2 * Copyright 2016 Advanced Micro Devices, Inc.
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:
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
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.
24 #ifndef SI_SHADER_PRIVATE_H
25 #define SI_SHADER_PRIVATE_H
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"
36 #include <llvm-c/Core.h>
37 #include <llvm-c/TargetMachine.h>
39 struct pipe_debug_callback
;
40 struct ac_shader_binary
;
42 #define RADEON_LLVM_MAX_INPUT_SLOTS 32
43 #define RADEON_LLVM_MAX_INPUTS 32 * 4
44 #define RADEON_LLVM_MAX_OUTPUTS 32 * 4
46 #define RADEON_LLVM_INITIAL_CF_DEPTH 4
48 #define RADEON_LLVM_MAX_SYSTEM_VALUES 11
49 #define RADEON_LLVM_MAX_ADDRS 16
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
;
60 unsigned type
; /* PIPE_SHADER_* specifies the type of shader. */
62 /* For clamping the non-constant index in resource indexing: */
63 unsigned num_const_buffers
;
64 unsigned num_shader_buffers
;
66 unsigned num_samplers
;
68 /* Whether the prolog will be compiled separately. */
71 struct ac_shader_abi abi
;
73 /** This function is responsible for initilizing the inputs array and will be
74 * called once for each input declared in the TGSI shader.
76 void (*load_input
)(struct si_shader_context
*,
78 const struct tgsi_full_declaration
*decl
,
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.
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
];
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.
96 LLVMValueRef system_values
[RADEON_LLVM_MAX_SYSTEM_VALUES
];
101 struct si_llvm_flow
*flow
;
103 unsigned flow_depth_max
;
105 struct lp_build_if_state merged_wrap_if_state
;
107 struct tgsi_array_info
*temp_arrays
;
108 LLVMValueRef
*temp_array_allocas
;
110 LLVMValueRef undef_alloca
;
112 LLVMValueRef main_fn
;
113 LLVMTypeRef return_type
;
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
;
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
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
136 int param_vs_state_bits
;
137 int param_vs_blit_inputs
;
139 int param_streamout_config
;
140 int param_streamout_write_index
;
141 int param_streamout_offset
[4];
144 /* Layout of TCS outputs in the offchip buffer
146 * [0:5] = the number of patches per threadgroup, max = NUM_PATCHES (40)
148 * [6:11] = the number of output vertices per patch, max = 32
150 * [12:31] = the offset of per patch attributes in the buffer in bytes.
151 * max = NUM_PATCHES*32*32*16
153 int param_tcs_offchip_layout
;
156 /* Offsets where TCS outputs and TCS patch outputs live in LDS:
157 * [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
158 * [16:31] = TCS output patch0 offset for per-patch / 16
159 * max = (NUM_PATCHES + 1) * 32*32
161 int param_tcs_out_lds_offsets
;
162 /* Layout of TCS outputs / TES inputs:
163 * [0:12] = stride between output patches in DW, num_outputs * num_vertices * 4
164 * max = 32*32*4 + 32*4
165 * [26:31] = gl_PatchVerticesIn, max = 32
167 int param_tcs_out_lds_layout
;
168 int param_tcs_offchip_addr_base64k
;
169 int param_tcs_factor_addr_base64k
;
170 int param_tcs_offchip_offset
;
171 int param_tcs_factor_offset
;
172 int param_tcs_patch_id
;
173 int param_tcs_rel_ids
;
178 int param_tes_rel_patch_id
;
179 int param_tes_patch_id
;
181 int param_es2gs_offset
;
183 int param_gs2vs_offset
;
184 int param_gs_wave_id
; /* GFX6 */
185 LLVMValueRef gs_vtx_offset
[6]; /* in dwords (GFX6) */
186 int param_gs_vtx01_offset
; /* in dwords (GFX9) */
187 int param_gs_vtx23_offset
; /* in dwords (GFX9) */
188 int param_gs_vtx45_offset
; /* in dwords (GFX9) */
191 int param_block_size
;
192 int param_block_id
[3];
195 LLVMTargetMachineRef tm
;
197 unsigned range_md_kind
;
198 unsigned fpmath_md_kind
;
199 LLVMValueRef fpmath_md_2p5_ulp
;
201 /* Preloaded descriptors. */
202 LLVMValueRef esgs_ring
;
203 LLVMValueRef gsvs_ring
[4];
205 LLVMValueRef invoc0_tess_factors
[6]; /* outer[4], inner[2] */
206 LLVMValueRef gs_next_vertex
[4];
207 LLVMValueRef postponed_kill
;
208 LLVMValueRef return_value
;
225 LLVMValueRef shared_memory
;
228 static inline struct si_shader_context
*
229 si_shader_context(struct lp_build_tgsi_context
*bld_base
)
231 return (struct si_shader_context
*)bld_base
;
234 static inline struct si_shader_context
*
235 si_shader_context_from_abi(struct ac_shader_abi
*abi
)
237 struct si_shader_context
*ctx
= NULL
;
238 return container_of(abi
, ctx
, abi
);
241 void si_llvm_add_attribute(LLVMValueRef F
, const char *name
, int value
);
243 unsigned si_llvm_compile(LLVMModuleRef M
, struct ac_shader_binary
*binary
,
244 LLVMTargetMachineRef tm
,
245 struct pipe_debug_callback
*debug
);
247 LLVMTypeRef
tgsi2llvmtype(struct lp_build_tgsi_context
*bld_base
,
248 enum tgsi_opcode_type type
);
250 LLVMValueRef
bitcast(struct lp_build_tgsi_context
*bld_base
,
251 enum tgsi_opcode_type type
, LLVMValueRef value
);
253 LLVMValueRef
si_llvm_bound_index(struct si_shader_context
*ctx
,
257 void si_llvm_context_init(struct si_shader_context
*ctx
,
258 struct si_screen
*sscreen
,
259 LLVMTargetMachineRef tm
);
260 void si_llvm_context_set_tgsi(struct si_shader_context
*ctx
,
261 struct si_shader
*shader
);
263 void si_llvm_create_func(struct si_shader_context
*ctx
,
265 LLVMTypeRef
*return_types
, unsigned num_return_elems
,
266 LLVMTypeRef
*ParamTypes
, unsigned ParamCount
);
268 void si_llvm_dispose(struct si_shader_context
*ctx
);
270 void si_llvm_optimize_module(struct si_shader_context
*ctx
);
272 LLVMValueRef
si_llvm_emit_fetch_64bit(struct lp_build_tgsi_context
*bld_base
,
273 enum tgsi_opcode_type type
,
277 LLVMValueRef
si_llvm_emit_fetch(struct lp_build_tgsi_context
*bld_base
,
278 const struct tgsi_full_src_register
*reg
,
279 enum tgsi_opcode_type type
,
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 void si_emit_waitcnt(struct si_shader_context
*ctx
, unsigned simm16
);
295 LLVMValueRef
si_get_indirect_index(struct si_shader_context
*ctx
,
296 const struct tgsi_ind_register
*ind
,
297 unsigned addr_mul
, int rel_index
);
298 LLVMValueRef
si_get_bounded_indirect_index(struct si_shader_context
*ctx
,
299 const struct tgsi_ind_register
*ind
,
300 int rel_index
, unsigned num
);
302 LLVMTypeRef
si_const_array(LLVMTypeRef elem_type
, int num_elements
);
304 void si_shader_context_init_alu(struct lp_build_tgsi_context
*bld_base
);
305 void si_shader_context_init_mem(struct si_shader_context
*ctx
);
307 LLVMValueRef
si_load_sampler_desc(struct si_shader_context
*ctx
,
308 LLVMValueRef list
, LLVMValueRef index
,
309 enum ac_descriptor_type type
);
310 LLVMValueRef
si_load_image_desc(struct si_shader_context
*ctx
,
311 LLVMValueRef list
, LLVMValueRef index
,
312 enum ac_descriptor_type desc_type
, bool dcc_off
);
314 void si_load_system_value(struct si_shader_context
*ctx
,
316 const struct tgsi_full_declaration
*decl
);
317 void si_declare_compute_memory(struct si_shader_context
*ctx
,
318 const struct tgsi_full_declaration
*decl
);
320 void si_llvm_load_input_vs(
321 struct si_shader_context
*ctx
,
322 unsigned input_index
,
323 LLVMValueRef out
[4]);
324 void si_llvm_load_input_fs(
325 struct si_shader_context
*ctx
,
326 unsigned input_index
,
327 LLVMValueRef out
[4]);
329 bool si_nir_build_llvm(struct si_shader_context
*ctx
, struct nir_shader
*nir
);