test/Transforms/LoopVectorize/bounded-load-scalable.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "scalar.ph:" --version 6
; RUN: opt -passes=loop-vectorize -force-target-supports-scalable-vectors=true -S %s | FileCheck %s

; FIXME:
; A bounded (i % 2^N) load can be widened for a scalable VF if the bound is
; known to be a multiple of the runtime VF. As vscale is a power of two, it is
; sufficient to check the maximum runtime VF.

; FIXME: The maximum runtime VF (vscale x 4 with vscale <= 16, i.e. 64) is equal to the
; bound, so the load is widened.
define i32 @bounded_load_bound64_max_vscale_16(ptr %A, i64 %N) #0 {
; CHECK-LABEL: define i32 @bounded_load_bound64_max_vscale_16(
; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 2
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK:       [[VECTOR_SCEVCHECK]]:
; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[N]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt i64 [[TMP2]], 63
; CHECK-NEXT:    br i1 [[TMP3]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK:       [[VECTOR_PH]]:
; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP1]]
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
; CHECK:       [[VECTOR_BODY]]:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP4:%.*]] = and i64 [[INDEX]], 63
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP4]]
; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP5]], align 4
; CHECK-NEXT:    [[TMP6]] = add <vscale x 4 x i32> [[VEC_PHI]], [[WIDE_LOAD]]
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP1]]
; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK:       [[MIDDLE_BLOCK]]:
; CHECK-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP6]])
; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK:       [[SCALAR_PH]]:
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %sum = phi i32 [ 0, %entry ], [ %sum.next, %loop ]
  %bounded = urem i64 %iv, 64
  %gep.A = getelementptr inbounds i32, ptr %A, i64 %bounded
  %lv = load i32, ptr %gep.A, align 4
  %sum.next = add i32 %sum, %lv
  %iv.next = add nuw nsw i64 %iv, 1
  %cond = icmp eq i64 %iv.next, %N
  br i1 %cond, label %exit, label %loop, !llvm.loop !0

exit:
  %r = phi i32 [ %sum.next, %loop ]
  ret i32 %r
}

; FIXME: The maximum runtime VF (64) exceeds the bound, so the load cannot be widened
; for vscale x 4 and a fixed VF is used instead.
define i32 @bounded_load_bound32_max_vscale_16(ptr %A, i64 %N) #0 {
; CHECK-LABEL: define i32 @bounded_load_bound32_max_vscale_16(
; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 2
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK:       [[VECTOR_SCEVCHECK]]:
; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[N]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt i64 [[TMP2]], 31
; CHECK-NEXT:    br i1 [[TMP3]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK:       [[VECTOR_PH]]:
; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP1]]
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
; CHECK:       [[VECTOR_BODY]]:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP4:%.*]] = and i64 [[INDEX]], 31
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP4]]
; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP5]], align 4
; CHECK-NEXT:    [[TMP6]] = add <vscale x 4 x i32> [[VEC_PHI]], [[WIDE_LOAD]]
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP1]]
; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK:       [[MIDDLE_BLOCK]]:
; CHECK-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP6]])
; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK:       [[SCALAR_PH]]:
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %sum = phi i32 [ 0, %entry ], [ %sum.next, %loop ]
  %bounded = urem i64 %iv, 32
  %gep.A = getelementptr inbounds i32, ptr %A, i64 %bounded
  %lv = load i32, ptr %gep.A, align 4
  %sum.next = add i32 %sum, %lv
  %iv.next = add nuw nsw i64 %iv, 1
  %cond = icmp eq i64 %iv.next, %N
  br i1 %cond, label %exit, label %loop, !llvm.loop !0

exit:
  %r = phi i32 [ %sum.next, %loop ]
  ret i32 %r
}

; FIXME: With vscale <= 2 the maximum runtime VF is 8 and divides the bound.
define i32 @bounded_load_bound8_max_vscale_2(ptr %A, i64 %N) #1 {
; CHECK-LABEL: define i32 @bounded_load_bound8_max_vscale_2(
; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) #[[ATTR1:[0-9]+]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 2
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK:       [[VECTOR_SCEVCHECK]]:
; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[N]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt i64 [[TMP2]], 7
; CHECK-NEXT:    br i1 [[TMP3]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK:       [[VECTOR_PH]]:
; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP1]]
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
; CHECK:       [[VECTOR_BODY]]:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP4:%.*]] = and i64 [[INDEX]], 7
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP4]]
; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP5]], align 4
; CHECK-NEXT:    [[TMP6]] = add <vscale x 4 x i32> [[VEC_PHI]], [[WIDE_LOAD]]
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP1]]
; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK:       [[MIDDLE_BLOCK]]:
; CHECK-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP6]])
; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK:       [[SCALAR_PH]]:
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %sum = phi i32 [ 0, %entry ], [ %sum.next, %loop ]
  %bounded = urem i64 %iv, 8
  %gep.A = getelementptr inbounds i32, ptr %A, i64 %bounded
  %lv = load i32, ptr %gep.A, align 4
  %sum.next = add i32 %sum, %lv
  %iv.next = add nuw nsw i64 %iv, 1
  %cond = icmp eq i64 %iv.next, %N
  br i1 %cond, label %exit, label %loop, !llvm.loop !0

exit:
  %r = phi i32 [ %sum.next, %loop ]
  ret i32 %r
}

; Without vscale_range the maximum value of vscale is unknown, so the load
; cannot be widened for a scalable VF.
define i32 @bounded_load_unknown_max_vscale(ptr %A, i64 %N) {
; CHECK-LABEL: define i32 @bounded_load_unknown_max_vscale(
; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 2
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK:       [[VECTOR_SCEVCHECK]]:
; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[N]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt i64 [[TMP2]], 63
; CHECK-NEXT:    br i1 [[TMP3]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK:       [[VECTOR_PH]]:
; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP1]]
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
; CHECK:       [[VECTOR_BODY]]:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP4:%.*]] = and i64 [[INDEX]], 63
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP4]]
; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP5]], align 4
; CHECK-NEXT:    [[TMP6]] = add <vscale x 4 x i32> [[VEC_PHI]], [[WIDE_LOAD]]
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP1]]
; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK:       [[MIDDLE_BLOCK]]:
; CHECK-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP6]])
; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK:       [[SCALAR_PH]]:
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %sum = phi i32 [ 0, %entry ], [ %sum.next, %loop ]
  %bounded = urem i64 %iv, 64
  %gep.A = getelementptr inbounds i32, ptr %A, i64 %bounded
  %lv = load i32, ptr %gep.A, align 4
  %sum.next = add i32 %sum, %lv
  %iv.next = add nuw nsw i64 %iv, 1
  %cond = icmp eq i64 %iv.next, %N
  br i1 %cond, label %exit, label %loop, !llvm.loop !0

exit:
  %r = phi i32 [ %sum.next, %loop ]
  ret i32 %r
}

define i32 @bounded_load_bound64_user_ic(ptr %A, i64 %N) #0 {
; CHECK-LABEL: define i32 @bounded_load_bound64_user_ic(
; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT:    [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 3
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK:       [[VECTOR_SCEVCHECK]]:
; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[N]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt i64 [[TMP2]], 63
; CHECK-NEXT:    br i1 [[TMP3]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK:       [[VECTOR_PH]]:
; CHECK-NEXT:    [[TMP4:%.*]] = shl nuw i64 [[TMP0]], 2
; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP1]]
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
; CHECK:       [[VECTOR_BODY]]:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP8:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi <vscale x 4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP9:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP5:%.*]] = and i64 [[INDEX]], 63
; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP5]]
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i64 [[TMP4]]
; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP6]], align 4
; CHECK-NEXT:    [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i32>, ptr [[TMP7]], align 4
; CHECK-NEXT:    [[TMP8]] = add <vscale x 4 x i32> [[VEC_PHI]], [[WIDE_LOAD]]
; CHECK-NEXT:    [[TMP9]] = add <vscale x 4 x i32> [[VEC_PHI1]], [[WIDE_LOAD2]]
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP1]]
; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; CHECK:       [[MIDDLE_BLOCK]]:
; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <vscale x 4 x i32> [[TMP9]], [[TMP8]]
; CHECK-NEXT:    [[TMP11:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[BIN_RDX]])
; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[CMP_N]], [[EXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK:       [[SCALAR_PH]]:
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %sum = phi i32 [ 0, %entry ], [ %sum.next, %loop ]
  %bounded = urem i64 %iv, 64
  %gep.A = getelementptr inbounds i32, ptr %A, i64 %bounded
  %lv = load i32, ptr %gep.A, align 4
  %sum.next = add i32 %sum, %lv
  %iv.next = add nuw nsw i64 %iv, 1
  %cond = icmp eq i64 %iv.next, %N
  br i1 %cond, label %exit, label %loop, !llvm.loop !4

exit:
  %r = phi i32 [ %sum.next, %loop ]
  ret i32 %r
}

attributes #0 = { vscale_range(1,16) }
attributes #1 = { vscale_range(1,2) }

!0 = distinct !{!0, !1, !2, !3}
!1 = !{!"llvm.loop.vectorize.width", i32 4}
!2 = !{!"llvm.loop.vectorize.scalable.enable"}
!3 = !{!"llvm.loop.interleave.count", i32 1}
!4 = distinct !{!4, !1, !2, !5}
!5 = !{!"llvm.loop.interleave.count", i32 2}