2
* Copyright 2016 Advanced Micro Devices, Inc.
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:
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
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.
25
#ifndef SI_SHADER_PRIVATE_H
26
#define SI_SHADER_PRIVATE_H
28
#include "ac_shader_abi.h"
29
#include "si_shader.h"
31
struct util_debug_callback;
33
struct si_shader_output_values {
34
LLVMValueRef values[4];
39
struct si_shader_context {
40
struct ac_llvm_context ac;
41
struct si_shader *shader;
42
struct si_shader_selector *next_shader_sel;
43
struct si_screen *screen;
45
gl_shader_stage stage;
47
/* For clamping the non-constant index in resource indexing: */
48
unsigned num_const_buffers;
49
unsigned num_shader_buffers;
51
unsigned num_samplers;
53
struct ac_shader_args args;
54
struct ac_shader_abi abi;
56
LLVMBasicBlockRef merged_wrap_if_entry_block;
57
int merged_wrap_if_label;
60
LLVMTypeRef return_type;
62
struct ac_arg const_and_shader_buffers;
63
struct ac_arg samplers_and_images;
65
/* For merged shaders, the per-stage descriptors for the stage other
66
* than the one we're processing, used to pass them through from the
67
* first stage to the second.
69
struct ac_arg other_const_and_shader_buffers;
70
struct ac_arg other_samplers_and_images;
72
struct ac_arg internal_bindings;
73
struct ac_arg bindless_samplers_and_images;
74
struct ac_arg small_prim_cull_info;
76
struct ac_arg vb_descriptors[5];
77
struct ac_arg vertex_index0;
78
/* VS states and layout of LS outputs / TCS inputs at the end
79
* [0] = clamp vertex color
81
* [2:3] = NGG: output primitive type
82
* [4:5] = NGG: provoking vertex index
83
* [6] = NGG: streamout queries enabled
84
* [7:10] = NGG: small prim filter precision = num_samples / quant_mode,
85
* but in reality it's: 1/2^n, from 1/16 to 1/4096 = 1/2^4 to 1/2^12
86
* Only the first 4 bits of the exponent are stored.
87
* Set it like this: (fui(num_samples / quant_mode) >> 23)
88
* Expand to FP32 like this: ((0x70 | value) << 23);
89
* With 0x70 = 112, we get 2^(112 + value - 127) = 2^(value - 15)
90
* = 1/2^(15 - value) in FP32
91
* [11:23] = stride between patches in DW = num_inputs * num_vertices * 4
92
* max = 32*32*4 + 32*4
93
* [24:31] = stride between vertices in DW = num_inputs * 4
96
struct ac_arg vs_state_bits;
97
struct ac_arg vs_blit_inputs;
100
/* Layout of TCS outputs in the offchip buffer
102
* [0:5] = the number of patches per threadgroup - 1, max = 63
104
* [6:10] = the number of output vertices per patch - 1, max = 31
106
* [11:31] = the offset of per patch attributes in the buffer in bytes.
107
* max = NUM_PATCHES*32*32*16 = 1M
109
struct ac_arg tcs_offchip_layout;
112
/* Offsets where TCS outputs and TCS patch outputs live in LDS:
113
* [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32 = 64K (TODO: not enough bits)
114
* [16:31] = TCS output patch0 offset for per-patch / 16
115
* max = (NUM_PATCHES + 1) * 32*32 = 66624 (TODO: not enough bits)
117
struct ac_arg tcs_out_lds_offsets;
118
/* Layout of TCS outputs / TES inputs:
119
* [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
120
* max = 32*32*4 + 32*4 = 4224
121
* [13:18] = gl_PatchVerticesIn, max = 32
122
* [19:31] = high 13 bits of the 32-bit address of tessellation ring buffers
124
struct ac_arg tcs_out_lds_layout;
127
struct ac_arg tes_offchip_addr;
129
struct ac_arg pos_fixed_pt;
131
struct ac_arg block_size;
132
struct ac_arg cs_user_data;
133
struct ac_arg cs_shaderbuf[3];
134
struct ac_arg cs_image[3];
136
struct ac_llvm_compiler *compiler;
138
/* GS vertex offsets unpacked with the gfx6-9 tristrip_adj bug workaround. */
139
LLVMValueRef gs_vtx_offset[6];
141
/* Preloaded descriptors. */
142
LLVMValueRef esgs_ring;
143
LLVMValueRef gsvs_ring[4];
144
LLVMValueRef tess_offchip_ring;
146
LLVMValueRef invoc0_tess_factors[6]; /* outer[4], inner[2] */
147
LLVMValueRef gs_next_vertex[4];
148
LLVMValueRef gs_curprim_verts[4];
149
LLVMValueRef gs_generated_prims[4];
150
LLVMValueRef gs_ngg_emit;
151
LLVMValueRef gs_ngg_scratch;
152
LLVMValueRef return_value;
155
static inline struct si_shader_context *si_shader_context_from_abi(struct ac_shader_abi *abi)
157
return container_of(abi, struct si_shader_context, abi);
161
bool si_is_multi_part_shader(struct si_shader *shader);
162
bool si_is_merged_shader(struct si_shader *shader);
163
void si_add_arg_checked(struct ac_shader_args *args, enum ac_arg_regfile file, unsigned registers,
164
enum ac_arg_type type, struct ac_arg *arg, unsigned idx);
165
void si_init_shader_args(struct si_shader_context *ctx, bool ngg_cull_shader);
166
unsigned si_get_max_workgroup_size(const struct si_shader *shader);
167
bool si_vs_needs_prolog(const struct si_shader_selector *sel,
168
const struct si_vs_prolog_bits *prolog_key,
169
const union si_shader_key *key, bool ngg_cull_shader, bool is_gs);
170
void si_get_vs_prolog_key(const struct si_shader_info *info, unsigned num_input_sgprs,
171
bool ngg_cull_shader, const struct si_vs_prolog_bits *prolog_key,
172
struct si_shader *shader_out, union si_shader_part_key *key);
173
struct nir_shader *si_get_nir_shader(struct si_shader_selector *sel,
174
const union si_shader_key *key,
176
bool si_need_ps_prolog(const union si_shader_part_key *key);
177
void si_get_ps_prolog_key(struct si_shader *shader, union si_shader_part_key *key,
178
bool separate_prolog);
179
void si_get_ps_epilog_key(struct si_shader *shader, union si_shader_part_key *key);
180
void si_fix_resource_usage(struct si_screen *sscreen, struct si_shader *shader);
182
/* gfx10_shader_ngg.c */
183
LLVMValueRef gfx10_get_thread_id_in_tg(struct si_shader_context *ctx);
184
bool gfx10_ngg_export_prim_early(struct si_shader *shader);
185
void gfx10_ngg_build_sendmsg_gs_alloc_req(struct si_shader_context *ctx);
186
void gfx10_ngg_build_export_prim(struct si_shader_context *ctx, LLVMValueRef user_edgeflags[3],
187
LLVMValueRef prim_passthrough);
188
void gfx10_emit_ngg_culling_epilogue(struct ac_shader_abi *abi);
189
void gfx10_emit_ngg_epilogue(struct ac_shader_abi *abi);
190
void gfx10_ngg_gs_emit_vertex(struct si_shader_context *ctx, unsigned stream, LLVMValueRef *addrs);
191
void gfx10_ngg_gs_emit_prologue(struct si_shader_context *ctx);
192
void gfx10_ngg_gs_emit_epilogue(struct si_shader_context *ctx);
193
unsigned gfx10_ngg_get_scratch_dw_size(struct si_shader *shader);
194
bool gfx10_ngg_calculate_subgroup_info(struct si_shader *shader);
196
/* si_shader_llvm.c */
197
bool si_compile_llvm(struct si_screen *sscreen, struct si_shader_binary *binary,
198
struct ac_shader_config *conf, struct ac_llvm_compiler *compiler,
199
struct ac_llvm_context *ac, struct util_debug_callback *debug,
200
gl_shader_stage stage, const char *name, bool less_optimized);
201
void si_llvm_context_init(struct si_shader_context *ctx, struct si_screen *sscreen,
202
struct ac_llvm_compiler *compiler, unsigned wave_size);
203
void si_llvm_create_func(struct si_shader_context *ctx, const char *name, LLVMTypeRef *return_types,
204
unsigned num_return_elems, unsigned max_workgroup_size);
205
void si_llvm_create_main_func(struct si_shader_context *ctx, bool ngg_cull_shader);
206
void si_llvm_optimize_module(struct si_shader_context *ctx);
207
void si_llvm_dispose(struct si_shader_context *ctx);
208
LLVMValueRef si_buffer_load_const(struct si_shader_context *ctx, LLVMValueRef resource,
209
LLVMValueRef offset);
210
void si_llvm_build_ret(struct si_shader_context *ctx, LLVMValueRef ret);
211
LLVMValueRef si_insert_input_ret(struct si_shader_context *ctx, LLVMValueRef ret,
212
struct ac_arg param, unsigned return_index);
213
LLVMValueRef si_insert_input_ret_float(struct si_shader_context *ctx, LLVMValueRef ret,
214
struct ac_arg param, unsigned return_index);
215
LLVMValueRef si_insert_input_ptr(struct si_shader_context *ctx, LLVMValueRef ret,
216
struct ac_arg param, unsigned return_index);
217
LLVMValueRef si_prolog_get_internal_bindings(struct si_shader_context *ctx);
218
void si_llvm_emit_barrier(struct si_shader_context *ctx);
219
void si_llvm_declare_esgs_ring(struct si_shader_context *ctx);
220
LLVMValueRef si_unpack_param(struct si_shader_context *ctx, struct ac_arg param, unsigned rshift,
222
LLVMValueRef si_get_primitive_id(struct si_shader_context *ctx, unsigned swizzle);
223
void si_build_wrapper_function(struct si_shader_context *ctx, LLVMValueRef *parts,
224
unsigned num_parts, unsigned main_part,
225
unsigned next_shader_first_part, bool same_thread_count);
226
bool si_llvm_translate_nir(struct si_shader_context *ctx, struct si_shader *shader,
227
struct nir_shader *nir, bool free_nir, bool ngg_cull_shader);
228
bool si_llvm_compile_shader(struct si_screen *sscreen, struct ac_llvm_compiler *compiler,
229
struct si_shader *shader, struct util_debug_callback *debug,
230
struct nir_shader *nir, bool free_nir);
232
/* si_shader_llvm_gs.c */
233
LLVMValueRef si_is_es_thread(struct si_shader_context *ctx);
234
LLVMValueRef si_is_gs_thread(struct si_shader_context *ctx);
235
void si_llvm_emit_es_epilogue(struct ac_shader_abi *abi);
236
void si_preload_esgs_ring(struct si_shader_context *ctx);
237
void si_preload_gs_rings(struct si_shader_context *ctx);
238
void si_llvm_init_gs_callbacks(struct si_shader_context *ctx);
240
/* si_shader_llvm_tess.c */
241
void si_llvm_preload_tes_rings(struct si_shader_context *ctx);
242
void si_llvm_emit_ls_epilogue(struct ac_shader_abi *abi);
243
void si_llvm_build_tcs_epilog(struct si_shader_context *ctx, union si_shader_part_key *key);
244
void si_llvm_init_tcs_callbacks(struct si_shader_context *ctx);
245
void si_llvm_init_tes_callbacks(struct si_shader_context *ctx, bool ngg_cull_shader);
247
/* si_shader_llvm_ps.c */
248
LLVMValueRef si_get_sample_id(struct si_shader_context *ctx);
249
void si_llvm_build_ps_prolog(struct si_shader_context *ctx, union si_shader_part_key *key);
250
void si_llvm_build_ps_epilog(struct si_shader_context *ctx, union si_shader_part_key *key);
251
void si_llvm_build_monolithic_ps(struct si_shader_context *ctx, struct si_shader *shader);
252
void si_llvm_init_ps_callbacks(struct si_shader_context *ctx);
254
/* si_shader_llvm_resources.c */
255
void si_llvm_init_resource_callbacks(struct si_shader_context *ctx);
257
/* si_shader_llvm_vs.c */
258
void si_llvm_clipvertex_to_clipdist(struct si_shader_context *ctx,
259
struct ac_export_args clipdist[2], LLVMValueRef clipvertex[4]);
260
void 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);
264
void si_llvm_emit_streamout(struct si_shader_context *ctx, struct si_shader_output_values *outputs,
265
unsigned noutput, unsigned stream);
266
void si_llvm_build_vs_exports(struct si_shader_context *ctx,
267
struct si_shader_output_values *outputs, unsigned noutput);
268
void si_llvm_emit_vs_epilogue(struct ac_shader_abi *abi);
269
void si_llvm_build_vs_prolog(struct si_shader_context *ctx, union si_shader_part_key *key);
270
void si_llvm_init_vs_callbacks(struct si_shader_context *ctx, bool ngg_cull_shader);