1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,SI %s
2 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI %s
4 ; GCN-LABEL: {{^}}trunc_i64_bitcast_v2i32:
5 ; GCN: buffer_load_dword v
6 ; GCN: buffer_store_dword v
7 define amdgpu_kernel void @trunc_i64_bitcast_v2i32(i32 addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
8 %ld = load <2 x i32>, <2 x i32> addrspace(1)* %in
9 %bc = bitcast <2 x i32> %ld to i64
10 %trunc = trunc i64 %bc to i32
11 store i32 %trunc, i32 addrspace(1)* %out
15 ; GCN-LABEL: {{^}}trunc_i96_bitcast_v3i32:
16 ; GCN: buffer_load_dword v
17 ; GCN: buffer_store_dword v
18 define amdgpu_kernel void @trunc_i96_bitcast_v3i32(i32 addrspace(1)* %out, <3 x i32> addrspace(1)* %in) {
19 %ld = load <3 x i32>, <3 x i32> addrspace(1)* %in
20 %bc = bitcast <3 x i32> %ld to i96
21 %trunc = trunc i96 %bc to i32
22 store i32 %trunc, i32 addrspace(1)* %out
26 ; GCN-LABEL: {{^}}trunc_i128_bitcast_v4i32:
27 ; GCN: buffer_load_dword v
28 ; GCN: buffer_store_dword v
29 define amdgpu_kernel void @trunc_i128_bitcast_v4i32(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
30 %ld = load <4 x i32>, <4 x i32> addrspace(1)* %in
31 %bc = bitcast <4 x i32> %ld to i128
32 %trunc = trunc i128 %bc to i32
33 store i32 %trunc, i32 addrspace(1)* %out
37 ; Don't want load width reduced in this case.
38 ; GCN-LABEL: {{^}}trunc_i16_bitcast_v2i16:
39 ; GCN: buffer_load_dword [[VAL:v[0-9]+]]
40 ; GCN: buffer_store_short [[VAL]]
41 define amdgpu_kernel void @trunc_i16_bitcast_v2i16(i16 addrspace(1)* %out, <2 x i16> addrspace(1)* %in) {
42 %ld = load <2 x i16>, <2 x i16> addrspace(1)* %in
43 %bc = bitcast <2 x i16> %ld to i32
44 %trunc = trunc i32 %bc to i16
45 store i16 %trunc, i16 addrspace(1)* %out
49 ; GCN-LABEL: {{^}}trunc_i16_bitcast_v4i16:
50 ; FIXME We need to teach the dagcombiner to reduce load width for:
51 ; t21: v2i32,ch = load<LD8[%in(addrspace=1)]> t12, t10, undef:i64
52 ; t23: i64 = bitcast t21
53 ; t30: i16 = truncate t23
54 ; SI: buffer_load_dword v[[VAL:[0-9]+]]
55 ; VI: buffer_load_dwordx2 v{{\[}}[[VAL:[0-9]+]]
56 ; GCN: buffer_store_short v[[VAL]], off
57 define amdgpu_kernel void @trunc_i16_bitcast_v4i16(i16 addrspace(1)* %out, <4 x i16> addrspace(1)* %in) {
58 %ld = load <4 x i16>, <4 x i16> addrspace(1)* %in
59 %bc = bitcast <4 x i16> %ld to i64
60 %trunc = trunc i64 %bc to i16
61 store i16 %trunc, i16 addrspace(1)* %out
65 ; FIXME: Consistently shrink or not here
66 ; GCN-LABEL: {{^}}trunc_i8_bitcast_v2i8:
67 ; SI: buffer_load_ubyte [[VAL:v[0-9]+]]
68 ; VI: buffer_load_ushort [[VAL:v[0-9]+]]
69 ; GCN: buffer_store_byte [[VAL]]
70 define amdgpu_kernel void @trunc_i8_bitcast_v2i8(i8 addrspace(1)* %out, <2 x i8> addrspace(1)* %in) {
71 %ld = load <2 x i8>, <2 x i8> addrspace(1)* %in
72 %bc = bitcast <2 x i8> %ld to i16
73 %trunc = trunc i16 %bc to i8
74 store i8 %trunc, i8 addrspace(1)* %out
78 ; GCN-LABEL: {{^}}trunc_i32_bitcast_v4i8:
79 ; GCN: buffer_load_dword [[VAL:v[0-9]+]]
80 ; GCN: buffer_store_byte [[VAL]]
81 define amdgpu_kernel void @trunc_i32_bitcast_v4i8(i8 addrspace(1)* %out, <4 x i8> addrspace(1)* %in) {
82 %ld = load <4 x i8>, <4 x i8> addrspace(1)* %in
83 %bc = bitcast <4 x i8> %ld to i32
84 %trunc = trunc i32 %bc to i8
85 store i8 %trunc, i8 addrspace(1)* %out
89 ; GCN-LABEL: {{^}}trunc_i24_bitcast_v3i8:
90 ; GCN: buffer_load_dword [[VAL:v[0-9]+]]
91 ; GCN: buffer_store_byte [[VAL]]
92 define amdgpu_kernel void @trunc_i24_bitcast_v3i8(i8 addrspace(1)* %out, <3 x i8> addrspace(1)* %in) {
93 %ld = load <3 x i8>, <3 x i8> addrspace(1)* %in
94 %bc = bitcast <3 x i8> %ld to i24
95 %trunc = trunc i24 %bc to i8
96 store i8 %trunc, i8 addrspace(1)* %out