; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -verify-machineinstrs -mattr=+simd128 | FileCheck %s

; Test experimental prefetch instructions

target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
target triple = "wasm32-unknown-unknown"

declare void @llvm.wasm.prefetch.t(i8*)
declare void @llvm.wasm.prefetch.nt(i8*)
@gv = global i8 0

;===----------------------------------------------------------------------------
; prefetch.t
;===----------------------------------------------------------------------------

define void @prefetch_t_no_offset(i8* %p) {
; CHECK-LABEL: prefetch_t_no_offset:
; CHECK:         .functype prefetch_t_no_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    prefetch.t 0
; CHECK-NEXT:    # fallthrough-return
  tail call void @llvm.wasm.prefetch.t(i8* %p)
  ret void
}

define void @prefetch_t_with_folded_offset(i8* %p) {
; CHECK-LABEL: prefetch_t_with_folded_offset:
; CHECK:         .functype prefetch_t_with_folded_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const 24
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.t 0
; CHECK-NEXT:    # fallthrough-return
  %q = ptrtoint i8* %p to i32
  %r = add nuw i32 %q, 24
  %s = inttoptr i32 %r to i8*
  tail call void @llvm.wasm.prefetch.t(i8* %s)
  ret void
}

define void @prefetch_t_with_folded_gep_offset(i8* %p) {
; CHECK-LABEL: prefetch_t_with_folded_gep_offset:
; CHECK:         .functype prefetch_t_with_folded_gep_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const 6
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.t 0
; CHECK-NEXT:    # fallthrough-return
  %s = getelementptr inbounds i8, i8* %p, i32 6
  tail call void @llvm.wasm.prefetch.t(i8* %s)
  ret void
}

define void @prefetch_t_with_unfolded_gep_negative_offset(i8* %p) {
; CHECK-LABEL: prefetch_t_with_unfolded_gep_negative_offset:
; CHECK:         .functype prefetch_t_with_unfolded_gep_negative_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const -6
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.t 0
; CHECK-NEXT:    # fallthrough-return
  %s = getelementptr inbounds i8, i8* %p, i32 -6
  tail call void @llvm.wasm.prefetch.t(i8* %s)
  ret void
}

define void @prefetch_t_with_unfolded_offset(i8* %p) {
; CHECK-LABEL: prefetch_t_with_unfolded_offset:
; CHECK:         .functype prefetch_t_with_unfolded_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const 24
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.t 0
; CHECK-NEXT:    # fallthrough-return
  %q = ptrtoint i8* %p to i32
  %r = add nsw i32 %q, 24
  %s = inttoptr i32 %r to i8*
  tail call void @llvm.wasm.prefetch.t(i8* %s)
  ret void
}

define void @prefetch_t_with_unfolded_gep_offset(i8* %p) {
; CHECK-LABEL: prefetch_t_with_unfolded_gep_offset:
; CHECK:         .functype prefetch_t_with_unfolded_gep_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const 6
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.t 0
; CHECK-NEXT:    # fallthrough-return
  %s = getelementptr i8, i8* %p, i32 6
  tail call void @llvm.wasm.prefetch.t(i8* %s)
  ret void
}

define void @prefetch_t_from_numeric_address() {
; CHECK-LABEL: prefetch_t_from_numeric_address:
; CHECK:         .functype prefetch_t_from_numeric_address () -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    i32.const 42
; CHECK-NEXT:    prefetch.t 0
; CHECK-NEXT:    # fallthrough-return
  %s = inttoptr i32 42 to i8*
  tail call void @llvm.wasm.prefetch.t(i8* %s)
  ret void
}

define void @prefetch_t_from_global_address() {
; CHECK-LABEL: prefetch_t_from_global_address:
; CHECK:         .functype prefetch_t_from_global_address () -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    i32.const gv
; CHECK-NEXT:    prefetch.t 0
; CHECK-NEXT:    # fallthrough-return
  tail call void @llvm.wasm.prefetch.t(i8* @gv)
  ret void
}

;===----------------------------------------------------------------------------
; prefetch.nt
;===----------------------------------------------------------------------------

define void @prefetch_nt_no_offset(i8* %p) {
; CHECK-LABEL: prefetch_nt_no_offset:
; CHECK:         .functype prefetch_nt_no_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    prefetch.nt 0
; CHECK-NEXT:    # fallthrough-return
  tail call void @llvm.wasm.prefetch.nt(i8* %p)
  ret void
}

define void @prefetch_nt_with_folded_offset(i8* %p) {
; CHECK-LABEL: prefetch_nt_with_folded_offset:
; CHECK:         .functype prefetch_nt_with_folded_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const 24
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.nt 0
; CHECK-NEXT:    # fallthrough-return
  %q = ptrtoint i8* %p to i32
  %r = add nuw i32 %q, 24
  %s = inttoptr i32 %r to i8*
  tail call void @llvm.wasm.prefetch.nt(i8* %s)
  ret void
}

define void @prefetch_nt_with_folded_gep_offset(i8* %p) {
; CHECK-LABEL: prefetch_nt_with_folded_gep_offset:
; CHECK:         .functype prefetch_nt_with_folded_gep_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const 6
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.nt 0
; CHECK-NEXT:    # fallthrough-return
  %s = getelementptr inbounds i8, i8* %p, i64 6
  tail call void @llvm.wasm.prefetch.nt(i8* %s)
  ret void
}

define void @prefetch_nt_with_unfolded_gep_negative_offset(i8* %p) {
; CHECK-LABEL: prefetch_nt_with_unfolded_gep_negative_offset:
; CHECK:         .functype prefetch_nt_with_unfolded_gep_negative_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const -6
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.nt 0
; CHECK-NEXT:    # fallthrough-return
  %s = getelementptr inbounds i8, i8* %p, i64 -6
  tail call void @llvm.wasm.prefetch.nt(i8* %s)
  ret void
}

define void @prefetch_nt_with_unfolded_offset(i8* %p) {
; CHECK-LABEL: prefetch_nt_with_unfolded_offset:
; CHECK:         .functype prefetch_nt_with_unfolded_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const 24
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.nt 0
; CHECK-NEXT:    # fallthrough-return
  %q = ptrtoint i8* %p to i32
  %r = add nsw i32 %q, 24
  %s = inttoptr i32 %r to i8*
  tail call void @llvm.wasm.prefetch.nt(i8* %s)
  ret void
}

define void @prefetch_nt_with_unfolded_gep_offset(i8* %p) {
; CHECK-LABEL: prefetch_nt_with_unfolded_gep_offset:
; CHECK:         .functype prefetch_nt_with_unfolded_gep_offset (i32) -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    local.get 0
; CHECK-NEXT:    i32.const 6
; CHECK-NEXT:    i32.add
; CHECK-NEXT:    prefetch.nt 0
; CHECK-NEXT:    # fallthrough-return
  %s = getelementptr i8, i8* %p, i64 6
  tail call void @llvm.wasm.prefetch.nt(i8* %s)
  ret void
}

define void @prefetch_nt_from_numeric_address() {
; CHECK-LABEL: prefetch_nt_from_numeric_address:
; CHECK:         .functype prefetch_nt_from_numeric_address () -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    i32.const 42
; CHECK-NEXT:    prefetch.nt 0
; CHECK-NEXT:    # fallthrough-return
  %s = inttoptr i32 42 to i8*
  tail call void @llvm.wasm.prefetch.nt(i8* %s)
  ret void
}

define void @prefetch_nt_from_global_address() {
; CHECK-LABEL: prefetch_nt_from_global_address:
; CHECK:         .functype prefetch_nt_from_global_address () -> ()
; CHECK-NEXT:  # %bb.0:
; CHECK-NEXT:    i32.const gv
; CHECK-NEXT:    prefetch.nt 0
; CHECK-NEXT:    # fallthrough-return
  tail call void @llvm.wasm.prefetch.nt(i8* @gv)
  ret void
}