caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
890c4bece26e005cd9fa5511fe0efa7307794de5
clang-p2996/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.memcpy.ll
Fabian Ritter a4fd3dba6e [AMDGPU] Use wider loop lowering type for LowerMemIntrinsics (#112332) 
When llvm.memcpy or llvm.memmove intrinsics are lowered as a loop in
LowerMemIntrinsics.cpp, the loop consists of a single load/store pair
per iteration. We can improve performance in some cases by emitting
multiple load/store pairs per iteration. This patch achieves that by
increasing the width of the loop lowering type in the GCN target and
letting legalization split the resulting too-wide access pairs into
multiple legal access pairs.

This change only affects lowered memcpys and memmoves with large (>=
1024 bytes) constant lengths. Smaller constant lengths are handled by
ISel directly; non-constant lengths would be slowed down by this change
if the dynamic length was smaller or slightly larger than what an
unrolled iteration copies.

The chosen default unroll factor is the result of microbenchmarks on
gfx1030. This change leads to speedups of 15-38% for global memory and
1.9-5.8x for scratch in these microbenchmarks.

Part of SWDEV-455845.
2024-10-28 09:04:19 +01:00

359 lines
19 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel -mtriple=amdgcn-- -verify-machineinstrs -amdgpu-memcpy-loop-unroll=2 -mem-intrinsic-expand-size=35 %s -o - | FileCheck -check-prefix=LOOP %s
; RUN: llc -global-isel -mtriple=amdgcn-- -verify-machineinstrs -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: v_mov_b32_e32 v5, s1
; LOOP-NEXT: v_mov_b32_e32 v4, s0
; LOOP-NEXT: .LBB0_1: ; %load-store-loop
; LOOP-NEXT: ; =>This Inner Loop Header: Depth=1
; LOOP-NEXT: s_waitcnt expcnt(2)
; LOOP-NEXT: v_add_i32_e32 v29, vcc, v2, v4
; LOOP-NEXT: v_addc_u32_e32 v30, vcc, v3, v5, vcc
; LOOP-NEXT: buffer_load_ubyte v24, v[29:30], s[0:3], 0 addr64
; LOOP-NEXT: buffer_load_ubyte v27, v[29:30], s[0:3], 0 addr64 offset:1
; LOOP-NEXT: buffer_load_ubyte v34, v[29:30], s[0:3], 0 addr64 offset:2
; LOOP-NEXT: buffer_load_ubyte v35, v[29:30], s[0:3], 0 addr64 offset:3
; LOOP-NEXT: buffer_load_ubyte v36, v[29:30], s[0:3], 0 addr64 offset:4
; LOOP-NEXT: buffer_load_ubyte v37, v[29:30], s[0:3], 0 addr64 offset:5
; LOOP-NEXT: buffer_load_ubyte v38, v[29:30], s[0:3], 0 addr64 offset:6
; LOOP-NEXT: buffer_load_ubyte v39, v[29:30], s[0:3], 0 addr64 offset:7
; LOOP-NEXT: buffer_load_ubyte v6, v[29:30], s[0:3], 0 addr64 offset:8
; LOOP-NEXT: buffer_load_ubyte v9, v[29:30], s[0:3], 0 addr64 offset:9
; LOOP-NEXT: buffer_load_ubyte v10, v[29:30], s[0:3], 0 addr64 offset:10
; LOOP-NEXT: s_waitcnt expcnt(0)
; LOOP-NEXT: buffer_load_ubyte v11, v[29:30], s[0:3], 0 addr64 offset:11
; LOOP-NEXT: buffer_load_ubyte v7, v[29:30], s[0:3], 0 addr64 offset:12
; LOOP-NEXT: buffer_load_ubyte v13, v[29:30], s[0:3], 0 addr64 offset:13
; LOOP-NEXT: buffer_load_ubyte v14, v[29:30], s[0:3], 0 addr64 offset:14
; LOOP-NEXT: buffer_load_ubyte v15, v[29:30], s[0:3], 0 addr64 offset:15
; LOOP-NEXT: buffer_load_ubyte v8, v[29:30], s[0:3], 0 addr64 offset:16
; LOOP-NEXT: buffer_load_ubyte v17, v[29:30], s[0:3], 0 addr64 offset:17
; LOOP-NEXT: buffer_load_ubyte v18, v[29:30], s[0:3], 0 addr64 offset:18
; LOOP-NEXT: buffer_load_ubyte v19, v[29:30], s[0:3], 0 addr64 offset:19
; LOOP-NEXT: buffer_load_ubyte v12, v[29:30], s[0:3], 0 addr64 offset:20
; LOOP-NEXT: buffer_load_ubyte v21, v[29:30], s[0:3], 0 addr64 offset:21
; LOOP-NEXT: buffer_load_ubyte v22, v[29:30], s[0:3], 0 addr64 offset:22
; LOOP-NEXT: buffer_load_ubyte v23, v[29:30], s[0:3], 0 addr64 offset:23
; LOOP-NEXT: buffer_load_ubyte v16, v[29:30], s[0:3], 0 addr64 offset:24
; LOOP-NEXT: buffer_load_ubyte v25, v[29:30], s[0:3], 0 addr64 offset:25
; LOOP-NEXT: buffer_load_ubyte v26, v[29:30], s[0:3], 0 addr64 offset:26
; LOOP-NEXT: buffer_load_ubyte v28, v[29:30], s[0:3], 0 addr64 offset:27
; LOOP-NEXT: buffer_load_ubyte v20, v[29:30], s[0:3], 0 addr64 offset:28
; LOOP-NEXT: buffer_load_ubyte v31, v[29:30], s[0:3], 0 addr64 offset:29
; LOOP-NEXT: buffer_load_ubyte v32, v[29:30], s[0:3], 0 addr64 offset:30
; LOOP-NEXT: buffer_load_ubyte v33, v[29:30], s[0:3], 0 addr64 offset:31
; LOOP-NEXT: s_waitcnt vmcnt(14)
; LOOP-NEXT: v_lshlrev_b32_e32 v27, 8, v27
; LOOP-NEXT: v_or_b32_e32 v24, v27, v24
; LOOP-NEXT: v_lshlrev_b32_e32 v27, 24, v35
; LOOP-NEXT: v_lshlrev_b32_e32 v29, 16, v34
; LOOP-NEXT: v_or_b32_e32 v27, v27, v29
; LOOP-NEXT: v_lshlrev_b32_e32 v29, 8, v37
; LOOP-NEXT: v_lshlrev_b32_e32 v30, 24, v39
; LOOP-NEXT: v_lshlrev_b32_e32 v34, 16, v38
; LOOP-NEXT: v_or_b32_e32 v29, v29, v36
; LOOP-NEXT: v_or_b32_e32 v30, v30, v34
; LOOP-NEXT: v_add_i32_e32 v34, vcc, v0, v4
; LOOP-NEXT: v_addc_u32_e32 v35, vcc, v1, v5, vcc
; LOOP-NEXT: v_add_i32_e32 v4, vcc, 32, v4
; LOOP-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
; LOOP-NEXT: v_cmp_gt_u32_e32 vcc, 32, v4
; LOOP-NEXT: v_lshlrev_b32_e32 v9, 8, v9
; LOOP-NEXT: v_lshlrev_b32_e32 v11, 24, v11
; LOOP-NEXT: v_lshlrev_b32_e32 v10, 16, v10
; LOOP-NEXT: v_lshlrev_b32_e32 v13, 8, v13
; LOOP-NEXT: v_lshlrev_b32_e32 v15, 24, v15
; LOOP-NEXT: v_lshlrev_b32_e32 v14, 16, v14
; LOOP-NEXT: v_lshlrev_b32_e32 v17, 8, v17
; LOOP-NEXT: s_waitcnt vmcnt(12)
; LOOP-NEXT: v_lshlrev_b32_e32 v19, 24, v19
; LOOP-NEXT: v_lshlrev_b32_e32 v18, 16, v18
; LOOP-NEXT: s_waitcnt vmcnt(10)
; LOOP-NEXT: v_lshlrev_b32_e32 v21, 8, v21
; LOOP-NEXT: s_waitcnt vmcnt(8)
; LOOP-NEXT: v_lshlrev_b32_e32 v23, 24, v23
; LOOP-NEXT: v_lshlrev_b32_e32 v22, 16, v22
; LOOP-NEXT: s_waitcnt vmcnt(6)
; LOOP-NEXT: v_lshlrev_b32_e32 v25, 8, v25
; LOOP-NEXT: s_waitcnt vmcnt(4)
; LOOP-NEXT: v_lshlrev_b32_e32 v28, 24, v28
; LOOP-NEXT: v_lshlrev_b32_e32 v26, 16, v26
; LOOP-NEXT: s_waitcnt vmcnt(2)
; LOOP-NEXT: v_lshlrev_b32_e32 v31, 8, v31
; LOOP-NEXT: s_waitcnt vmcnt(0)
; LOOP-NEXT: v_lshlrev_b32_e32 v33, 24, v33
; LOOP-NEXT: v_lshlrev_b32_e32 v32, 16, v32
; LOOP-NEXT: v_or_b32_e32 v6, v9, v6
; LOOP-NEXT: v_or_b32_e32 v9, v11, v10
; LOOP-NEXT: v_or_b32_e32 v7, v13, v7
; LOOP-NEXT: v_or_b32_e32 v10, v15, v14
; LOOP-NEXT: v_or_b32_e32 v8, v17, v8
; LOOP-NEXT: v_or_b32_e32 v11, v19, v18
; LOOP-NEXT: v_or_b32_e32 v12, v21, v12
; LOOP-NEXT: v_or_b32_e32 v13, v23, v22
; LOOP-NEXT: v_or_b32_e32 v14, v25, v16
; LOOP-NEXT: v_or_b32_e32 v15, v28, v26
; LOOP-NEXT: v_or_b32_e32 v16, v31, v20
; LOOP-NEXT: v_or_b32_e32 v17, v33, v32
; LOOP-NEXT: v_or_b32_e32 v18, v27, v24
; LOOP-NEXT: v_or_b32_e32 v19, v30, v29
; LOOP-NEXT: v_or_b32_e32 v6, v9, v6
; LOOP-NEXT: v_or_b32_e32 v7, v10, v7
; LOOP-NEXT: v_or_b32_e32 v8, v11, v8
; LOOP-NEXT: v_or_b32_e32 v9, v13, v12
; LOOP-NEXT: v_or_b32_e32 v10, v15, v14
; LOOP-NEXT: v_or_b32_e32 v11, v17, v16
; LOOP-NEXT: v_lshrrev_b32_e32 v12, 16, v18
; LOOP-NEXT: v_bfe_u32 v13, v18, 8, 8
; LOOP-NEXT: buffer_store_byte v18, v[34:35], s[0:3], 0 addr64
; LOOP-NEXT: v_lshrrev_b32_e32 v14, 24, v18
; LOOP-NEXT: v_lshrrev_b32_e32 v15, 16, v19
; LOOP-NEXT: v_bfe_u32 v16, v19, 8, 8
; LOOP-NEXT: buffer_store_byte v19, v[34:35], s[0:3], 0 addr64 offset:4
; LOOP-NEXT: v_lshrrev_b32_e32 v17, 24, v19
; LOOP-NEXT: s_waitcnt expcnt(1)
; LOOP-NEXT: v_lshrrev_b32_e32 v18, 16, v6
; LOOP-NEXT: s_waitcnt expcnt(0)
; LOOP-NEXT: v_bfe_u32 v19, v6, 8, 8
; LOOP-NEXT: buffer_store_byte v6, v[34:35], s[0:3], 0 addr64 offset:8
; LOOP-NEXT: s_waitcnt expcnt(0)
; LOOP-NEXT: v_lshrrev_b32_e32 v6, 24, v6
; LOOP-NEXT: v_lshrrev_b32_e32 v20, 16, v7
; LOOP-NEXT: v_bfe_u32 v21, v7, 8, 8
; LOOP-NEXT: buffer_store_byte v7, v[34:35], s[0:3], 0 addr64 offset:12
; LOOP-NEXT: s_waitcnt expcnt(0)
; LOOP-NEXT: v_lshrrev_b32_e32 v7, 24, v7
; LOOP-NEXT: v_lshrrev_b32_e32 v22, 16, v8
; LOOP-NEXT: v_bfe_u32 v23, v8, 8, 8
; LOOP-NEXT: buffer_store_byte v8, v[34:35], s[0:3], 0 addr64 offset:16
; LOOP-NEXT: s_waitcnt expcnt(0)
; LOOP-NEXT: v_lshrrev_b32_e32 v8, 24, v8
; LOOP-NEXT: v_lshrrev_b32_e32 v24, 16, v9
; LOOP-NEXT: v_bfe_u32 v25, v9, 8, 8
; LOOP-NEXT: buffer_store_byte v9, v[34:35], s[0:3], 0 addr64 offset:20
; LOOP-NEXT: s_waitcnt expcnt(0)
; LOOP-NEXT: v_lshrrev_b32_e32 v9, 24, v9
; LOOP-NEXT: v_lshrrev_b32_e32 v26, 16, v10
; LOOP-NEXT: v_bfe_u32 v27, v10, 8, 8
; LOOP-NEXT: buffer_store_byte v10, v[34:35], s[0:3], 0 addr64 offset:24
; LOOP-NEXT: s_waitcnt expcnt(0)
; LOOP-NEXT: v_lshrrev_b32_e32 v10, 24, v10
; LOOP-NEXT: v_lshrrev_b32_e32 v28, 16, v11
; LOOP-NEXT: v_bfe_u32 v29, v11, 8, 8
; LOOP-NEXT: buffer_store_byte v11, v[34:35], 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 v13, v[34:35], s[0:3], 0 addr64 offset:1
; LOOP-NEXT: buffer_store_byte v12, v[34:35], s[0:3], 0 addr64 offset:2
; LOOP-NEXT: buffer_store_byte v14, v[34:35], s[0:3], 0 addr64 offset:3
; LOOP-NEXT: buffer_store_byte v16, v[34:35], s[0:3], 0 addr64 offset:5
; LOOP-NEXT: buffer_store_byte v15, v[34:35], s[0:3], 0 addr64 offset:6
; LOOP-NEXT: buffer_store_byte v17, v[34:35], s[0:3], 0 addr64 offset:7
; LOOP-NEXT: buffer_store_byte v19, v[34:35], s[0:3], 0 addr64 offset:9
; LOOP-NEXT: buffer_store_byte v18, v[34:35], s[0:3], 0 addr64 offset:10
; LOOP-NEXT: buffer_store_byte v6, v[34:35], s[0:3], 0 addr64 offset:11
; LOOP-NEXT: buffer_store_byte v21, v[34:35], s[0:3], 0 addr64 offset:13
; LOOP-NEXT: buffer_store_byte v20, v[34:35], s[0:3], 0 addr64 offset:14
; LOOP-NEXT: buffer_store_byte v7, v[34:35], s[0:3], 0 addr64 offset:15
; LOOP-NEXT: buffer_store_byte v23, v[34:35], s[0:3], 0 addr64 offset:17
; LOOP-NEXT: buffer_store_byte v22, v[34:35], s[0:3], 0 addr64 offset:18
; LOOP-NEXT: buffer_store_byte v8, v[34:35], s[0:3], 0 addr64 offset:19
; LOOP-NEXT: buffer_store_byte v25, v[34:35], s[0:3], 0 addr64 offset:21
; LOOP-NEXT: buffer_store_byte v24, v[34:35], s[0:3], 0 addr64 offset:22
; LOOP-NEXT: buffer_store_byte v9, v[34:35], s[0:3], 0 addr64 offset:23
; LOOP-NEXT: buffer_store_byte v27, v[34:35], s[0:3], 0 addr64 offset:25
; LOOP-NEXT: buffer_store_byte v26, v[34:35], s[0:3], 0 addr64 offset:26
; LOOP-NEXT: buffer_store_byte v10, v[34:35], s[0:3], 0 addr64 offset:27
; LOOP-NEXT: buffer_store_byte v29, v[34:35], s[0:3], 0 addr64 offset:29
; LOOP-NEXT: buffer_store_byte v28, v[34:35], s[0:3], 0 addr64 offset:30
; LOOP-NEXT: buffer_store_byte v11, v[34:35], s[0:3], 0 addr64 offset:31
; LOOP-NEXT: s_cbranch_vccnz .LBB0_1
; LOOP-NEXT: ; %bb.2: ; %memcpy-split
; 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: 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
}
Reference in New Issue View Git Blame Copy Permalink