155 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 computed correctly5# when barrier instructions exist in the loop.6# The original code is as follows.7# 8# ```9# volatile int x = 0;10# void f(int * restrict a, int * restrict b, int * restrict c, int n) {11# for (int i = 0; i < n; i++) {12# a[i] *= c[i];13# b[i] *= c[i];14# x += i;15# a[i + 1] *= i;16# x += i;17# b[i + 1] *= i;18# }19# }20# ```21#22# FIXME: Currently the following dependencies are missed.23# Loop carried edges from SU(16)24# Order25# SU(6)26# SU(8)27# SU(10)28# SU(11)29# Loop carried edges from SU(17)30# Order31# SU(10)32# SU(11)33# Loop carried edges from SU(19)34# Order35# SU(10)36# SU(11)37 38# CHECK: ===== Loop Carried Edges Begin =====39# CHECK-NEXT: ===== Loop Carried Edges End =====40 41--- |42 @x = dso_local global i32 0, align 443 44 define dso_local void @f(ptr noalias nocapture noundef %a, ptr noalias nocapture noundef %b, ptr noalias nocapture noundef readonly %c, i32 noundef %n) {45 entry:46 %cmp26 = icmp sgt i32 %n, 047 br i1 %cmp26, label %for.body.preheader, label %for.cond.cleanup48 49 for.body.preheader:50 %.pre = load i32, ptr %a, align 4, !tbaa !551 %.pre28 = load i32, ptr %b, align 4, !tbaa !552 %cgep = getelementptr i8, ptr %b, i32 453 %cgep37 = getelementptr i8, ptr %a, i32 454 br label %for.body55 56 for.cond.cleanup:57 ret void58 59 for.body:60 %lsr.iv35 = phi ptr [ %c, %for.body.preheader ], [ %cgep42, %for.body ]61 %lsr.iv31 = phi ptr [ %cgep37, %for.body.preheader ], [ %cgep41, %for.body ]62 %lsr.iv = phi ptr [ %cgep, %for.body.preheader ], [ %cgep40, %for.body ]63 %0 = phi i32 [ %mul11, %for.body ], [ %.pre28, %for.body.preheader ]64 %1 = phi i32 [ %mul7, %for.body ], [ %.pre, %for.body.preheader ]65 %i.027 = phi i32 [ %add5, %for.body ], [ 0, %for.body.preheader ]66 %2 = load i32, ptr %lsr.iv35, align 4, !tbaa !567 %mul = mul nsw i32 %1, %268 %cgep38 = getelementptr i8, ptr %lsr.iv31, i32 -469 store i32 %mul, ptr %cgep38, align 4, !tbaa !570 %mul4 = mul nsw i32 %0, %271 %cgep39 = getelementptr i8, ptr %lsr.iv, i32 -472 store i32 %mul4, ptr %cgep39, align 4, !tbaa !573 %3 = load volatile i32, ptr @x, align 4, !tbaa !574 %4 = add i32 %i.027, %375 store volatile i32 %4, ptr @x, align 4, !tbaa !576 %add5 = add nuw nsw i32 %i.027, 177 %5 = load i32, ptr %lsr.iv31, align 4, !tbaa !578 %mul7 = mul nsw i32 %5, %i.02779 store i32 %mul7, ptr %lsr.iv31, align 4, !tbaa !580 %6 = load volatile i32, ptr @x, align 4, !tbaa !581 %7 = add i32 %i.027, %682 store volatile i32 %7, ptr @x, align 4, !tbaa !583 %8 = load i32, ptr %lsr.iv, align 4, !tbaa !584 %mul11 = mul nsw i32 %8, %i.02785 store i32 %mul11, ptr %lsr.iv, align 4, !tbaa !586 %exitcond.not = icmp eq i32 %n, %add587 %cgep40 = getelementptr i8, ptr %lsr.iv, i32 488 %cgep41 = getelementptr i8, ptr %lsr.iv31, i32 489 %cgep42 = getelementptr i8, ptr %lsr.iv35, i32 490 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body91 }92 93 !5 = !{!6, !6, i64 0}94 !6 = !{!"int", !7, i64 0}95 !7 = !{!"omnipotent char", !8, i64 0}96 !8 = !{!"Simple C/C++ TBAA"}97 98...99---100name: f101tracksRegLiveness: true102body: |103 bb.0.entry:104 successors: %bb.1, %bb.2105 liveins: $r0, $r1, $r2, $r3106 107 %19:intregs = COPY $r3108 %18:intregs = COPY $r2109 %17:intregs = COPY $r1110 %16:intregs = COPY $r0111 %20:predregs = C2_cmpgti %19, 0112 J2_jumpf %20, %bb.2, implicit-def dead $pc113 J2_jump %bb.1, implicit-def dead $pc114 115 bb.1.for.body.preheader:116 %0:intregs, %3:intregs = L2_loadri_pi %16, 4 :: (load (s32) from %ir.a, !tbaa !5)117 %1:intregs, %2:intregs = L2_loadri_pi %17, 4 :: (load (s32) from %ir.b, !tbaa !5)118 %22:intregs = A2_tfrsi 0119 %26:intregs = C4_addipc target-flags(hexagon-pcrel) @x120 %30:intregs = COPY %19121 J2_loop0r %bb.3, %30, implicit-def $lc0, implicit-def $sa0, implicit-def $usr122 J2_jump %bb.3, implicit-def dead $pc123 124 bb.2.for.cond.cleanup:125 PS_jmpret $r31, implicit-def dead $pc126 127 bb.3.for.body:128 successors: %bb.2, %bb.3129 130 %4:intregs = PHI %18, %bb.1, %15, %bb.3131 %5:intregs = PHI %3, %bb.1, %14, %bb.3132 %6:intregs = PHI %2, %bb.1, %13, %bb.3133 %7:intregs = PHI %1, %bb.1, %12, %bb.3134 %8:intregs = PHI %0, %bb.1, %11, %bb.3135 %9:intregs = PHI %22, %bb.1, %10, %bb.3136 %23:intregs, %15:intregs = L2_loadri_pi %4, 4 :: (load (s32) from %ir.lsr.iv35, !tbaa !5)137 %24:intregs = nsw M2_mpyi %8, %23138 S2_storeri_io %5, -4, killed %24 :: (store (s32) into %ir.cgep38, !tbaa !5)139 %25:intregs = nsw M2_mpyi %7, %23140 S2_storeri_io %6, -4, killed %25 :: (store (s32) into %ir.cgep39, !tbaa !5)141 L4_add_memopw_io %26, 0, %9 :: (volatile store (s32) into @x, !tbaa !5), (volatile dereferenceable load (s32) from @x, !tbaa !5)142 %10:intregs = nuw nsw A2_addi %9, 1143 %27:intregs = L2_loadri_io %5, 0 :: (load (s32) from %ir.lsr.iv31, !tbaa !5)144 %11:intregs = nsw M2_mpyi killed %27, %9145 S2_storeri_io %5, 0, %11 :: (store (s32) into %ir.lsr.iv31, !tbaa !5)146 L4_add_memopw_io %26, 0, %9 :: (volatile store (s32) into @x, !tbaa !5), (volatile dereferenceable load (s32) from @x, !tbaa !5)147 %28:intregs = L2_loadri_io %6, 0 :: (load (s32) from %ir.lsr.iv, !tbaa !5)148 %12:intregs = nsw M2_mpyi killed %28, %9149 S2_storeri_io %6, 0, %12 :: (store (s32) into %ir.lsr.iv, !tbaa !5)150 %13:intregs = A2_addi %6, 4151 %14:intregs = A2_addi %5, 4152 ENDLOOP0 %bb.3, implicit-def $pc, implicit-def $lc0, implicit $sa0, implicit $lc0153 J2_jump %bb.2, implicit-def $pc154...155