blob: b01932f69b06c8aa47bc9bacd89297a692bc69da [file] [log] [blame]
Konstantin Zhuravlyovf86e4b72016-11-13 07:01:11 +00001; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
Matt Arsenault7aad8fd2017-01-24 22:02:15 +00002; RUN: llc -march=amdgcn -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
Konstantin Zhuravlyovf86e4b72016-11-13 07:01:11 +00003
4declare half @llvm.sin.f16(half %a)
5declare <2 x half> @llvm.sin.v2f16(<2 x half> %a)
6
7; GCN-LABEL: {{^}}sin_f16
8; GCN: buffer_load_ushort v[[A_F16:[0-9]+]]
9; GCN: v_cvt_f32_f16_e32 v[[A_F32:[0-9]+]], v[[A_F16]]
Matt Arsenault972034b2016-11-15 00:04:33 +000010; GCN: v_mul_f32_e32 v[[M_F32:[0-9]+]], {{0.15915494|0x3e22f983}}, v[[A_F32]]
Konstantin Zhuravlyovf86e4b72016-11-13 07:01:11 +000011; GCN: v_fract_f32_e32 v[[F_F32:[0-9]+]], v[[M_F32]]
12; GCN: v_sin_f32_e32 v[[R_F32:[0-9]+]], v[[F_F32]]
13; GCN: v_cvt_f16_f32_e32 v[[R_F16:[0-9]+]], v[[R_F32]]
14; GCN: buffer_store_short v[[R_F16]]
15; GCN: s_endpgm
16define void @sin_f16(
17 half addrspace(1)* %r,
18 half addrspace(1)* %a) {
19entry:
20 %a.val = load half, half addrspace(1)* %a
21 %r.val = call half @llvm.sin.f16(half %a.val)
22 store half %r.val, half addrspace(1)* %r
23 ret void
24}
25
26; GCN-LABEL: {{^}}sin_v2f16
27; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
28; SI: v_mov_b32_e32 v[[HALF_PIE:[0-9]+]], 0x3e22f983{{$}}
29; GCN: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
30; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
31; GCN: v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
32; SI: v_mul_f32_e32 v[[M_F32_0:[0-9]+]], v[[HALF_PIE]], v[[A_F32_0]]
Matt Arsenault972034b2016-11-15 00:04:33 +000033; VI: v_mul_f32_e32 v[[M_F32_0:[0-9]+]], 0.15915494, v[[A_F32_0]]
Konstantin Zhuravlyovf86e4b72016-11-13 07:01:11 +000034; GCN: v_fract_f32_e32 v[[F_F32_0:[0-9]+]], v[[M_F32_0]]
35; SI: v_mul_f32_e32 v[[M_F32_1:[0-9]+]], v[[HALF_PIE]], v[[A_F32_1]]
Matt Arsenault972034b2016-11-15 00:04:33 +000036; VI: v_mul_f32_e32 v[[M_F32_1:[0-9]+]], 0.15915494, v[[A_F32_1]]
Konstantin Zhuravlyovf86e4b72016-11-13 07:01:11 +000037; GCN: v_fract_f32_e32 v[[F_F32_1:[0-9]+]], v[[M_F32_1]]
38; GCN: v_sin_f32_e32 v[[R_F32_0:[0-9]+]], v[[F_F32_0]]
39; GCN: v_sin_f32_e32 v[[R_F32_1:[0-9]+]], v[[F_F32_1]]
40; GCN: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
41; GCN: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
42; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
43; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
44; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
45; GCN: buffer_store_dword v[[R_V2_F16]]
46; GCN: s_endpgm
47define void @sin_v2f16(
48 <2 x half> addrspace(1)* %r,
49 <2 x half> addrspace(1)* %a) {
50entry:
51 %a.val = load <2 x half>, <2 x half> addrspace(1)* %a
52 %r.val = call <2 x half> @llvm.sin.v2f16(<2 x half> %a.val)
53 store <2 x half> %r.val, <2 x half> addrspace(1)* %r
54 ret void
55}