1 | //==- BuiltinsAMDGPU.def - AMDGPU Builtin function database ------*- C++ -*-==// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | // |
9 | // This file defines the AMDGPU-specific builtin function database. Users of |
10 | // this file must define the BUILTIN macro to make use of this information. |
11 | // |
12 | //===----------------------------------------------------------------------===// |
13 | |
14 | // The format of this database matches clang/Basic/Builtins.def. |
15 | |
16 | #if defined(BUILTIN) && !defined(TARGET_BUILTIN) |
17 | # define TARGET_BUILTIN(ID, TYPE, ATTRS, FEATURE) BUILTIN(ID, TYPE, ATTRS) |
18 | #endif |
19 | //===----------------------------------------------------------------------===// |
20 | // SI+ only builtins. |
21 | //===----------------------------------------------------------------------===// |
22 | |
23 | BUILTIN(__builtin_amdgcn_dispatch_ptr, "v*4", "nc") |
24 | BUILTIN(__builtin_amdgcn_kernarg_segment_ptr, "v*4", "nc") |
25 | BUILTIN(__builtin_amdgcn_implicitarg_ptr, "v*4", "nc") |
26 | BUILTIN(__builtin_amdgcn_queue_ptr, "v*4", "nc") |
27 | |
28 | BUILTIN(__builtin_amdgcn_workgroup_id_x, "Ui", "nc") |
29 | BUILTIN(__builtin_amdgcn_workgroup_id_y, "Ui", "nc") |
30 | BUILTIN(__builtin_amdgcn_workgroup_id_z, "Ui", "nc") |
31 | |
32 | BUILTIN(__builtin_amdgcn_workitem_id_x, "Ui", "nc") |
33 | BUILTIN(__builtin_amdgcn_workitem_id_y, "Ui", "nc") |
34 | BUILTIN(__builtin_amdgcn_workitem_id_z, "Ui", "nc") |
35 | |
36 | //===----------------------------------------------------------------------===// |
37 | // Instruction builtins. |
38 | //===----------------------------------------------------------------------===// |
39 | BUILTIN(__builtin_amdgcn_s_getreg, "UiIi", "n") |
40 | BUILTIN(__builtin_amdgcn_s_getpc, "LUi", "n") |
41 | BUILTIN(__builtin_amdgcn_s_waitcnt, "vIi", "n") |
42 | BUILTIN(__builtin_amdgcn_s_sendmsg, "vIiUi", "n") |
43 | BUILTIN(__builtin_amdgcn_s_sendmsghalt, "vIiUi", "n") |
44 | BUILTIN(__builtin_amdgcn_s_barrier, "v", "n") |
45 | BUILTIN(__builtin_amdgcn_wave_barrier, "v", "n") |
46 | BUILTIN(__builtin_amdgcn_s_dcache_inv, "v", "n") |
47 | BUILTIN(__builtin_amdgcn_buffer_wbinvl1, "v", "n") |
48 | |
49 | // FIXME: Need to disallow constant address space. |
50 | BUILTIN(__builtin_amdgcn_div_scale, "dddbb*", "n") |
51 | BUILTIN(__builtin_amdgcn_div_scalef, "fffbb*", "n") |
52 | BUILTIN(__builtin_amdgcn_div_fmas, "ddddb", "nc") |
53 | BUILTIN(__builtin_amdgcn_div_fmasf, "ffffb", "nc") |
54 | BUILTIN(__builtin_amdgcn_div_fixup, "dddd", "nc") |
55 | BUILTIN(__builtin_amdgcn_div_fixupf, "ffff", "nc") |
56 | BUILTIN(__builtin_amdgcn_trig_preop, "ddi", "nc") |
57 | BUILTIN(__builtin_amdgcn_trig_preopf, "ffi", "nc") |
58 | BUILTIN(__builtin_amdgcn_rcp, "dd", "nc") |
59 | BUILTIN(__builtin_amdgcn_rcpf, "ff", "nc") |
60 | BUILTIN(__builtin_amdgcn_rsq, "dd", "nc") |
61 | BUILTIN(__builtin_amdgcn_rsqf, "ff", "nc") |
62 | BUILTIN(__builtin_amdgcn_rsq_clamp, "dd", "nc") |
63 | BUILTIN(__builtin_amdgcn_rsq_clampf, "ff", "nc") |
64 | BUILTIN(__builtin_amdgcn_sinf, "ff", "nc") |
65 | BUILTIN(__builtin_amdgcn_cosf, "ff", "nc") |
66 | BUILTIN(__builtin_amdgcn_log_clampf, "ff", "nc") |
67 | BUILTIN(__builtin_amdgcn_ldexp, "ddi", "nc") |
68 | BUILTIN(__builtin_amdgcn_ldexpf, "ffi", "nc") |
69 | BUILTIN(__builtin_amdgcn_frexp_mant, "dd", "nc") |
70 | BUILTIN(__builtin_amdgcn_frexp_mantf, "ff", "nc") |
71 | BUILTIN(__builtin_amdgcn_frexp_exp, "id", "nc") |
72 | BUILTIN(__builtin_amdgcn_frexp_expf, "if", "nc") |
73 | BUILTIN(__builtin_amdgcn_fract, "dd", "nc") |
74 | BUILTIN(__builtin_amdgcn_fractf, "ff", "nc") |
75 | BUILTIN(__builtin_amdgcn_lerp, "UiUiUiUi", "nc") |
76 | BUILTIN(__builtin_amdgcn_class, "bdi", "nc") |
77 | BUILTIN(__builtin_amdgcn_classf, "bfi", "nc") |
78 | BUILTIN(__builtin_amdgcn_cubeid, "ffff", "nc") |
79 | BUILTIN(__builtin_amdgcn_cubesc, "ffff", "nc") |
80 | BUILTIN(__builtin_amdgcn_cubetc, "ffff", "nc") |
81 | BUILTIN(__builtin_amdgcn_cubema, "ffff", "nc") |
82 | BUILTIN(__builtin_amdgcn_s_memtime, "LUi", "n") |
83 | BUILTIN(__builtin_amdgcn_s_sleep, "vIi", "n") |
84 | BUILTIN(__builtin_amdgcn_s_incperflevel, "vIi", "n") |
85 | BUILTIN(__builtin_amdgcn_s_decperflevel, "vIi", "n") |
86 | BUILTIN(__builtin_amdgcn_uicmp, "LUiUiUiIi", "nc") |
87 | BUILTIN(__builtin_amdgcn_uicmpl, "LUiLUiLUiIi", "nc") |
88 | BUILTIN(__builtin_amdgcn_sicmp, "LUiiiIi", "nc") |
89 | BUILTIN(__builtin_amdgcn_sicmpl, "LUiLiLiIi", "nc") |
90 | BUILTIN(__builtin_amdgcn_fcmp, "LUiddIi", "nc") |
91 | BUILTIN(__builtin_amdgcn_fcmpf, "LUiffIi", "nc") |
92 | BUILTIN(__builtin_amdgcn_ds_swizzle, "iiIi", "nc") |
93 | BUILTIN(__builtin_amdgcn_ds_permute, "iii", "nc") |
94 | BUILTIN(__builtin_amdgcn_ds_bpermute, "iii", "nc") |
95 | BUILTIN(__builtin_amdgcn_readfirstlane, "ii", "nc") |
96 | BUILTIN(__builtin_amdgcn_readlane, "iii", "nc") |
97 | BUILTIN(__builtin_amdgcn_fmed3f, "ffff", "nc") |
98 | BUILTIN(__builtin_amdgcn_ds_faddf, "ff*3fIiIiIb", "n") |
99 | BUILTIN(__builtin_amdgcn_ds_fminf, "ff*3fIiIiIb", "n") |
100 | BUILTIN(__builtin_amdgcn_ds_fmaxf, "ff*3fIiIiIb", "n") |
101 | BUILTIN(__builtin_amdgcn_ds_append, "ii*3", "n") |
102 | BUILTIN(__builtin_amdgcn_ds_consume, "ii*3", "n") |
103 | |
104 | //===----------------------------------------------------------------------===// |
105 | // CI+ only builtins. |
106 | //===----------------------------------------------------------------------===// |
107 | TARGET_BUILTIN(__builtin_amdgcn_s_dcache_inv_vol, "v", "n", "ci-insts") |
108 | TARGET_BUILTIN(__builtin_amdgcn_buffer_wbinvl1_vol, "v", "n", "ci-insts") |
109 | |
110 | //===----------------------------------------------------------------------===// |
111 | // Interpolation builtins. |
112 | //===----------------------------------------------------------------------===// |
113 | BUILTIN(__builtin_amdgcn_interp_p1_f16, "ffUiUibUi", "nc") |
114 | BUILTIN(__builtin_amdgcn_interp_p2_f16, "hffUiUibUi", "nc") |
115 | BUILTIN(__builtin_amdgcn_interp_p1, "ffUiUiUi", "nc") |
116 | BUILTIN(__builtin_amdgcn_interp_p2, "fffUiUiUi", "nc") |
117 | BUILTIN(__builtin_amdgcn_interp_mov, "fUiUiUiUi", "nc") |
118 | |
119 | //===----------------------------------------------------------------------===// |
120 | // VI+ only builtins. |
121 | //===----------------------------------------------------------------------===// |
122 | |
123 | TARGET_BUILTIN(__builtin_amdgcn_div_fixuph, "hhhh", "nc", "16-bit-insts") |
124 | TARGET_BUILTIN(__builtin_amdgcn_rcph, "hh", "nc", "16-bit-insts") |
125 | TARGET_BUILTIN(__builtin_amdgcn_rsqh, "hh", "nc", "16-bit-insts") |
126 | TARGET_BUILTIN(__builtin_amdgcn_sinh, "hh", "nc", "16-bit-insts") |
127 | TARGET_BUILTIN(__builtin_amdgcn_cosh, "hh", "nc", "16-bit-insts") |
128 | TARGET_BUILTIN(__builtin_amdgcn_ldexph, "hhi", "nc", "16-bit-insts") |
129 | TARGET_BUILTIN(__builtin_amdgcn_frexp_manth, "hh", "nc", "16-bit-insts") |
130 | TARGET_BUILTIN(__builtin_amdgcn_frexp_exph, "sh", "nc", "16-bit-insts") |
131 | TARGET_BUILTIN(__builtin_amdgcn_fracth, "hh", "nc", "16-bit-insts") |
132 | TARGET_BUILTIN(__builtin_amdgcn_classh, "bhi", "nc", "16-bit-insts") |
133 | TARGET_BUILTIN(__builtin_amdgcn_s_memrealtime, "LUi", "n", "s-memrealtime") |
134 | TARGET_BUILTIN(__builtin_amdgcn_mov_dpp, "iiIiIiIiIb", "nc", "dpp") |
135 | TARGET_BUILTIN(__builtin_amdgcn_update_dpp, "iiiIiIiIiIb", "nc", "dpp") |
136 | TARGET_BUILTIN(__builtin_amdgcn_s_dcache_wb, "v", "n", "vi-insts") |
137 | |
138 | //===----------------------------------------------------------------------===// |
139 | // GFX9+ only builtins. |
140 | //===----------------------------------------------------------------------===// |
141 | |
142 | TARGET_BUILTIN(__builtin_amdgcn_fmed3h, "hhhh", "nc", "gfx9-insts") |
143 | |
144 | //===----------------------------------------------------------------------===// |
145 | // Deep learning builtins. |
146 | //===----------------------------------------------------------------------===// |
147 | |
148 | TARGET_BUILTIN(__builtin_amdgcn_fdot2, "fV2hV2hfIb", "nc", "dot2-insts") |
149 | TARGET_BUILTIN(__builtin_amdgcn_sdot2, "SiV2SsV2SsSiIb", "nc", "dot2-insts") |
150 | TARGET_BUILTIN(__builtin_amdgcn_udot2, "UiV2UsV2UsUiIb", "nc", "dot2-insts") |
151 | TARGET_BUILTIN(__builtin_amdgcn_sdot4, "SiSiSiSiIb", "nc", "dot1-insts") |
152 | TARGET_BUILTIN(__builtin_amdgcn_udot4, "UiUiUiUiIb", "nc", "dot2-insts") |
153 | TARGET_BUILTIN(__builtin_amdgcn_sdot8, "SiSiSiSiIb", "nc", "dot1-insts") |
154 | TARGET_BUILTIN(__builtin_amdgcn_udot8, "UiUiUiUiIb", "nc", "dot2-insts") |
155 | |
156 | //===----------------------------------------------------------------------===// |
157 | // Special builtins. |
158 | //===----------------------------------------------------------------------===// |
159 | BUILTIN(__builtin_amdgcn_read_exec, "LUi", "nc") |
160 | BUILTIN(__builtin_amdgcn_read_exec_lo, "Ui", "nc") |
161 | BUILTIN(__builtin_amdgcn_read_exec_hi, "Ui", "nc") |
162 | |
163 | //===----------------------------------------------------------------------===// |
164 | // R600-NI only builtins. |
165 | //===----------------------------------------------------------------------===// |
166 | |
167 | BUILTIN(__builtin_r600_implicitarg_ptr, "Uc*7", "nc") |
168 | |
169 | BUILTIN(__builtin_r600_read_tgid_x, "Ui", "nc") |
170 | BUILTIN(__builtin_r600_read_tgid_y, "Ui", "nc") |
171 | BUILTIN(__builtin_r600_read_tgid_z, "Ui", "nc") |
172 | |
173 | BUILTIN(__builtin_r600_read_tidig_x, "Ui", "nc") |
174 | BUILTIN(__builtin_r600_read_tidig_y, "Ui", "nc") |
175 | BUILTIN(__builtin_r600_read_tidig_z, "Ui", "nc") |
176 | |
177 | BUILTIN(__builtin_r600_recipsqrt_ieee, "dd", "nc") |
178 | BUILTIN(__builtin_r600_recipsqrt_ieeef, "ff", "nc") |
179 | |
180 | #undef BUILTIN |
181 | #undef TARGET_BUILTIN |
182 | |