src/gallium/drivers/radeonsi/si_shader_internal.h

   1 /*
   2  * Copyright 2016 Advanced Micro Devices, Inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * on the rights to use, copy, modify, merge, publish, distribute, sub
   9  * license, and/or sell copies of the Software, and to permit persons to whom
  10  * the Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  20  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  21  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  22  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  */
  24
  25 #ifndef SI_SHADER_PRIVATE_H
  26 #define SI_SHADER_PRIVATE_H
  27
  28 #include "si_shader.h"
  29 #include "ac_shader_abi.h"
  30
  31 #include <llvm-c/Core.h>
  32 #include <llvm-c/TargetMachine.h>
  33
  34 struct pipe_debug_callback;
  35
  36 #define RADEON_LLVM_MAX_INPUTS 32 * 4
  37
  38 struct si_shader_output_values {
  39         LLVMValueRef values[4];
  40         unsigned semantic_name;
  41         unsigned semantic_index;
  42         ubyte vertex_stream[4];
  43 };
  44
  45 struct si_shader_context {
  46         struct ac_llvm_context ac;
  47         struct si_shader *shader;
  48         struct si_screen *screen;
  49
  50         unsigned type; /* PIPE_SHADER_* specifies the type of shader. */
  51
  52         /* For clamping the non-constant index in resource indexing: */
  53         unsigned num_const_buffers;
  54         unsigned num_shader_buffers;
  55         unsigned num_images;
  56         unsigned num_samplers;
  57
  58         struct ac_shader_args args;
  59         struct ac_shader_abi abi;
  60
  61         LLVMValueRef inputs[RADEON_LLVM_MAX_INPUTS];
  62
  63         LLVMBasicBlockRef merged_wrap_if_entry_block;
  64         int merged_wrap_if_label;
  65
  66         LLVMValueRef main_fn;
  67         LLVMTypeRef return_type;
  68
  69         struct ac_arg const_and_shader_buffers;
  70         struct ac_arg samplers_and_images;
  71
  72         /* For merged shaders, the per-stage descriptors for the stage other
  73          * than the one we're processing, used to pass them through from the
  74          * first stage to the second.
  75          */
  76         struct ac_arg other_const_and_shader_buffers;
  77         struct ac_arg other_samplers_and_images;
  78
  79         struct ac_arg rw_buffers;
  80         struct ac_arg bindless_samplers_and_images;
  81         /* Common inputs for merged shaders. */
  82         struct ac_arg merged_wave_info;
  83         struct ac_arg merged_scratch_offset;
  84         /* API VS */
  85         struct ac_arg vertex_buffers;
  86         struct ac_arg vb_descriptors[5];
  87         struct ac_arg rel_auto_id;
  88         struct ac_arg vs_prim_id;
  89         struct ac_arg vertex_index0;
  90         /* VS states and layout of LS outputs / TCS inputs at the end
  91          *   [0] = clamp vertex color
  92          *   [1] = indexed
  93          *   [8:20] = stride between patches in DW = num_inputs * num_vertices * 4
  94          *            max = 32*32*4 + 32*4
  95          *   [24:31] = stride between vertices in DW = num_inputs * 4
  96          *             max = 32*4
  97          */
  98         struct ac_arg vs_state_bits;
  99         struct ac_arg vs_blit_inputs;
 100         /* HW VS */
 101         struct ac_arg streamout_config;
 102         struct ac_arg streamout_write_index;
 103         struct ac_arg streamout_offset[4];
 104
 105         /* API TCS & TES */
 106         /* Layout of TCS outputs in the offchip buffer
 107          * # 6 bits
 108          *   [0:5] = the number of patches per threadgroup, max = NUM_PATCHES (40)
 109          * # 6 bits
 110          *   [6:11] = the number of output vertices per patch, max = 32
 111          * # 20 bits
 112          *   [12:31] = the offset of per patch attributes in the buffer in bytes.
 113          *             max = NUM_PATCHES*32*32*16
 114          */
 115         struct ac_arg tcs_offchip_layout;
 116
 117         /* API TCS */
 118         /* Offsets where TCS outputs and TCS patch outputs live in LDS:
 119          *   [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
 120          *   [16:31] = TCS output patch0 offset for per-patch / 16
 121          *             max = (NUM_PATCHES + 1) * 32*32
 122          */
 123         struct ac_arg tcs_out_lds_offsets;
 124         /* Layout of TCS outputs / TES inputs:
 125          *   [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
 126          *            max = 32*32*4 + 32*4
 127          *   [13:18] = gl_PatchVerticesIn, max = 32
 128          *   [19:31] = high 13 bits of the 32-bit address of tessellation ring buffers
 129          */
 130         struct ac_arg tcs_out_lds_layout;
 131         struct ac_arg tcs_offchip_offset;
 132         struct ac_arg tcs_factor_offset;
 133
 134         /* API TES */
 135         struct ac_arg tes_offchip_addr;
 136         struct ac_arg tes_u;
 137         struct ac_arg tes_v;
 138         struct ac_arg tes_rel_patch_id;
 139         /* HW ES */
 140         struct ac_arg es2gs_offset;
 141         /* HW GS */
 142         /* On gfx10:
 143          *  - bits 0..11: ordered_wave_id
 144          *  - bits 12..20: number of vertices in group
 145          *  - bits 22..30: number of primitives in group
 146          */
 147         struct ac_arg gs_tg_info;
 148         /* API GS */
 149         struct ac_arg gs2vs_offset;
 150         struct ac_arg gs_wave_id; /* GFX6 */
 151         struct ac_arg gs_vtx_offset[6]; /* in dwords (GFX6) */
 152         struct ac_arg gs_vtx01_offset; /* in dwords (GFX9) */
 153         struct ac_arg gs_vtx23_offset; /* in dwords (GFX9) */
 154         struct ac_arg gs_vtx45_offset; /* in dwords (GFX9) */
 155         /* PS */
 156         struct ac_arg pos_fixed_pt;
 157         /* CS */
 158         struct ac_arg block_size;
 159         struct ac_arg cs_user_data;
 160
 161         struct ac_llvm_compiler *compiler;
 162
 163         /* Preloaded descriptors. */
 164         LLVMValueRef esgs_ring;
 165         LLVMValueRef gsvs_ring[4];
 166         LLVMValueRef tess_offchip_ring;
 167
 168         LLVMValueRef invoc0_tess_factors[6]; /* outer[4], inner[2] */
 169         LLVMValueRef gs_next_vertex[4];
 170         LLVMValueRef gs_curprim_verts[4];
 171         LLVMValueRef gs_generated_prims[4];
 172         LLVMValueRef gs_ngg_emit;
 173         LLVMValueRef gs_ngg_scratch;
 174         LLVMValueRef postponed_kill;
 175         LLVMValueRef return_value;
 176
 177         LLVMTypeRef voidt;
 178         LLVMTypeRef i1;
 179         LLVMTypeRef i8;
 180         LLVMTypeRef i32;
 181         LLVMTypeRef i64;
 182         LLVMTypeRef i128;
 183         LLVMTypeRef f32;
 184         LLVMTypeRef v2i32;
 185         LLVMTypeRef v4i32;
 186         LLVMTypeRef v4f32;
 187         LLVMTypeRef v8i32;
 188
 189         LLVMValueRef i32_0;
 190         LLVMValueRef i32_1;
 191         LLVMValueRef i1false;
 192         LLVMValueRef i1true;
 193 };
 194
 195 static inline struct si_shader_context *
 196 si_shader_context_from_abi(struct ac_shader_abi *abi)
 197 {
 198         struct si_shader_context *ctx = NULL;
 199         return container_of(abi, ctx, abi);
 200 }
 201
 202 unsigned si_llvm_compile(LLVMModuleRef M, struct si_shader_binary *binary,
 203                          struct ac_llvm_compiler *compiler,
 204                          struct pipe_debug_callback *debug,
 205                          bool less_optimized, unsigned wave_size);
 206
 207 LLVMValueRef si_llvm_bound_index(struct si_shader_context *ctx,
 208                                  LLVMValueRef index,
 209                                  unsigned num);
 210
 211 void si_llvm_context_init(struct si_shader_context *ctx,
 212                           struct si_screen *sscreen,
 213                           struct ac_llvm_compiler *compiler,
 214                           unsigned wave_size);
 215 void si_llvm_context_set_ir(struct si_shader_context *ctx,
 216                             struct si_shader *shader);
 217
 218 void si_llvm_create_func(struct si_shader_context *ctx, const char *name,
 219                          LLVMTypeRef *return_types, unsigned num_return_elems,
 220                          unsigned max_workgroup_size);
 221
 222 void si_llvm_dispose(struct si_shader_context *ctx);
 223
 224 void si_llvm_optimize_module(struct si_shader_context *ctx);
 225
 226 LLVMValueRef si_nir_load_input_tes(struct ac_shader_abi *abi,
 227                                    LLVMTypeRef type,
 228                                    LLVMValueRef vertex_index,
 229                                    LLVMValueRef param_index,
 230                                    unsigned const_index,
 231                                    unsigned location,
 232                                    unsigned driver_location,
 233                                    unsigned component,
 234                                    unsigned num_components,
 235                                    bool is_patch,
 236                                    bool is_compact,
 237                                    bool load_input);
 238 LLVMValueRef si_nir_lookup_interp_param(struct ac_shader_abi *abi,
 239                                         enum glsl_interp_mode interp,
 240                                         unsigned location);
 241 LLVMValueRef si_get_sample_id(struct si_shader_context *ctx);
 242 LLVMValueRef si_load_sampler_desc(struct si_shader_context *ctx,
 243                                   LLVMValueRef list, LLVMValueRef index,
 244                                   enum ac_descriptor_type type);
 245 LLVMValueRef si_load_image_desc(struct si_shader_context *ctx,
 246                                 LLVMValueRef list, LLVMValueRef index,
 247                                 enum ac_descriptor_type desc_type,
 248                                 bool uses_store, bool bindless);
 249 LLVMValueRef si_nir_emit_fbfetch(struct ac_shader_abi *abi);
 250 void si_declare_compute_memory(struct si_shader_context *ctx);
 251 LLVMValueRef si_get_primitive_id(struct si_shader_context *ctx,
 252                                  unsigned swizzle);
 253 void si_llvm_export_vs(struct si_shader_context *ctx,
 254                        struct si_shader_output_values *outputs,
 255                        unsigned noutput);
 256 void si_emit_streamout_output(struct si_shader_context *ctx,
 257                               LLVMValueRef const *so_buffers,
 258                               LLVMValueRef const *so_write_offsets,
 259                               struct pipe_stream_output *stream_out,
 260                               struct si_shader_output_values *shader_out);
 261
 262 void si_llvm_load_input_vs(
 263         struct si_shader_context *ctx,
 264         unsigned input_index,
 265         LLVMValueRef out[4]);
 266
 267 bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *nir);
 268
 269 LLVMValueRef si_unpack_param(struct si_shader_context *ctx,
 270                              struct ac_arg param, unsigned rshift,
 271                              unsigned bitwidth);
 272 LLVMValueRef si_is_es_thread(struct si_shader_context *ctx);
 273 LLVMValueRef si_is_gs_thread(struct si_shader_context *ctx);
 274
 275 void gfx10_emit_ngg_epilogue(struct ac_shader_abi *abi,
 276                              unsigned max_outputs,
 277                              LLVMValueRef *addrs);
 278 void gfx10_ngg_gs_emit_vertex(struct si_shader_context *ctx,
 279                               unsigned stream,
 280                               LLVMValueRef *addrs);
 281 void gfx10_ngg_gs_emit_prologue(struct si_shader_context *ctx);
 282 void gfx10_ngg_gs_emit_epilogue(struct si_shader_context *ctx);
 283 void gfx10_ngg_calculate_subgroup_info(struct si_shader *shader);
 284
 285 #endif