================ @@ -0,0 +1,1017 @@ +//===--- BuiltinsAMDGPU.td - AMDGPU Builtin function defs -------*- C++ -*-===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// +// +// This file defines the AMDGPU-specific builtin function database. +// +//===----------------------------------------------------------------------===// + +include "clang/Basic/BuiltinsBase.td" + +//===----------------------------------------------------------------------===// +// AMDGPU builtin base classes +//===----------------------------------------------------------------------===// + +class AMDGPUBuiltin<string prototype, list<Attribute> Attr = [], string Feat = ""> : TargetBuiltin { + let Spellings = [NAME]; + let Prototype = prototype; + let Attributes = !listconcat([NoThrow], Attr); + let Features = Feat; +} + +//===----------------------------------------------------------------------===// +// SI+ only builtins. +//===----------------------------------------------------------------------===// + +def __builtin_amdgcn_dispatch_ptr : AMDGPUBuiltin<"void address_space<4> *()", [Const]>; +def __builtin_amdgcn_kernarg_segment_ptr : AMDGPUBuiltin<"void address_space<4> *()", [Const]>; +def __builtin_amdgcn_implicitarg_ptr : AMDGPUBuiltin<"void address_space<4> *()", [Const]>; +def __builtin_amdgcn_queue_ptr : AMDGPUBuiltin<"void address_space<4> *()", [Const]>; + +def __builtin_amdgcn_workgroup_id_x : AMDGPUBuiltin<"unsigned int()", [Const]>; +def __builtin_amdgcn_workgroup_id_y : AMDGPUBuiltin<"unsigned int()", [Const]>; +def __builtin_amdgcn_workgroup_id_z : AMDGPUBuiltin<"unsigned int()", [Const]>; + +def __builtin_amdgcn_cluster_id_x : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; +def __builtin_amdgcn_cluster_id_y : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; +def __builtin_amdgcn_cluster_id_z : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; + +def __builtin_amdgcn_cluster_workgroup_id_x : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; +def __builtin_amdgcn_cluster_workgroup_id_y : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; +def __builtin_amdgcn_cluster_workgroup_id_z : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; +def __builtin_amdgcn_cluster_workgroup_flat_id : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; + +def __builtin_amdgcn_cluster_workgroup_max_id_x : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; +def __builtin_amdgcn_cluster_workgroup_max_id_y : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; +def __builtin_amdgcn_cluster_workgroup_max_id_z : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; +def __builtin_amdgcn_cluster_workgroup_max_flat_id : AMDGPUBuiltin<"unsigned int()", [Const], "gfx1250-insts">; + +def __builtin_amdgcn_workitem_id_x : AMDGPUBuiltin<"unsigned int()", [Const]>; +def __builtin_amdgcn_workitem_id_y : AMDGPUBuiltin<"unsigned int()", [Const]>; +def __builtin_amdgcn_workitem_id_z : AMDGPUBuiltin<"unsigned int()", [Const]>; + +def __builtin_amdgcn_workgroup_size_x : AMDGPUBuiltin<"unsigned short()", [Const]>; +def __builtin_amdgcn_workgroup_size_y : AMDGPUBuiltin<"unsigned short()", [Const]>; +def __builtin_amdgcn_workgroup_size_z : AMDGPUBuiltin<"unsigned short()", [Const]>; + +def __builtin_amdgcn_grid_size_x : AMDGPUBuiltin<"unsigned int()", [Const]>; +def __builtin_amdgcn_grid_size_y : AMDGPUBuiltin<"unsigned int()", [Const]>; +def __builtin_amdgcn_grid_size_z : AMDGPUBuiltin<"unsigned int()", [Const]>; + +def __builtin_amdgcn_mbcnt_hi : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_mbcnt_lo : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int)", [Const]>; + +def __builtin_amdgcn_s_memtime : AMDGPUBuiltin<"uint64_t()", [], "s-memtime-inst">; + +//===----------------------------------------------------------------------===// +// Instruction builtins. +//===----------------------------------------------------------------------===// +def __builtin_amdgcn_s_getreg : AMDGPUBuiltin<"unsigned int(_Constant int)">; +def __builtin_amdgcn_s_setreg : AMDGPUBuiltin<"void(_Constant int, unsigned int)">; +def __builtin_amdgcn_s_getpc : AMDGPUBuiltin<"uint64_t()">; +def __builtin_amdgcn_s_waitcnt : AMDGPUBuiltin<"void(_Constant int)">; +def __builtin_amdgcn_s_sendmsg : AMDGPUBuiltin<"void(_Constant int, unsigned int)">; +def __builtin_amdgcn_s_sendmsghalt : AMDGPUBuiltin<"void(_Constant int, unsigned int)">; +def __builtin_amdgcn_s_barrier : AMDGPUBuiltin<"void()">; +def __builtin_amdgcn_s_ttracedata : AMDGPUBuiltin<"void(int)">; +def __builtin_amdgcn_wave_barrier : AMDGPUBuiltin<"void()">; +def __builtin_amdgcn_sched_barrier : AMDGPUBuiltin<"void(_Constant int)">; +def __builtin_amdgcn_sched_group_barrier : AMDGPUBuiltin<"void(_Constant int, _Constant int, _Constant int)">; +def __builtin_amdgcn_iglp_opt : AMDGPUBuiltin<"void(_Constant int)">; +def __builtin_amdgcn_s_dcache_inv : AMDGPUBuiltin<"void()">; +def __builtin_amdgcn_buffer_wbinvl1 : AMDGPUBuiltin<"void()">; +def __builtin_amdgcn_fence : AMDGPUBuiltin<"void(unsigned int, char const *, ...)">; +def __builtin_amdgcn_groupstaticsize : AMDGPUBuiltin<"unsigned int()">; +def __builtin_amdgcn_wavefrontsize : AMDGPUBuiltin<"unsigned int()", [Const]>; + +def __builtin_amdgcn_atomic_inc32 : AMDGPUBuiltin<"uint32_t(uint32_t volatile *, uint32_t, unsigned int, char const *)">; +def __builtin_amdgcn_atomic_inc64 : AMDGPUBuiltin<"uint64_t(uint64_t volatile *, uint64_t, unsigned int, char const *)">; + +def __builtin_amdgcn_atomic_dec32 : AMDGPUBuiltin<"uint32_t(uint32_t volatile *, uint32_t, unsigned int, char const *)">; +def __builtin_amdgcn_atomic_dec64 : AMDGPUBuiltin<"uint64_t(uint64_t volatile *, uint64_t, unsigned int, char const *)">; + +// FIXME: Need to disallow constant address space. +def __builtin_amdgcn_div_scale : AMDGPUBuiltin<"double(double, double, bool, bool *)">; +def __builtin_amdgcn_div_scalef : AMDGPUBuiltin<"float(float, float, bool, bool *)">; +def __builtin_amdgcn_div_fmas : AMDGPUBuiltin<"double(double, double, double, bool)", [Const]>; +def __builtin_amdgcn_div_fmasf : AMDGPUBuiltin<"float(float, float, float, bool)", [Const]>; +def __builtin_amdgcn_div_fixup : AMDGPUBuiltin<"double(double, double, double)", [Const]>; +def __builtin_amdgcn_div_fixupf : AMDGPUBuiltin<"float(float, float, float)", [Const]>; +def __builtin_amdgcn_trig_preop : AMDGPUBuiltin<"double(double, int)", [Const]>; +def __builtin_amdgcn_trig_preopf : AMDGPUBuiltin<"float(float, int)", [Const]>; +def __builtin_amdgcn_rcp : AMDGPUBuiltin<"double(double)", [Const]>; +def __builtin_amdgcn_rcpf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_sqrt : AMDGPUBuiltin<"double(double)", [Const]>; +def __builtin_amdgcn_sqrtf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_rsq : AMDGPUBuiltin<"double(double)", [Const]>; +def __builtin_amdgcn_rsqf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_rsq_clamp : AMDGPUBuiltin<"double(double)", [Const]>; +def __builtin_amdgcn_rsq_clampf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_sinf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_cosf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_logf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_exp2f : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_log_clampf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_ldexp : AMDGPUBuiltin<"double(double, int)", [Const]>; +def __builtin_amdgcn_ldexpf : AMDGPUBuiltin<"float(float, int)", [Const]>; +def __builtin_amdgcn_frexp_mant : AMDGPUBuiltin<"double(double)", [Const]>; +def __builtin_amdgcn_frexp_mantf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_frexp_exp : AMDGPUBuiltin<"int(double)", [Const]>; +def __builtin_amdgcn_frexp_expf : AMDGPUBuiltin<"int(float)", [Const]>; +def __builtin_amdgcn_fract : AMDGPUBuiltin<"double(double)", [Const]>; +def __builtin_amdgcn_fractf : AMDGPUBuiltin<"float(float)", [Const]>; +def __builtin_amdgcn_lerp : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const], "lerp-inst">; +def __builtin_amdgcn_class : AMDGPUBuiltin<"bool(double, int)", [Const]>; +def __builtin_amdgcn_classf : AMDGPUBuiltin<"bool(float, int)", [Const]>; +def __builtin_amdgcn_cubeid : AMDGPUBuiltin<"float(float, float, float)", [Const], "cube-insts">; +def __builtin_amdgcn_cubesc : AMDGPUBuiltin<"float(float, float, float)", [Const], "cube-insts">; +def __builtin_amdgcn_cubetc : AMDGPUBuiltin<"float(float, float, float)", [Const], "cube-insts">; +def __builtin_amdgcn_cubema : AMDGPUBuiltin<"float(float, float, float)", [Const], "cube-insts">; +def __builtin_amdgcn_s_sleep : AMDGPUBuiltin<"void(_Constant int)">; +def __builtin_amdgcn_s_incperflevel : AMDGPUBuiltin<"void(_Constant int)">; +def __builtin_amdgcn_s_decperflevel : AMDGPUBuiltin<"void(_Constant int)">; +def __builtin_amdgcn_s_setprio : AMDGPUBuiltin<"void(_Constant short)">; +def __builtin_amdgcn_ds_swizzle : AMDGPUBuiltin<"int(int, _Constant int)", [Const]>; +def __builtin_amdgcn_ds_permute : AMDGPUBuiltin<"int(int, int)", [Const]>; +def __builtin_amdgcn_ds_bpermute : AMDGPUBuiltin<"int(int, int)", [Const]>; +def __builtin_amdgcn_readfirstlane : AMDGPUBuiltin<"int(int)", [Const]>; +def __builtin_amdgcn_readlane : AMDGPUBuiltin<"int(int, int)", [Const]>; +def __builtin_amdgcn_fmed3f : AMDGPUBuiltin<"float(float, float, float)", [Const]>; +def __builtin_amdgcn_ds_faddf : AMDGPUBuiltin<"float(float address_space<3> *, float, _Constant int, _Constant int, _Constant bool)">; +def __builtin_amdgcn_ds_fminf : AMDGPUBuiltin<"float(float address_space<3> *, float, _Constant int, _Constant int, _Constant bool)">; +def __builtin_amdgcn_ds_fmaxf : AMDGPUBuiltin<"float(float address_space<3> *, float, _Constant int, _Constant int, _Constant bool)">; +def __builtin_amdgcn_ds_append : AMDGPUBuiltin<"int(int address_space<3> *)">; +def __builtin_amdgcn_ds_consume : AMDGPUBuiltin<"int(int address_space<3> *)">; +def __builtin_amdgcn_alignbit : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_alignbyte : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_ubfe : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_sbfe : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_cvt_pkrtz : AMDGPUBuiltin<"_ExtVector<2, __fp16>(float, float)", [Const]>; +def __builtin_amdgcn_cvt_pknorm_i16 : AMDGPUBuiltin<"_ExtVector<2, short>(float, float)", [Const], "cvt-pknorm-vop2-insts">; +def __builtin_amdgcn_cvt_pknorm_u16 : AMDGPUBuiltin<"_ExtVector<2, unsigned short>(float, float)", [Const], "cvt-pknorm-vop2-insts">; +def __builtin_amdgcn_cvt_pk_i16 : AMDGPUBuiltin<"_ExtVector<2, short>(int, int)", [Const]>; +def __builtin_amdgcn_cvt_pk_u16 : AMDGPUBuiltin<"_ExtVector<2, unsigned short>(unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_cvt_pk_u8_f32 : AMDGPUBuiltin<"unsigned int(float, unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_cvt_off_f32_i4 : AMDGPUBuiltin<"float(int)", [Const]>; +def __builtin_amdgcn_msad_u8 : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_sad_u8 : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const], "sad-insts">; +def __builtin_amdgcn_sad_hi_u8 : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const], "sad-insts">; +def __builtin_amdgcn_sad_u16 : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const], "sad-insts">; +def __builtin_amdgcn_qsad_pk_u16_u8 : AMDGPUBuiltin<"uint64_t(uint64_t, unsigned int, uint64_t)", [Const], "qsad-insts">; +def __builtin_amdgcn_mqsad_pk_u16_u8 : AMDGPUBuiltin<"uint64_t(uint64_t, unsigned int, uint64_t)", [Const]>; +def __builtin_amdgcn_mqsad_u32_u8 : AMDGPUBuiltin<"_Vector<4, unsigned int>(uint64_t, unsigned int, _Vector<4, unsigned int>)", [Const]>; + +def __builtin_amdgcn_make_buffer_rsrc : AMDGPUBuiltin<"__amdgpu_buffer_rsrc_t(void *, short, int64_t, int)", [Const]>; +def __builtin_amdgcn_raw_buffer_store_b8 : AMDGPUBuiltin<"void(unsigned char, __amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_store_b16 : AMDGPUBuiltin<"void(unsigned short, __amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_store_b32 : AMDGPUBuiltin<"void(unsigned int, __amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_store_b64 : AMDGPUBuiltin<"void(_Vector<2, unsigned int>, __amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_store_b96 : AMDGPUBuiltin<"void(_Vector<3, unsigned int>, __amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_store_b128 : AMDGPUBuiltin<"void(_Vector<4, unsigned int>, __amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_load_b8 : AMDGPUBuiltin<"unsigned char(__amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_load_b16 : AMDGPUBuiltin<"unsigned short(__amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_load_b32 : AMDGPUBuiltin<"unsigned int(__amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_load_b64 : AMDGPUBuiltin<"_Vector<2, unsigned int>(__amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_load_b96 : AMDGPUBuiltin<"_Vector<3, unsigned int>(__amdgpu_buffer_rsrc_t, int, int, _Constant int)">; +def __builtin_amdgcn_raw_buffer_load_b128 : AMDGPUBuiltin<"_Vector<4, unsigned int>(__amdgpu_buffer_rsrc_t, int, int, _Constant int)">; + +def __builtin_amdgcn_raw_ptr_buffer_atomic_add_i32 : AMDGPUBuiltin<"int(int, __amdgpu_buffer_rsrc_t, int, int, _Constant int)">; + +def __builtin_amdgcn_raw_ptr_buffer_atomic_fadd_f32 : AMDGPUBuiltin<"float(float, __amdgpu_buffer_rsrc_t, int, int, _Constant int)", [], "atomic-fadd-rtn-insts">; +def __builtin_amdgcn_raw_ptr_buffer_atomic_fadd_v2f16 : AMDGPUBuiltin<"_Vector<2, _Float16>(_Vector<2, _Float16>, __amdgpu_buffer_rsrc_t, int, int, _Constant int)", [], "atomic-buffer-global-pk-add-f16-insts">; + +def __builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f32 : AMDGPUBuiltin<"float(float, __amdgpu_buffer_rsrc_t, int, int, _Constant int)", [], "atomic-fmin-fmax-global-f32">; +def __builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f32 : AMDGPUBuiltin<"float(float, __amdgpu_buffer_rsrc_t, int, int, _Constant int)", [], "atomic-fmin-fmax-global-f32">; +def __builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f64 : AMDGPUBuiltin<"double(double, __amdgpu_buffer_rsrc_t, int, int, _Constant int)", [], "atomic-fmin-fmax-global-f64">; +def __builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f64 : AMDGPUBuiltin<"double(double, __amdgpu_buffer_rsrc_t, int, int, _Constant int)", [], "atomic-fmin-fmax-global-f64">; + +def __builtin_amdgcn_raw_ptr_buffer_load_lds : AMDGPUBuiltin<"void(__amdgpu_buffer_rsrc_t, void address_space<3> *, _Constant unsigned int, int, int, _Constant int, _Constant int)", [], "vmem-to-lds-load-insts">; +def __builtin_amdgcn_struct_ptr_buffer_load_lds : AMDGPUBuiltin<"void(__amdgpu_buffer_rsrc_t, void address_space<3> *, _Constant unsigned int, int, int, int, _Constant int, _Constant int)", [], "vmem-to-lds-load-insts">; + +//===----------------------------------------------------------------------===// +// Ballot builtins. +//===----------------------------------------------------------------------===// + +def __builtin_amdgcn_ballot_w32 : AMDGPUBuiltin<"uint32_t(bool)", [Const], "wavefrontsize32">; +def __builtin_amdgcn_ballot_w64 : AMDGPUBuiltin<"uint64_t(bool)", [Const]>; + +def __builtin_amdgcn_inverse_ballot_w32 : AMDGPUBuiltin<"bool(uint32_t)", [Const], "wavefrontsize32">; +def __builtin_amdgcn_inverse_ballot_w64 : AMDGPUBuiltin<"bool(uint64_t)", [Const], "wavefrontsize64">; + +// Deprecated intrinsics in favor of __builtin_amdgn_ballot_{w32|w64} +def __builtin_amdgcn_uicmp : AMDGPUBuiltin<"uint64_t(unsigned int, unsigned int, _Constant int)", [Const]>; +def __builtin_amdgcn_uicmpl : AMDGPUBuiltin<"uint64_t(uint64_t, uint64_t, _Constant int)", [Const]>; +def __builtin_amdgcn_sicmp : AMDGPUBuiltin<"uint64_t(int, int, _Constant int)", [Const]>; +def __builtin_amdgcn_sicmpl : AMDGPUBuiltin<"uint64_t(int64_t, int64_t, _Constant int)", [Const]>; +def __builtin_amdgcn_fcmp : AMDGPUBuiltin<"uint64_t(double, double, _Constant int)", [Const]>; +def __builtin_amdgcn_fcmpf : AMDGPUBuiltin<"uint64_t(float, float, _Constant int)", [Const]>; + +//===----------------------------------------------------------------------===// +// Flat addressing builtins. +//===----------------------------------------------------------------------===// +def __builtin_amdgcn_is_shared : AMDGPUBuiltin<"bool(void const address_space<0> *)", [Const]>; +def __builtin_amdgcn_is_private : AMDGPUBuiltin<"bool(void const address_space<0> *)", [Const]>; + +//===----------------------------------------------------------------------===// +// GWS builtins. +//===----------------------------------------------------------------------===// +def __builtin_amdgcn_ds_gws_init : AMDGPUBuiltin<"void(unsigned int, unsigned int)", [], "gws">; +def __builtin_amdgcn_ds_gws_barrier : AMDGPUBuiltin<"void(unsigned int, unsigned int)", [], "gws">; +def __builtin_amdgcn_ds_gws_sema_v : AMDGPUBuiltin<"void(unsigned int)", [], "gws">; +def __builtin_amdgcn_ds_gws_sema_br : AMDGPUBuiltin<"void(unsigned int, unsigned int)", [], "gws">; +def __builtin_amdgcn_ds_gws_sema_p : AMDGPUBuiltin<"void(unsigned int)", [], "gws">; + +//===----------------------------------------------------------------------===// +// CI+ only builtins. +//===----------------------------------------------------------------------===// +def __builtin_amdgcn_s_dcache_inv_vol : AMDGPUBuiltin<"void()", [], "ci-insts">; +def __builtin_amdgcn_buffer_wbinvl1_vol : AMDGPUBuiltin<"void()", [], "ci-insts">; +def __builtin_amdgcn_ds_gws_sema_release_all : AMDGPUBuiltin<"void(unsigned int)", [], "ci-insts">; + +//===----------------------------------------------------------------------===// +// Interpolation builtins. +//===----------------------------------------------------------------------===// +def __builtin_amdgcn_interp_p1_f16 : AMDGPUBuiltin<"float(float, unsigned int, unsigned int, bool, unsigned int)", [Const]>; +def __builtin_amdgcn_interp_p2_f16 : AMDGPUBuiltin<"__fp16(float, float, unsigned int, unsigned int, bool, unsigned int)", [Const]>; +def __builtin_amdgcn_interp_p1 : AMDGPUBuiltin<"float(float, unsigned int, unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_interp_p2 : AMDGPUBuiltin<"float(float, float, unsigned int, unsigned int, unsigned int)", [Const]>; +def __builtin_amdgcn_interp_mov : AMDGPUBuiltin<"float(unsigned int, unsigned int, unsigned int, unsigned int)", [Const]>; + +//===----------------------------------------------------------------------===// +// VI+ only builtins. +//===----------------------------------------------------------------------===// + +def __builtin_amdgcn_div_fixuph : AMDGPUBuiltin<"__fp16(__fp16, __fp16, __fp16)", [Const], "16-bit-insts">; +def __builtin_amdgcn_rcph : AMDGPUBuiltin<"__fp16(__fp16)", [Const], "16-bit-insts">; +def __builtin_amdgcn_sqrth : AMDGPUBuiltin<"__fp16(__fp16)", [Const], "16-bit-insts">; +def __builtin_amdgcn_rsqh : AMDGPUBuiltin<"__fp16(__fp16)", [Const], "16-bit-insts">; +def __builtin_amdgcn_sinh : AMDGPUBuiltin<"__fp16(__fp16)", [Const], "16-bit-insts">; +def __builtin_amdgcn_cosh : AMDGPUBuiltin<"__fp16(__fp16)", [Const], "16-bit-insts">; +def __builtin_amdgcn_ldexph : AMDGPUBuiltin<"__fp16(__fp16, int)", [Const], "16-bit-insts">; +def __builtin_amdgcn_frexp_manth : AMDGPUBuiltin<"__fp16(__fp16)", [Const], "16-bit-insts">; +def __builtin_amdgcn_frexp_exph : AMDGPUBuiltin<"short(__fp16)", [Const], "16-bit-insts">; +def __builtin_amdgcn_fracth : AMDGPUBuiltin<"__fp16(__fp16)", [Const], "16-bit-insts">; +def __builtin_amdgcn_classh : AMDGPUBuiltin<"bool(__fp16, int)", [Const], "16-bit-insts">; +def __builtin_amdgcn_s_memrealtime : AMDGPUBuiltin<"uint64_t()", [], "s-memrealtime">; +def __builtin_amdgcn_mov_dpp : AMDGPUBuiltin<"int(int, _Constant int, _Constant int, _Constant int, _Constant bool)", [Const, CustomTypeChecking], "dpp">; +def __builtin_amdgcn_update_dpp : AMDGPUBuiltin<"int(int, int, _Constant int, _Constant int, _Constant int, _Constant bool)", [Const, CustomTypeChecking], "dpp">; +def __builtin_amdgcn_s_dcache_wb : AMDGPUBuiltin<"void()", [], "gfx8-insts">; +def __builtin_amdgcn_perm : AMDGPUBuiltin<"unsigned int(unsigned int, unsigned int, unsigned int)", [Const], "gfx8-insts">; + +//===----------------------------------------------------------------------===// +// GFX9+ only builtins. +//===----------------------------------------------------------------------===// + +def __builtin_amdgcn_fmed3h : AMDGPUBuiltin<"__fp16(__fp16, __fp16, __fp16)", [Const], "gfx9-insts">; + +def __builtin_amdgcn_global_atomic_fadd_f64 : AMDGPUBuiltin<"double(double address_space<1> *, double)", [], "gfx90a-insts">; +def __builtin_amdgcn_global_atomic_fadd_f32 : AMDGPUBuiltin<"float(float address_space<1> *, float)", [], "atomic-fadd-rtn-insts">; +def __builtin_amdgcn_global_atomic_fadd_v2f16 : AMDGPUBuiltin<"_Vector<2, _Float16>(_Vector<2, _Float16 address_space<1> *>, _Vector<2, _Float16>)", [CustomTypeChecking], "atomic-buffer-global-pk-add-f16-insts">; +def __builtin_amdgcn_global_atomic_fmin_f64 : AMDGPUBuiltin<"double(double address_space<1> *, double)", [], "gfx90a-insts">; +def __builtin_amdgcn_global_atomic_fmax_f64 : AMDGPUBuiltin<"double(double address_space<1> *, double)", [], "gfx90a-insts">; + +def __builtin_amdgcn_flat_atomic_fadd_f64 : AMDGPUBuiltin<"double(double address_space<0> *, double)", [], "gfx90a-insts">; +def __builtin_amdgcn_flat_atomic_fmin_f64 : AMDGPUBuiltin<"double(double address_space<0> *, double)", [], "gfx90a-insts">; +def __builtin_amdgcn_flat_atomic_fmax_f64 : AMDGPUBuiltin<"double(double address_space<0> *, double)", [], "gfx90a-insts">; + +def __builtin_amdgcn_ds_atomic_fadd_f64 : AMDGPUBuiltin<"double(double address_space<3> *, double)", [], "gfx90a-insts">; +def __builtin_amdgcn_ds_atomic_fadd_f32 : AMDGPUBuiltin<"float(float address_space<3> *, float)", [], "gfx8-insts">; + +def __builtin_amdgcn_flat_atomic_fadd_f32 : AMDGPUBuiltin<"float(float address_space<0> *, float)", [], "gfx940-insts">; +def __builtin_amdgcn_flat_atomic_fadd_v2f16 : AMDGPUBuiltin<"_Vector<2, _Float16>(_Vector<2, _Float16 address_space<0> *>, _Vector<2, _Float16>)", [CustomTypeChecking], "atomic-flat-pk-add-16-insts">; +def __builtin_amdgcn_flat_atomic_fadd_v2bf16 : AMDGPUBuiltin<"_Vector<2, short>(_Vector<2, short address_space<0> *>, _Vector<2, short>)", [CustomTypeChecking], "atomic-flat-pk-add-16-insts">; +def __builtin_amdgcn_global_atomic_fadd_v2bf16 : AMDGPUBuiltin<"_Vector<2, short>(_Vector<2, short address_space<1> *>, _Vector<2, short>)", [CustomTypeChecking], "atomic-global-pk-add-bf16-inst">; +def __builtin_amdgcn_ds_atomic_fadd_v2bf16 : AMDGPUBuiltin<"_Vector<2, short>(_Vector<2, short address_space<3> *>, _Vector<2, short>)", [CustomTypeChecking], "atomic-ds-pk-add-16-insts">; +def __builtin_amdgcn_ds_atomic_fadd_v2f16 : AMDGPUBuiltin<"_Vector<2, _Float16>(_Vector<2, _Float16 address_space<3> *>, _Vector<2, _Float16>)", [CustomTypeChecking], "atomic-ds-pk-add-16-insts">; +def __builtin_amdgcn_load_to_lds : AMDGPUBuiltin<"void(void *, void address_space<3> *, _Constant unsigned int, _Constant int, _Constant unsigned int)", [], "vmem-to-lds-load-insts">; +def __builtin_amdgcn_global_load_lds : AMDGPUBuiltin<"void(void address_space<1> *, void address_space<3> *, _Constant unsigned int, _Constant int, _Constant unsigned int)", [], "vmem-to-lds-load-insts">; + +//===----------------------------------------------------------------------===// +// Deep learning builtins. +//===----------------------------------------------------------------------===// + +def __builtin_amdgcn_fdot2 : AMDGPUBuiltin<"float(_Vector<2, _Float16>, _Vector<2, _Float16>, float, _Constant bool)", [Const], "dot10-insts">; +def __builtin_amdgcn_fdot2_f16_f16 : AMDGPUBuiltin<"_Float16(_Vector<2, _Float16>, _Vector<2, _Float16>, _Float16)", [Const], "dot9-insts">; +def __builtin_amdgcn_fdot2_bf16_bf16 : AMDGPUBuiltin<"short(_Vector<2, short>, _Vector<2, short>, short)", [Const], "dot9-insts">; +def __builtin_amdgcn_fdot2_f32_bf16 : AMDGPUBuiltin<"float(_Vector<2, short>, _Vector<2, short>, float, _Constant bool)", [Const], "dot12-insts">; +def __builtin_amdgcn_sdot2 : AMDGPUBuiltin<"signed int(_Vector<2, signed short>, _Vector<2, signed short>, signed int, _Constant bool)", [Const], "dot2-insts">; ---------------- shiltian wrote:
`signed short` is `short`. `unsigned int` is `unsigned`. Any particular reason to make it this way? https://github.com/llvm/llvm-project/pull/175873 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
