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// Note: (unsigned) long int type should be avoided in builtin definitions
13// since it has different size on Linux (64 bit) and Windows (32 bit).
14// (unsigned) long long int type should also be avoided, which is 64 bit for
15// C/C++/HIP but is 128 bit for OpenCL. Use `W` as width modifier in builtin
16// definitions since it is fixed for 64 bit.
17//===----------------------------------------------------------------------===//
18
19// The format of this database matches clang/Basic/Builtins.def.
20
21#if defined(BUILTIN) && !defined(TARGET_BUILTIN)
22# define TARGET_BUILTIN(ID, TYPE, ATTRS, FEATURE) BUILTIN(ID, TYPE, ATTRS)
23#endif
24//===----------------------------------------------------------------------===//
25// SI+ only builtins.
26//===----------------------------------------------------------------------===//
27
28BUILTIN(__builtin_amdgcn_dispatch_ptr, "v*4", "nc")
29BUILTIN(__builtin_amdgcn_kernarg_segment_ptr, "v*4", "nc")
30BUILTIN(__builtin_amdgcn_implicitarg_ptr, "v*4", "nc")
31BUILTIN(__builtin_amdgcn_queue_ptr, "v*4", "nc")
32
33BUILTIN(__builtin_amdgcn_workgroup_id_x, "Ui", "nc")
34BUILTIN(__builtin_amdgcn_workgroup_id_y, "Ui", "nc")
35BUILTIN(__builtin_amdgcn_workgroup_id_z, "Ui", "nc")
36
37BUILTIN(__builtin_amdgcn_workitem_id_x, "Ui", "nc")
38BUILTIN(__builtin_amdgcn_workitem_id_y, "Ui", "nc")
39BUILTIN(__builtin_amdgcn_workitem_id_z, "Ui", "nc")
40
41BUILTIN(__builtin_amdgcn_workgroup_size_x, "Us", "nc")
42BUILTIN(__builtin_amdgcn_workgroup_size_y, "Us", "nc")
43BUILTIN(__builtin_amdgcn_workgroup_size_z, "Us", "nc")
44
45BUILTIN(__builtin_amdgcn_grid_size_x, "Ui", "nc")
46BUILTIN(__builtin_amdgcn_grid_size_y, "Ui", "nc")
47BUILTIN(__builtin_amdgcn_grid_size_z, "Ui", "nc")
48
49BUILTIN(__builtin_amdgcn_mbcnt_hi, "UiUiUi", "nc")
50BUILTIN(__builtin_amdgcn_mbcnt_lo, "UiUiUi", "nc")
51
52TARGET_BUILTIN(__builtin_amdgcn_s_memtime, "WUi", "n", "s-memtime-inst")
53
54//===----------------------------------------------------------------------===//
55// Instruction builtins.
56//===----------------------------------------------------------------------===//
57BUILTIN(__builtin_amdgcn_s_getreg, "UiIi", "n")
58BUILTIN(__builtin_amdgcn_s_setreg, "vIiUi", "n")
59BUILTIN(__builtin_amdgcn_s_getpc, "WUi", "n")
60BUILTIN(__builtin_amdgcn_s_waitcnt, "vIi", "n")
61BUILTIN(__builtin_amdgcn_s_sendmsg, "vIiUi", "n")
62BUILTIN(__builtin_amdgcn_s_sendmsghalt, "vIiUi", "n")
63BUILTIN(__builtin_amdgcn_s_barrier, "v", "n")
64BUILTIN(__builtin_amdgcn_s_ttracedata, "vi", "n")
65BUILTIN(__builtin_amdgcn_wave_barrier, "v", "n")
66BUILTIN(__builtin_amdgcn_sched_barrier, "vIi", "n")
67BUILTIN(__builtin_amdgcn_sched_group_barrier, "vIiIiIi", "n")
68BUILTIN(__builtin_amdgcn_iglp_opt, "vIi", "n")
69BUILTIN(__builtin_amdgcn_s_dcache_inv, "v", "n")
70BUILTIN(__builtin_amdgcn_buffer_wbinvl1, "v", "n")
71BUILTIN(__builtin_amdgcn_fence, "vUicC*.", "n")
72BUILTIN(__builtin_amdgcn_groupstaticsize, "Ui", "n")
73BUILTIN(__builtin_amdgcn_wavefrontsize, "Ui", "nc")
74
75BUILTIN(__builtin_amdgcn_atomic_inc32, "UZiUZiD*UZiUicC*", "n")
76BUILTIN(__builtin_amdgcn_atomic_inc64, "UWiUWiD*UWiUicC*", "n")
77
78BUILTIN(__builtin_amdgcn_atomic_dec32, "UZiUZiD*UZiUicC*", "n")
79BUILTIN(__builtin_amdgcn_atomic_dec64, "UWiUWiD*UWiUicC*", "n")
80
81// FIXME: Need to disallow constant address space.
82BUILTIN(__builtin_amdgcn_div_scale, "dddbb*", "n")
83BUILTIN(__builtin_amdgcn_div_scalef, "fffbb*", "n")
84BUILTIN(__builtin_amdgcn_div_fmas, "ddddb", "nc")
85BUILTIN(__builtin_amdgcn_div_fmasf, "ffffb", "nc")
86BUILTIN(__builtin_amdgcn_div_fixup, "dddd", "nc")
87BUILTIN(__builtin_amdgcn_div_fixupf, "ffff", "nc")
88BUILTIN(__builtin_amdgcn_trig_preop, "ddi", "nc")
89BUILTIN(__builtin_amdgcn_trig_preopf, "ffi", "nc")
90BUILTIN(__builtin_amdgcn_rcp, "dd", "nc")
91BUILTIN(__builtin_amdgcn_rcpf, "ff", "nc")
92BUILTIN(__builtin_amdgcn_sqrt, "dd", "nc")
93BUILTIN(__builtin_amdgcn_sqrtf, "ff", "nc")
94BUILTIN(__builtin_amdgcn_rsq, "dd", "nc")
95BUILTIN(__builtin_amdgcn_rsqf, "ff", "nc")
96BUILTIN(__builtin_amdgcn_rsq_clamp, "dd", "nc")
97BUILTIN(__builtin_amdgcn_rsq_clampf, "ff", "nc")
98BUILTIN(__builtin_amdgcn_sinf, "ff", "nc")
99BUILTIN(__builtin_amdgcn_cosf, "ff", "nc")
100BUILTIN(__builtin_amdgcn_logf, "ff", "nc")
101BUILTIN(__builtin_amdgcn_exp2f, "ff", "nc")
102BUILTIN(__builtin_amdgcn_log_clampf, "ff", "nc")
103BUILTIN(__builtin_amdgcn_ldexp, "ddi", "nc")
104BUILTIN(__builtin_amdgcn_ldexpf, "ffi", "nc")
105BUILTIN(__builtin_amdgcn_frexp_mant, "dd", "nc")
106BUILTIN(__builtin_amdgcn_frexp_mantf, "ff", "nc")
107BUILTIN(__builtin_amdgcn_frexp_exp, "id", "nc")
108BUILTIN(__builtin_amdgcn_frexp_expf, "if", "nc")
109BUILTIN(__builtin_amdgcn_fract, "dd", "nc")
110BUILTIN(__builtin_amdgcn_fractf, "ff", "nc")
111BUILTIN(__builtin_amdgcn_lerp, "UiUiUiUi", "nc")
112BUILTIN(__builtin_amdgcn_class, "bdi", "nc")
113BUILTIN(__builtin_amdgcn_classf, "bfi", "nc")
114BUILTIN(__builtin_amdgcn_cubeid, "ffff", "nc")
115BUILTIN(__builtin_amdgcn_cubesc, "ffff", "nc")
116BUILTIN(__builtin_amdgcn_cubetc, "ffff", "nc")
117BUILTIN(__builtin_amdgcn_cubema, "ffff", "nc")
118BUILTIN(__builtin_amdgcn_s_sleep, "vIi", "n")
119BUILTIN(__builtin_amdgcn_s_incperflevel, "vIi", "n")
120BUILTIN(__builtin_amdgcn_s_decperflevel, "vIi", "n")
121BUILTIN(__builtin_amdgcn_s_setprio, "vIs", "n")
122BUILTIN(__builtin_amdgcn_ds_swizzle, "iiIi", "nc")
123BUILTIN(__builtin_amdgcn_ds_permute, "iii", "nc")
124BUILTIN(__builtin_amdgcn_ds_bpermute, "iii", "nc")
125BUILTIN(__builtin_amdgcn_readfirstlane, "ii", "nc")
126BUILTIN(__builtin_amdgcn_readlane, "iii", "nc")
127BUILTIN(__builtin_amdgcn_fmed3f, "ffff", "nc")
128BUILTIN(__builtin_amdgcn_ds_faddf, "ff*3fIiIiIb", "n")
129BUILTIN(__builtin_amdgcn_ds_fminf, "ff*3fIiIiIb", "n")
130BUILTIN(__builtin_amdgcn_ds_fmaxf, "ff*3fIiIiIb", "n")
131BUILTIN(__builtin_amdgcn_ds_append, "ii*3", "n")
132BUILTIN(__builtin_amdgcn_ds_consume, "ii*3", "n")
133BUILTIN(__builtin_amdgcn_alignbit, "UiUiUiUi", "nc")
134BUILTIN(__builtin_amdgcn_alignbyte, "UiUiUiUi", "nc")
135BUILTIN(__builtin_amdgcn_ubfe, "UiUiUiUi", "nc")
136BUILTIN(__builtin_amdgcn_sbfe, "UiUiUiUi", "nc")
137BUILTIN(__builtin_amdgcn_cvt_pkrtz, "E2hff", "nc")
138BUILTIN(__builtin_amdgcn_cvt_pknorm_i16, "E2sff", "nc")
139BUILTIN(__builtin_amdgcn_cvt_pknorm_u16, "E2Usff", "nc")
140BUILTIN(__builtin_amdgcn_cvt_pk_i16, "E2sii", "nc")
141BUILTIN(__builtin_amdgcn_cvt_pk_u16, "E2UsUiUi", "nc")
142BUILTIN(__builtin_amdgcn_cvt_pk_u8_f32, "UifUiUi", "nc")
143BUILTIN(__builtin_amdgcn_cvt_off_f32_i4, "fi", "nc")
144BUILTIN(__builtin_amdgcn_sad_u8, "UiUiUiUi", "nc")
145BUILTIN(__builtin_amdgcn_msad_u8, "UiUiUiUi", "nc")
146BUILTIN(__builtin_amdgcn_sad_hi_u8, "UiUiUiUi", "nc")
147BUILTIN(__builtin_amdgcn_sad_u16, "UiUiUiUi", "nc")
148BUILTIN(__builtin_amdgcn_qsad_pk_u16_u8, "WUiWUiUiWUi", "nc")
149BUILTIN(__builtin_amdgcn_mqsad_pk_u16_u8, "WUiWUiUiWUi", "nc")
150BUILTIN(__builtin_amdgcn_mqsad_u32_u8, "V4UiWUiUiV4Ui", "nc")
151
152BUILTIN(__builtin_amdgcn_make_buffer_rsrc, "Qbv*sii", "nc")
153BUILTIN(__builtin_amdgcn_raw_buffer_store_b8, "vUcQbiiIi", "n")
154BUILTIN(__builtin_amdgcn_raw_buffer_store_b16, "vUsQbiiIi", "n")
155BUILTIN(__builtin_amdgcn_raw_buffer_store_b32, "vUiQbiiIi", "n")
156BUILTIN(__builtin_amdgcn_raw_buffer_store_b64, "vV2UiQbiiIi", "n")
157BUILTIN(__builtin_amdgcn_raw_buffer_store_b96, "vV3UiQbiiIi", "n")
158BUILTIN(__builtin_amdgcn_raw_buffer_store_b128, "vV4UiQbiiIi", "n")
159BUILTIN(__builtin_amdgcn_raw_buffer_load_b8, "UcQbiiIi", "n")
160BUILTIN(__builtin_amdgcn_raw_buffer_load_b16, "UsQbiiIi", "n")
161BUILTIN(__builtin_amdgcn_raw_buffer_load_b32, "UiQbiiIi", "n")
162BUILTIN(__builtin_amdgcn_raw_buffer_load_b64, "V2UiQbiiIi", "n")
163BUILTIN(__builtin_amdgcn_raw_buffer_load_b96, "V3UiQbiiIi", "n")
164BUILTIN(__builtin_amdgcn_raw_buffer_load_b128, "V4UiQbiiIi", "n")
165
166TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_load_lds, "vQbv*3IUiiiIiIi", "t", "vmem-to-lds-load-insts")
167
168//===----------------------------------------------------------------------===//
169// Ballot builtins.
170//===----------------------------------------------------------------------===//
171
172TARGET_BUILTIN(__builtin_amdgcn_ballot_w32, "ZUib", "nc", "wavefrontsize32")
173BUILTIN(__builtin_amdgcn_ballot_w64, "WUib", "nc")
174
175// Deprecated intrinsics in favor of __builtin_amdgn_ballot_{w32|w64}
176BUILTIN(__builtin_amdgcn_uicmp, "WUiUiUiIi", "nc")
177BUILTIN(__builtin_amdgcn_uicmpl, "WUiWUiWUiIi", "nc")
178BUILTIN(__builtin_amdgcn_sicmp, "WUiiiIi", "nc")
179BUILTIN(__builtin_amdgcn_sicmpl, "WUiWiWiIi", "nc")
180BUILTIN(__builtin_amdgcn_fcmp, "WUiddIi", "nc")
181BUILTIN(__builtin_amdgcn_fcmpf, "WUiffIi", "nc")
182
183//===----------------------------------------------------------------------===//
184// Flat addressing builtins.
185//===----------------------------------------------------------------------===//
186BUILTIN(__builtin_amdgcn_is_shared, "bvC*0", "nc")
187BUILTIN(__builtin_amdgcn_is_private, "bvC*0", "nc")
188
189//===----------------------------------------------------------------------===//
190// GWS builtins.
191//===----------------------------------------------------------------------===//
192TARGET_BUILTIN(__builtin_amdgcn_ds_gws_init, "vUiUi", "n", "gws")
193TARGET_BUILTIN(__builtin_amdgcn_ds_gws_barrier, "vUiUi", "n", "gws")
194TARGET_BUILTIN(__builtin_amdgcn_ds_gws_sema_v, "vUi", "n", "gws")
195TARGET_BUILTIN(__builtin_amdgcn_ds_gws_sema_br, "vUiUi", "n", "gws")
196TARGET_BUILTIN(__builtin_amdgcn_ds_gws_sema_p, "vUi", "n", "gws")
197
198//===----------------------------------------------------------------------===//
199// CI+ only builtins.
200//===----------------------------------------------------------------------===//
201TARGET_BUILTIN(__builtin_amdgcn_s_dcache_inv_vol, "v", "n", "ci-insts")
202TARGET_BUILTIN(__builtin_amdgcn_buffer_wbinvl1_vol, "v", "n", "ci-insts")
203TARGET_BUILTIN(__builtin_amdgcn_ds_gws_sema_release_all, "vUi", "n", "ci-insts")
204
205//===----------------------------------------------------------------------===//
206// Interpolation builtins.
207//===----------------------------------------------------------------------===//
208BUILTIN(__builtin_amdgcn_interp_p1_f16, "ffUiUibUi", "nc")
209BUILTIN(__builtin_amdgcn_interp_p2_f16, "hffUiUibUi", "nc")
210BUILTIN(__builtin_amdgcn_interp_p1, "ffUiUiUi", "nc")
211BUILTIN(__builtin_amdgcn_interp_p2, "fffUiUiUi", "nc")
212BUILTIN(__builtin_amdgcn_interp_mov, "fUiUiUiUi", "nc")
213
214//===----------------------------------------------------------------------===//
215// VI+ only builtins.
216//===----------------------------------------------------------------------===//
217
218TARGET_BUILTIN(__builtin_amdgcn_div_fixuph, "hhhh", "nc", "16-bit-insts")
219TARGET_BUILTIN(__builtin_amdgcn_rcph, "hh", "nc", "16-bit-insts")
220TARGET_BUILTIN(__builtin_amdgcn_sqrth, "hh", "nc", "16-bit-insts")
221TARGET_BUILTIN(__builtin_amdgcn_rsqh, "hh", "nc", "16-bit-insts")
222TARGET_BUILTIN(__builtin_amdgcn_sinh, "hh", "nc", "16-bit-insts")
223TARGET_BUILTIN(__builtin_amdgcn_cosh, "hh", "nc", "16-bit-insts")
224TARGET_BUILTIN(__builtin_amdgcn_ldexph, "hhi", "nc", "16-bit-insts")
225TARGET_BUILTIN(__builtin_amdgcn_frexp_manth, "hh", "nc", "16-bit-insts")
226TARGET_BUILTIN(__builtin_amdgcn_frexp_exph, "sh", "nc", "16-bit-insts")
227TARGET_BUILTIN(__builtin_amdgcn_fracth, "hh", "nc", "16-bit-insts")
228TARGET_BUILTIN(__builtin_amdgcn_classh, "bhi", "nc", "16-bit-insts")
229TARGET_BUILTIN(__builtin_amdgcn_s_memrealtime, "WUi", "n", "s-memrealtime")
230TARGET_BUILTIN(__builtin_amdgcn_mov_dpp, "iiIiIiIiIb", "nct", "dpp")
231TARGET_BUILTIN(__builtin_amdgcn_update_dpp, "iiiIiIiIiIb", "nct", "dpp")
232TARGET_BUILTIN(__builtin_amdgcn_s_dcache_wb, "v", "n", "gfx8-insts")
233TARGET_BUILTIN(__builtin_amdgcn_perm, "UiUiUiUi", "nc", "gfx8-insts")
234
235//===----------------------------------------------------------------------===//
236// GFX9+ only builtins.
237//===----------------------------------------------------------------------===//
238
239TARGET_BUILTIN(__builtin_amdgcn_fmed3h, "hhhh", "nc", "gfx9-insts")
240
241TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_f64, "dd*1d", "t", "gfx90a-insts")
242TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_f32, "ff*1f", "t", "atomic-fadd-rtn-insts")
243TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_v2f16, "V2hV2h*1V2h", "t", "atomic-buffer-global-pk-add-f16-insts")
244TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fmin_f64, "dd*1d", "t", "gfx90a-insts")
245TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fmax_f64, "dd*1d", "t", "gfx90a-insts")
246
247TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_f64, "dd*0d", "t", "gfx90a-insts")
248TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fmin_f64, "dd*0d", "t", "gfx90a-insts")
249TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fmax_f64, "dd*0d", "t", "gfx90a-insts")
250
251TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_f64, "dd*3d", "t", "gfx90a-insts")
252TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_f32, "ff*3f", "t", "gfx8-insts")
253
254TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_f32, "ff*0f", "t", "gfx940-insts")
255TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_v2f16, "V2hV2h*0V2h", "t", "atomic-flat-pk-add-16-insts")
256TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_v2bf16, "V2sV2s*0V2s", "t", "atomic-flat-pk-add-16-insts")
257TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_v2bf16, "V2sV2s*1V2s", "t", "atomic-global-pk-add-bf16-inst")
258TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_v2bf16, "V2sV2s*3V2s", "t", "atomic-ds-pk-add-16-insts")
259TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_v2f16, "V2hV2h*3V2h", "t", "atomic-ds-pk-add-16-insts")
260TARGET_BUILTIN(__builtin_amdgcn_load_to_lds, "vv*v*3IUiIiIUi", "", "vmem-to-lds-load-insts")
261TARGET_BUILTIN(__builtin_amdgcn_global_load_lds, "vv*1v*3IUiIiIUi", "t", "vmem-to-lds-load-insts")
262
263//===----------------------------------------------------------------------===//
264// Deep learning builtins.
265//===----------------------------------------------------------------------===//
266
267TARGET_BUILTIN(__builtin_amdgcn_fdot2, "fV2hV2hfIb", "nc", "dot10-insts")
268TARGET_BUILTIN(__builtin_amdgcn_fdot2_f16_f16, "hV2hV2hh", "nc", "dot9-insts")
269TARGET_BUILTIN(__builtin_amdgcn_fdot2_bf16_bf16, "sV2sV2ss", "nc", "dot9-insts")
270TARGET_BUILTIN(__builtin_amdgcn_fdot2_f32_bf16, "fV2sV2sfIb", "nc", "dot12-insts")
271TARGET_BUILTIN(__builtin_amdgcn_sdot2, "SiV2SsV2SsSiIb", "nc", "dot2-insts")
272TARGET_BUILTIN(__builtin_amdgcn_udot2, "UiV2UsV2UsUiIb", "nc", "dot2-insts")
273TARGET_BUILTIN(__builtin_amdgcn_sdot4, "SiSiSiSiIb", "nc", "dot1-insts")
274TARGET_BUILTIN(__builtin_amdgcn_udot4, "UiUiUiUiIb", "nc", "dot7-insts")
275TARGET_BUILTIN(__builtin_amdgcn_sudot4, "iIbiIbiiIb", "nc", "dot8-insts")
276TARGET_BUILTIN(__builtin_amdgcn_sdot8, "SiSiSiSiIb", "nc", "dot1-insts")
277TARGET_BUILTIN(__builtin_amdgcn_udot8, "UiUiUiUiIb", "nc", "dot7-insts")
278TARGET_BUILTIN(__builtin_amdgcn_sudot8, "iIbiIbiiIb", "nc", "dot8-insts")
279TARGET_BUILTIN(__builtin_amdgcn_dot4_f32_fp8_bf8, "fUiUif", "nc", "dot11-insts")
280TARGET_BUILTIN(__builtin_amdgcn_dot4_f32_bf8_fp8, "fUiUif", "nc", "dot11-insts")
281TARGET_BUILTIN(__builtin_amdgcn_dot4_f32_fp8_fp8, "fUiUif", "nc", "dot11-insts")
282TARGET_BUILTIN(__builtin_amdgcn_dot4_f32_bf8_bf8, "fUiUif", "nc", "dot11-insts")
283TARGET_BUILTIN(__builtin_amdgcn_fdot2c_f32_bf16, "fV2yV2yfIb", "nc", "dot13-insts")
284
285//===----------------------------------------------------------------------===//
286// GFX10+ only builtins.
287//===----------------------------------------------------------------------===//
288TARGET_BUILTIN(__builtin_amdgcn_permlane16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts")
289TARGET_BUILTIN(__builtin_amdgcn_permlanex16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts")
290TARGET_BUILTIN(__builtin_amdgcn_mov_dpp8, "UiUiIUi", "nct", "gfx10-insts")
291TARGET_BUILTIN(__builtin_amdgcn_s_ttracedata_imm, "vIs", "n", "gfx10-insts")
292
293//===----------------------------------------------------------------------===//
294// Raytracing builtins.
295// By default the 1st argument is i32 and the 4/5-th arguments are float4.
296// Postfix l indicates the 1st argument is i64.
297// Postfix h indicates the 4/5-th arguments are half4.
298//===----------------------------------------------------------------------===//
299TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray, "V4UiUifV4fV4fV4fV4Ui", "nc", "gfx10-insts")
300TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_h, "V4UiUifV4fV4hV4hV4Ui", "nc", "gfx10-insts")
301TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_l, "V4UiWUifV4fV4fV4fV4Ui", "nc", "gfx10-insts")
302TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_lh, "V4UiWUifV4fV4hV4hV4Ui", "nc", "gfx10-insts")
303
304
305//===----------------------------------------------------------------------===//
306// GFX11+ only builtins.
307//===----------------------------------------------------------------------===//
308
309// TODO: This is a no-op in wave32. Should the builtin require wavefrontsize64?
310TARGET_BUILTIN(__builtin_amdgcn_permlane64, "UiUi", "nc", "gfx11-insts")
311TARGET_BUILTIN(__builtin_amdgcn_s_wait_event_export_ready, "v", "n", "gfx11-insts")
312
313//===----------------------------------------------------------------------===//
314// WMMA builtins.
315// Postfix w32 indicates the builtin requires wavefront size of 32.
316// Postfix w64 indicates the builtin requires wavefront size of 64.
317//===----------------------------------------------------------------------===//
318TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_f16_w32, "V8fV16hV16hV8f", "nc", "gfx11-insts,wavefrontsize32")
319TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf16_w32, "V8fV16sV16sV8f", "nc", "gfx11-insts,wavefrontsize32")
320TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x16_f16_w32, "V16hV16hV16hV16hIb", "nc", "gfx11-insts,wavefrontsize32")
321TARGET_BUILTIN(__builtin_amdgcn_wmma_bf16_16x16x16_bf16_w32, "V16sV16sV16sV16sIb", "nc", "gfx11-insts,wavefrontsize32")
322TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x16_f16_tied_w32, "V16hV16hV16hV16hIb", "nc", "gfx11-insts,wavefrontsize32")
323TARGET_BUILTIN(__builtin_amdgcn_wmma_bf16_16x16x16_bf16_tied_w32, "V16sV16sV16sV16sIb", "nc", "gfx11-insts,wavefrontsize32")
324TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu8_w32, "V8iIbV4iIbV4iV8iIb", "nc", "gfx11-insts,wavefrontsize32")
325TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu4_w32, "V8iIbV2iIbV2iV8iIb", "nc", "gfx11-insts,wavefrontsize32")
326
327TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_f16_w64, "V4fV16hV16hV4f", "nc", "gfx11-insts,wavefrontsize64")
328TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf16_w64, "V4fV16sV16sV4f", "nc", "gfx11-insts,wavefrontsize64")
329TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x16_f16_w64, "V8hV16hV16hV8hIb", "nc", "gfx11-insts,wavefrontsize64")
330TARGET_BUILTIN(__builtin_amdgcn_wmma_bf16_16x16x16_bf16_w64, "V8sV16sV16sV8sIb", "nc", "gfx11-insts,wavefrontsize64")
331TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x16_f16_tied_w64, "V8hV16hV16hV8hIb", "nc", "gfx11-insts,wavefrontsize64")
332TARGET_BUILTIN(__builtin_amdgcn_wmma_bf16_16x16x16_bf16_tied_w64, "V8sV16sV16sV8sIb", "nc", "gfx11-insts,wavefrontsize64")
333TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu8_w64, "V4iIbV4iIbV4iV4iIb", "nc", "gfx11-insts,wavefrontsize64")
334TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu4_w64, "V4iIbV2iIbV2iV4iIb", "nc", "gfx11-insts,wavefrontsize64")
335
336TARGET_BUILTIN(__builtin_amdgcn_s_sendmsg_rtn, "UiUIi", "n", "gfx11-insts")
337TARGET_BUILTIN(__builtin_amdgcn_s_sendmsg_rtnl, "UWiUIi", "n", "gfx11-insts")
338
339TARGET_BUILTIN(__builtin_amdgcn_ds_bvh_stack_rtn, "V2UiUiUiV4UiIi", "n", "gfx11-insts")
340
341//===----------------------------------------------------------------------===//
342// Special builtins.
343//===----------------------------------------------------------------------===//
344BUILTIN(__builtin_amdgcn_read_exec, "WUi", "nc")
345BUILTIN(__builtin_amdgcn_read_exec_lo, "Ui", "nc")
346BUILTIN(__builtin_amdgcn_read_exec_hi, "Ui", "nc")
347
348BUILTIN(__builtin_amdgcn_endpgm, "v", "nr")
349
350BUILTIN(__builtin_amdgcn_get_fpenv, "WUi", "n")
351BUILTIN(__builtin_amdgcn_set_fpenv, "vWUi", "n")
352
353//===----------------------------------------------------------------------===//
354// R600-NI only builtins.
355//===----------------------------------------------------------------------===//
356
357BUILTIN(__builtin_r600_implicitarg_ptr, "Uc*7", "nc")
358
359BUILTIN(__builtin_r600_read_tgid_x, "Ui", "nc")
360BUILTIN(__builtin_r600_read_tgid_y, "Ui", "nc")
361BUILTIN(__builtin_r600_read_tgid_z, "Ui", "nc")
362
363BUILTIN(__builtin_r600_read_tidig_x, "Ui", "nc")
364BUILTIN(__builtin_r600_read_tidig_y, "Ui", "nc")
365BUILTIN(__builtin_r600_read_tidig_z, "Ui", "nc")
366
367BUILTIN(__builtin_r600_recipsqrt_ieee, "dd", "nc")
368BUILTIN(__builtin_r600_recipsqrt_ieeef, "ff", "nc")
369
370//===----------------------------------------------------------------------===//
371// MFMA builtins.
372//===----------------------------------------------------------------------===//
373
374TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x1f32, "V32fffV32fIiIiIi", "nc", "mai-insts")
375TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x1f32, "V16fffV16fIiIiIi", "nc", "mai-insts")
376TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x1f32, "V4fffV4fIiIiIi", "nc", "mai-insts")
377TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x2f32, "V16fffV16fIiIiIi", "nc", "mai-insts")
378TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4f32, "V4fffV4fIiIiIi", "nc", "mai-insts")
379TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4f16, "V32fV4hV4hV32fIiIiIi", "nc", "mai-insts")
380TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4f16, "V16fV4hV4hV16fIiIiIi", "nc", "mai-insts")
381TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x4f16, "V4fV4hV4hV4fIiIiIi", "nc", "mai-insts")
382TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x8f16, "V16fV4hV4hV16fIiIiIi", "nc", "mai-insts")
383TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x16f16, "V4fV4hV4hV4fIiIiIi", "nc", "mai-insts")
384TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x4i8, "V32iiiV32iIiIiIi", "nc", "mai-insts")
385TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x4i8, "V16iiiV16iIiIiIi", "nc", "mai-insts")
386TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_4x4x4i8, "V4iiiV4iIiIiIi", "nc", "mai-insts")
387TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x8i8, "V16iiiV16iIiIiIi", "nc", "mai-insts")
388TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x16i8, "V4iiiV4iIiIiIi", "nc", "mai-insts")
389TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x2bf16, "V32fV2sV2sV32fIiIiIi", "nc", "mai-insts")
390TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x2bf16, "V16fV2sV2sV16fIiIiIi", "nc", "mai-insts")
391TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x2bf16, "V4fV2sV2sV4fIiIiIi", "nc", "mai-insts")
392TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4bf16, "V16fV2sV2sV16fIiIiIi", "nc", "mai-insts")
393TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x8bf16, "V4fV2sV2sV4fIiIiIi", "nc", "mai-insts")
394
395TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4bf16_1k, "V32fV4sV4sV32fIiIiIi", "nc", "mai-insts")
396TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4bf16_1k, "V16fV4sV4sV16fIiIiIi", "nc", "mai-insts")
397TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x4bf16_1k, "V4fV4sV4sV4fIiIiIi", "nc", "mai-insts")
398TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x8bf16_1k, "V16fV4sV4sV16fIiIiIi", "nc", "mai-insts")
399TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x16bf16_1k, "V4fV4sV4sV4fIiIiIi", "nc", "mai-insts")
400TARGET_BUILTIN(__builtin_amdgcn_mfma_f64_16x16x4f64, "V4dddV4dIiIiIi", "nc", "mai-insts")
401TARGET_BUILTIN(__builtin_amdgcn_mfma_f64_4x4x4f64, "ddddIiIiIi", "nc", "mai-insts")
402
403TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x32_i8, "V4iWiWiV4iIiIiIi", "nc", "mai-insts")
404TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x16_i8, "V16iWiWiV16iIiIiIi", "nc", "mai-insts")
405TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x8_xf32, "V4fV2fV2fV4fIiIiIi", "nc", "mai-insts")
406TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4_xf32, "V16fV2fV2fV16fIiIiIi", "nc", "mai-insts")
407TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_bf8_bf8, "V4fWiWiV4fIiIiIi", "nc", "fp8-insts")
408TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_bf8_fp8, "V4fWiWiV4fIiIiIi", "nc", "fp8-insts")
409TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_fp8_bf8, "V4fWiWiV4fIiIiIi", "nc", "fp8-insts")
410TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_fp8_fp8, "V4fWiWiV4fIiIiIi", "nc", "fp8-insts")
411TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_bf8_bf8, "V16fWiWiV16fIiIiIi", "nc", "fp8-insts")
412TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_bf8_fp8, "V16fWiWiV16fIiIiIi", "nc", "fp8-insts")
413TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_fp8_bf8, "V16fWiWiV16fIiIiIi", "nc", "fp8-insts")
414TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_fp8_fp8, "V16fWiWiV16fIiIiIi", "nc", "fp8-insts")
415TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x32_f16, "V4fV4hV8hV4fiIiIi", "nc", "mai-insts")
416TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x16_f16, "V16fV4hV8hV16fiIiIi", "nc", "mai-insts")
417TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x32_bf16, "V4fV4sV8sV4fiIiIi", "nc", "mai-insts")
418TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x16_bf16, "V16fV4sV8sV16fiIiIi", "nc", "mai-insts")
419TARGET_BUILTIN(__builtin_amdgcn_smfmac_i32_16x16x64_i8, "V4iV2iV4iV4iiIiIi", "nc", "mai-insts")
420TARGET_BUILTIN(__builtin_amdgcn_smfmac_i32_32x32x32_i8, "V16iV2iV4iV16iiIiIi", "nc", "mai-insts")
421TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_bf8_bf8, "V4fV2iV4iV4fiIiIi", "nc", "fp8-insts")
422TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_bf8_fp8, "V4fV2iV4iV4fiIiIi", "nc", "fp8-insts")
423TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_fp8_bf8, "V4fV2iV4iV4fiIiIi", "nc", "fp8-insts")
424TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_fp8_fp8, "V4fV2iV4iV4fiIiIi", "nc", "fp8-insts")
425TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_bf8_bf8, "V16fV2iV4iV16fiIiIi", "nc", "fp8-insts")
426TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_bf8_fp8, "V16fV2iV4iV16fiIiIi", "nc", "fp8-insts")
427TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_fp8_bf8, "V16fV2iV4iV16fiIiIi", "nc", "fp8-insts")
428TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_fp8_fp8, "V16fV2iV4iV16fiIiIi", "nc", "fp8-insts")
429
430TARGET_BUILTIN(__builtin_amdgcn_cvt_f32_bf8, "fiIi", "nc", "fp8-conversion-insts")
431TARGET_BUILTIN(__builtin_amdgcn_cvt_f32_fp8, "fiIi", "nc", "fp8-conversion-insts")
432TARGET_BUILTIN(__builtin_amdgcn_cvt_pk_f32_bf8, "V2fiIb", "nc", "fp8-conversion-insts")
433TARGET_BUILTIN(__builtin_amdgcn_cvt_pk_f32_fp8, "V2fiIb", "nc", "fp8-conversion-insts")
434TARGET_BUILTIN(__builtin_amdgcn_cvt_pk_bf8_f32, "iffiIb", "nc", "fp8-conversion-insts")
435TARGET_BUILTIN(__builtin_amdgcn_cvt_pk_fp8_f32, "iffiIb", "nc", "fp8-conversion-insts")
436TARGET_BUILTIN(__builtin_amdgcn_cvt_sr_bf8_f32, "ifiiIi", "nc", "fp8-conversion-insts")
437TARGET_BUILTIN(__builtin_amdgcn_cvt_sr_fp8_f32, "ifiiIi", "nc", "fp8-conversion-insts")
438
439//===----------------------------------------------------------------------===//
440// GFX950 only builtins.
441//===----------------------------------------------------------------------===//
442TARGET_BUILTIN(__builtin_amdgcn_mfma_scale_f32_16x16x128_f8f6f4, "V4fV8ZiV8ZiV4fIiIiIiiIii", "nc", "gfx950-insts")
443TARGET_BUILTIN(__builtin_amdgcn_mfma_scale_f32_32x32x64_f8f6f4, "V16fV8ZiV8ZiV16fIiIiIiiIii", "nc", "gfx950-insts")
444
445TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_f16, "V4fV8hV8hV4fIiIiIi", "nc", "gfx950-insts")
446TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_bf16, "V4fV8yV8yV4fIiIiIi", "nc", "gfx950-insts")
447TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_f16, "V16fV8hV8hV16fIiIiIi", "nc", "gfx950-insts")
448TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_bf16, "V16fV8yV8yV16fIiIiIi", "nc", "gfx950-insts")
449TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x64_i8, "V4iV4iV4iV4iIiIiIi", "nc", "gfx950-insts")
450TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x32_i8, "V16iV4iV4iV16iIiIiIi", "nc", "gfx950-insts")
451
452TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_f16, "V4fV8hV16hV4fiIiIi", "nc", "gfx950-insts")
453TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_f16, "V16fV8hV16hV16fiIiIi", "nc", "gfx950-insts")
454TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_bf16, "V4fV8yV16yV4fiIiIi", "nc", "gfx950-insts")
455TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_bf16, "V16fV8yV16yV16fiIiIi", "nc", "gfx950-insts")
456TARGET_BUILTIN(__builtin_amdgcn_smfmac_i32_16x16x128_i8, "V4iV4iV8iV4iiIiIi", "nc", "gfx950-insts")
457TARGET_BUILTIN(__builtin_amdgcn_smfmac_i32_32x32x64_i8, "V16iV4iV8iV16iiIiIi", "nc", "gfx950-insts")
458TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x128_bf8_bf8, "V4fV4iV8iV4fiIiIi", "nc", "gfx950-insts")
459TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x128_bf8_fp8, "V4fV4iV8iV4fiIiIi", "nc", "gfx950-insts")
460TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x128_fp8_bf8, "V4fV4iV8iV4fiIiIi", "nc", "gfx950-insts")
461TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x128_fp8_fp8, "V4fV4iV8iV4fiIiIi", "nc", "gfx950-insts")
462TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x64_bf8_bf8, "V16fV4iV8iV16fiIiIi", "nc", "gfx950-insts")
463TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x64_bf8_fp8, "V16fV4iV8iV16fiIiIi", "nc", "gfx950-insts")
464TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x64_fp8_bf8, "V16fV4iV8iV16fiIiIi", "nc", "gfx950-insts")
465TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x64_fp8_fp8, "V16fV4iV8iV16fiIiIi", "nc", "gfx950-insts")
466
467TARGET_BUILTIN(__builtin_amdgcn_permlane16_swap, "V2UiUiUiIbIb", "nc", "permlane16-swap")
468TARGET_BUILTIN(__builtin_amdgcn_permlane32_swap, "V2UiUiUiIbIb", "nc", "permlane32-swap")
469
470TARGET_BUILTIN(__builtin_amdgcn_ds_read_tr4_b64_v2i32, "V2iV2i*3", "nc", "gfx950-insts")
471TARGET_BUILTIN(__builtin_amdgcn_ds_read_tr6_b96_v3i32, "V3iV3i*3", "nc", "gfx950-insts")
472TARGET_BUILTIN(__builtin_amdgcn_ds_read_tr8_b64_v2i32, "V2iV2i*3", "nc", "gfx950-insts")
473TARGET_BUILTIN(__builtin_amdgcn_ds_read_tr16_b64_v4i16, "V4sV4s*3", "nc", "gfx950-insts")
474TARGET_BUILTIN(__builtin_amdgcn_ds_read_tr16_b64_v4f16, "V4hV4h*3", "nc", "gfx950-insts")
475TARGET_BUILTIN(__builtin_amdgcn_ds_read_tr16_b64_v4bf16, "V4yV4y*3", "nc", "gfx950-insts")
476
477TARGET_BUILTIN(__builtin_amdgcn_ashr_pk_i8_i32, "UsUiUiUi", "nc", "ashr-pk-insts")
478TARGET_BUILTIN(__builtin_amdgcn_ashr_pk_u8_i32, "UsUiUiUi", "nc", "ashr-pk-insts")
479
480TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_2xpk16_fp6_f32, "V6UiV16fV16ff", "nc", "gfx950-insts")
481TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_2xpk16_bf6_f32, "V6UiV16fV16ff", "nc", "gfx950-insts")
482
483//===----------------------------------------------------------------------===//
484// GFX12+ only builtins.
485//===----------------------------------------------------------------------===//
486
487TARGET_BUILTIN(__builtin_amdgcn_s_sleep_var, "vUi", "n", "gfx12-insts")
488TARGET_BUILTIN(__builtin_amdgcn_permlane16_var, "UiUiUiUiIbIb", "nc", "gfx12-insts")
489TARGET_BUILTIN(__builtin_amdgcn_permlanex16_var, "UiUiUiUiIbIb", "nc", "gfx12-insts")
490TARGET_BUILTIN(__builtin_amdgcn_s_barrier_signal, "vIi", "n", "gfx12-insts")
491TARGET_BUILTIN(__builtin_amdgcn_s_barrier_signal_var, "vv*i", "n", "gfx12-insts")
492TARGET_BUILTIN(__builtin_amdgcn_s_barrier_wait, "vIs", "n", "gfx12-insts")
493TARGET_BUILTIN(__builtin_amdgcn_s_barrier_signal_isfirst, "bIi", "n", "gfx12-insts")
494TARGET_BUILTIN(__builtin_amdgcn_s_get_barrier_state, "Uii", "n", "gfx12-insts")
495TARGET_BUILTIN(__builtin_amdgcn_s_get_named_barrier_state, "Uiv*", "n", "gfx12-insts")
496TARGET_BUILTIN(__builtin_amdgcn_s_prefetch_data, "vvC*Ui", "nc", "gfx12-insts")
497TARGET_BUILTIN(__builtin_amdgcn_s_buffer_prefetch_data, "vQbIiUi", "nc", "gfx12-insts")
498
499TARGET_BUILTIN(__builtin_amdgcn_global_load_tr_b64_v2i32, "V2iV2i*1", "nc", "gfx12-insts,wavefrontsize32")
500TARGET_BUILTIN(__builtin_amdgcn_global_load_tr_b128_v8i16, "V8sV8s*1", "nc", "gfx12-insts,wavefrontsize32")
501TARGET_BUILTIN(__builtin_amdgcn_global_load_tr_b128_v8f16, "V8hV8h*1", "nc", "gfx12-insts,wavefrontsize32")
502TARGET_BUILTIN(__builtin_amdgcn_global_load_tr_b128_v8bf16, "V8yV8y*1", "nc", "gfx12-insts,wavefrontsize32")
503TARGET_BUILTIN(__builtin_amdgcn_global_load_tr_b64_i32, "ii*1", "nc", "gfx12-insts,wavefrontsize64")
504TARGET_BUILTIN(__builtin_amdgcn_global_load_tr_b128_v4i16, "V4sV4s*1", "nc", "gfx12-insts,wavefrontsize64")
505TARGET_BUILTIN(__builtin_amdgcn_global_load_tr_b128_v4f16, "V4hV4h*1", "nc", "gfx12-insts,wavefrontsize64")
506TARGET_BUILTIN(__builtin_amdgcn_global_load_tr_b128_v4bf16, "V4yV4y*1", "nc", "gfx12-insts,wavefrontsize64")
507
508TARGET_BUILTIN(__builtin_amdgcn_ds_bpermute_fi_b32, "iii", "nc", "gfx12-insts")
509
510// For the following two builtins, the second and third return values of the
511// intrinsics are returned through the last two pointer-type function arguments.
512TARGET_BUILTIN(__builtin_amdgcn_image_bvh8_intersect_ray, "V10UiWUifUcV3fV3fUiV4UiV3f*V3f*", "nc", "gfx12-insts")
513TARGET_BUILTIN(__builtin_amdgcn_image_bvh_dual_intersect_ray, "V10UiWUifUcV3fV3fV2UiV4UiV3f*V3f*", "nc", "gfx12-insts")
514
515TARGET_BUILTIN(__builtin_amdgcn_ds_bvh_stack_push4_pop1_rtn, "V2UiUiUiV4UiIi", "n", "gfx11-insts")
516TARGET_BUILTIN(__builtin_amdgcn_ds_bvh_stack_push8_pop1_rtn, "V2UiUiUiV8UiIi", "n", "gfx12-insts")
517
518// The intrinsic returns {i64, i32}, the builtin returns <2 x i64>.
519// The second return value of the intrinsic is zext'ed.
520TARGET_BUILTIN(__builtin_amdgcn_ds_bvh_stack_push8_pop2_rtn, "V2WUiUiUiV8UiIi", "n", "gfx12-insts")
521
522//===----------------------------------------------------------------------===//
523// WMMA builtins.
524// Postfix w32 indicates the builtin requires wavefront size of 32.
525// Postfix w64 indicates the builtin requires wavefront size of 64.
526//
527// Some of these are very similar to their GFX11 counterparts, but they don't
528// require replication of the A,B matrices, so they use fewer vector elements.
529// Therefore, we add an "_gfx12" suffix to distinguish them from the existing
530// builtins.
531//===----------------------------------------------------------------------===//
532TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_f16_w32_gfx12, "V8fV8hV8hV8f", "nc", "gfx12-insts,wavefrontsize32")
533TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf16_w32_gfx12, "V8fV8sV8sV8f", "nc", "gfx12-insts,wavefrontsize32")
534TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x16_f16_w32_gfx12, "V8hV8hV8hV8h", "nc", "gfx12-insts,wavefrontsize32")
535TARGET_BUILTIN(__builtin_amdgcn_wmma_bf16_16x16x16_bf16_w32_gfx12, "V8sV8sV8sV8s", "nc", "gfx12-insts,wavefrontsize32")
536TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu8_w32_gfx12, "V8iIbV2iIbV2iV8iIb", "nc", "gfx12-insts,wavefrontsize32")
537TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu4_w32_gfx12, "V8iIbiIbiV8iIb", "nc", "gfx12-insts,wavefrontsize32")
538// These are gfx12-only, but for consistency with the other WMMA variants we're
539// keeping the "_gfx12" suffix.
540TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_fp8_fp8_w32_gfx12, "V8fV2iV2iV8f", "nc", "gfx12-insts,wavefrontsize32")
541TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_fp8_bf8_w32_gfx12, "V8fV2iV2iV8f", "nc", "gfx12-insts,wavefrontsize32")
542TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf8_fp8_w32_gfx12, "V8fV2iV2iV8f", "nc", "gfx12-insts,wavefrontsize32")
543TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf8_bf8_w32_gfx12, "V8fV2iV2iV8f", "nc", "gfx12-insts,wavefrontsize32")
544TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x32_iu4_w32_gfx12, "V8iIbV2iIbV2iV8iIb", "nc", "gfx12-insts,wavefrontsize32")
545
546TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_f16_w64_gfx12, "V4fV4hV4hV4f", "nc", "gfx12-insts,wavefrontsize64")
547TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf16_w64_gfx12, "V4fV4sV4sV4f", "nc", "gfx12-insts,wavefrontsize64")
548TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x16_f16_w64_gfx12, "V4hV4hV4hV4h", "nc", "gfx12-insts,wavefrontsize64")
549TARGET_BUILTIN(__builtin_amdgcn_wmma_bf16_16x16x16_bf16_w64_gfx12, "V4sV4sV4sV4s", "nc", "gfx12-insts,wavefrontsize64")
550TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu8_w64_gfx12, "V4iIbiIbiV4iIb", "nc", "gfx12-insts,wavefrontsize64")
551TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu4_w64_gfx12, "V4iIbiIbiV4iIb", "nc", "gfx12-insts,wavefrontsize64")
552// These are gfx12-only, but for consistency with the other WMMA variants we're
553// keeping the "_gfx12" suffix.
554TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_fp8_fp8_w64_gfx12, "V4fiiV4f", "nc", "gfx12-insts,wavefrontsize64")
555TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_fp8_bf8_w64_gfx12, "V4fiiV4f", "nc", "gfx12-insts,wavefrontsize64")
556TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf8_fp8_w64_gfx12, "V4fiiV4f", "nc", "gfx12-insts,wavefrontsize64")
557TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf8_bf8_w64_gfx12, "V4fiiV4f", "nc", "gfx12-insts,wavefrontsize64")
558TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x32_iu4_w64_gfx12, "V4iIbiIbiV4iIb", "nc", "gfx12-insts,wavefrontsize64")
559
560TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_f16_w32, "V8fV8hV16hV8fi", "nc", "gfx12-insts,wavefrontsize32")
561TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_bf16_w32, "V8fV8sV16sV8fi", "nc", "gfx12-insts,wavefrontsize32")
562TARGET_BUILTIN(__builtin_amdgcn_swmmac_f16_16x16x32_f16_w32, "V8hV8hV16hV8hi", "nc", "gfx12-insts,wavefrontsize32")
563TARGET_BUILTIN(__builtin_amdgcn_swmmac_bf16_16x16x32_bf16_w32, "V8sV8sV16sV8si", "nc", "gfx12-insts,wavefrontsize32")
564TARGET_BUILTIN(__builtin_amdgcn_swmmac_i32_16x16x32_iu8_w32, "V8iIbV2iIbV4iV8iiIb", "nc", "gfx12-insts,wavefrontsize32")
565TARGET_BUILTIN(__builtin_amdgcn_swmmac_i32_16x16x32_iu4_w32, "V8iIbiIbV2iV8iiIb", "nc", "gfx12-insts,wavefrontsize32")
566TARGET_BUILTIN(__builtin_amdgcn_swmmac_i32_16x16x64_iu4_w32, "V8iIbV2iIbV4iV8iiIb", "nc", "gfx12-insts,wavefrontsize32")
567TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_fp8_fp8_w32, "V8fV2iV4iV8fi", "nc", "gfx12-insts,wavefrontsize32")
568TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_fp8_bf8_w32, "V8fV2iV4iV8fi", "nc", "gfx12-insts,wavefrontsize32")
569TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_bf8_fp8_w32, "V8fV2iV4iV8fi", "nc", "gfx12-insts,wavefrontsize32")
570TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_bf8_bf8_w32, "V8fV2iV4iV8fi", "nc", "gfx12-insts,wavefrontsize32")
571
572TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_f16_w64, "V4fV4hV8hV4fi", "nc", "gfx12-insts,wavefrontsize64")
573TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_bf16_w64, "V4fV4sV8sV4fi", "nc", "gfx12-insts,wavefrontsize64")
574TARGET_BUILTIN(__builtin_amdgcn_swmmac_f16_16x16x32_f16_w64, "V4hV4hV8hV4hi", "nc", "gfx12-insts,wavefrontsize64")
575TARGET_BUILTIN(__builtin_amdgcn_swmmac_bf16_16x16x32_bf16_w64, "V4sV4sV8sV4si", "nc", "gfx12-insts,wavefrontsize64")
576TARGET_BUILTIN(__builtin_amdgcn_swmmac_i32_16x16x32_iu8_w64, "V4iIbiIbV2iV4iiIb", "nc", "gfx12-insts,wavefrontsize64")
577TARGET_BUILTIN(__builtin_amdgcn_swmmac_i32_16x16x32_iu4_w64, "V4iIbiIbiV4iiIb", "nc", "gfx12-insts,wavefrontsize64")
578TARGET_BUILTIN(__builtin_amdgcn_swmmac_i32_16x16x64_iu4_w64, "V4iIbiIbV2iV4iiIb", "nc", "gfx12-insts,wavefrontsize64")
579TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_fp8_fp8_w64, "V4fiV2iV4fi", "nc", "gfx12-insts,wavefrontsize64")
580TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_fp8_bf8_w64, "V4fiV2iV4fi", "nc", "gfx12-insts,wavefrontsize64")
581TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_bf8_fp8_w64, "V4fiV2iV4fi", "nc", "gfx12-insts,wavefrontsize64")
582TARGET_BUILTIN(__builtin_amdgcn_swmmac_f32_16x16x32_bf8_bf8_w64, "V4fiV2iV4fi", "nc", "gfx12-insts,wavefrontsize64")
583
584TARGET_BUILTIN(__builtin_amdgcn_prng_b32, "UiUi", "nc", "prng-inst")
585TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_fp6_f16, "V6UiV32hf", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
586TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_bf6_f16, "V6UiV32hf", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
587TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_fp6_bf16, "V6UiV32yf", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
588TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_bf6_bf16, "V6UiV32yf", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
589TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_f16_fp8, "V2hV2hifIiIb", "nc", "fp8-cvt-scale-insts")
590TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_f16_bf8, "V2hV2hifIiIb", "nc", "bf8-cvt-scale-insts")
591TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_f32_fp8, "fifIi", "nc", "fp8-cvt-scale-insts")
592TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_f32_bf8, "fifIi", "nc", "bf8-cvt-scale-insts")
593TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_fp8_f32, "V2sV2sfffIb", "nc", "fp8-cvt-scale-insts")
594TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_bf8_f32, "V2sV2sfffIb", "nc", "bf8-cvt-scale-insts")
595TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_f32_fp8, "V2fUifIb", "nc", "fp8-cvt-scale-insts")
596TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_f32_bf8, "V2fUifIb", "nc", "bf8-cvt-scale-insts")
597TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_fp8_f16, "V2sV2sV2hfIb", "nc", "fp8-cvt-scale-insts")
598TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_fp8_bf16, "V2sV2sV2yfIb", "nc", "fp8-cvt-scale-insts")
599TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_bf8_f16, "V2sV2sV2hfIb", "nc", "bf8-cvt-scale-insts")
600TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_bf8_bf16, "V2sV2sV2yfIb", "nc", "bf8-cvt-scale-insts")
601TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_f32_fp4, "V2fUifIi", "nc", "fp4-cvt-scale-insts")
602TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_fp4_f32, "UiUifffIi", "nc", "fp4-cvt-scale-insts")
603TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_f16_fp4, "V2hUifIi", "nc", "fp4-cvt-scale-insts")
604TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_bf16_fp4, "V2yUifIi", "nc", "fp4-cvt-scale-insts")
605TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_f32_fp6, "V32fV6Uif", "nc", "fp6bf6-cvt-scale-insts")
606TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_f32_bf6, "V32fV6Uif", "nc", "fp6bf6-cvt-scale-insts")
607TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_f16_fp6, "V32hV6Uif", "nc", "fp6bf6-cvt-scale-insts")
608TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_bf16_fp6, "V32yV6Uif", "nc", "fp6bf6-cvt-scale-insts")
609TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_f16_bf6, "V32hV6Uif", "nc", "fp6bf6-cvt-scale-insts")
610TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk32_bf16_bf6, "V32yV6Uif", "nc", "fp6bf6-cvt-scale-insts")
611TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_f16_fp8, "V2hUifIb", "nc", "fp8-cvt-scale-insts")
612TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_bf16_fp8, "V2yUifIb", "nc", "fp8-cvt-scale-insts")
613TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_f16_bf8, "V2hUifIb", "nc", "bf8-cvt-scale-insts")
614TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_bf16_bf8, "V2yUifIb", "nc", "bf8-cvt-scale-insts")
615TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_fp4_f16, "UiUiV2hfIi", "nc", "fp4-cvt-scale-insts")
616TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_pk_fp4_bf16, "UiUiV2yfIi", "nc", "fp4-cvt-scale-insts")
617TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_pk_fp4_f16, "UiUiV2hUifIi", "nc", "fp4-cvt-scale-insts")
618TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_pk_fp4_bf16, "UiUiV2yUifIi", "nc", "fp4-cvt-scale-insts")
619TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_pk_fp4_f32, "UiUiV2fUifIi", "nc", "fp4-cvt-scale-insts")
620TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_bf8_bf16, "iiyUifIi", "nc", "bf8-cvt-scale-insts")
621TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_bf8_f16, "iihUifIi", "nc", "bf8-cvt-scale-insts")
622TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_bf8_f32, "iifUifIi", "nc", "bf8-cvt-scale-insts")
623TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_fp8_bf16, "iiyUifIi", "nc", "fp8-cvt-scale-insts")
624TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_fp8_f16, "iihUifIi", "nc", "fp8-cvt-scale-insts")
625TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_fp8_f32, "iifUifIi", "nc", "fp8-cvt-scale-insts")
626
627TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_pk32_bf6_bf16, "V6UiV32yUif", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
628TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_pk32_bf6_f16, "V6UiV32hUif", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
629TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_pk32_bf6_f32, "V6UiV32fUif", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
630TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_pk32_fp6_bf16, "V6UiV32yUif", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
631TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_pk32_fp6_f16, "V6UiV32hUif", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
632TARGET_BUILTIN(__builtin_amdgcn_cvt_scalef32_sr_pk32_fp6_f32, "V6UiV32fUif", "nc", "f16bf16-to-fp6bf6-cvt-scale-insts")
633TARGET_BUILTIN(__builtin_amdgcn_bitop3_b32, "iiiiIUi", "nc", "bitop3-insts")
634TARGET_BUILTIN(__builtin_amdgcn_bitop3_b16, "ssssIUi", "nc", "bitop3-insts")
635
636TARGET_BUILTIN(__builtin_amdgcn_cvt_sr_bf16_f32, "V2yV2yfUiIb", "nc", "f32-to-f16bf16-cvt-sr-insts")
637TARGET_BUILTIN(__builtin_amdgcn_cvt_sr_f16_f32, "V2hV2hfUiIb", "nc", "f32-to-f16bf16-cvt-sr-insts")
638
639#undef BUILTIN
640#undef TARGET_BUILTIN
641

Provided by KDAB

Privacy Policy
Improve your Profiling and Debugging skills
Find out more

source code of clang/include/clang/Basic/BuiltinsAMDGPU.def