1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
2 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
4 ; GCN-LABEL: {{^}}uitofp_i16_to_f16
5 ; GCN: buffer_load_ushort v[[A_I16:[0-9]+]]
6 ; SI: v_cvt_f32_u32_e32 v[[A_F32:[0-9]+]], v[[A_I16]]
7 ; SI: v_cvt_f16_f32_e32 v[[R_F16:[0-9]+]], v[[A_F32]]
9 ; VI: v_cvt_f16_u16_e32 v[[R_F16:[0-9]+]], v[[A_I16]]
11 ; GCN: buffer_store_short v[[R_F16]]
13 define amdgpu_kernel void @uitofp_i16_to_f16(
14 half addrspace(1)* %r,
15 i16 addrspace(1)* %a) {
17 %a.val = load i16, i16 addrspace(1)* %a
18 %r.val = uitofp i16 %a.val to half
19 store half %r.val, half addrspace(1)* %r
23 ; GCN-LABEL: {{^}}uitofp_i32_to_f16
24 ; GCN: buffer_load_dword v[[A_I32:[0-9]+]]
25 ; GCN: v_cvt_f32_u32_e32 v[[A_I16:[0-9]+]], v[[A_I32]]
26 ; GCN: v_cvt_f16_f32_e32 v[[R_F16:[0-9]+]], v[[A_I16]]
27 ; GCN: buffer_store_short v[[R_F16]]
29 define amdgpu_kernel void @uitofp_i32_to_f16(
30 half addrspace(1)* %r,
31 i32 addrspace(1)* %a) {
33 %a.val = load i32, i32 addrspace(1)* %a
34 %r.val = uitofp i32 %a.val to half
35 store half %r.val, half addrspace(1)* %r
39 ; f16 = uitofp i64 is in uint_to_fp.i64.ll
41 ; GCN-LABEL: {{^}}uitofp_v2i16_to_v2f16
42 ; GCN: buffer_load_dword
44 ; SI: v_cvt_f32_u32_e32
45 ; SI: v_cvt_f32_u32_e32
46 ; SI: v_cvt_f16_f32_e32
47 ; SI: v_cvt_f16_f32_e32
48 ; SI-DAG: v_lshlrev_b32_e32
52 ; VI-DAG: v_cvt_f16_u16_e32
53 ; VI-DAG: v_cvt_f16_u16_sdwa v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1
56 ; GCN: buffer_store_dword
58 define amdgpu_kernel void @uitofp_v2i16_to_v2f16(
59 <2 x half> addrspace(1)* %r,
60 <2 x i16> addrspace(1)* %a) {
62 %a.val = load <2 x i16>, <2 x i16> addrspace(1)* %a
63 %r.val = uitofp <2 x i16> %a.val to <2 x half>
64 store <2 x half> %r.val, <2 x half> addrspace(1)* %r
68 ; GCN-LABEL: {{^}}uitofp_v2i32_to_v2f16
69 ; GCN: buffer_load_dwordx2
71 ; SI: v_cvt_f32_u32_e32
72 ; SI: v_cvt_f32_u32_e32
73 ; SI: v_cvt_f16_f32_e32
74 ; SI: v_cvt_f16_f32_e32
75 ; SI-DAG: v_lshlrev_b32_e32
78 ; VI-DAG: v_cvt_f32_u32_e32
79 ; VI-DAG: v_cvt_f32_u32_e32
80 ; VI-DAG: v_cvt_f16_f32_e32
81 ; VI-DAG: v_cvt_f16_f32_sdwa
84 ; GCN: buffer_store_dword
86 define amdgpu_kernel void @uitofp_v2i32_to_v2f16(
87 <2 x half> addrspace(1)* %r,
88 <2 x i32> addrspace(1)* %a) {
90 %a.val = load <2 x i32>, <2 x i32> addrspace(1)* %a
91 %r.val = uitofp <2 x i32> %a.val to <2 x half>
92 store <2 x half> %r.val, <2 x half> addrspace(1)* %r
96 ; GCN-LABEL: {{^}}s_uint_to_fp_i1_to_f16:
97 ; GCN-DAG: v_cmp_le_f32_e32 [[CMP0:vcc]], 1.0, {{v[0-9]+}}
98 ; GCN-DAG: v_cmp_le_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]], 0, {{v[0-9]+}}
99 ; GCN: s_xor_b64 [[R_CMP:s\[[0-9]+:[0-9]+\]]], [[CMP1]], [[CMP0]]
100 ; GCN: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1.0, [[R_CMP]]
101 ; GCN-NEXT: v_cvt_f16_f32_e32 [[R_F16:v[0-9]+]], [[RESULT]]
102 ; GCN: buffer_store_short
104 define amdgpu_kernel void @s_uint_to_fp_i1_to_f16(half addrspace(1)* %out, float addrspace(1)* %in0, float addrspace(1)* %in1) {
105 %a = load float, float addrspace(1) * %in0
106 %b = load float, float addrspace(1) * %in1
107 %acmp = fcmp oge float %a, 0.000000e+00
108 %bcmp = fcmp oge float %b, 1.000000e+00
109 %result = xor i1 %acmp, %bcmp
110 %fp = uitofp i1 %result to half
111 store half %fp, half addrspace(1)* %out
115 ; f16 = uitofp i64 is in uint_to_fp.i64.ll