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 "ac_shader_abi.h"
29#include "si_shader.h"
30
31struct pipe_debug_callback;
32
33/* Ideally pass the sample mask input to the PS epilog as v14, which
34 * is its usual location, so that the shader doesn't have to add v_mov.
35 */
36#define PS_EPILOG_SAMPLEMASK_MIN_LOC 14
37
38struct si_shader_output_values {
39   LLVMValueRef values[4];
40   ubyte vertex_stream[4];
41   ubyte semantic;
42};
43
44struct si_shader_context {
45   struct ac_llvm_context ac;
46   struct si_shader *shader;
47   struct si_shader_selector *next_shader_sel;
48   struct si_screen *screen;
49
50   gl_shader_stage stage;
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   LLVMBasicBlockRef merged_wrap_if_entry_block;
62   int merged_wrap_if_label;
63
64   LLVMValueRef main_fn;
65   LLVMTypeRef return_type;
66
67   struct ac_arg const_and_shader_buffers;
68   struct ac_arg samplers_and_images;
69
70   /* For merged shaders, the per-stage descriptors for the stage other
71    * than the one we're processing, used to pass them through from the
72    * first stage to the second.
73    */
74   struct ac_arg other_const_and_shader_buffers;
75   struct ac_arg other_samplers_and_images;
76
77   struct ac_arg internal_bindings;
78   struct ac_arg bindless_samplers_and_images;
79   struct ac_arg small_prim_cull_info;
80   /* API VS */
81   struct ac_arg vb_descriptors[5];
82   struct ac_arg vertex_index0;
83   /* VS states and layout of LS outputs / TCS inputs at the end
84    *   [0] = clamp vertex color
85    *   [1] = indexed
86    *   [2:3] = NGG: output primitive type
87    *   [4:5] = NGG: provoking vertex index
88    *   [6]   = NGG: streamout queries enabled
89    *   [7:10] = NGG: small prim filter precision = num_samples / quant_mode,
90    *            but in reality it's: 1/2^n, from 1/16 to 1/4096 = 1/2^4 to 1/2^12
91    *            Only the first 4 bits of the exponent are stored.
92    *            Set it like this: (fui(num_samples / quant_mode) >> 23)
93    *            Expand to FP32 like this: ((0x70 | value) << 23);
94    *            With 0x70 = 112, we get 2^(112 + value - 127) = 2^(value - 15)
95    *            = 1/2^(15 - value) in FP32
96    *   [11:23] = stride between patches in DW = num_inputs * num_vertices * 4
97    *             max = 32*32*4 + 32*4
98    *   [24:31] = stride between vertices in DW = num_inputs * 4
99    *             max = 32*4
100    */
101   struct ac_arg vs_state_bits;
102   struct ac_arg vs_blit_inputs;
103
104   /* API TCS & TES */
105   /* Layout of TCS outputs in the offchip buffer
106    * # 6 bits
107    *   [0:5] = the number of patches per threadgroup - 1, max = 63
108    * # 5 bits
109    *   [6:10] = the number of output vertices per patch - 1, max = 31
110    * # 21 bits
111    *   [11:31] = the offset of per patch attributes in the buffer in bytes.
112    *             max = NUM_PATCHES*32*32*16 = 1M
113    */
114   struct ac_arg tcs_offchip_layout;
115
116   /* API TCS */
117   /* Offsets where TCS outputs and TCS patch outputs live in LDS:
118    *   [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32 = 64K (TODO: not enough bits)
119    *   [16:31] = TCS output patch0 offset for per-patch / 16
120    *             max = (NUM_PATCHES + 1) * 32*32 = 66624 (TODO: not enough bits)
121    */
122   struct ac_arg tcs_out_lds_offsets;
123   /* Layout of TCS outputs / TES inputs:
124    *   [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
125    *            max = 32*32*4 + 32*4 = 4224
126    *   [13:18] = gl_PatchVerticesIn, max = 32
127    *   [19:31] = high 13 bits of the 32-bit address of tessellation ring buffers
128    */
129   struct ac_arg tcs_out_lds_layout;
130
131   /* API TES */
132   struct ac_arg tes_offchip_addr;
133   /* PS */
134   struct ac_arg pos_fixed_pt;
135   /* CS */
136   struct ac_arg block_size;
137   struct ac_arg cs_user_data;
138   struct ac_arg cs_shaderbuf[3];
139   struct ac_arg cs_image[3];
140
141   struct ac_llvm_compiler *compiler;
142
143   /* Preloaded descriptors. */
144   LLVMValueRef esgs_ring;
145   LLVMValueRef gsvs_ring[4];
146   LLVMValueRef tess_offchip_ring;
147
148   LLVMValueRef invoc0_tess_factors[6]; /* outer[4], inner[2] */
149   LLVMValueRef gs_next_vertex[4];
150   LLVMValueRef gs_curprim_verts[4];
151   LLVMValueRef gs_generated_prims[4];
152   LLVMValueRef gs_ngg_emit;
153   LLVMValueRef gs_ngg_scratch;
154   LLVMValueRef return_value;
155};
156
157static inline struct si_shader_context *si_shader_context_from_abi(struct ac_shader_abi *abi)
158{
159   return container_of(abi, struct si_shader_context, abi);
160}
161
162/* si_shader.c */
163bool si_is_multi_part_shader(struct si_shader *shader);
164bool si_is_merged_shader(struct si_shader *shader);
165void si_add_arg_checked(struct ac_shader_args *args, enum ac_arg_regfile file, unsigned registers,
166                        enum ac_arg_type type, struct ac_arg *arg, unsigned idx);
167void si_init_shader_args(struct si_shader_context *ctx, bool ngg_cull_shader);
168unsigned si_get_max_workgroup_size(const struct si_shader *shader);
169bool si_vs_needs_prolog(const struct si_shader_selector *sel,
170                        const struct si_vs_prolog_bits *prolog_key,
171                        const struct si_shader_key *key, bool ngg_cull_shader);
172void si_get_vs_prolog_key(const struct si_shader_info *info, unsigned num_input_sgprs,
173                          bool ngg_cull_shader, const struct si_vs_prolog_bits *prolog_key,
174                          struct si_shader *shader_out, union si_shader_part_key *key);
175struct nir_shader *si_get_nir_shader(struct si_shader_selector *sel,
176                                     const struct si_shader_key *key,
177                                     bool *free_nir);
178bool si_need_ps_prolog(const union si_shader_part_key *key);
179void si_get_ps_prolog_key(struct si_shader *shader, union si_shader_part_key *key,
180                          bool separate_prolog);
181void si_get_ps_epilog_key(struct si_shader *shader, union si_shader_part_key *key);
182void si_fix_resource_usage(struct si_screen *sscreen, struct si_shader *shader);
183
184/* gfx10_shader_ngg.c */
185bool gfx10_ngg_export_prim_early(struct si_shader *shader);
186void gfx10_ngg_build_sendmsg_gs_alloc_req(struct si_shader_context *ctx);
187void gfx10_ngg_build_export_prim(struct si_shader_context *ctx, LLVMValueRef user_edgeflags[3],
188                                 LLVMValueRef prim_passthrough);
189void gfx10_emit_ngg_culling_epilogue(struct ac_shader_abi *abi);
190void gfx10_emit_ngg_epilogue(struct ac_shader_abi *abi);
191void gfx10_ngg_gs_emit_vertex(struct si_shader_context *ctx, unsigned stream, LLVMValueRef *addrs);
192void gfx10_ngg_gs_emit_prologue(struct si_shader_context *ctx);
193void gfx10_ngg_gs_emit_epilogue(struct si_shader_context *ctx);
194unsigned gfx10_ngg_get_scratch_dw_size(struct si_shader *shader);
195bool gfx10_ngg_calculate_subgroup_info(struct si_shader *shader);
196
197/* si_shader_llvm.c */
198bool si_compile_llvm(struct si_screen *sscreen, struct si_shader_binary *binary,
199                     struct ac_shader_config *conf, struct ac_llvm_compiler *compiler,
200                     struct ac_llvm_context *ac, struct pipe_debug_callback *debug,
201                     gl_shader_stage stage, const char *name, bool less_optimized);
202void si_llvm_context_init(struct si_shader_context *ctx, struct si_screen *sscreen,
203                          struct ac_llvm_compiler *compiler, unsigned wave_size);
204void si_llvm_create_func(struct si_shader_context *ctx, const char *name, LLVMTypeRef *return_types,
205                         unsigned num_return_elems, unsigned max_workgroup_size);
206void si_llvm_create_main_func(struct si_shader_context *ctx, bool ngg_cull_shader);
207void si_llvm_optimize_module(struct si_shader_context *ctx);
208void si_llvm_dispose(struct si_shader_context *ctx);
209LLVMValueRef si_buffer_load_const(struct si_shader_context *ctx, LLVMValueRef resource,
210                                  LLVMValueRef offset);
211void si_llvm_build_ret(struct si_shader_context *ctx, LLVMValueRef ret);
212LLVMValueRef si_insert_input_ret(struct si_shader_context *ctx, LLVMValueRef ret,
213                                 struct ac_arg param, unsigned return_index);
214LLVMValueRef si_insert_input_ret_float(struct si_shader_context *ctx, LLVMValueRef ret,
215                                       struct ac_arg param, unsigned return_index);
216LLVMValueRef si_insert_input_ptr(struct si_shader_context *ctx, LLVMValueRef ret,
217                                 struct ac_arg param, unsigned return_index);
218LLVMValueRef si_prolog_get_internal_bindings(struct si_shader_context *ctx);
219void si_llvm_emit_barrier(struct si_shader_context *ctx);
220void si_llvm_declare_esgs_ring(struct si_shader_context *ctx);
221LLVMValueRef si_unpack_param(struct si_shader_context *ctx, struct ac_arg param, unsigned rshift,
222                             unsigned bitwidth);
223LLVMValueRef si_get_primitive_id(struct si_shader_context *ctx, unsigned swizzle);
224void si_build_wrapper_function(struct si_shader_context *ctx, LLVMValueRef *parts,
225                               unsigned num_parts, unsigned main_part,
226                               unsigned next_shader_first_part, bool same_thread_count);
227bool si_llvm_translate_nir(struct si_shader_context *ctx, struct si_shader *shader,
228                           struct nir_shader *nir, bool free_nir, bool ngg_cull_shader);
229bool si_llvm_compile_shader(struct si_screen *sscreen, struct ac_llvm_compiler *compiler,
230                            struct si_shader *shader, struct pipe_debug_callback *debug,
231                            struct nir_shader *nir, bool free_nir);
232
233/* si_shader_llvm_gs.c */
234LLVMValueRef si_is_es_thread(struct si_shader_context *ctx);
235LLVMValueRef si_is_gs_thread(struct si_shader_context *ctx);
236void si_llvm_emit_es_epilogue(struct ac_shader_abi *abi);
237void si_preload_esgs_ring(struct si_shader_context *ctx);
238void si_preload_gs_rings(struct si_shader_context *ctx);
239void si_llvm_build_gs_prolog(struct si_shader_context *ctx, union si_shader_part_key *key);
240void si_llvm_init_gs_callbacks(struct si_shader_context *ctx);
241
242/* si_shader_llvm_tess.c */
243void si_llvm_preload_tes_rings(struct si_shader_context *ctx);
244void si_llvm_emit_ls_epilogue(struct ac_shader_abi *abi);
245void si_llvm_build_tcs_epilog(struct si_shader_context *ctx, union si_shader_part_key *key);
246void si_llvm_init_tcs_callbacks(struct si_shader_context *ctx);
247void si_llvm_init_tes_callbacks(struct si_shader_context *ctx, bool ngg_cull_shader);
248
249/* si_shader_llvm_ps.c */
250LLVMValueRef si_get_sample_id(struct si_shader_context *ctx);
251void si_llvm_build_ps_prolog(struct si_shader_context *ctx, union si_shader_part_key *key);
252void si_llvm_build_ps_epilog(struct si_shader_context *ctx, union si_shader_part_key *key);
253void si_llvm_build_monolithic_ps(struct si_shader_context *ctx, struct si_shader *shader);
254void si_llvm_init_ps_callbacks(struct si_shader_context *ctx);
255
256/* si_shader_llvm_resources.c */
257void si_llvm_init_resource_callbacks(struct si_shader_context *ctx);
258
259/* si_shader_llvm_vs.c */
260void si_llvm_streamout_store_output(struct si_shader_context *ctx, LLVMValueRef const *so_buffers,
261                                    LLVMValueRef const *so_write_offsets,
262                                    struct pipe_stream_output *stream_out,
263                                    struct si_shader_output_values *shader_out);
264void si_llvm_emit_streamout(struct si_shader_context *ctx, struct si_shader_output_values *outputs,
265                            unsigned noutput, unsigned stream);
266void si_llvm_build_vs_exports(struct si_shader_context *ctx,
267                              struct si_shader_output_values *outputs, unsigned noutput);
268void si_llvm_emit_vs_epilogue(struct ac_shader_abi *abi);
269void si_llvm_build_vs_prolog(struct si_shader_context *ctx, union si_shader_part_key *key);
270void si_llvm_init_vs_callbacks(struct si_shader_context *ctx, bool ngg_cull_shader);
271
272#endif
273