src/amd/common/ac_llvm_build.h

   1 /*
   2  * Copyright 2016 Bas Nieuwenhuizen
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the
   6  * "Software"), to deal in the Software without restriction, including
   7  * without limitation the rights to use, copy, modify, merge, publish,
   8  * distribute, sub license, and/or sell copies of the Software, and to
   9  * permit persons to whom the Software is furnished to do so, subject to
  10  * the following conditions:
  11  *
  12  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  13  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  14  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  15  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
  16  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  17  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  18  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  19  *
  20  * The above copyright notice and this permission notice (including the
  21  * next paragraph) shall be included in all copies or substantial portions
  22  * of the Software.
  23  *
  24  */
  25 #ifndef AC_LLVM_BUILD_H
  26 #define AC_LLVM_BUILD_H
  27
  28 #include <stdbool.h>
  29 #include <llvm-c/TargetMachine.h>
  30
  31 #include "amd_family.h"
  32
  33 #ifdef __cplusplus
  34 extern "C" {
  35 #endif
  36
  37 enum {
  38         AC_CONST_ADDR_SPACE = 2, /* CONST is the only address space that selects SMEM loads */
  39         AC_LOCAL_ADDR_SPACE = 3,
  40 };
  41
  42 struct ac_llvm_context {
  43         LLVMContextRef context;
  44         LLVMModuleRef module;
  45         LLVMBuilderRef builder;
  46
  47         LLVMTypeRef voidt;
  48         LLVMTypeRef i1;
  49         LLVMTypeRef i8;
  50         LLVMTypeRef i16;
  51         LLVMTypeRef i32;
  52         LLVMTypeRef i64;
  53         LLVMTypeRef f16;
  54         LLVMTypeRef f32;
  55         LLVMTypeRef f64;
  56         LLVMTypeRef v2i16;
  57         LLVMTypeRef v2i32;
  58         LLVMTypeRef v3i32;
  59         LLVMTypeRef v4i32;
  60         LLVMTypeRef v2f32;
  61         LLVMTypeRef v4f32;
  62         LLVMTypeRef v8i32;
  63
  64         LLVMValueRef i32_0;
  65         LLVMValueRef i32_1;
  66         LLVMValueRef i64_0;
  67         LLVMValueRef i64_1;
  68         LLVMValueRef f32_0;
  69         LLVMValueRef f32_1;
  70         LLVMValueRef f64_0;
  71         LLVMValueRef f64_1;
  72         LLVMValueRef i1true;
  73         LLVMValueRef i1false;
  74
  75         unsigned range_md_kind;
  76         unsigned invariant_load_md_kind;
  77         unsigned uniform_md_kind;
  78         unsigned fpmath_md_kind;
  79         LLVMValueRef fpmath_md_2p5_ulp;
  80         LLVMValueRef empty_md;
  81
  82         enum chip_class chip_class;
  83         enum radeon_family family;
  84
  85         LLVMValueRef lds;
  86 };
  87
  88 void
  89 ac_llvm_context_init(struct ac_llvm_context *ctx, LLVMContextRef context,
  90                      enum chip_class chip_class, enum radeon_family family);
  91
  92 int
  93 ac_get_llvm_num_components(LLVMValueRef value);
  94
  95 LLVMValueRef
  96 ac_llvm_extract_elem(struct ac_llvm_context *ac,
  97                      LLVMValueRef value,
  98                      int index);
  99
 100 unsigned ac_get_type_size(LLVMTypeRef type);
 101
 102 LLVMTypeRef ac_to_integer_type(struct ac_llvm_context *ctx, LLVMTypeRef t);
 103 LLVMValueRef ac_to_integer(struct ac_llvm_context *ctx, LLVMValueRef v);
 104 LLVMTypeRef ac_to_float_type(struct ac_llvm_context *ctx, LLVMTypeRef t);
 105 LLVMValueRef ac_to_float(struct ac_llvm_context *ctx, LLVMValueRef v);
 106
 107 LLVMValueRef
 108 ac_build_intrinsic(struct ac_llvm_context *ctx, const char *name,
 109                    LLVMTypeRef return_type, LLVMValueRef *params,
 110                    unsigned param_count, unsigned attrib_mask);
 111
 112 void ac_build_type_name_for_intr(LLVMTypeRef type, char *buf, unsigned bufsize);
 113
 114 LLVMValueRef
 115 ac_build_phi(struct ac_llvm_context *ctx, LLVMTypeRef type,
 116              unsigned count_incoming, LLVMValueRef *values,
 117              LLVMBasicBlockRef *blocks);
 118
 119 void ac_build_optimization_barrier(struct ac_llvm_context *ctx,
 120                                    LLVMValueRef *pvgpr);
 121
 122 LLVMValueRef ac_build_shader_clock(struct ac_llvm_context *ctx);
 123
 124 LLVMValueRef ac_build_ballot(struct ac_llvm_context *ctx, LLVMValueRef value);
 125
 126 LLVMValueRef ac_build_vote_all(struct ac_llvm_context *ctx, LLVMValueRef value);
 127
 128 LLVMValueRef ac_build_vote_any(struct ac_llvm_context *ctx, LLVMValueRef value);
 129
 130 LLVMValueRef ac_build_vote_eq(struct ac_llvm_context *ctx, LLVMValueRef value);
 131
 132 LLVMValueRef
 133 ac_build_varying_gather_values(struct ac_llvm_context *ctx, LLVMValueRef *values,
 134                                unsigned value_count, unsigned component);
 135
 136 LLVMValueRef
 137 ac_build_gather_values_extended(struct ac_llvm_context *ctx,
 138                                 LLVMValueRef *values,
 139                                 unsigned value_count,
 140                                 unsigned value_stride,
 141                                 bool load,
 142                                 bool always_vector);
 143 LLVMValueRef
 144 ac_build_gather_values(struct ac_llvm_context *ctx,
 145                        LLVMValueRef *values,
 146                        unsigned value_count);
 147 LLVMValueRef ac_build_expand_to_vec4(struct ac_llvm_context *ctx,
 148                                      LLVMValueRef value,
 149                                      unsigned num_channels);
 150
 151 LLVMValueRef
 152 ac_build_fdiv(struct ac_llvm_context *ctx,
 153               LLVMValueRef num,
 154               LLVMValueRef den);
 155
 156 void
 157 ac_prepare_cube_coords(struct ac_llvm_context *ctx,
 158                        bool is_deriv, bool is_array, bool is_lod,
 159                        LLVMValueRef *coords_arg,
 160                        LLVMValueRef *derivs_arg);
 161
 162
 163 LLVMValueRef
 164 ac_build_fs_interp(struct ac_llvm_context *ctx,
 165                    LLVMValueRef llvm_chan,
 166                    LLVMValueRef attr_number,
 167                    LLVMValueRef params,
 168                    LLVMValueRef i,
 169                    LLVMValueRef j);
 170
 171 LLVMValueRef
 172 ac_build_fs_interp_mov(struct ac_llvm_context *ctx,
 173                        LLVMValueRef parameter,
 174                        LLVMValueRef llvm_chan,
 175                        LLVMValueRef attr_number,
 176                        LLVMValueRef params);
 177
 178 LLVMValueRef
 179 ac_build_gep0(struct ac_llvm_context *ctx,
 180               LLVMValueRef base_ptr,
 181               LLVMValueRef index);
 182
 183 void
 184 ac_build_indexed_store(struct ac_llvm_context *ctx,
 185                        LLVMValueRef base_ptr, LLVMValueRef index,
 186                        LLVMValueRef value);
 187
 188 LLVMValueRef ac_build_load(struct ac_llvm_context *ctx, LLVMValueRef base_ptr,
 189                            LLVMValueRef index);
 190 LLVMValueRef ac_build_load_invariant(struct ac_llvm_context *ctx,
 191                                      LLVMValueRef base_ptr, LLVMValueRef index);
 192 LLVMValueRef ac_build_load_to_sgpr(struct ac_llvm_context *ctx,
 193                                    LLVMValueRef base_ptr, LLVMValueRef index);
 194
 195 void
 196 ac_build_buffer_store_dword(struct ac_llvm_context *ctx,
 197                             LLVMValueRef rsrc,
 198                             LLVMValueRef vdata,
 199                             unsigned num_channels,
 200                             LLVMValueRef voffset,
 201                             LLVMValueRef soffset,
 202                             unsigned inst_offset,
 203                             bool glc,
 204                             bool slc,
 205                             bool writeonly_memory,
 206                             bool swizzle_enable_hint);
 207 LLVMValueRef
 208 ac_build_buffer_load(struct ac_llvm_context *ctx,
 209                      LLVMValueRef rsrc,
 210                      int num_channels,
 211                      LLVMValueRef vindex,
 212                      LLVMValueRef voffset,
 213                      LLVMValueRef soffset,
 214                      unsigned inst_offset,
 215                      unsigned glc,
 216                      unsigned slc,
 217                      bool can_speculate,
 218                      bool allow_smem);
 219
 220 LLVMValueRef ac_build_buffer_load_format(struct ac_llvm_context *ctx,
 221                                          LLVMValueRef rsrc,
 222                                          LLVMValueRef vindex,
 223                                          LLVMValueRef voffset,
 224                                          unsigned num_channels,
 225                                          bool glc,
 226                                          bool can_speculate);
 227
 228 LLVMValueRef
 229 ac_get_thread_id(struct ac_llvm_context *ctx);
 230
 231 #define AC_TID_MASK_TOP_LEFT 0xfffffffc
 232 #define AC_TID_MASK_TOP      0xfffffffd
 233 #define AC_TID_MASK_LEFT     0xfffffffe
 234
 235 LLVMValueRef
 236 ac_build_ddxy(struct ac_llvm_context *ctx,
 237               uint32_t mask,
 238               int idx,
 239               LLVMValueRef val);
 240
 241 #define AC_SENDMSG_GS 2
 242 #define AC_SENDMSG_GS_DONE 3
 243
 244 #define AC_SENDMSG_GS_OP_NOP      (0 << 4)
 245 #define AC_SENDMSG_GS_OP_CUT      (1 << 4)
 246 #define AC_SENDMSG_GS_OP_EMIT     (2 << 4)
 247 #define AC_SENDMSG_GS_OP_EMIT_CUT (3 << 4)
 248
 249 void ac_build_sendmsg(struct ac_llvm_context *ctx,
 250                       uint32_t msg,
 251                       LLVMValueRef wave_id);
 252
 253 LLVMValueRef ac_build_imsb(struct ac_llvm_context *ctx,
 254                            LLVMValueRef arg,
 255                            LLVMTypeRef dst_type);
 256
 257 LLVMValueRef ac_build_umsb(struct ac_llvm_context *ctx,
 258                           LLVMValueRef arg,
 259                           LLVMTypeRef dst_type);
 260 LLVMValueRef ac_build_fmin(struct ac_llvm_context *ctx, LLVMValueRef a,
 261                            LLVMValueRef b);
 262 LLVMValueRef ac_build_fmax(struct ac_llvm_context *ctx, LLVMValueRef a,
 263                            LLVMValueRef b);
 264 LLVMValueRef ac_build_imin(struct ac_llvm_context *ctx, LLVMValueRef a,
 265                            LLVMValueRef b);
 266 LLVMValueRef ac_build_imax(struct ac_llvm_context *ctx, LLVMValueRef a,
 267                            LLVMValueRef b);
 268 LLVMValueRef ac_build_umin(struct ac_llvm_context *ctx, LLVMValueRef a, LLVMValueRef b);
 269 LLVMValueRef ac_build_clamp(struct ac_llvm_context *ctx, LLVMValueRef value);
 270
 271 struct ac_export_args {
 272         LLVMValueRef out[4];
 273         unsigned target;
 274         unsigned enabled_channels;
 275         bool compr;
 276         bool done;
 277         bool valid_mask;
 278 };
 279
 280 void ac_build_export(struct ac_llvm_context *ctx, struct ac_export_args *a);
 281
 282 void ac_build_export_null(struct ac_llvm_context *ctx);
 283
 284 enum ac_image_opcode {
 285         ac_image_sample,
 286         ac_image_gather4,
 287         ac_image_load,
 288         ac_image_load_mip,
 289         ac_image_get_lod,
 290         ac_image_get_resinfo,
 291 };
 292
 293 struct ac_image_args {
 294         enum ac_image_opcode opcode;
 295         bool level_zero;
 296         bool bias;
 297         bool lod;
 298         bool deriv;
 299         bool compare;
 300         bool offset;
 301
 302         LLVMValueRef resource;
 303         LLVMValueRef sampler;
 304         LLVMValueRef addr;
 305         unsigned dmask;
 306         bool unorm;
 307         bool da;
 308 };
 309
 310 LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx,
 311                                    struct ac_image_args *a);
 312 LLVMValueRef ac_build_cvt_pkrtz_f16(struct ac_llvm_context *ctx,
 313                                     LLVMValueRef args[2]);
 314 LLVMValueRef ac_build_cvt_pknorm_i16(struct ac_llvm_context *ctx,
 315                                      LLVMValueRef args[2]);
 316 LLVMValueRef ac_build_cvt_pknorm_u16(struct ac_llvm_context *ctx,
 317                                      LLVMValueRef args[2]);
 318 LLVMValueRef ac_build_cvt_pk_i16(struct ac_llvm_context *ctx,
 319                                  LLVMValueRef args[2], unsigned bits, bool hi);
 320 LLVMValueRef ac_build_cvt_pk_u16(struct ac_llvm_context *ctx,
 321                                  LLVMValueRef args[2], unsigned bits, bool hi);
 322 LLVMValueRef ac_build_wqm_vote(struct ac_llvm_context *ctx, LLVMValueRef i1);
 323 void ac_build_kill_if_false(struct ac_llvm_context *ctx, LLVMValueRef i1);
 324 LLVMValueRef ac_build_bfe(struct ac_llvm_context *ctx, LLVMValueRef input,
 325                           LLVMValueRef offset, LLVMValueRef width,
 326                           bool is_signed);
 327
 328 void ac_build_waitcnt(struct ac_llvm_context *ctx, unsigned simm16);
 329
 330 void ac_get_image_intr_name(const char *base_name,
 331                             LLVMTypeRef data_type,
 332                             LLVMTypeRef coords_type,
 333                             LLVMTypeRef rsrc_type,
 334                             char *out_name, unsigned out_len);
 335
 336 void ac_optimize_vs_outputs(struct ac_llvm_context *ac,
 337                             LLVMValueRef main_fn,
 338                             uint8_t *vs_output_param_offset,
 339                             uint32_t num_outputs,
 340                             uint8_t *num_param_exports);
 341 void ac_init_exec_full_mask(struct ac_llvm_context *ctx);
 342
 343 void ac_declare_lds_as_pointer(struct ac_llvm_context *ac);
 344 LLVMValueRef ac_lds_load(struct ac_llvm_context *ctx,
 345                          LLVMValueRef dw_addr);
 346 void ac_lds_store(struct ac_llvm_context *ctx,
 347                   LLVMValueRef dw_addr, LLVMValueRef value);
 348
 349 LLVMValueRef ac_find_lsb(struct ac_llvm_context *ctx,
 350                          LLVMTypeRef dst_type,
 351                          LLVMValueRef src0);
 352
 353 LLVMTypeRef ac_array_in_const_addr_space(LLVMTypeRef elem_type);
 354
 355 #ifdef __cplusplus
 356 }
 357 #endif
 358
 359 #endif