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 23BUILTIN(__builtin_amdgcn_dispatch_ptr, "v*4", "nc") 24BUILTIN(__builtin_amdgcn_kernarg_segment_ptr, "v*4", "nc") 25BUILTIN(__builtin_amdgcn_implicitarg_ptr, "v*4", "nc") 26BUILTIN(__builtin_amdgcn_queue_ptr, "v*4", "nc") 27 28BUILTIN(__builtin_amdgcn_workgroup_id_x, "Ui", "nc") 29BUILTIN(__builtin_amdgcn_workgroup_id_y, "Ui", "nc") 30BUILTIN(__builtin_amdgcn_workgroup_id_z, "Ui", "nc") 31 32BUILTIN(__builtin_amdgcn_workitem_id_x, "Ui", "nc") 33BUILTIN(__builtin_amdgcn_workitem_id_y, "Ui", "nc") 34BUILTIN(__builtin_amdgcn_workitem_id_z, "Ui", "nc") 35 36BUILTIN(__builtin_amdgcn_workgroup_size_x, "Us", "nc") 37BUILTIN(__builtin_amdgcn_workgroup_size_y, "Us", "nc") 38BUILTIN(__builtin_amdgcn_workgroup_size_z, "Us", "nc") 39 40BUILTIN(__builtin_amdgcn_grid_size_x, "Ui", "nc") 41BUILTIN(__builtin_amdgcn_grid_size_y, "Ui", "nc") 42BUILTIN(__builtin_amdgcn_grid_size_z, "Ui", "nc") 43 44BUILTIN(__builtin_amdgcn_mbcnt_hi, "UiUiUi", "nc") 45BUILTIN(__builtin_amdgcn_mbcnt_lo, "UiUiUi", "nc") 46 47TARGET_BUILTIN(__builtin_amdgcn_s_memtime, "LUi", "n", "s-memtime-inst") 48 49//===----------------------------------------------------------------------===// 50// Instruction builtins. 51//===----------------------------------------------------------------------===// 52BUILTIN(__builtin_amdgcn_s_getreg, "UiIi", "n") 53BUILTIN(__builtin_amdgcn_s_setreg, "vIiUi", "n") 54BUILTIN(__builtin_amdgcn_s_getpc, "LUi", "n") 55BUILTIN(__builtin_amdgcn_s_waitcnt, "vIi", "n") 56BUILTIN(__builtin_amdgcn_s_sendmsg, "vIiUi", "n") 57BUILTIN(__builtin_amdgcn_s_sendmsghalt, "vIiUi", "n") 58BUILTIN(__builtin_amdgcn_s_barrier, "v", "n") 59BUILTIN(__builtin_amdgcn_wave_barrier, "v", "n") 60BUILTIN(__builtin_amdgcn_s_dcache_inv, "v", "n") 61BUILTIN(__builtin_amdgcn_buffer_wbinvl1, "v", "n") 62BUILTIN(__builtin_amdgcn_ds_gws_init, "vUiUi", "n") 63BUILTIN(__builtin_amdgcn_ds_gws_barrier, "vUiUi", "n") 64BUILTIN(__builtin_amdgcn_ds_gws_sema_v, "vUi", "n") 65BUILTIN(__builtin_amdgcn_ds_gws_sema_br, "vUiUi", "n") 66BUILTIN(__builtin_amdgcn_ds_gws_sema_p, "vUi", "n") 67BUILTIN(__builtin_amdgcn_fence, "vUicC*", "n") 68BUILTIN(__builtin_amdgcn_groupstaticsize, "Ui", "n") 69 70BUILTIN(__builtin_amdgcn_atomic_inc32, "UZiUZiD*UZiUicC*", "n") 71BUILTIN(__builtin_amdgcn_atomic_inc64, "UWiUWiD*UWiUicC*", "n") 72 73BUILTIN(__builtin_amdgcn_atomic_dec32, "UZiUZiD*UZiUicC*", "n") 74BUILTIN(__builtin_amdgcn_atomic_dec64, "UWiUWiD*UWiUicC*", "n") 75 76// FIXME: Need to disallow constant address space. 77BUILTIN(__builtin_amdgcn_div_scale, "dddbb*", "n") 78BUILTIN(__builtin_amdgcn_div_scalef, "fffbb*", "n") 79BUILTIN(__builtin_amdgcn_div_fmas, "ddddb", "nc") 80BUILTIN(__builtin_amdgcn_div_fmasf, "ffffb", "nc") 81BUILTIN(__builtin_amdgcn_div_fixup, "dddd", "nc") 82BUILTIN(__builtin_amdgcn_div_fixupf, "ffff", "nc") 83BUILTIN(__builtin_amdgcn_trig_preop, "ddi", "nc") 84BUILTIN(__builtin_amdgcn_trig_preopf, "ffi", "nc") 85BUILTIN(__builtin_amdgcn_rcp, "dd", "nc") 86BUILTIN(__builtin_amdgcn_rcpf, "ff", "nc") 87BUILTIN(__builtin_amdgcn_sqrt, "dd", "nc") 88BUILTIN(__builtin_amdgcn_sqrtf, "ff", "nc") 89BUILTIN(__builtin_amdgcn_rsq, "dd", "nc") 90BUILTIN(__builtin_amdgcn_rsqf, "ff", "nc") 91BUILTIN(__builtin_amdgcn_rsq_clamp, "dd", "nc") 92BUILTIN(__builtin_amdgcn_rsq_clampf, "ff", "nc") 93BUILTIN(__builtin_amdgcn_sinf, "ff", "nc") 94BUILTIN(__builtin_amdgcn_cosf, "ff", "nc") 95BUILTIN(__builtin_amdgcn_log_clampf, "ff", "nc") 96BUILTIN(__builtin_amdgcn_ldexp, "ddi", "nc") 97BUILTIN(__builtin_amdgcn_ldexpf, "ffi", "nc") 98BUILTIN(__builtin_amdgcn_frexp_mant, "dd", "nc") 99BUILTIN(__builtin_amdgcn_frexp_mantf, "ff", "nc") 100BUILTIN(__builtin_amdgcn_frexp_exp, "id", "nc") 101BUILTIN(__builtin_amdgcn_frexp_expf, "if", "nc") 102BUILTIN(__builtin_amdgcn_fract, "dd", "nc") 103BUILTIN(__builtin_amdgcn_fractf, "ff", "nc") 104BUILTIN(__builtin_amdgcn_lerp, "UiUiUiUi", "nc") 105BUILTIN(__builtin_amdgcn_class, "bdi", "nc") 106BUILTIN(__builtin_amdgcn_classf, "bfi", "nc") 107BUILTIN(__builtin_amdgcn_cubeid, "ffff", "nc") 108BUILTIN(__builtin_amdgcn_cubesc, "ffff", "nc") 109BUILTIN(__builtin_amdgcn_cubetc, "ffff", "nc") 110BUILTIN(__builtin_amdgcn_cubema, "ffff", "nc") 111BUILTIN(__builtin_amdgcn_s_sleep, "vIi", "n") 112BUILTIN(__builtin_amdgcn_s_incperflevel, "vIi", "n") 113BUILTIN(__builtin_amdgcn_s_decperflevel, "vIi", "n") 114BUILTIN(__builtin_amdgcn_uicmp, "LUiUiUiIi", "nc") 115BUILTIN(__builtin_amdgcn_uicmpl, "LUiLUiLUiIi", "nc") 116BUILTIN(__builtin_amdgcn_sicmp, "LUiiiIi", "nc") 117BUILTIN(__builtin_amdgcn_sicmpl, "LUiLiLiIi", "nc") 118BUILTIN(__builtin_amdgcn_fcmp, "LUiddIi", "nc") 119BUILTIN(__builtin_amdgcn_fcmpf, "LUiffIi", "nc") 120BUILTIN(__builtin_amdgcn_ds_swizzle, "iiIi", "nc") 121BUILTIN(__builtin_amdgcn_ds_permute, "iii", "nc") 122BUILTIN(__builtin_amdgcn_ds_bpermute, "iii", "nc") 123BUILTIN(__builtin_amdgcn_readfirstlane, "ii", "nc") 124BUILTIN(__builtin_amdgcn_readlane, "iii", "nc") 125BUILTIN(__builtin_amdgcn_fmed3f, "ffff", "nc") 126BUILTIN(__builtin_amdgcn_ds_faddf, "ff*3fIiIiIb", "n") 127BUILTIN(__builtin_amdgcn_ds_fminf, "ff*3fIiIiIb", "n") 128BUILTIN(__builtin_amdgcn_ds_fmaxf, "ff*3fIiIiIb", "n") 129BUILTIN(__builtin_amdgcn_ds_append, "ii*3", "n") 130BUILTIN(__builtin_amdgcn_ds_consume, "ii*3", "n") 131BUILTIN(__builtin_amdgcn_alignbit, "UiUiUiUi", "nc") 132BUILTIN(__builtin_amdgcn_alignbyte, "UiUiUiUi", "nc") 133BUILTIN(__builtin_amdgcn_ubfe, "UiUiUiUi", "nc") 134BUILTIN(__builtin_amdgcn_sbfe, "UiUiUiUi", "nc") 135BUILTIN(__builtin_amdgcn_cvt_pkrtz, "E2hff", "nc") 136BUILTIN(__builtin_amdgcn_cvt_pknorm_i16, "E2sff", "nc") 137BUILTIN(__builtin_amdgcn_cvt_pknorm_u16, "E2Usff", "nc") 138BUILTIN(__builtin_amdgcn_cvt_pk_i16, "E2sii", "nc") 139BUILTIN(__builtin_amdgcn_cvt_pk_u16, "E2UsUiUi", "nc") 140BUILTIN(__builtin_amdgcn_cvt_pk_u8_f32, "UifUiUi", "nc") 141BUILTIN(__builtin_amdgcn_sad_u8, "UiUiUiUi", "nc") 142BUILTIN(__builtin_amdgcn_msad_u8, "UiUiUiUi", "nc") 143BUILTIN(__builtin_amdgcn_sad_hi_u8, "UiUiUiUi", "nc") 144BUILTIN(__builtin_amdgcn_sad_u16, "UiUiUiUi", "nc") 145BUILTIN(__builtin_amdgcn_qsad_pk_u16_u8, "LUiLUiUiLUi", "nc") 146BUILTIN(__builtin_amdgcn_mqsad_pk_u16_u8, "LUiLUiUiLUi", "nc") 147BUILTIN(__builtin_amdgcn_mqsad_u32_u8, "V4UiLUiUiV4Ui", "nc") 148 149//===----------------------------------------------------------------------===// 150// CI+ only builtins. 151//===----------------------------------------------------------------------===// 152TARGET_BUILTIN(__builtin_amdgcn_s_dcache_inv_vol, "v", "n", "ci-insts") 153TARGET_BUILTIN(__builtin_amdgcn_buffer_wbinvl1_vol, "v", "n", "ci-insts") 154TARGET_BUILTIN(__builtin_amdgcn_ds_gws_sema_release_all, "vUi", "n", "ci-insts") 155TARGET_BUILTIN(__builtin_amdgcn_is_shared, "bvC*0", "nc", "flat-address-space") 156TARGET_BUILTIN(__builtin_amdgcn_is_private, "bvC*0", "nc", "flat-address-space") 157 158//===----------------------------------------------------------------------===// 159// Interpolation builtins. 160//===----------------------------------------------------------------------===// 161BUILTIN(__builtin_amdgcn_interp_p1_f16, "ffUiUibUi", "nc") 162BUILTIN(__builtin_amdgcn_interp_p2_f16, "hffUiUibUi", "nc") 163BUILTIN(__builtin_amdgcn_interp_p1, "ffUiUiUi", "nc") 164BUILTIN(__builtin_amdgcn_interp_p2, "fffUiUiUi", "nc") 165BUILTIN(__builtin_amdgcn_interp_mov, "fUiUiUiUi", "nc") 166 167//===----------------------------------------------------------------------===// 168// VI+ only builtins. 169//===----------------------------------------------------------------------===// 170 171TARGET_BUILTIN(__builtin_amdgcn_div_fixuph, "hhhh", "nc", "16-bit-insts") 172TARGET_BUILTIN(__builtin_amdgcn_rcph, "hh", "nc", "16-bit-insts") 173TARGET_BUILTIN(__builtin_amdgcn_sqrth, "hh", "nc", "16-bit-insts") 174TARGET_BUILTIN(__builtin_amdgcn_rsqh, "hh", "nc", "16-bit-insts") 175TARGET_BUILTIN(__builtin_amdgcn_sinh, "hh", "nc", "16-bit-insts") 176TARGET_BUILTIN(__builtin_amdgcn_cosh, "hh", "nc", "16-bit-insts") 177TARGET_BUILTIN(__builtin_amdgcn_ldexph, "hhi", "nc", "16-bit-insts") 178TARGET_BUILTIN(__builtin_amdgcn_frexp_manth, "hh", "nc", "16-bit-insts") 179TARGET_BUILTIN(__builtin_amdgcn_frexp_exph, "sh", "nc", "16-bit-insts") 180TARGET_BUILTIN(__builtin_amdgcn_fracth, "hh", "nc", "16-bit-insts") 181TARGET_BUILTIN(__builtin_amdgcn_classh, "bhi", "nc", "16-bit-insts") 182TARGET_BUILTIN(__builtin_amdgcn_s_memrealtime, "LUi", "n", "s-memrealtime") 183TARGET_BUILTIN(__builtin_amdgcn_mov_dpp, "iiIiIiIiIb", "nc", "dpp") 184TARGET_BUILTIN(__builtin_amdgcn_update_dpp, "iiiIiIiIiIb", "nc", "dpp") 185TARGET_BUILTIN(__builtin_amdgcn_s_dcache_wb, "v", "n", "gfx8-insts") 186TARGET_BUILTIN(__builtin_amdgcn_perm, "UiUiUiUi", "nc", "gfx8-insts") 187 188//===----------------------------------------------------------------------===// 189// GFX9+ only builtins. 190//===----------------------------------------------------------------------===// 191 192TARGET_BUILTIN(__builtin_amdgcn_fmed3h, "hhhh", "nc", "gfx9-insts") 193 194//===----------------------------------------------------------------------===// 195// Deep learning builtins. 196//===----------------------------------------------------------------------===// 197 198TARGET_BUILTIN(__builtin_amdgcn_fdot2, "fV2hV2hfIb", "nc", "dot7-insts") 199TARGET_BUILTIN(__builtin_amdgcn_sdot2, "SiV2SsV2SsSiIb", "nc", "dot2-insts") 200TARGET_BUILTIN(__builtin_amdgcn_udot2, "UiV2UsV2UsUiIb", "nc", "dot2-insts") 201TARGET_BUILTIN(__builtin_amdgcn_sdot4, "SiSiSiSiIb", "nc", "dot1-insts") 202TARGET_BUILTIN(__builtin_amdgcn_udot4, "UiUiUiUiIb", "nc", "dot7-insts") 203TARGET_BUILTIN(__builtin_amdgcn_sdot8, "SiSiSiSiIb", "nc", "dot1-insts") 204TARGET_BUILTIN(__builtin_amdgcn_udot8, "UiUiUiUiIb", "nc", "dot7-insts") 205 206//===----------------------------------------------------------------------===// 207// GFX10+ only builtins. 208//===----------------------------------------------------------------------===// 209TARGET_BUILTIN(__builtin_amdgcn_permlane16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts") 210TARGET_BUILTIN(__builtin_amdgcn_permlanex16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts") 211TARGET_BUILTIN(__builtin_amdgcn_mov_dpp8, "UiUiIUi", "nc", "gfx10-insts") 212 213//===----------------------------------------------------------------------===// 214// Special builtins. 215//===----------------------------------------------------------------------===// 216BUILTIN(__builtin_amdgcn_read_exec, "LUi", "nc") 217BUILTIN(__builtin_amdgcn_read_exec_lo, "Ui", "nc") 218BUILTIN(__builtin_amdgcn_read_exec_hi, "Ui", "nc") 219 220BUILTIN(__builtin_amdgcn_endpgm, "v", "nr") 221 222//===----------------------------------------------------------------------===// 223// R600-NI only builtins. 224//===----------------------------------------------------------------------===// 225 226BUILTIN(__builtin_r600_implicitarg_ptr, "Uc*7", "nc") 227 228BUILTIN(__builtin_r600_read_tgid_x, "Ui", "nc") 229BUILTIN(__builtin_r600_read_tgid_y, "Ui", "nc") 230BUILTIN(__builtin_r600_read_tgid_z, "Ui", "nc") 231 232BUILTIN(__builtin_r600_read_tidig_x, "Ui", "nc") 233BUILTIN(__builtin_r600_read_tidig_y, "Ui", "nc") 234BUILTIN(__builtin_r600_read_tidig_z, "Ui", "nc") 235 236BUILTIN(__builtin_r600_recipsqrt_ieee, "dd", "nc") 237BUILTIN(__builtin_r600_recipsqrt_ieeef, "ff", "nc") 238 239//===----------------------------------------------------------------------===// 240// MFMA builtins. 241//===----------------------------------------------------------------------===// 242 243TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x1f32, "V32fffV32fIiIiIi", "nc", "mai-insts") 244TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x1f32, "V16fffV16fIiIiIi", "nc", "mai-insts") 245TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x1f32, "V4fffV4fIiIiIi", "nc", "mai-insts") 246TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x2f32, "V16fffV16fIiIiIi", "nc", "mai-insts") 247TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4f32, "V4fffV4fIiIiIi", "nc", "mai-insts") 248TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4f16, "V32fV4hV4hV32fIiIiIi", "nc", "mai-insts") 249TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4f16, "V16fV4hV4hV16fIiIiIi", "nc", "mai-insts") 250TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x4f16, "V4fV4hV4hV4fIiIiIi", "nc", "mai-insts") 251TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x8f16, "V16fV4hV4hV16fIiIiIi", "nc", "mai-insts") 252TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x16f16, "V4fV4hV4hV4fIiIiIi", "nc", "mai-insts") 253TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x4i8, "V32iiiV32iIiIiIi", "nc", "mai-insts") 254TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x4i8, "V16iiiV16iIiIiIi", "nc", "mai-insts") 255TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_4x4x4i8, "V4iiiV4iIiIiIi", "nc", "mai-insts") 256TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x8i8, "V16iiiV16iIiIiIi", "nc", "mai-insts") 257TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x16i8, "V4iiiV4iIiIiIi", "nc", "mai-insts") 258TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x2bf16, "V32fV2sV2sV32fIiIiIi", "nc", "mai-insts") 259TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x2bf16, "V16fV2sV2sV16fIiIiIi", "nc", "mai-insts") 260TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x2bf16, "V4fV2sV2sV4fIiIiIi", "nc", "mai-insts") 261TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4bf16, "V16fV2sV2sV16fIiIiIi", "nc", "mai-insts") 262TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x8bf16, "V4fV2sV2sV4fIiIiIi", "nc", "mai-insts") 263 264TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4bf16_1k, "V32fV4sV4sV32fIiIiIi", "nc", "mai-insts") 265TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4bf16_1k, "V16fV4sV4sV16fIiIiIi", "nc", "mai-insts") 266TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x4bf16_1k, "V4fV4sV4sV4fIiIiIi", "nc", "mai-insts") 267TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x8bf16_1k, "V16fV4sV4sV16fIiIiIi", "nc", "mai-insts") 268TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x16bf16_1k, "V4fV4sV4sV4fIiIiIi", "nc", "mai-insts") 269TARGET_BUILTIN(__builtin_amdgcn_mfma_f64_16x16x4f64, "V4dddV4dIiIiIi", "nc", "mai-insts") 270TARGET_BUILTIN(__builtin_amdgcn_mfma_f64_4x4x4f64, "ddddIiIiIi", "nc", "mai-insts") 271 272#undef BUILTIN 273#undef TARGET_BUILTIN 274