[LV] Update check to find epilogue resume value to check all incoming.

This fixes a crash where all incoming values for the epilogue resume
value are zero, because there are no remaining iterations to execute for
the epilogue loop.

Author: fhahn

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -9765,7 +9765,10 @@ preparePlanForEpilogueVectorLoop(VPlan &Plan, Loop *L,
                 match(
                     P.getIncomingValueForBlock(EPI.MainLoopIterationCountCheck),
                     m_SpecificInt(0)) &&
-                is_contained(P.incoming_values(), EPI.VectorTripCount))
+                all_of(P.incoming_values(), [&EPI](Value *Inc) {
+                  return Inc == EPI.VectorTripCount ||
+                         match(Inc, m_SpecificInt(0));
+                }))
               return &P;
             return nullptr;
           });

llvm/test/Transforms/LoopVectorize/AArch64/sve-epilog-vect-no-remaining-iterations.ll

@@ -0,0 +1,146 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 5
+; RUN: opt -passes=loop-vectorize -S %s | FileCheck %s
+
+target triple = "aarch64-linux-gnu"
+
+define i64 @main_vector_loop_fixed_with_no_remaining_iterations(ptr %src, ptr noalias %dst, i32 %x) #0 {
+; CHECK-LABEL: define i64 @main_vector_loop_fixed_with_no_remaining_iterations(
+; CHECK-SAME: ptr [[SRC:%.*]], ptr noalias [[DST:%.*]], i32 [[X:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ITER_CHECK:.*]]:
+; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP3:%.*]] = mul i64 [[TMP2]], 2
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 16, [[TMP3]]
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]]
+; CHECK:       [[VECTOR_MAIN_LOOP_ITER_CHECK]]:
+; CHECK-NEXT:    br i1 true, label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i32> poison, i32 [[X]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT]], <16 x i32> poison, <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP0:%.*]] = call <16 x i32> @llvm.abs.v16i32(<16 x i32> [[BROADCAST_SPLAT]], i1 false)
+; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i32> @llvm.abs.v16i32(<16 x i32> [[BROADCAST_SPLAT]], i1 false)
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi <16 x i64> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP17:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr { [4 x i8] }, ptr [[SRC]], i64 [[INDEX]], i32 0, i64 3
+; CHECK-NEXT:    [[WIDE_VEC2:%.*]] = load <64 x i8>, ptr [[TMP4]], align 1
+; CHECK-NEXT:    [[STRIDED_VEC3:%.*]] = shufflevector <64 x i8> [[WIDE_VEC2]], <64 x i8> poison, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>
+; CHECK-NEXT:    [[TMP6:%.*]] = zext <16 x i8> [[STRIDED_VEC3]] to <16 x i32>
+; CHECK-NEXT:    [[TMP8:%.*]] = call <16 x i32> @llvm.umin.v16i32(<16 x i32> [[TMP0]], <16 x i32> [[TMP6]])
+; CHECK-NEXT:    [[TMP10:%.*]] = call <16 x i32> @llvm.umin.v16i32(<16 x i32> [[TMP1]], <16 x i32> [[TMP8]])
+; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[INDEX]]
+; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[TMP11]], i32 0
+; CHECK-NEXT:    store <16 x i8> zeroinitializer, ptr [[TMP12]], align 1
+; CHECK-NEXT:    [[TMP15:%.*]] = zext <16 x i32> [[TMP10]] to <16 x i64>
+; CHECK-NEXT:    [[TMP17]] = or <16 x i64> [[VEC_PHI1]], [[TMP15]]
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
+; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[TMP18:%.*]] = call i64 @llvm.vector.reduce.or.v16i64(<16 x i64> [[TMP17]])
+; CHECK-NEXT:    br label %[[VEC_EPILOG_ITER_CHECK:.*]]
+; CHECK:       [[VEC_EPILOG_ITER_CHECK]]:
+; CHECK-NEXT:    [[TMP13:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP14:%.*]] = mul i64 [[TMP13]], 2
+; CHECK-NEXT:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ule i64 16, [[TMP14]]
+; CHECK-NEXT:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]]
+; CHECK:       [[VEC_EPILOG_PH]]:
+; CHECK-NEXT:    [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 0, %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
+; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP18]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
+; CHECK-NEXT:    [[TMP31:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP16:%.*]] = mul i64 [[TMP31]], 2
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 16, [[TMP16]]
+; CHECK-NEXT:    [[TMP32:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
+; CHECK-NEXT:    [[TMP36:%.*]] = select i1 [[TMP32]], i64 [[TMP16]], i64 [[N_MOD_VF]]
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 16, [[TMP36]]
+; CHECK-NEXT:    [[TMP19:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT:    [[TMP20:%.*]] = mul i64 [[TMP19]], 2
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i32> poison, i32 [[X]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i32> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i32> poison, <vscale x 2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <vscale x 2 x i64> zeroinitializer, i64 [[BC_MERGE_RDX]], i32 0
+; CHECK-NEXT:    [[TMP22:%.*]] = call <vscale x 2 x i32> @llvm.abs.nxv2i32(<vscale x 2 x i32> [[BROADCAST_SPLAT2]], i1 false)
+; CHECK-NEXT:    [[TMP23:%.*]] = call <vscale x 2 x i32> @llvm.abs.nxv2i32(<vscale x 2 x i32> [[BROADCAST_SPLAT2]], i1 false)
+; CHECK-NEXT:    [[TMP24:%.*]] = call <vscale x 2 x i64> @llvm.stepvector.nxv2i64()
+; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[VEC_EPILOG_RESUME_VAL]], i64 0
+; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[DOTSPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP25:%.*]] = mul <vscale x 2 x i64> [[TMP24]], splat (i64 1)
+; CHECK-NEXT:    [[INDUCTION:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT]], [[TMP25]]
+; CHECK-NEXT:    [[TMP37:%.*]] = mul i64 1, [[TMP20]]
+; CHECK-NEXT:    [[DOTSPLATINSERT4:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP37]], i64 0
+; CHECK-NEXT:    [[DOTSPLAT5:%.*]] = shufflevector <vscale x 2 x i64> [[DOTSPLATINSERT4]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
+; CHECK-NEXT:    br label %[[VEC_EPILOG_VECTOR_BODY:.*]]
+; CHECK:       [[VEC_EPILOG_VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX6:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], %[[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT10:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 2 x i64> [ [[INDUCTION]], %[[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI6:%.*]] = phi <vscale x 2 x i64> [ [[TMP21]], %[[VEC_EPILOG_PH]] ], [ [[TMP34:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP38:%.*]] = getelementptr { [4 x i8] }, ptr [[SRC]], <vscale x 2 x i64> [[VEC_IND]], i32 0, i64 3
+; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 2 x i8> @llvm.masked.gather.nxv2i8.nxv2p0(<vscale x 2 x ptr> [[TMP38]], i32 1, <vscale x 2 x i1> splat (i1 true), <vscale x 2 x i8> poison)
+; CHECK-NEXT:    [[TMP28:%.*]] = zext <vscale x 2 x i8> [[WIDE_MASKED_GATHER]] to <vscale x 2 x i32>
+; CHECK-NEXT:    [[TMP29:%.*]] = call <vscale x 2 x i32> @llvm.umin.nxv2i32(<vscale x 2 x i32> [[TMP22]], <vscale x 2 x i32> [[TMP28]])
+; CHECK-NEXT:    [[TMP39:%.*]] = call <vscale x 2 x i32> @llvm.umin.nxv2i32(<vscale x 2 x i32> [[TMP23]], <vscale x 2 x i32> [[TMP29]])
+; CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[INDEX6]]
+; CHECK-NEXT:    [[TMP27:%.*]] = getelementptr inbounds i8, ptr [[TMP26]], i32 0
+; CHECK-NEXT:    store <vscale x 2 x i8> zeroinitializer, ptr [[TMP27]], align 1
+; CHECK-NEXT:    [[TMP33:%.*]] = zext <vscale x 2 x i32> [[TMP39]] to <vscale x 2 x i64>
+; CHECK-NEXT:    [[TMP34]] = or <vscale x 2 x i64> [[VEC_PHI6]], [[TMP33]]
+; CHECK-NEXT:    [[INDEX_NEXT10]] = add nuw i64 [[INDEX6]], [[TMP20]]
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[DOTSPLAT5]]
+; CHECK-NEXT:    [[TMP35:%.*]] = icmp eq i64 [[INDEX_NEXT10]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP35]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK:       [[VEC_EPILOG_MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[TMP30:%.*]] = call i64 @llvm.vector.reduce.or.nxv2i64(<vscale x 2 x i64> [[TMP34]])
+; CHECK-NEXT:    br label %[[VEC_EPILOG_SCALAR_PH]]
+; CHECK:       [[VEC_EPILOG_SCALAR_PH]]:
+; CHECK-NEXT:    [[BC_RESUME_VAL8:%.*]] = phi i64 [ [[N_VEC]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 0, %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[ITER_CHECK]] ]
+; CHECK-NEXT:    [[BC_MERGE_RDX9:%.*]] = phi i64 [ [[TMP30]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[TMP18]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[ITER_CHECK]] ]
+; CHECK-NEXT:    br label %[[LOOP:.*]]
+; CHECK:       [[LOOP]]:
+; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL8]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[RED:%.*]] = phi i64 [ [[BC_MERGE_RDX9]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[RED_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[GEP_SRC_I_I:%.*]] = getelementptr { [4 x i8] }, ptr [[SRC]], i64 [[IV]], i32 0, i64 3
+; CHECK-NEXT:    [[L:%.*]] = load i8, ptr [[GEP_SRC_I_I]], align 1
+; CHECK-NEXT:    [[L_EXT:%.*]] = zext i8 [[L]] to i32
+; CHECK-NEXT:    [[ABS_0:%.*]] = call i32 @llvm.abs.i32(i32 [[X]], i1 false)
+; CHECK-NEXT:    [[MIN_0:%.*]] = call i32 @llvm.umin.i32(i32 [[ABS_0]], i32 [[L_EXT]])
+; CHECK-NEXT:    [[ABS_1:%.*]] = call i32 @llvm.abs.i32(i32 [[X]], i1 false)
+; CHECK-NEXT:    [[MIN_1:%.*]] = call i32 @llvm.umin.i32(i32 [[ABS_1]], i32 [[MIN_0]])
+; CHECK-NEXT:    [[GEP_DST:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[IV]]
+; CHECK-NEXT:    store i8 0, ptr [[GEP_DST]], align 1
+; CHECK-NEXT:    [[MIN_EXT:%.*]] = zext i32 [[MIN_1]] to i64
+; CHECK-NEXT:    [[RED_NEXT]] = or i64 [[RED]], [[MIN_EXT]]
+; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
+; CHECK-NEXT:    [[EXITCOND_NOT_I_I:%.*]] = icmp eq i64 [[IV_NEXT]], 16
+; CHECK-NEXT:    br i1 [[EXITCOND_NOT_I_I]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    [[RED_NEXT_LCSSA:%.*]] = phi i64 [ [[RED_NEXT]], %[[LOOP]] ]
+; CHECK-NEXT:    ret i64 [[RED_NEXT_LCSSA]]
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+  %red = phi i64 [ 0, %entry ], [ %red.next, %loop ]
+  %gep.src.i.i = getelementptr { [4 x i8] }, ptr %src, i64 %iv, i32 0, i64 3
+  %l = load i8, ptr %gep.src.i.i, align 1
+  %l.ext = zext i8 %l to i32
+  %abs.0 = call i32 @llvm.abs.i32(i32 %x, i1 false)
+  %min.0 = call i32 @llvm.umin.i32(i32 %abs.0, i32 %l.ext)
+  %abs.1 = call i32 @llvm.abs.i32(i32 %x, i1 false)
+  %min.1 = call i32 @llvm.umin.i32(i32 %abs.1, i32 %min.0)
+  %gep.dst = getelementptr inbounds i8, ptr  %dst, i64 %iv
+  store i8 0, ptr %gep.dst, align 1
+  %min.ext = zext i32 %min.1 to i64
+  %red.next = or i64 %red, %min.ext
+  %iv.next = add i64 %iv, 1
+  %exitcond.not.i.i = icmp eq i64 %iv.next, 16
+  br i1 %exitcond.not.i.i, label %exit, label %loop
+
+exit:
+  ret i64 %red.next
+}
+
+declare i32 @llvm.umin.i32(i32, i32)
+
+declare i32 @llvm.abs.i32(i32, i1 immarg)
+
+attributes #0 = { "target-cpu"="neoverse-512tvb" }