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_sched_barrier, "vIi", "n") 66BUILTIN(__builtin_amdgcn_s_dcache_inv, "v", "n") 67BUILTIN(__builtin_amdgcn_buffer_wbinvl1, "v", "n") 68BUILTIN(__builtin_amdgcn_ds_gws_init, "vUiUi", "n") 69BUILTIN(__builtin_amdgcn_ds_gws_barrier, "vUiUi", "n") 70BUILTIN(__builtin_amdgcn_ds_gws_sema_v, "vUi", "n") 71BUILTIN(__builtin_amdgcn_ds_gws_sema_br, "vUiUi", "n") 72BUILTIN(__builtin_amdgcn_ds_gws_sema_p, "vUi", "n") 73BUILTIN(__builtin_amdgcn_fence, "vUicC*", "n") 74BUILTIN(__builtin_amdgcn_groupstaticsize, "Ui", "n") 75 76BUILTIN(__builtin_amdgcn_atomic_inc32, "UZiUZiD*UZiUicC*", "n") 77BUILTIN(__builtin_amdgcn_atomic_inc64, "UWiUWiD*UWiUicC*", "n") 78 79BUILTIN(__builtin_amdgcn_atomic_dec32, "UZiUZiD*UZiUicC*", "n") 80BUILTIN(__builtin_amdgcn_atomic_dec64, "UWiUWiD*UWiUicC*", "n") 81 82// FIXME: Need to disallow constant address space. 83BUILTIN(__builtin_amdgcn_div_scale, "dddbb*", "n") 84BUILTIN(__builtin_amdgcn_div_scalef, "fffbb*", "n") 85BUILTIN(__builtin_amdgcn_div_fmas, "ddddb", "nc") 86BUILTIN(__builtin_amdgcn_div_fmasf, "ffffb", "nc") 87BUILTIN(__builtin_amdgcn_div_fixup, "dddd", "nc") 88BUILTIN(__builtin_amdgcn_div_fixupf, "ffff", "nc") 89BUILTIN(__builtin_amdgcn_trig_preop, "ddi", "nc") 90BUILTIN(__builtin_amdgcn_trig_preopf, "ffi", "nc") 91BUILTIN(__builtin_amdgcn_rcp, "dd", "nc") 92BUILTIN(__builtin_amdgcn_rcpf, "ff", "nc") 93BUILTIN(__builtin_amdgcn_sqrt, "dd", "nc") 94BUILTIN(__builtin_amdgcn_sqrtf, "ff", "nc") 95BUILTIN(__builtin_amdgcn_rsq, "dd", "nc") 96BUILTIN(__builtin_amdgcn_rsqf, "ff", "nc") 97BUILTIN(__builtin_amdgcn_rsq_clamp, "dd", "nc") 98BUILTIN(__builtin_amdgcn_rsq_clampf, "ff", "nc") 99BUILTIN(__builtin_amdgcn_sinf, "ff", "nc") 100BUILTIN(__builtin_amdgcn_cosf, "ff", "nc") 101BUILTIN(__builtin_amdgcn_log_clampf, "ff", "nc") 102BUILTIN(__builtin_amdgcn_ldexp, "ddi", "nc") 103BUILTIN(__builtin_amdgcn_ldexpf, "ffi", "nc") 104BUILTIN(__builtin_amdgcn_frexp_mant, "dd", "nc") 105BUILTIN(__builtin_amdgcn_frexp_mantf, "ff", "nc") 106BUILTIN(__builtin_amdgcn_frexp_exp, "id", "nc") 107BUILTIN(__builtin_amdgcn_frexp_expf, "if", "nc") 108BUILTIN(__builtin_amdgcn_fract, "dd", "nc") 109BUILTIN(__builtin_amdgcn_fractf, "ff", "nc") 110BUILTIN(__builtin_amdgcn_lerp, "UiUiUiUi", "nc") 111BUILTIN(__builtin_amdgcn_class, "bdi", "nc") 112BUILTIN(__builtin_amdgcn_classf, "bfi", "nc") 113BUILTIN(__builtin_amdgcn_cubeid, "ffff", "nc") 114BUILTIN(__builtin_amdgcn_cubesc, "ffff", "nc") 115BUILTIN(__builtin_amdgcn_cubetc, "ffff", "nc") 116BUILTIN(__builtin_amdgcn_cubema, "ffff", "nc") 117BUILTIN(__builtin_amdgcn_s_sleep, "vIi", "n") 118BUILTIN(__builtin_amdgcn_s_incperflevel, "vIi", "n") 119BUILTIN(__builtin_amdgcn_s_decperflevel, "vIi", "n") 120BUILTIN(__builtin_amdgcn_s_setprio, "vIs", "n") 121BUILTIN(__builtin_amdgcn_uicmp, "WUiUiUiIi", "nc") 122BUILTIN(__builtin_amdgcn_uicmpl, "WUiWUiWUiIi", "nc") 123BUILTIN(__builtin_amdgcn_sicmp, "WUiiiIi", "nc") 124BUILTIN(__builtin_amdgcn_sicmpl, "WUiWiWiIi", "nc") 125BUILTIN(__builtin_amdgcn_fcmp, "WUiddIi", "nc") 126BUILTIN(__builtin_amdgcn_fcmpf, "WUiffIi", "nc") 127BUILTIN(__builtin_amdgcn_ds_swizzle, "iiIi", "nc") 128BUILTIN(__builtin_amdgcn_ds_permute, "iii", "nc") 129BUILTIN(__builtin_amdgcn_ds_bpermute, "iii", "nc") 130BUILTIN(__builtin_amdgcn_readfirstlane, "ii", "nc") 131BUILTIN(__builtin_amdgcn_readlane, "iii", "nc") 132BUILTIN(__builtin_amdgcn_fmed3f, "ffff", "nc") 133BUILTIN(__builtin_amdgcn_ds_faddf, "ff*3fIiIiIb", "n") 134BUILTIN(__builtin_amdgcn_ds_fminf, "ff*3fIiIiIb", "n") 135BUILTIN(__builtin_amdgcn_ds_fmaxf, "ff*3fIiIiIb", "n") 136BUILTIN(__builtin_amdgcn_ds_append, "ii*3", "n") 137BUILTIN(__builtin_amdgcn_ds_consume, "ii*3", "n") 138BUILTIN(__builtin_amdgcn_alignbit, "UiUiUiUi", "nc") 139BUILTIN(__builtin_amdgcn_alignbyte, "UiUiUiUi", "nc") 140BUILTIN(__builtin_amdgcn_ubfe, "UiUiUiUi", "nc") 141BUILTIN(__builtin_amdgcn_sbfe, "UiUiUiUi", "nc") 142BUILTIN(__builtin_amdgcn_cvt_pkrtz, "E2hff", "nc") 143BUILTIN(__builtin_amdgcn_cvt_pknorm_i16, "E2sff", "nc") 144BUILTIN(__builtin_amdgcn_cvt_pknorm_u16, "E2Usff", "nc") 145BUILTIN(__builtin_amdgcn_cvt_pk_i16, "E2sii", "nc") 146BUILTIN(__builtin_amdgcn_cvt_pk_u16, "E2UsUiUi", "nc") 147BUILTIN(__builtin_amdgcn_cvt_pk_u8_f32, "UifUiUi", "nc") 148BUILTIN(__builtin_amdgcn_sad_u8, "UiUiUiUi", "nc") 149BUILTIN(__builtin_amdgcn_msad_u8, "UiUiUiUi", "nc") 150BUILTIN(__builtin_amdgcn_sad_hi_u8, "UiUiUiUi", "nc") 151BUILTIN(__builtin_amdgcn_sad_u16, "UiUiUiUi", "nc") 152BUILTIN(__builtin_amdgcn_qsad_pk_u16_u8, "WUiWUiUiWUi", "nc") 153BUILTIN(__builtin_amdgcn_mqsad_pk_u16_u8, "WUiWUiUiWUi", "nc") 154BUILTIN(__builtin_amdgcn_mqsad_u32_u8, "V4UiWUiUiV4Ui", "nc") 155 156//===----------------------------------------------------------------------===// 157// CI+ only builtins. 158//===----------------------------------------------------------------------===// 159TARGET_BUILTIN(__builtin_amdgcn_s_dcache_inv_vol, "v", "n", "ci-insts") 160TARGET_BUILTIN(__builtin_amdgcn_buffer_wbinvl1_vol, "v", "n", "ci-insts") 161TARGET_BUILTIN(__builtin_amdgcn_ds_gws_sema_release_all, "vUi", "n", "ci-insts") 162TARGET_BUILTIN(__builtin_amdgcn_is_shared, "bvC*0", "nc", "flat-address-space") 163TARGET_BUILTIN(__builtin_amdgcn_is_private, "bvC*0", "nc", "flat-address-space") 164 165//===----------------------------------------------------------------------===// 166// Interpolation builtins. 167//===----------------------------------------------------------------------===// 168BUILTIN(__builtin_amdgcn_interp_p1_f16, "ffUiUibUi", "nc") 169BUILTIN(__builtin_amdgcn_interp_p2_f16, "hffUiUibUi", "nc") 170BUILTIN(__builtin_amdgcn_interp_p1, "ffUiUiUi", "nc") 171BUILTIN(__builtin_amdgcn_interp_p2, "fffUiUiUi", "nc") 172BUILTIN(__builtin_amdgcn_interp_mov, "fUiUiUiUi", "nc") 173 174//===----------------------------------------------------------------------===// 175// VI+ only builtins. 176//===----------------------------------------------------------------------===// 177 178TARGET_BUILTIN(__builtin_amdgcn_div_fixuph, "hhhh", "nc", "16-bit-insts") 179TARGET_BUILTIN(__builtin_amdgcn_rcph, "hh", "nc", "16-bit-insts") 180TARGET_BUILTIN(__builtin_amdgcn_sqrth, "hh", "nc", "16-bit-insts") 181TARGET_BUILTIN(__builtin_amdgcn_rsqh, "hh", "nc", "16-bit-insts") 182TARGET_BUILTIN(__builtin_amdgcn_sinh, "hh", "nc", "16-bit-insts") 183TARGET_BUILTIN(__builtin_amdgcn_cosh, "hh", "nc", "16-bit-insts") 184TARGET_BUILTIN(__builtin_amdgcn_ldexph, "hhi", "nc", "16-bit-insts") 185TARGET_BUILTIN(__builtin_amdgcn_frexp_manth, "hh", "nc", "16-bit-insts") 186TARGET_BUILTIN(__builtin_amdgcn_frexp_exph, "sh", "nc", "16-bit-insts") 187TARGET_BUILTIN(__builtin_amdgcn_fracth, "hh", "nc", "16-bit-insts") 188TARGET_BUILTIN(__builtin_amdgcn_classh, "bhi", "nc", "16-bit-insts") 189TARGET_BUILTIN(__builtin_amdgcn_s_memrealtime, "WUi", "n", "s-memrealtime") 190TARGET_BUILTIN(__builtin_amdgcn_mov_dpp, "iiIiIiIiIb", "nc", "dpp") 191TARGET_BUILTIN(__builtin_amdgcn_update_dpp, "iiiIiIiIiIb", "nc", "dpp") 192TARGET_BUILTIN(__builtin_amdgcn_s_dcache_wb, "v", "n", "gfx8-insts") 193TARGET_BUILTIN(__builtin_amdgcn_perm, "UiUiUiUi", "nc", "gfx8-insts") 194 195//===----------------------------------------------------------------------===// 196// GFX9+ only builtins. 197//===----------------------------------------------------------------------===// 198 199TARGET_BUILTIN(__builtin_amdgcn_fmed3h, "hhhh", "nc", "gfx9-insts") 200 201TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_f64, "dd*1d", "t", "gfx90a-insts") 202TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_f32, "ff*1f", "t", "gfx90a-insts") 203TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_v2f16, "V2hV2h*1V2h", "t", "gfx90a-insts") 204TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fmin_f64, "dd*1d", "t", "gfx90a-insts") 205TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fmax_f64, "dd*1d", "t", "gfx90a-insts") 206 207TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_f64, "dd*0d", "t", "gfx90a-insts") 208TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fmin_f64, "dd*0d", "t", "gfx90a-insts") 209TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fmax_f64, "dd*0d", "t", "gfx90a-insts") 210 211TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_f64, "dd*3d", "t", "gfx90a-insts") 212TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_f32, "ff*3f", "t", "gfx8-insts") 213 214TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_f32, "ff*0f", "t", "gfx940-insts") 215TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_v2f16, "V2hV2h*0V2h", "t", "gfx940-insts") 216TARGET_BUILTIN(__builtin_amdgcn_flat_atomic_fadd_v2bf16, "V2sV2s*0V2s", "t", "gfx940-insts") 217TARGET_BUILTIN(__builtin_amdgcn_global_atomic_fadd_v2bf16, "V2sV2s*1V2s", "t", "gfx940-insts") 218TARGET_BUILTIN(__builtin_amdgcn_ds_atomic_fadd_v2bf16, "V2sV2s*3V2s", "t", "gfx940-insts") 219 220//===----------------------------------------------------------------------===// 221// Deep learning builtins. 222//===----------------------------------------------------------------------===// 223 224TARGET_BUILTIN(__builtin_amdgcn_fdot2, "fV2hV2hfIb", "nc", "dot7-insts") 225TARGET_BUILTIN(__builtin_amdgcn_fdot2_f16_f16, "hV2hV2hh", "nc", "dot8-insts") 226TARGET_BUILTIN(__builtin_amdgcn_fdot2_bf16_bf16, "sV2sV2ss", "nc", "dot8-insts") 227TARGET_BUILTIN(__builtin_amdgcn_fdot2_f32_bf16, "fV2sV2sfIb", "nc", "dot8-insts") 228TARGET_BUILTIN(__builtin_amdgcn_sdot2, "SiV2SsV2SsSiIb", "nc", "dot2-insts") 229TARGET_BUILTIN(__builtin_amdgcn_udot2, "UiV2UsV2UsUiIb", "nc", "dot2-insts") 230TARGET_BUILTIN(__builtin_amdgcn_sdot4, "SiSiSiSiIb", "nc", "dot1-insts") 231TARGET_BUILTIN(__builtin_amdgcn_udot4, "UiUiUiUiIb", "nc", "dot7-insts") 232TARGET_BUILTIN(__builtin_amdgcn_sudot4, "iIbiIbiiIb", "nc", "dot8-insts") 233TARGET_BUILTIN(__builtin_amdgcn_sdot8, "SiSiSiSiIb", "nc", "dot1-insts") 234TARGET_BUILTIN(__builtin_amdgcn_udot8, "UiUiUiUiIb", "nc", "dot7-insts") 235TARGET_BUILTIN(__builtin_amdgcn_sudot8, "iIbiIbiiIb", "nc", "dot8-insts") 236 237//===----------------------------------------------------------------------===// 238// GFX10+ only builtins. 239//===----------------------------------------------------------------------===// 240TARGET_BUILTIN(__builtin_amdgcn_permlane16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts") 241TARGET_BUILTIN(__builtin_amdgcn_permlanex16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts") 242TARGET_BUILTIN(__builtin_amdgcn_mov_dpp8, "UiUiIUi", "nc", "gfx10-insts") 243 244//===----------------------------------------------------------------------===// 245// Raytracing builtins. 246// By default the 1st argument is i32 and the 4/5-th arguments are float4. 247// Postfix l indicates the 1st argument is i64. 248// Postfix h indicates the 4/5-th arguments are half4. 249//===----------------------------------------------------------------------===// 250TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray, "V4UiUifV4fV4fV4fV4Ui", "nc", "gfx10-insts") 251TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_h, "V4UiUifV4fV4hV4hV4Ui", "nc", "gfx10-insts") 252TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_l, "V4UiWUifV4fV4fV4fV4Ui", "nc", "gfx10-insts") 253TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_lh, "V4UiWUifV4fV4hV4hV4Ui", "nc", "gfx10-insts") 254 255 256//===----------------------------------------------------------------------===// 257// GFX11+ only builtins. 258//===----------------------------------------------------------------------===// 259 260//===----------------------------------------------------------------------===// 261// WMMA builtins. 262// Postfix w32 indicates the builtin requires wavefront size of 32. 263// Postfix w64 indicates the builtin requires wavefront size of 64. 264//===----------------------------------------------------------------------===// 265TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_f16_w32, "V8fV16hV16hV8f", "nc", "gfx11-insts") 266TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf16_w32, "V8fV16sV16sV8f", "nc", "gfx11-insts") 267TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x16_f16_w32, "V16hV16hV16hV16hIb", "nc", "gfx11-insts") 268TARGET_BUILTIN(__builtin_amdgcn_wmma_bf16_16x16x16_bf16_w32, "V16sV16sV16sV16sIb", "nc", "gfx11-insts") 269TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu8_w32, "V8iIbV4iIbV4iV8iIb", "nc", "gfx11-insts") 270TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu4_w32, "V8iIbV2iIbV2iV8iIb", "nc", "gfx11-insts") 271 272TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_f16_w64, "V4fV16hV16hV4f", "nc", "gfx11-insts") 273TARGET_BUILTIN(__builtin_amdgcn_wmma_f32_16x16x16_bf16_w64, "V4fV16sV16sV4f", "nc", "gfx11-insts") 274TARGET_BUILTIN(__builtin_amdgcn_wmma_f16_16x16x16_f16_w64, "V8hV16hV16hV8hIb", "nc", "gfx11-insts") 275TARGET_BUILTIN(__builtin_amdgcn_wmma_bf16_16x16x16_bf16_w64, "V8sV16sV16sV8sIb", "nc", "gfx11-insts") 276TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu8_w64, "V4iIbV4iIbV4iV4iIb", "nc", "gfx11-insts") 277TARGET_BUILTIN(__builtin_amdgcn_wmma_i32_16x16x16_iu4_w64, "V4iIbV2iIbV2iV4iIb", "nc", "gfx11-insts") 278 279//===----------------------------------------------------------------------===// 280// Special builtins. 281//===----------------------------------------------------------------------===// 282BUILTIN(__builtin_amdgcn_read_exec, "WUi", "nc") 283BUILTIN(__builtin_amdgcn_read_exec_lo, "Ui", "nc") 284BUILTIN(__builtin_amdgcn_read_exec_hi, "Ui", "nc") 285 286BUILTIN(__builtin_amdgcn_endpgm, "v", "nr") 287 288//===----------------------------------------------------------------------===// 289// R600-NI only builtins. 290//===----------------------------------------------------------------------===// 291 292BUILTIN(__builtin_r600_implicitarg_ptr, "Uc*7", "nc") 293 294BUILTIN(__builtin_r600_read_tgid_x, "Ui", "nc") 295BUILTIN(__builtin_r600_read_tgid_y, "Ui", "nc") 296BUILTIN(__builtin_r600_read_tgid_z, "Ui", "nc") 297 298BUILTIN(__builtin_r600_read_tidig_x, "Ui", "nc") 299BUILTIN(__builtin_r600_read_tidig_y, "Ui", "nc") 300BUILTIN(__builtin_r600_read_tidig_z, "Ui", "nc") 301 302BUILTIN(__builtin_r600_recipsqrt_ieee, "dd", "nc") 303BUILTIN(__builtin_r600_recipsqrt_ieeef, "ff", "nc") 304 305//===----------------------------------------------------------------------===// 306// MFMA builtins. 307//===----------------------------------------------------------------------===// 308 309TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x1f32, "V32fffV32fIiIiIi", "nc", "mai-insts") 310TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x1f32, "V16fffV16fIiIiIi", "nc", "mai-insts") 311TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x1f32, "V4fffV4fIiIiIi", "nc", "mai-insts") 312TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x2f32, "V16fffV16fIiIiIi", "nc", "mai-insts") 313TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4f32, "V4fffV4fIiIiIi", "nc", "mai-insts") 314TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4f16, "V32fV4hV4hV32fIiIiIi", "nc", "mai-insts") 315TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4f16, "V16fV4hV4hV16fIiIiIi", "nc", "mai-insts") 316TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x4f16, "V4fV4hV4hV4fIiIiIi", "nc", "mai-insts") 317TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x8f16, "V16fV4hV4hV16fIiIiIi", "nc", "mai-insts") 318TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x16f16, "V4fV4hV4hV4fIiIiIi", "nc", "mai-insts") 319TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x4i8, "V32iiiV32iIiIiIi", "nc", "mai-insts") 320TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x4i8, "V16iiiV16iIiIiIi", "nc", "mai-insts") 321TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_4x4x4i8, "V4iiiV4iIiIiIi", "nc", "mai-insts") 322TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x8i8, "V16iiiV16iIiIiIi", "nc", "mai-insts") 323TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x16i8, "V4iiiV4iIiIiIi", "nc", "mai-insts") 324TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x2bf16, "V32fV2sV2sV32fIiIiIi", "nc", "mai-insts") 325TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x2bf16, "V16fV2sV2sV16fIiIiIi", "nc", "mai-insts") 326TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x2bf16, "V4fV2sV2sV4fIiIiIi", "nc", "mai-insts") 327TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4bf16, "V16fV2sV2sV16fIiIiIi", "nc", "mai-insts") 328TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x8bf16, "V4fV2sV2sV4fIiIiIi", "nc", "mai-insts") 329 330TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4bf16_1k, "V32fV4sV4sV32fIiIiIi", "nc", "mai-insts") 331TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x4bf16_1k, "V16fV4sV4sV16fIiIiIi", "nc", "mai-insts") 332TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_4x4x4bf16_1k, "V4fV4sV4sV4fIiIiIi", "nc", "mai-insts") 333TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x8bf16_1k, "V16fV4sV4sV16fIiIiIi", "nc", "mai-insts") 334TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x16bf16_1k, "V4fV4sV4sV4fIiIiIi", "nc", "mai-insts") 335TARGET_BUILTIN(__builtin_amdgcn_mfma_f64_16x16x4f64, "V4dddV4dIiIiIi", "nc", "mai-insts") 336TARGET_BUILTIN(__builtin_amdgcn_mfma_f64_4x4x4f64, "ddddIiIiIi", "nc", "mai-insts") 337 338TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x32_i8, "V4iWiWiV4iIiIiIi", "nc", "mai-insts") 339TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x16_i8, "V16iWiWiV16iIiIiIi", "nc", "mai-insts") 340TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x8_xf32, "V4fV2fV2fV4fIiIiIi", "nc", "mai-insts") 341TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x4_xf32, "V16fV2fV2fV16fIiIiIi", "nc", "mai-insts") 342TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_bf8_bf8, "V4fWiWiV4fIiIiIi", "nc", "fp8-insts") 343TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_bf8_fp8, "V4fWiWiV4fIiIiIi", "nc", "fp8-insts") 344TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_fp8_bf8, "V4fWiWiV4fIiIiIi", "nc", "fp8-insts") 345TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_16x16x32_fp8_fp8, "V4fWiWiV4fIiIiIi", "nc", "fp8-insts") 346TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_bf8_bf8, "V16fWiWiV16fIiIiIi", "nc", "fp8-insts") 347TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_bf8_fp8, "V16fWiWiV16fIiIiIi", "nc", "fp8-insts") 348TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_fp8_bf8, "V16fWiWiV16fIiIiIi", "nc", "fp8-insts") 349TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_fp8_fp8, "V16fWiWiV16fIiIiIi", "nc", "fp8-insts") 350TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x32_f16, "V4fV4hV8hV4fiIiIi", "nc", "mai-insts") 351TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x16_f16, "V16fV4hV8hV16fiIiIi", "nc", "mai-insts") 352TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x32_bf16, "V4fV4sV8sV4fiIiIi", "nc", "mai-insts") 353TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x16_bf16, "V16fV4sV8sV16fiIiIi", "nc", "mai-insts") 354TARGET_BUILTIN(__builtin_amdgcn_smfmac_i32_16x16x64_i8, "V4iV2iV4iV4iiIiIi", "nc", "mai-insts") 355TARGET_BUILTIN(__builtin_amdgcn_smfmac_i32_32x32x32_i8, "V16iV2iV4iV16iiIiIi", "nc", "mai-insts") 356TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_bf8_bf8, "V4fV2iV4iV4fiIiIi", "nc", "fp8-insts") 357TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_bf8_fp8, "V4fV2iV4iV4fiIiIi", "nc", "fp8-insts") 358TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_fp8_bf8, "V4fV2iV4iV4fiIiIi", "nc", "fp8-insts") 359TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_fp8_fp8, "V4fV2iV4iV4fiIiIi", "nc", "fp8-insts") 360TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_bf8_bf8, "V16fV2iV4iV16fiIiIi", "nc", "fp8-insts") 361TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_bf8_fp8, "V16fV2iV4iV16fiIiIi", "nc", "fp8-insts") 362TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_fp8_bf8, "V16fV2iV4iV16fiIiIi", "nc", "fp8-insts") 363TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_32x32x32_fp8_fp8, "V16fV2iV4iV16fiIiIi", "nc", "fp8-insts") 364 365TARGET_BUILTIN(__builtin_amdgcn_cvt_f32_bf8, "fiIi", "nc", "fp8-insts") 366TARGET_BUILTIN(__builtin_amdgcn_cvt_f32_fp8, "fiIi", "nc", "fp8-insts") 367TARGET_BUILTIN(__builtin_amdgcn_cvt_pk_f32_bf8, "V2fiIb", "nc", "fp8-insts") 368TARGET_BUILTIN(__builtin_amdgcn_cvt_pk_f32_fp8, "V2fiIb", "nc", "fp8-insts") 369TARGET_BUILTIN(__builtin_amdgcn_cvt_pk_bf8_f32, "iffiIb", "nc", "fp8-insts") 370TARGET_BUILTIN(__builtin_amdgcn_cvt_pk_fp8_f32, "iffiIb", "nc", "fp8-insts") 371TARGET_BUILTIN(__builtin_amdgcn_cvt_sr_bf8_f32, "ifiiIi", "nc", "fp8-insts") 372TARGET_BUILTIN(__builtin_amdgcn_cvt_sr_fp8_f32, "ifiiIi", "nc", "fp8-insts") 373 374#undef BUILTIN 375#undef TARGET_BUILTIN 376