llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.memcpy.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-- -amdgpu-memcpy-loop-unroll=2 -mem-intrinsic-expand-size=35 %s -o - | FileCheck -check-prefix=LOOP %s
; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-- -amdgpu-memcpy-loop-unroll=2 -mem-intrinsic-expand-size=37 %s -o - | FileCheck -check-prefix=UNROLL %s

declare void @llvm.memcpy.p1.p1.i32(ptr addrspace(1), ptr addrspace(1), i32, i1 immarg)

define amdgpu_cs void @memcpy_p1i8(ptr addrspace(1) %dst, ptr addrspace(1) %src) {
; LOOP-LABEL: memcpy_p1i8:
; LOOP:       ; %bb.0:
; LOOP-NEXT:    s_mov_b64 s[0:1], 0
; LOOP-NEXT:    s_mov_b32 s2, 0
; LOOP-NEXT:    s_mov_b32 s3, 0xf000
; LOOP-NEXT:  .LBB0_1: ; %static-memcpy-expansion-main-body
; LOOP-NEXT:    ; =>This Inner Loop Header: Depth=1
; LOOP-NEXT:    s_waitcnt expcnt(1)
; LOOP-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64
; LOOP-NEXT:    buffer_load_ubyte v5, v[2:3], s[0:3], 0 addr64 offset:1
; LOOP-NEXT:    buffer_load_ubyte v6, v[2:3], s[0:3], 0 addr64 offset:2
; LOOP-NEXT:    buffer_load_ubyte v7, v[2:3], s[0:3], 0 addr64 offset:3
; LOOP-NEXT:    buffer_load_ubyte v8, v[2:3], s[0:3], 0 addr64 offset:4
; LOOP-NEXT:    buffer_load_ubyte v9, v[2:3], s[0:3], 0 addr64 offset:5
; LOOP-NEXT:    buffer_load_ubyte v10, v[2:3], s[0:3], 0 addr64 offset:6
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    buffer_load_ubyte v11, v[2:3], s[0:3], 0 addr64 offset:7
; LOOP-NEXT:    buffer_load_ubyte v12, v[2:3], s[0:3], 0 addr64 offset:8
; LOOP-NEXT:    s_waitcnt vmcnt(7)
; LOOP-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
; LOOP-NEXT:    v_or_b32_e32 v4, v5, v4
; LOOP-NEXT:    buffer_load_ubyte v5, v[2:3], s[0:3], 0 addr64 offset:9
; LOOP-NEXT:    s_waitcnt vmcnt(6)
; LOOP-NEXT:    v_lshlrev_b32_e32 v7, 24, v7
; LOOP-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
; LOOP-NEXT:    v_or_b32_e32 v6, v7, v6
; LOOP-NEXT:    buffer_load_ubyte v7, v[2:3], s[0:3], 0 addr64 offset:10
; LOOP-NEXT:    s_waitcnt vmcnt(5)
; LOOP-NEXT:    v_lshlrev_b32_e32 v9, 8, v9
; LOOP-NEXT:    v_or_b32_e32 v8, v9, v8
; LOOP-NEXT:    buffer_load_ubyte v9, v[2:3], s[0:3], 0 addr64 offset:11
; LOOP-NEXT:    s_waitcnt vmcnt(4)
; LOOP-NEXT:    v_lshlrev_b32_e32 v11, 24, v11
; LOOP-NEXT:    v_lshlrev_b32_e32 v10, 16, v10
; LOOP-NEXT:    v_or_b32_e32 v10, v11, v10
; LOOP-NEXT:    buffer_load_ubyte v11, v[2:3], s[0:3], 0 addr64 offset:12
; LOOP-NEXT:    s_waitcnt vmcnt(3)
; LOOP-NEXT:    v_lshlrev_b32_e32 v5, 8, v5
; LOOP-NEXT:    v_or_b32_e32 v5, v5, v12
; LOOP-NEXT:    buffer_load_ubyte v12, v[2:3], s[0:3], 0 addr64 offset:13
; LOOP-NEXT:    s_waitcnt vmcnt(2)
; LOOP-NEXT:    v_lshlrev_b32_e32 v9, 24, v9
; LOOP-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
; LOOP-NEXT:    v_or_b32_e32 v7, v9, v7
; LOOP-NEXT:    buffer_load_ubyte v9, v[2:3], s[0:3], 0 addr64 offset:15
; LOOP-NEXT:    buffer_load_ubyte v13, v[2:3], s[0:3], 0 addr64 offset:14
; LOOP-NEXT:    s_waitcnt vmcnt(2)
; LOOP-NEXT:    v_lshlrev_b32_e32 v12, 8, v12
; LOOP-NEXT:    v_or_b32_e32 v11, v12, v11
; LOOP-NEXT:    s_waitcnt vmcnt(1)
; LOOP-NEXT:    v_lshlrev_b32_e32 v9, 24, v9
; LOOP-NEXT:    s_waitcnt vmcnt(0)
; LOOP-NEXT:    v_lshlrev_b32_e32 v12, 16, v13
; LOOP-NEXT:    v_or_b32_e32 v9, v9, v12
; LOOP-NEXT:    buffer_load_ubyte v12, v[2:3], s[0:3], 0 addr64 offset:17
; LOOP-NEXT:    buffer_load_ubyte v13, v[2:3], s[0:3], 0 addr64 offset:16
; LOOP-NEXT:    buffer_load_ubyte v14, v[2:3], s[0:3], 0 addr64 offset:19
; LOOP-NEXT:    buffer_load_ubyte v15, v[2:3], s[0:3], 0 addr64 offset:18
; LOOP-NEXT:    s_waitcnt vmcnt(3)
; LOOP-NEXT:    v_lshlrev_b32_e32 v12, 8, v12
; LOOP-NEXT:    s_waitcnt vmcnt(2)
; LOOP-NEXT:    v_or_b32_e32 v12, v12, v13
; LOOP-NEXT:    s_waitcnt vmcnt(1)
; LOOP-NEXT:    v_lshlrev_b32_e32 v13, 24, v14
; LOOP-NEXT:    s_waitcnt vmcnt(0)
; LOOP-NEXT:    v_lshlrev_b32_e32 v14, 16, v15
; LOOP-NEXT:    v_or_b32_e32 v13, v13, v14
; LOOP-NEXT:    buffer_load_ubyte v14, v[2:3], s[0:3], 0 addr64 offset:21
; LOOP-NEXT:    buffer_load_ubyte v15, v[2:3], s[0:3], 0 addr64 offset:20
; LOOP-NEXT:    buffer_load_ubyte v16, v[2:3], s[0:3], 0 addr64 offset:23
; LOOP-NEXT:    buffer_load_ubyte v17, v[2:3], s[0:3], 0 addr64 offset:22
; LOOP-NEXT:    s_waitcnt vmcnt(3)
; LOOP-NEXT:    v_lshlrev_b32_e32 v14, 8, v14
; LOOP-NEXT:    s_waitcnt vmcnt(2)
; LOOP-NEXT:    v_or_b32_e32 v14, v14, v15
; LOOP-NEXT:    s_waitcnt vmcnt(1)
; LOOP-NEXT:    v_lshlrev_b32_e32 v15, 24, v16
; LOOP-NEXT:    s_waitcnt vmcnt(0)
; LOOP-NEXT:    v_lshlrev_b32_e32 v16, 16, v17
; LOOP-NEXT:    v_or_b32_e32 v15, v15, v16
; LOOP-NEXT:    buffer_load_ubyte v16, v[2:3], s[0:3], 0 addr64 offset:25
; LOOP-NEXT:    buffer_load_ubyte v17, v[2:3], s[0:3], 0 addr64 offset:24
; LOOP-NEXT:    buffer_load_ubyte v18, v[2:3], s[0:3], 0 addr64 offset:27
; LOOP-NEXT:    buffer_load_ubyte v19, v[2:3], s[0:3], 0 addr64 offset:26
; LOOP-NEXT:    s_waitcnt vmcnt(3)
; LOOP-NEXT:    v_lshlrev_b32_e32 v16, 8, v16
; LOOP-NEXT:    s_waitcnt vmcnt(2)
; LOOP-NEXT:    v_or_b32_e32 v16, v16, v17
; LOOP-NEXT:    s_waitcnt vmcnt(1)
; LOOP-NEXT:    v_lshlrev_b32_e32 v17, 24, v18
; LOOP-NEXT:    s_waitcnt vmcnt(0)
; LOOP-NEXT:    v_lshlrev_b32_e32 v18, 16, v19
; LOOP-NEXT:    v_or_b32_e32 v17, v17, v18
; LOOP-NEXT:    buffer_load_ubyte v18, v[2:3], s[0:3], 0 addr64 offset:29
; LOOP-NEXT:    buffer_load_ubyte v19, v[2:3], s[0:3], 0 addr64 offset:28
; LOOP-NEXT:    buffer_load_ubyte v20, v[2:3], s[0:3], 0 addr64 offset:31
; LOOP-NEXT:    buffer_load_ubyte v21, v[2:3], s[0:3], 0 addr64 offset:30
; LOOP-NEXT:    s_waitcnt vmcnt(3)
; LOOP-NEXT:    v_lshlrev_b32_e32 v18, 8, v18
; LOOP-NEXT:    s_waitcnt vmcnt(2)
; LOOP-NEXT:    v_or_b32_e32 v18, v18, v19
; LOOP-NEXT:    s_waitcnt vmcnt(1)
; LOOP-NEXT:    v_lshlrev_b32_e32 v19, 24, v20
; LOOP-NEXT:    s_waitcnt vmcnt(0)
; LOOP-NEXT:    v_lshlrev_b32_e32 v20, 16, v21
; LOOP-NEXT:    v_or_b32_e32 v19, v19, v20
; LOOP-NEXT:    v_or_b32_e32 v4, v6, v4
; LOOP-NEXT:    v_or_b32_e32 v6, v10, v8
; LOOP-NEXT:    v_or_b32_e32 v5, v7, v5
; LOOP-NEXT:    v_or_b32_e32 v7, v9, v11
; LOOP-NEXT:    v_or_b32_e32 v8, v13, v12
; LOOP-NEXT:    v_or_b32_e32 v9, v15, v14
; LOOP-NEXT:    v_or_b32_e32 v10, v17, v16
; LOOP-NEXT:    v_or_b32_e32 v11, v19, v18
; LOOP-NEXT:    v_bfe_u32 v12, v4, 8, 8
; LOOP-NEXT:    buffer_store_byte v12, v[0:1], s[0:3], 0 addr64 offset:1
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v12, 16, v4
; LOOP-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v4, 24, v4
; LOOP-NEXT:    buffer_store_byte v12, v[0:1], s[0:3], 0 addr64 offset:2
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v12, 16, v6
; LOOP-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:3
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_bfe_u32 v4, v6, 8, 8
; LOOP-NEXT:    buffer_store_byte v6, v[0:1], s[0:3], 0 addr64 offset:4
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v6, 24, v6
; LOOP-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:5
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v4, 16, v5
; LOOP-NEXT:    buffer_store_byte v12, v[0:1], s[0:3], 0 addr64 offset:6
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_bfe_u32 v12, v5, 8, 8
; LOOP-NEXT:    buffer_store_byte v5, v[0:1], s[0:3], 0 addr64 offset:8
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v5, 24, v5
; LOOP-NEXT:    buffer_store_byte v6, v[0:1], s[0:3], 0 addr64 offset:7
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v6, 16, v7
; LOOP-NEXT:    buffer_store_byte v12, v[0:1], s[0:3], 0 addr64 offset:9
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_bfe_u32 v12, v7, 8, 8
; LOOP-NEXT:    buffer_store_byte v7, v[0:1], s[0:3], 0 addr64 offset:12
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v7, 24, v7
; LOOP-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:10
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v4, 16, v8
; LOOP-NEXT:    buffer_store_byte v5, v[0:1], s[0:3], 0 addr64 offset:11
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_bfe_u32 v5, v8, 8, 8
; LOOP-NEXT:    buffer_store_byte v8, v[0:1], s[0:3], 0 addr64 offset:16
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v8, 24, v8
; LOOP-NEXT:    buffer_store_byte v12, v[0:1], s[0:3], 0 addr64 offset:13
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v12, 16, v9
; LOOP-NEXT:    buffer_store_byte v6, v[0:1], s[0:3], 0 addr64 offset:14
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_bfe_u32 v6, v9, 8, 8
; LOOP-NEXT:    buffer_store_byte v9, v[0:1], s[0:3], 0 addr64 offset:20
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v9, 24, v9
; LOOP-NEXT:    buffer_store_byte v7, v[0:1], s[0:3], 0 addr64 offset:15
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v7, 16, v10
; LOOP-NEXT:    buffer_store_byte v5, v[0:1], s[0:3], 0 addr64 offset:17
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_bfe_u32 v5, v10, 8, 8
; LOOP-NEXT:    buffer_store_byte v10, v[0:1], s[0:3], 0 addr64 offset:24
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v10, 24, v10
; LOOP-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:18
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v4, 16, v11
; LOOP-NEXT:    buffer_store_byte v8, v[0:1], s[0:3], 0 addr64 offset:19
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_bfe_u32 v8, v11, 8, 8
; LOOP-NEXT:    buffer_store_byte v11, v[0:1], s[0:3], 0 addr64 offset:28
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v11, 24, v11
; LOOP-NEXT:    buffer_store_byte v6, v[0:1], s[0:3], 0 addr64 offset:21
; LOOP-NEXT:    buffer_store_byte v12, v[0:1], s[0:3], 0 addr64 offset:22
; LOOP-NEXT:    buffer_store_byte v9, v[0:1], s[0:3], 0 addr64 offset:23
; LOOP-NEXT:    buffer_store_byte v5, v[0:1], s[0:3], 0 addr64 offset:25
; LOOP-NEXT:    buffer_store_byte v7, v[0:1], s[0:3], 0 addr64 offset:26
; LOOP-NEXT:    buffer_store_byte v10, v[0:1], s[0:3], 0 addr64 offset:27
; LOOP-NEXT:    buffer_store_byte v8, v[0:1], s[0:3], 0 addr64 offset:29
; LOOP-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:30
; LOOP-NEXT:    buffer_store_byte v11, v[0:1], s[0:3], 0 addr64 offset:31
; LOOP-NEXT:    s_add_u32 s0, s0, 32
; LOOP-NEXT:    s_addc_u32 s1, s1, 0
; LOOP-NEXT:    s_cmp_lt_u32 s0, 32
; LOOP-NEXT:    s_cbranch_scc1 .LBB0_1
; LOOP-NEXT:  ; %bb.2: ; %static-memcpy-post-expansion
; LOOP-NEXT:    s_mov_b32 s2, 0
; LOOP-NEXT:    s_mov_b32 s3, 0xf000
; LOOP-NEXT:    s_mov_b64 s[0:1], 0
; LOOP-NEXT:    s_waitcnt expcnt(1)
; LOOP-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:33
; LOOP-NEXT:    buffer_load_ubyte v5, v[2:3], s[0:3], 0 addr64 offset:35
; LOOP-NEXT:    buffer_load_ubyte v6, v[2:3], s[0:3], 0 addr64 offset:34
; LOOP-NEXT:    buffer_load_ubyte v2, v[2:3], s[0:3], 0 addr64 offset:32
; LOOP-NEXT:    s_waitcnt vmcnt(3)
; LOOP-NEXT:    v_lshlrev_b32_e32 v3, 8, v4
; LOOP-NEXT:    s_waitcnt vmcnt(2)
; LOOP-NEXT:    v_lshlrev_b32_e32 v4, 24, v5
; LOOP-NEXT:    s_waitcnt vmcnt(1)
; LOOP-NEXT:    v_lshlrev_b32_e32 v5, 16, v6
; LOOP-NEXT:    s_waitcnt vmcnt(0)
; LOOP-NEXT:    v_or_b32_e32 v2, v3, v2
; LOOP-NEXT:    v_or_b32_e32 v3, v4, v5
; LOOP-NEXT:    v_or_b32_e32 v2, v3, v2
; LOOP-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
; LOOP-NEXT:    v_bfe_u32 v4, v2, 8, 8
; LOOP-NEXT:    buffer_store_byte v2, v[0:1], s[0:3], 0 addr64 offset:32
; LOOP-NEXT:    s_waitcnt expcnt(0)
; LOOP-NEXT:    v_lshrrev_b32_e32 v2, 24, v2
; LOOP-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:33
; LOOP-NEXT:    buffer_store_byte v3, v[0:1], s[0:3], 0 addr64 offset:34
; LOOP-NEXT:    buffer_store_byte v2, v[0:1], s[0:3], 0 addr64 offset:35
; LOOP-NEXT:    s_endpgm
;
; UNROLL-LABEL: memcpy_p1i8:
; UNROLL:       ; %bb.0:
; UNROLL-NEXT:    s_mov_b32 s2, 0
; UNROLL-NEXT:    s_mov_b32 s3, 0xf000
; UNROLL-NEXT:    s_mov_b64 s[0:1], 0
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:1
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:1
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:2
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:2
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:3
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:3
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:4
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:4
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:5
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:5
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:6
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:6
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:7
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:7
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:8
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:8
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:9
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:9
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:10
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:10
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:11
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:11
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:12
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:12
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:13
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:13
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:14
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:14
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:15
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:15
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:16
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:16
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:17
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:17
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:18
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:18
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:19
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:19
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:20
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:20
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:21
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:21
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:22
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:22
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:23
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:23
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:24
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:24
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:25
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:25
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:26
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:26
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:27
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:27
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:28
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:28
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:29
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:29
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:30
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:30
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:31
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:31
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:32
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:32
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:33
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:33
; UNROLL-NEXT:    s_waitcnt expcnt(0)
; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64 offset:34
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64 offset:34
; UNROLL-NEXT:    buffer_load_ubyte v2, v[2:3], s[0:3], 0 addr64 offset:35
; UNROLL-NEXT:    s_waitcnt vmcnt(0)
; UNROLL-NEXT:    buffer_store_byte v2, v[0:1], s[0:3], 0 addr64 offset:35
; UNROLL-NEXT:    s_endpgm
  call void @llvm.memcpy.p1.p1.i32(ptr addrspace(1) %dst, ptr addrspace(1) %src, i32 36, i1 false)
  ret void
}