; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=ve-unknown-unknown -mattr=+vpu | FileCheck %s

@v256i64 = common dso_local local_unnamed_addr global <256 x i64> zeroinitializer, align 16

; Function Attrs: norecurse nounwind readonly
define fastcc <256 x i64> @loadv256i64(ptr nocapture readonly) {
; CHECK-LABEL: loadv256i64:
; CHECK:       # %bb.0:
; CHECK-NEXT:    lea %s1, 256
; CHECK-NEXT:    lvl %s1
; CHECK-NEXT:    vld %v0, 8, %s0
; CHECK-NEXT:    b.l.t (, %s10)
  %2 = load <256 x i64>, ptr %0, align 16
  ret <256 x i64> %2
}

; Function Attrs: norecurse nounwind readonly
define fastcc <256 x double> @loadv256f64(ptr nocapture readonly) {
; CHECK-LABEL: loadv256f64:
; CHECK:       # %bb.0:
; CHECK-NEXT:    lea %s1, 256
; CHECK-NEXT:    lvl %s1
; CHECK-NEXT:    vld %v0, 8, %s0
; CHECK-NEXT:    b.l.t (, %s10)
  %2 = load <256 x double>, ptr %0, align 16
  ret <256 x double> %2
}

; Function Attrs: norecurse nounwind readonly
define fastcc <256 x i32> @loadv256i32(ptr nocapture readonly) {
; CHECK-LABEL: loadv256i32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    lea %s1, 256
; CHECK-NEXT:    lvl %s1
; CHECK-NEXT:    vldl.zx %v0, 4, %s0
; CHECK-NEXT:    b.l.t (, %s10)
  %2 = load <256 x i32>, ptr %0, align 16
  ret <256 x i32> %2
}

; Function Attrs: norecurse nounwind readonly
define fastcc <256 x float> @loadv256f32(ptr nocapture readonly) {
; CHECK-LABEL: loadv256f32:
; CHECK:       # %bb.0:
; CHECK-NEXT:    lea %s1, 256
; CHECK-NEXT:    lvl %s1
; CHECK-NEXT:    vldu %v0, 4, %s0
; CHECK-NEXT:    b.l.t (, %s10)
  %2 = load <256 x float>, ptr %0, align 16
  ret <256 x float> %2
}

; Function Attrs: norecurse nounwind readonly
define fastcc <256 x i64> @loadv256i64stk() {
; CHECK-LABEL: loadv256i64stk:
; CHECK:       # %bb.0:
; CHECK-NEXT:    lea %s11, -2048(, %s11)
; CHECK-NEXT:    brge.l.t %s11, %s8, .LBB4_2
; CHECK-NEXT:  # %bb.1:
; CHECK-NEXT:    ld %s61, 24(, %s14)
; CHECK-NEXT:    or %s62, 0, %s0
; CHECK-NEXT:    lea %s63, 315
; CHECK-NEXT:    shm.l %s63, (%s61)
; CHECK-NEXT:    shm.l %s8, 8(%s61)
; CHECK-NEXT:    shm.l %s11, 16(%s61)
; CHECK-NEXT:    monc
; CHECK-NEXT:    or %s0, 0, %s62
; CHECK-NEXT:  .LBB4_2:
; CHECK-NEXT:    lea %s0, 256
; CHECK-NEXT:    lea %s1, (, %s11)
; CHECK-NEXT:    lvl %s0
; CHECK-NEXT:    vld %v0, 8, %s1
; CHECK-NEXT:    lea %s11, 2048(, %s11)
; CHECK-NEXT:    b.l.t (, %s10)
  %addr = alloca <256 x i64>, align 16
  %1 = load <256 x i64>, ptr %addr, align 16
  ret <256 x i64> %1
}

; Function Attrs: norecurse nounwind readonly
define fastcc <256 x i64> @loadv256i64com() {
; CHECK-LABEL: loadv256i64com:
; CHECK:       # %bb.0:
; CHECK-NEXT:    lea %s0, v256i64@lo
; CHECK-NEXT:    and %s0, %s0, (32)0
; CHECK-NEXT:    lea.sl %s0, v256i64@hi(, %s0)
; CHECK-NEXT:    lea %s1, 256
; CHECK-NEXT:    lvl %s1
; CHECK-NEXT:    vld %v0, 8, %s0
; CHECK-NEXT:    b.l.t (, %s10)
  %1 = load <256 x i64>, ptr @v256i64, align 16
  ret <256 x i64> %1
}