src/gallium/drivers/radeonsi/si_shader_internal.h

   1 /*
   2  * Copyright 2016 Advanced Micro Devices, Inc.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #ifndef SI_SHADER_PRIVATE_H
  25 #define SI_SHADER_PRIVATE_H
  26
  27 #include "si_shader.h"
  28 #include "gallivm/lp_bld_flow.h"
  29 #include "gallivm/lp_bld_init.h"
  30 #include "gallivm/lp_bld_tgsi.h"
  31 #include "tgsi/tgsi_parse.h"
  32 #include "ac_shader_abi.h"
  33 #include "ac_llvm_util.h"
  34 #include "ac_llvm_build.h"
  35
  36 #include <llvm-c/Core.h>
  37 #include <llvm-c/TargetMachine.h>
  38
  39 struct pipe_debug_callback;
  40 struct ac_shader_binary;
  41
  42 #define RADEON_LLVM_MAX_INPUT_SLOTS 32
  43 #define RADEON_LLVM_MAX_INPUTS 32 * 4
  44 #define RADEON_LLVM_MAX_OUTPUTS 32 * 4
  45
  46 #define RADEON_LLVM_INITIAL_CF_DEPTH 4
  47
  48 #define RADEON_LLVM_MAX_SYSTEM_VALUES 11
  49 #define RADEON_LLVM_MAX_ADDRS 16
  50
  51 struct si_llvm_flow;
  52
  53 struct si_shader_context {
  54         struct lp_build_tgsi_context bld_base;
  55         struct gallivm_state gallivm;
  56         struct ac_llvm_context ac;
  57         struct si_shader *shader;
  58         struct si_screen *screen;
  59
  60         unsigned type; /* PIPE_SHADER_* specifies the type of shader. */
  61
  62         /* For clamping the non-constant index in resource indexing: */
  63         unsigned num_const_buffers;
  64         unsigned num_shader_buffers;
  65         unsigned num_images;
  66         unsigned num_samplers;
  67
  68         /* Whether the prolog will be compiled separately. */
  69         bool separate_prolog;
  70
  71         struct ac_shader_abi abi;
  72
  73         /** This function is responsible for initilizing the inputs array and will be
  74           * called once for each input declared in the TGSI shader.
  75           */
  76         void (*load_input)(struct si_shader_context *,
  77                            unsigned input_index,
  78                            const struct tgsi_full_declaration *decl,
  79                            LLVMValueRef out[4]);
  80
  81         /** This array contains the input values for the shader.  Typically these
  82           * values will be in the form of a target intrinsic that will inform the
  83           * backend how to load the actual inputs to the shader.
  84           */
  85         struct tgsi_full_declaration input_decls[RADEON_LLVM_MAX_INPUT_SLOTS];
  86         LLVMValueRef inputs[RADEON_LLVM_MAX_INPUTS];
  87         LLVMValueRef outputs[RADEON_LLVM_MAX_OUTPUTS][TGSI_NUM_CHANNELS];
  88         LLVMValueRef addrs[RADEON_LLVM_MAX_ADDRS][TGSI_NUM_CHANNELS];
  89
  90         /** This pointer is used to contain the temporary values.
  91           * The amount of temporary used in tgsi can't be bound to a max value and
  92           * thus we must allocate this array at runtime.
  93           */
  94         LLVMValueRef *temps;
  95         unsigned temps_count;
  96         LLVMValueRef system_values[RADEON_LLVM_MAX_SYSTEM_VALUES];
  97
  98         LLVMValueRef *imms;
  99         unsigned imms_num;
 100
 101         struct si_llvm_flow *flow;
 102         unsigned flow_depth;
 103         unsigned flow_depth_max;
 104
 105         struct lp_build_if_state merged_wrap_if_state;
 106
 107         struct tgsi_array_info *temp_arrays;
 108         LLVMValueRef *temp_array_allocas;
 109
 110         LLVMValueRef undef_alloca;
 111
 112         LLVMValueRef main_fn;
 113         LLVMTypeRef return_type;
 114
 115         /* Parameter indices for LLVMGetParam. */
 116         int param_rw_buffers;
 117         int param_const_and_shader_buffers;
 118         int param_samplers_and_images;
 119         int param_bindless_samplers_and_images;
 120         /* Common inputs for merged shaders. */
 121         int param_merged_wave_info;
 122         int param_merged_scratch_offset;
 123         /* API VS */
 124         int param_vertex_buffers;
 125         int param_rel_auto_id;
 126         int param_vs_prim_id;
 127         int param_vertex_index0;
 128         /* VS states and layout of LS outputs / TCS inputs at the end
 129          *   [0] = clamp vertex color
 130          *   [1] = indexed
 131          *   [8:20] = stride between patches in DW = num_inputs * num_vertices * 4
 132          *            max = 32*32*4 + 32*4
 133          *   [24:31] = stride between vertices in DW = num_inputs * 4
 134          *             max = 32*4
 135          */
 136         int param_vs_state_bits;
 137         /* HW VS */
 138         int param_streamout_config;
 139         int param_streamout_write_index;
 140         int param_streamout_offset[4];
 141
 142         /* API TCS & TES */
 143         /* Layout of TCS outputs in the offchip buffer
 144          * # 6 bits
 145          *   [0:5] = the number of patches per threadgroup, max = NUM_PATCHES (40)
 146          * # 6 bits
 147          *   [6:11] = the number of output vertices per patch, max = 32
 148          * # 20 bits
 149          *   [12:31] = the offset of per patch attributes in the buffer in bytes.
 150          *             max = NUM_PATCHES*32*32*16
 151          */
 152         int param_tcs_offchip_layout;
 153
 154         /* API TCS */
 155         /* Offsets where TCS outputs and TCS patch outputs live in LDS:
 156          *   [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
 157          *   [16:31] = TCS output patch0 offset for per-patch / 16
 158          *             max = (NUM_PATCHES + 1) * 32*32
 159          */
 160         int param_tcs_out_lds_offsets;
 161         /* Layout of TCS outputs / TES inputs:
 162          *   [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
 163          *            max = 32*32*4 + 32*4
 164          *   [13:20] = stride between output vertices in DW = num_inputs * 4
 165          *             max = 32*4
 166          *   [26:31] = gl_PatchVerticesIn, max = 32
 167          */
 168         int param_tcs_out_lds_layout;
 169         int param_tcs_offchip_addr_base64k;
 170         int param_tcs_factor_addr_base64k;
 171         int param_tcs_offchip_offset;
 172         int param_tcs_factor_offset;
 173         int param_tcs_patch_id;
 174         int param_tcs_rel_ids;
 175
 176         /* API TES */
 177         int param_tes_u;
 178         int param_tes_v;
 179         int param_tes_rel_patch_id;
 180         int param_tes_patch_id;
 181         /* HW ES */
 182         int param_es2gs_offset;
 183         /* API GS */
 184         int param_gs2vs_offset;
 185         int param_gs_wave_id; /* GFX6 */
 186         int param_gs_vtx0_offset; /* in dwords (GFX6) */
 187         int param_gs_vtx1_offset; /* in dwords (GFX6) */
 188         int param_gs_prim_id;
 189         int param_gs_vtx2_offset; /* in dwords (GFX6) */
 190         int param_gs_vtx3_offset; /* in dwords (GFX6) */
 191         int param_gs_vtx4_offset; /* in dwords (GFX6) */
 192         int param_gs_vtx5_offset; /* in dwords (GFX6) */
 193         int param_gs_instance_id;
 194         int param_gs_vtx01_offset; /* in dwords (GFX9) */
 195         int param_gs_vtx23_offset; /* in dwords (GFX9) */
 196         int param_gs_vtx45_offset; /* in dwords (GFX9) */
 197         /* CS */
 198         int param_grid_size;
 199         int param_block_size;
 200         int param_block_id[3];
 201         int param_thread_id;
 202
 203         LLVMTargetMachineRef tm;
 204
 205         unsigned range_md_kind;
 206         unsigned fpmath_md_kind;
 207         LLVMValueRef fpmath_md_2p5_ulp;
 208
 209         /* Preloaded descriptors. */
 210         LLVMValueRef esgs_ring;
 211         LLVMValueRef gsvs_ring[4];
 212
 213         LLVMValueRef lds;
 214         LLVMValueRef gs_next_vertex[4];
 215         LLVMValueRef postponed_kill;
 216         LLVMValueRef return_value;
 217
 218         LLVMTypeRef voidt;
 219         LLVMTypeRef i1;
 220         LLVMTypeRef i8;
 221         LLVMTypeRef i32;
 222         LLVMTypeRef i64;
 223         LLVMTypeRef i128;
 224         LLVMTypeRef f32;
 225         LLVMTypeRef v2i32;
 226         LLVMTypeRef v4i32;
 227         LLVMTypeRef v4f32;
 228         LLVMTypeRef v8i32;
 229
 230         LLVMValueRef i32_0;
 231         LLVMValueRef i32_1;
 232
 233         LLVMValueRef shared_memory;
 234 };
 235
 236 static inline struct si_shader_context *
 237 si_shader_context(struct lp_build_tgsi_context *bld_base)
 238 {
 239         return (struct si_shader_context*)bld_base;
 240 }
 241
 242 static inline struct si_shader_context *
 243 si_shader_context_from_abi(struct ac_shader_abi *abi)
 244 {
 245         struct si_shader_context *ctx = NULL;
 246         return container_of(abi, ctx, abi);
 247 }
 248
 249 void si_llvm_add_attribute(LLVMValueRef F, const char *name, int value);
 250
 251 unsigned si_llvm_compile(LLVMModuleRef M, struct ac_shader_binary *binary,
 252                          LLVMTargetMachineRef tm,
 253                          struct pipe_debug_callback *debug);
 254
 255 LLVMTypeRef tgsi2llvmtype(struct lp_build_tgsi_context *bld_base,
 256                           enum tgsi_opcode_type type);
 257
 258 LLVMValueRef bitcast(struct lp_build_tgsi_context *bld_base,
 259                      enum tgsi_opcode_type type, LLVMValueRef value);
 260
 261 LLVMValueRef si_llvm_bound_index(struct si_shader_context *ctx,
 262                                  LLVMValueRef index,
 263                                  unsigned num);
 264
 265 void si_llvm_context_init(struct si_shader_context *ctx,
 266                           struct si_screen *sscreen,
 267                           LLVMTargetMachineRef tm);
 268 void si_llvm_context_set_tgsi(struct si_shader_context *ctx,
 269                               struct si_shader *shader);
 270
 271 void si_llvm_create_func(struct si_shader_context *ctx,
 272                          const char *name,
 273                          LLVMTypeRef *return_types, unsigned num_return_elems,
 274                          LLVMTypeRef *ParamTypes, unsigned ParamCount);
 275
 276 void si_llvm_dispose(struct si_shader_context *ctx);
 277
 278 void si_llvm_optimize_module(struct si_shader_context *ctx);
 279
 280 LLVMValueRef si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context *bld_base,
 281                                       enum tgsi_opcode_type type,
 282                                       LLVMValueRef ptr,
 283                                       LLVMValueRef ptr2);
 284
 285 LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
 286                                 const struct tgsi_full_src_register *reg,
 287                                 enum tgsi_opcode_type type,
 288                                 unsigned swizzle);
 289
 290 void si_llvm_emit_store(struct lp_build_tgsi_context *bld_base,
 291                         const struct tgsi_full_instruction *inst,
 292                         const struct tgsi_opcode_info *info,
 293                         LLVMValueRef dst[4]);
 294
 295 /* Combine these with & instead of |. */
 296 #define NOOP_WAITCNT 0xf7f
 297 #define LGKM_CNT 0x07f
 298 #define VM_CNT 0xf70
 299
 300 void si_emit_waitcnt(struct si_shader_context *ctx, unsigned simm16);
 301
 302 LLVMValueRef si_get_indirect_index(struct si_shader_context *ctx,
 303                                    const struct tgsi_ind_register *ind,
 304                                    int rel_index);
 305 LLVMValueRef si_get_bounded_indirect_index(struct si_shader_context *ctx,
 306                                            const struct tgsi_ind_register *ind,
 307                                            int rel_index, unsigned num);
 308
 309 LLVMTypeRef si_const_array(LLVMTypeRef elem_type, int num_elements);
 310
 311 void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base);
 312 void si_shader_context_init_mem(struct si_shader_context *ctx);
 313
 314 LLVMValueRef si_load_sampler_desc(struct si_shader_context *ctx,
 315                                   LLVMValueRef list, LLVMValueRef index,
 316                                   enum ac_descriptor_type type);
 317 LLVMValueRef si_load_image_desc(struct si_shader_context *ctx,
 318                                 LLVMValueRef list, LLVMValueRef index,
 319                                 enum ac_descriptor_type desc_type, bool dcc_off);
 320
 321 void si_load_system_value(struct si_shader_context *ctx,
 322                           unsigned index,
 323                           const struct tgsi_full_declaration *decl);
 324 void si_declare_compute_memory(struct si_shader_context *ctx,
 325                                const struct tgsi_full_declaration *decl);
 326
 327 void si_llvm_load_input_vs(
 328         struct si_shader_context *ctx,
 329         unsigned input_index,
 330         LLVMValueRef out[4]);
 331 void si_llvm_load_input_fs(
 332         struct si_shader_context *ctx,
 333         unsigned input_index,
 334         LLVMValueRef out[4]);
 335
 336 bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir);
 337
 338 #endif