; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
;
; RUN: opt < %s -S -nvptx-lower-aggr-copies | FileCheck %s

target triple = "nvptx64-nvidia-cuda"

@global_data = global [256 x i8] zeroinitializer

declare void @llvm.memmove.p3.p0.i64(ptr addrspace(3) nocapture, ptr nocapture readonly, i64, i1)
declare void @llvm.memmove.p0.p3.i64(ptr nocapture, ptr addrspace(3) nocapture readonly, i64, i1)

; Test function that copies from alloca to shared memory, then from shared memory to global memory
define void @two_stage_copy() {
; CHECK-LABEL: define void @two_stage_copy() {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[LOCAL_BUF:%.*]] = alloca [256 x i8], align 1
; CHECK-NEXT:    [[SHARED_BUF:%.*]] = alloca [256 x i8], align 1, addrspace(3)
; CHECK-NEXT:    [[LOCAL_PTR:%.*]] = getelementptr inbounds [256 x i8], ptr [[LOCAL_BUF]], i64 0, i64 0
; CHECK-NEXT:    store i8 42, ptr [[LOCAL_PTR]], align 1
; CHECK-NEXT:    call void @llvm.memmove.p3.p0.i64(ptr addrspace(3) [[SHARED_BUF]], ptr [[LOCAL_BUF]], i64 256, i1 false)
; CHECK-NEXT:    call void @llvm.memmove.p0.p3.i64(ptr @global_data, ptr addrspace(3) [[SHARED_BUF]], i64 256, i1 false)
; CHECK-NEXT:    ret void
;
entry:
  %local_buf = alloca [256 x i8], align 1
  %shared_buf = alloca [256 x i8], align 1, addrspace(3)

  ; Initialize local buffer
  %local_ptr = getelementptr inbounds [256 x i8], ptr %local_buf, i64 0, i64 0
  store i8 42, ptr %local_ptr, align 1

  call void @llvm.memmove.p3.p0.i64(ptr addrspace(3) %shared_buf, ptr %local_buf, i64 256, i1 false)

  call void @llvm.memmove.p0.p3.i64(ptr @global_data, ptr addrspace(3) %shared_buf, i64 256, i1 false)

  ret void
}