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_wave_barrier, "v", "n") 65BUILTIN(__builtin_amdgcn_s_dcache_inv, "v", "n") 66BUILTIN(__builtin_amdgcn_buffer_wbinvl1, "v", "n") 67BUILTIN(__builtin_amdgcn_ds_gws_init, "vUiUi", "n") 68BUILTIN(__builtin_amdgcn_ds_gws_barrier, "vUiUi", "n") 69BUILTIN(__builtin_amdgcn_ds_gws_sema_v, "vUi", "n") 70BUILTIN(__builtin_amdgcn_ds_gws_sema_br, "vUiUi", "n") 71BUILTIN(__builtin_amdgcn_ds_gws_sema_p, "vUi", "n") 72BUILTIN(__builtin_amdgcn_fence, "vUicC*", "n") 73BUILTIN(__builtin_amdgcn_groupstaticsize, "Ui", "n") 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_log_clampf, "ff", "nc") 101BUILTIN(__builtin_amdgcn_ldexp, "ddi", "nc") 102BUILTIN(__builtin_amdgcn_ldexpf, "ffi", "nc") 103BUILTIN(__builtin_amdgcn_frexp_mant, "dd", "nc") 104BUILTIN(__builtin_amdgcn_frexp_mantf, "ff", "nc") 105BUILTIN(__builtin_amdgcn_frexp_exp, "id", "nc") 106BUILTIN(__builtin_amdgcn_frexp_expf, "if", "nc") 107BUILTIN(__builtin_amdgcn_fract, "dd", "nc") 108BUILTIN(__builtin_amdgcn_fractf, "ff", "nc") 109BUILTIN(__builtin_amdgcn_lerp, "UiUiUiUi", "nc") 110BUILTIN(__builtin_amdgcn_class, "bdi", "nc") 111BUILTIN(__builtin_amdgcn_classf, "bfi", "nc") 112BUILTIN(__builtin_amdgcn_cubeid, "ffff", "nc") 113BUILTIN(__builtin_amdgcn_cubesc, "ffff", "nc") 114BUILTIN(__builtin_amdgcn_cubetc, "ffff", "nc") 115BUILTIN(__builtin_amdgcn_cubema, "ffff", "nc") 116BUILTIN(__builtin_amdgcn_s_sleep, "vIi", "n") 117BUILTIN(__builtin_amdgcn_s_incperflevel, "vIi", "n") 118BUILTIN(__builtin_amdgcn_s_decperflevel, "vIi", "n") 119BUILTIN(__builtin_amdgcn_uicmp, "WUiUiUiIi", "nc") 120BUILTIN(__builtin_amdgcn_uicmpl, "WUiWUiWUiIi", "nc") 121BUILTIN(__builtin_amdgcn_sicmp, "WUiiiIi", "nc") 122BUILTIN(__builtin_amdgcn_sicmpl, "WUiWiWiIi", "nc") 123BUILTIN(__builtin_amdgcn_fcmp, "WUiddIi", "nc") 124BUILTIN(__builtin_amdgcn_fcmpf, "WUiffIi", "nc") 125BUILTIN(__builtin_amdgcn_ds_swizzle, "iiIi", "nc") 126BUILTIN(__builtin_amdgcn_ds_permute, "iii", "nc") 127BUILTIN(__builtin_amdgcn_ds_bpermute, "iii", "nc") 128BUILTIN(__builtin_amdgcn_readfirstlane, "ii", "nc") 129BUILTIN(__builtin_amdgcn_readlane, "iii", "nc") 130BUILTIN(__builtin_amdgcn_fmed3f, "ffff", "nc") 131BUILTIN(__builtin_amdgcn_ds_faddf, "ff*3fIiIiIb", "n") 132BUILTIN(__builtin_amdgcn_ds_fminf, "ff*3fIiIiIb", "n") 133BUILTIN(__builtin_amdgcn_ds_fmaxf, "ff*3fIiIiIb", "n") 134BUILTIN(__builtin_amdgcn_ds_append, "ii*3", "n") 135BUILTIN(__builtin_amdgcn_ds_consume, "ii*3", "n") 136BUILTIN(__builtin_amdgcn_alignbit, "UiUiUiUi", "nc") 137BUILTIN(__builtin_amdgcn_alignbyte, "UiUiUiUi", "nc") 138BUILTIN(__builtin_amdgcn_ubfe, "UiUiUiUi", "nc") 139BUILTIN(__builtin_amdgcn_sbfe, "UiUiUiUi", "nc") 140BUILTIN(__builtin_amdgcn_cvt_pkrtz, "E2hff", "nc") 141BUILTIN(__builtin_amdgcn_cvt_pknorm_i16, "E2sff", "nc") 142BUILTIN(__builtin_amdgcn_cvt_pknorm_u16, "E2Usff", "nc") 143BUILTIN(__builtin_amdgcn_cvt_pk_i16, "E2sii", "nc") 144BUILTIN(__builtin_amdgcn_cvt_pk_u16, "E2UsUiUi", "nc") 145BUILTIN(__builtin_amdgcn_cvt_pk_u8_f32, "UifUiUi", "nc") 146BUILTIN(__builtin_amdgcn_sad_u8, "UiUiUiUi", "nc") 147BUILTIN(__builtin_amdgcn_msad_u8, "UiUiUiUi", "nc") 148BUILTIN(__builtin_amdgcn_sad_hi_u8, "UiUiUiUi", "nc") 149BUILTIN(__builtin_amdgcn_sad_u16, "UiUiUiUi", "nc") 150BUILTIN(__builtin_amdgcn_qsad_pk_u16_u8, "WUiWUiUiWUi", "nc") 151BUILTIN(__builtin_amdgcn_mqsad_pk_u16_u8, "WUiWUiUiWUi", "nc") 152BUILTIN(__builtin_amdgcn_mqsad_u32_u8, "V4UiWUiUiV4Ui", "nc") 153 154//===----------------------------------------------------------------------===// 155// CI+ only builtins. 156//===----------------------------------------------------------------------===// 157TARGET_BUILTIN(__builtin_amdgcn_s_dcache_inv_vol, "v", "n", "ci-insts") 158TARGET_BUILTIN(__builtin_amdgcn_buffer_wbinvl1_vol, "v", "n", "ci-insts") 159TARGET_BUILTIN(__builtin_amdgcn_ds_gws_sema_release_all, "vUi", "n", "ci-insts") 160TARGET_BUILTIN(__builtin_amdgcn_is_shared, "bvC*0", "nc", "flat-address-space") 161TARGET_BUILTIN(__builtin_amdgcn_is_private, "bvC*0", "nc", "flat-address-space") 162 163//===----------------------------------------------------------------------===// 164// Interpolation builtins. 165//===----------------------------------------------------------------------===// 166BUILTIN(__builtin_amdgcn_interp_p1_f16, "ffUiUibUi", "nc") 167BUILTIN(__builtin_amdgcn_interp_p2_f16, "hffUiUibUi", "nc") 168BUILTIN(__builtin_amdgcn_interp_p1, "ffUiUiUi", "nc") 169BUILTIN(__builtin_amdgcn_interp_p2, "fffUiUiUi", "nc") 170BUILTIN(__builtin_amdgcn_interp_mov, "fUiUiUiUi", "nc") 171 172//===----------------------------------------------------------------------===// 173// VI+ only builtins. 174//===----------------------------------------------------------------------===// 175 176TARGET_BUILTIN(__builtin_amdgcn_div_fixuph, "hhhh", "nc", "16-bit-insts") 177TARGET_BUILTIN(__builtin_amdgcn_rcph, "hh", "nc", "16-bit-insts") 178TARGET_BUILTIN(__builtin_amdgcn_sqrth, "hh", "nc", "16-bit-insts") 179TARGET_BUILTIN(__builtin_amdgcn_rsqh, "hh", "nc", "16-bit-insts") 180TARGET_BUILTIN(__builtin_amdgcn_sinh, "hh", "nc", "16-bit-insts") 181TARGET_BUILTIN(__builtin_amdgcn_cosh, "hh", "nc", "16-bit-insts") 182TARGET_BUILTIN(__builtin_amdgcn_ldexph, "hhi", "nc", "16-bit-insts") 183TARGET_BUILTIN(__builtin_amdgcn_frexp_manth, "hh", "nc", "16-bit-insts") 184TARGET_BUILTIN(__builtin_amdgcn_frexp_exph, "sh", "nc", "16-bit-insts") 185TARGET_BUILTIN(__builtin_amdgcn_fracth, "hh", "nc", "16-bit-insts") 186TARGET_BUILTIN(__builtin_amdgcn_classh, "bhi", "nc", "16-bit-insts") 187TARGET_BUILTIN(__builtin_amdgcn_s_memrealtime, "WUi", "n", "s-memrealtime") 188TARGET_BUILTIN(__builtin_amdgcn_mov_dpp, "iiIiIiIiIb", "nc", "dpp") 189TARGET_BUILTIN(__builtin_amdgcn_update_dpp, "iiiIiIiIiIb", "nc", "dpp") 190TARGET_BUILTIN(__builtin_amdgcn_s_dcache_wb, "v", "n", "gfx8-insts") 191TARGET_BUILTIN(__builtin_amdgcn_perm, "UiUiUiUi", "nc", "gfx8-insts") 192 193//===----------------------------------------------------------------------===// 194// GFX9+ only builtins. 195//===----------------------------------------------------------------------===// 196 197TARGET_BUILTIN(__builtin_amdgcn_fmed3h, "hhhh", "nc", "gfx9-insts") 198 199//===----------------------------------------------------------------------===// 200// Deep learning builtins. 201//===----------------------------------------------------------------------===// 202 203TARGET_BUILTIN(__builtin_amdgcn_fdot2, "fV2hV2hfIb", "nc", "dot7-insts") 204TARGET_BUILTIN(__builtin_amdgcn_sdot2, "SiV2SsV2SsSiIb", "nc", "dot2-insts") 205TARGET_BUILTIN(__builtin_amdgcn_udot2, "UiV2UsV2UsUiIb", "nc", "dot2-insts") 206TARGET_BUILTIN(__builtin_amdgcn_sdot4, "SiSiSiSiIb", "nc", "dot1-insts") 207TARGET_BUILTIN(__builtin_amdgcn_udot4, "UiUiUiUiIb", "nc", "dot7-insts") 208TARGET_BUILTIN(__builtin_amdgcn_sdot8, "SiSiSiSiIb", "nc", "dot1-insts") 209TARGET_BUILTIN(__builtin_amdgcn_udot8, "UiUiUiUiIb", "nc", "dot7-insts") 210 211//===----------------------------------------------------------------------===// 212// GFX10+ only builtins. 213//===----------------------------------------------------------------------===// 214TARGET_BUILTIN(__builtin_amdgcn_permlane16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts") 215TARGET_BUILTIN(__builtin_amdgcn_permlanex16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts") 216TARGET_BUILTIN(__builtin_amdgcn_mov_dpp8, "UiUiIUi", "nc", "gfx10-insts") 217 218//===----------------------------------------------------------------------===// 219// Raytracing builtins. 220// By default the 1st argument is i32 and the 4/5-th arguments are float4. 221// Postfix l indicates the 1st argument is i64. 222// Postfix h indicates the 4/5-th arguments are half4. 223//===----------------------------------------------------------------------===// 224TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray, "V4UiUifV4fV4fV4fV4Ui", "nc", "gfx10-insts") 225TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_h, "V4UiUifV4fV4hV4hV4Ui", "nc", "gfx10-insts") 226TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_l, "V4UiWUifV4fV4fV4fV4Ui", "nc", "gfx10-insts") 227TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_lh, "V4UiWUifV4fV4hV4hV4Ui", "nc", "gfx10-insts") 228 229//===----------------------------------------------------------------------===// 230// Special builtins. 231//===----------------------------------------------------------------------===// 232BUILTIN(__builtin_amdgcn_read_exec, "WUi", "nc") 233BUILTIN(__builtin_amdgcn_read_exec_lo, "Ui", "nc") 234BUILTIN(__builtin_amdgcn_read_exec_hi, "Ui", "nc") 235 236BUILTIN(__builtin_amdgcn_endpgm, "v", "nr") 237 238//===----------------------------------------------------------------------===// 239// R600-NI only builtins. 240//===----------------------------------------------------------------------===// 241 242BUILTIN(__builtin_r600_implicitarg_ptr, "Uc*7", "nc") 243 244BUILTIN(__builtin_r600_read_tgid_x, "Ui", "nc") 245BUILTIN(__builtin_r600_read_tgid_y, "Ui", "nc") 246BUILTIN(__builtin_r600_read_tgid_z, "Ui", "nc") 247 248BUILTIN(__builtin_r600_read_tidig_x, "Ui", "nc") 249BUILTIN(__builtin_r600_read_tidig_y, "Ui", "nc") 250BUILTIN(__builtin_r600_read_tidig_z, "Ui", "nc") 251 252BUILTIN(__builtin_r600_recipsqrt_ieee, "dd", "nc") 253BUILTIN(__builtin_r600_recipsqrt_ieeef, "ff", "nc") 254 255//===----------------------------------------------------------------------===// 256// MFMA builtins. 257//===----------------------------------------------------------------------===// 258 259TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x1f32, "V32fffV32fIiIiIi", "nc", "mai-insts") 260TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x1f32, "V16fffV16fIiIiIi", "nc", "mai-insts") 261TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x1f32, "V4fffV4fIiIiIi", "nc", "mai-insts") 262TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x2f32, "V16fffV16fIiIiIi", "nc", "mai-insts") 263TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4f32, "V4fffV4fIiIiIi", "nc", "mai-insts") 264TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4f16, "V32fV4hV4hV32fIiIiIi", "nc", "mai-insts") 265TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4f16, "V16fV4hV4hV16fIiIiIi", "nc", "mai-insts") 266TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x4f16, "V4fV4hV4hV4fIiIiIi", "nc", "mai-insts") 267TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x8f16, "V16fV4hV4hV16fIiIiIi", "nc", "mai-insts") 268TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x16f16, "V4fV4hV4hV4fIiIiIi", "nc", "mai-insts") 269TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x4i8, "V32iiiV32iIiIiIi", "nc", "mai-insts") 270TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x4i8, "V16iiiV16iIiIiIi", "nc", "mai-insts") 271TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_4x4x4i8, "V4iiiV4iIiIiIi", "nc", "mai-insts") 272TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x8i8, "V16iiiV16iIiIiIi", "nc", "mai-insts") 273TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x16i8, "V4iiiV4iIiIiIi", "nc", "mai-insts") 274TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x2bf16, "V32fV2sV2sV32fIiIiIi", "nc", "mai-insts") 275TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x2bf16, "V16fV2sV2sV16fIiIiIi", "nc", "mai-insts") 276TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x2bf16, "V4fV2sV2sV4fIiIiIi", "nc", "mai-insts") 277TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4bf16, "V16fV2sV2sV16fIiIiIi", "nc", "mai-insts") 278TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x8bf16, "V4fV2sV2sV4fIiIiIi", "nc", "mai-insts") 279 280TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4bf16_1k, "V32fV4sV4sV32fIiIiIi", "nc", "mai-insts") 281TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4bf16_1k, "V16fV4sV4sV16fIiIiIi", "nc", "mai-insts") 282TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x4bf16_1k, "V4fV4sV4sV4fIiIiIi", "nc", "mai-insts") 283TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x8bf16_1k, "V16fV4sV4sV16fIiIiIi", "nc", "mai-insts") 284TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x16bf16_1k, "V4fV4sV4sV4fIiIiIi", "nc", "mai-insts") 285TARGET_BUILTIN(__builtin_amdgcn_mfma_f64_16x16x4f64, "V4dddV4dIiIiIi", "nc", "mai-insts") 286TARGET_BUILTIN(__builtin_amdgcn_mfma_f64_4x4x4f64, "ddddIiIiIi", "nc", "mai-insts") 287 288#undef BUILTIN 289#undef TARGET_BUILTIN 290