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 "si_shader.h"
29 #include "gallivm/lp_bld_flow.h"
30 #include "gallivm/lp_bld_init.h"
31 #include "gallivm/lp_bld_tgsi.h"
32 #include "tgsi/tgsi_parse.h"
33 #include "ac_shader_abi.h"
35 #include <llvm-c/Core.h>
36 #include <llvm-c/TargetMachine.h>
38 struct pipe_debug_callback
;
39 struct ac_shader_binary
;
41 #define RADEON_LLVM_MAX_INPUT_SLOTS 32
42 #define RADEON_LLVM_MAX_INPUTS 32 * 4
43 #define RADEON_LLVM_MAX_OUTPUTS 32 * 4
45 #define RADEON_LLVM_MAX_SYSTEM_VALUES 11
46 #define RADEON_LLVM_MAX_ADDRS 16
48 struct si_shader_context
{
49 struct lp_build_tgsi_context bld_base
;
50 struct gallivm_state gallivm
;
51 struct ac_llvm_context ac
;
52 struct si_shader
*shader
;
53 struct si_screen
*screen
;
55 unsigned type
; /* PIPE_SHADER_* specifies the type of shader. */
57 /* For clamping the non-constant index in resource indexing: */
58 unsigned num_const_buffers
;
59 unsigned num_shader_buffers
;
61 unsigned num_samplers
;
63 struct ac_shader_abi abi
;
65 /** This function is responsible for initilizing the inputs array and will be
66 * called once for each input declared in the TGSI shader.
68 void (*load_input
)(struct si_shader_context
*,
70 const struct tgsi_full_declaration
*decl
,
73 /** This array contains the input values for the shader. Typically these
74 * values will be in the form of a target intrinsic that will inform the
75 * backend how to load the actual inputs to the shader.
77 struct tgsi_full_declaration input_decls
[RADEON_LLVM_MAX_INPUT_SLOTS
];
78 LLVMValueRef inputs
[RADEON_LLVM_MAX_INPUTS
];
79 LLVMValueRef outputs
[RADEON_LLVM_MAX_OUTPUTS
][TGSI_NUM_CHANNELS
];
80 LLVMValueRef addrs
[RADEON_LLVM_MAX_ADDRS
][TGSI_NUM_CHANNELS
];
82 /** This pointer is used to contain the temporary values.
83 * The amount of temporary used in tgsi can't be bound to a max value and
84 * thus we must allocate this array at runtime.
88 LLVMValueRef system_values
[RADEON_LLVM_MAX_SYSTEM_VALUES
];
93 struct lp_build_if_state merged_wrap_if_state
;
95 struct tgsi_array_info
*temp_arrays
;
96 LLVMValueRef
*temp_array_allocas
;
98 LLVMValueRef undef_alloca
;
100 LLVMValueRef main_fn
;
101 LLVMTypeRef return_type
;
103 /* Parameter indices for LLVMGetParam. */
104 int param_rw_buffers
;
105 int param_const_and_shader_buffers
;
106 int param_samplers_and_images
;
107 int param_bindless_samplers_and_images
;
108 /* Common inputs for merged shaders. */
109 int param_merged_wave_info
;
110 int param_merged_scratch_offset
;
112 int param_vertex_buffers
;
113 int param_rel_auto_id
;
114 int param_vs_prim_id
;
115 int param_vertex_index0
;
116 /* VS states and layout of LS outputs / TCS inputs at the end
117 * [0] = clamp vertex color
119 * [8:20] = stride between patches in DW = num_inputs * num_vertices * 4
120 * max = 32*32*4 + 32*4
121 * [24:31] = stride between vertices in DW = num_inputs * 4
124 int param_vs_state_bits
;
125 int param_vs_blit_inputs
;
127 int param_streamout_config
;
128 int param_streamout_write_index
;
129 int param_streamout_offset
[4];
132 /* Layout of TCS outputs in the offchip buffer
134 * [0:5] = the number of patches per threadgroup, max = NUM_PATCHES (40)
136 * [6:11] = the number of output vertices per patch, max = 32
138 * [12:31] = the offset of per patch attributes in the buffer in bytes.
139 * max = NUM_PATCHES*32*32*16
141 int param_tcs_offchip_layout
;
144 /* Offsets where TCS outputs and TCS patch outputs live in LDS:
145 * [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
146 * [16:31] = TCS output patch0 offset for per-patch / 16
147 * max = (NUM_PATCHES + 1) * 32*32
149 int param_tcs_out_lds_offsets
;
150 /* Layout of TCS outputs / TES inputs:
151 * [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
152 * max = 32*32*4 + 32*4
153 * [13:18] = gl_PatchVerticesIn, max = 32
154 * [19:31] = high 13 bits of the 32-bit address of tessellation ring buffers
156 int param_tcs_out_lds_layout
;
157 int param_tcs_offchip_offset
;
158 int param_tcs_factor_offset
;
161 int param_tes_offchip_addr
;
164 int param_tes_rel_patch_id
;
166 int param_es2gs_offset
;
168 int param_gs2vs_offset
;
169 int param_gs_wave_id
; /* GFX6 */
170 LLVMValueRef gs_vtx_offset
[6]; /* in dwords (GFX6) */
171 int param_gs_vtx01_offset
; /* in dwords (GFX9) */
172 int param_gs_vtx23_offset
; /* in dwords (GFX9) */
173 int param_gs_vtx45_offset
; /* in dwords (GFX9) */
175 int param_block_size
;
177 struct si_compiler
*compiler
;
179 /* Preloaded descriptors. */
180 LLVMValueRef esgs_ring
;
181 LLVMValueRef gsvs_ring
[4];
182 LLVMValueRef tess_offchip_ring
;
184 LLVMValueRef invoc0_tess_factors
[6]; /* outer[4], inner[2] */
185 LLVMValueRef gs_next_vertex
[4];
186 LLVMValueRef postponed_kill
;
187 LLVMValueRef return_value
;
205 static inline struct si_shader_context
*
206 si_shader_context(struct lp_build_tgsi_context
*bld_base
)
208 return (struct si_shader_context
*)bld_base
;
211 static inline struct si_shader_context
*
212 si_shader_context_from_abi(struct ac_shader_abi
*abi
)
214 struct si_shader_context
*ctx
= NULL
;
215 return container_of(abi
, ctx
, abi
);
218 unsigned si_llvm_compile(LLVMModuleRef M
, struct ac_shader_binary
*binary
,
219 struct si_compiler
*compiler
,
220 struct pipe_debug_callback
*debug
);
222 LLVMTypeRef
tgsi2llvmtype(struct lp_build_tgsi_context
*bld_base
,
223 enum tgsi_opcode_type type
);
225 LLVMValueRef
bitcast(struct lp_build_tgsi_context
*bld_base
,
226 enum tgsi_opcode_type type
, LLVMValueRef value
);
228 LLVMValueRef
si_llvm_bound_index(struct si_shader_context
*ctx
,
232 void si_llvm_context_init(struct si_shader_context
*ctx
,
233 struct si_screen
*sscreen
,
234 struct si_compiler
*compiler
);
235 void si_llvm_context_set_tgsi(struct si_shader_context
*ctx
,
236 struct si_shader
*shader
);
238 void si_llvm_create_func(struct si_shader_context
*ctx
,
240 LLVMTypeRef
*return_types
, unsigned num_return_elems
,
241 LLVMTypeRef
*ParamTypes
, unsigned ParamCount
);
243 void si_llvm_dispose(struct si_shader_context
*ctx
);
245 void si_llvm_optimize_module(struct si_shader_context
*ctx
);
247 LLVMValueRef
si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context
*bld_base
,
252 LLVMValueRef
si_llvm_emit_fetch(struct lp_build_tgsi_context
*bld_base
,
253 const struct tgsi_full_src_register
*reg
,
254 enum tgsi_opcode_type type
,
257 void si_llvm_emit_kill(struct ac_shader_abi
*abi
, LLVMValueRef visible
);
259 LLVMValueRef
si_nir_load_input_tes(struct ac_shader_abi
*abi
,
261 LLVMValueRef vertex_index
,
262 LLVMValueRef param_index
,
263 unsigned const_index
,
265 unsigned driver_location
,
267 unsigned num_components
,
272 LLVMValueRef
si_llvm_load_input_gs(struct ac_shader_abi
*abi
,
273 unsigned input_index
,
274 unsigned vtx_offset_param
,
278 LLVMValueRef
si_nir_lookup_interp_param(struct ac_shader_abi
*abi
,
279 enum glsl_interp_mode interp
,
282 void si_llvm_emit_store(struct lp_build_tgsi_context
*bld_base
,
283 const struct tgsi_full_instruction
*inst
,
284 const struct tgsi_opcode_info
*info
,
286 LLVMValueRef dst
[4]);
288 /* Combine these with & instead of |. */
289 #define NOOP_WAITCNT 0xf7f
290 #define LGKM_CNT 0x07f
293 LLVMValueRef
si_get_indirect_index(struct si_shader_context
*ctx
,
294 const struct tgsi_ind_register
*ind
,
295 unsigned addr_mul
, int rel_index
);
296 LLVMValueRef
si_get_bounded_indirect_index(struct si_shader_context
*ctx
,
297 const struct tgsi_ind_register
*ind
,
298 int rel_index
, unsigned num
);
299 LLVMValueRef
si_get_sample_id(struct si_shader_context
*ctx
);
301 void si_shader_context_init_alu(struct lp_build_tgsi_context
*bld_base
);
302 void si_shader_context_init_mem(struct si_shader_context
*ctx
);
304 LLVMValueRef
si_load_sampler_desc(struct si_shader_context
*ctx
,
305 LLVMValueRef list
, LLVMValueRef index
,
306 enum ac_descriptor_type type
);
307 LLVMValueRef
si_load_image_desc(struct si_shader_context
*ctx
,
308 LLVMValueRef list
, LLVMValueRef index
,
309 enum ac_descriptor_type desc_type
, bool dcc_off
);
311 void si_load_system_value(struct si_shader_context
*ctx
,
313 const struct tgsi_full_declaration
*decl
);
314 void si_declare_compute_memory(struct si_shader_context
*ctx
);
315 void si_tgsi_declare_compute_memory(struct si_shader_context
*ctx
,
316 const struct tgsi_full_declaration
*decl
);
318 void si_llvm_load_input_vs(
319 struct si_shader_context
*ctx
,
320 unsigned input_index
,
321 LLVMValueRef out
[4]);
322 void si_llvm_load_input_fs(
323 struct si_shader_context
*ctx
,
324 unsigned input_index
,
325 LLVMValueRef out
[4]);
327 bool si_nir_build_llvm(struct si_shader_context
*ctx
, struct nir_shader
*nir
);
329 LLVMValueRef
si_unpack_param(struct si_shader_context
*ctx
,
330 unsigned param
, unsigned rshift
,