pc94 95 bb.2.for.body:96 successors: %bb.1, %bb.297 98 %0:intregs = PHI %7, %bb.3, %5, %bb.299 %1:intregs = PHI %6, %bb.3, %4, %bb.2100 %10:intregs = L2_loadri_io %0, 0 :: (load (s32) from %ir.lsr.iv15, !tbaa !5)101 %11:intregs = L2_loadri_io %1, 0 :: (load (s32) from %ir.lsr.iv14, !tbaa !5)102 %12:intregs = nsw A2_add killed %11, killed %10103 %4:intregs = S2_storeri_pi %1, 4, %12 :: (store (s32) into %ir.lsr.iv14, !tbaa !5)104 %13:intregs = L2_loadri_io %0, 0 :: (load (s32) from %ir.lsr.iv15, !tbaa !5)105 %14:intregs = nsw A2_add killed %13, %12106 %5:intregs = S2_storeri_pi %0, 4, killed %14 :: (store (s32) into %ir.lsr.iv15, !tbaa !5)107 ENDLOOP0 %bb.2, implicit-def $pc, implicit-def $lc0, implicit $sa0, implicit $lc0108 J2_jump %bb.1, implicit-def $pc109...110