2
* Copyright (C) 2015-2018 Rob Clark <robclark@freedesktop.org>
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
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
8
* and/or sell copies of the Software, and to permit persons to whom the
9
* 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 NONINFRINGEMENT. IN NO EVENT SHALL
18
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24
* Rob Clark <robclark@freedesktop.org>
27
#ifndef IR3_CONTEXT_H_
28
#define IR3_CONTEXT_H_
31
#include "ir3_compiler.h"
34
/* for conditionally setting boolean flag(s): */
35
#define COND(bool, val) ((bool) ? (val) : 0)
37
#define DBG(fmt, ...) \
39
mesa_logd("%s:%d: " fmt, __FUNCTION__, __LINE__, ##__VA_ARGS__); \
43
* The context for compilation of a single shader.
46
struct ir3_compiler *compiler;
47
const struct ir3_context_funcs *funcs;
51
struct nir_instr *cur_instr; /* current instruction, just for debug */
54
struct ir3_shader_variant *so;
56
/* Tables of scalar inputs/outputs. Because of the way varying packing
57
* works, we could have inputs w/ fractional location, which is a bit
58
* awkward to deal with unless we keep track of the split scalar in/
61
* These *only* have inputs/outputs that are touched by load_*input and
64
unsigned ninputs, noutputs;
65
struct ir3_instruction **inputs;
66
struct ir3_instruction **outputs;
68
struct ir3_block *block; /* the current block */
69
struct ir3_block *in_block; /* block created for shader inputs */
71
nir_function_impl *impl;
73
/* For fragment shaders, varyings are not actual shader inputs,
74
* instead the hw passes a ij coord which is used with
77
* But NIR doesn't know that, it still declares varyings as
78
* inputs. So we do all the input tracking normally and fix
79
* things up after compile_instructions()
81
struct ir3_instruction *ij[IJ_COUNT];
83
/* for fragment shaders, for gl_FrontFacing and gl_FragCoord: */
84
struct ir3_instruction *frag_face, *frag_coord;
86
/* For vertex shaders, keep track of the system values sources */
87
struct ir3_instruction *vertex_id, *basevertex, *instance_id, *base_instance,
88
*draw_id, *view_index;
90
/* For fragment shaders: */
91
struct ir3_instruction *samp_id, *samp_mask_in;
93
/* For geometry shaders: */
94
struct ir3_instruction *primitive_id;
95
struct ir3_instruction *gs_header;
97
/* For tessellation shaders: */
98
struct ir3_instruction *patch_vertices_in;
99
struct ir3_instruction *tcs_header;
100
struct ir3_instruction *tess_coord;
101
struct ir3_instruction *rel_patch_id;
103
/* Compute shader inputs: */
104
struct ir3_instruction *local_invocation_id, *work_group_id;
106
/* mapping from nir_register to defining instruction: */
107
struct hash_table *def_ht;
111
/* Tracking for max level of flowcontrol (branchstack) needed
114
unsigned stack, max_stack;
119
/* a common pattern for indirect addressing is to request the
120
* same address register multiple times. To avoid generating
121
* duplicate instruction sequences (which our backend does not
122
* try to clean up, since that should be done as the NIR stage)
123
* we cache the address value generated for a given src value:
125
* Note that we have to cache these per alignment, since same
126
* src used for an array of vec1 cannot be also used for an
129
struct hash_table *addr0_ht[4];
131
/* The same for a1.x. We only support immediate values for a1.x, as this
132
* is the only use so far.
134
struct hash_table_u64 *addr1_ht;
136
struct hash_table *sel_cond_conversions;
138
/* last dst array, for indirect we need to insert a var-store.
140
struct ir3_instruction **last_dst;
143
/* maps nir_block to ir3_block, mostly for the purposes of
144
* figuring out the blocks successors
146
struct hash_table *block_ht;
148
/* maps nir_block at the top of a loop to ir3_block collecting continue
151
struct hash_table *continue_block_ht;
153
/* on a4xx, bitmask of samplers which need astc+srgb workaround: */
156
/* on a4xx, per-sampler per-component swizzles, for tg4: */
157
uint16_t sampler_swizzles[16];
159
unsigned samples; /* bitmask of x,y sample shifts */
161
unsigned max_texture_index;
163
unsigned prefetch_limit;
165
/* set if we encounter something we can't handle yet, so we
166
* can bail cleanly and fallback to TGSI compiler f/e
171
struct ir3_context_funcs {
172
void (*emit_intrinsic_load_ssbo)(struct ir3_context *ctx,
173
nir_intrinsic_instr *intr,
174
struct ir3_instruction **dst);
175
void (*emit_intrinsic_store_ssbo)(struct ir3_context *ctx,
176
nir_intrinsic_instr *intr);
177
struct ir3_instruction *(*emit_intrinsic_atomic_ssbo)(
178
struct ir3_context *ctx, nir_intrinsic_instr *intr);
179
void (*emit_intrinsic_load_image)(struct ir3_context *ctx,
180
nir_intrinsic_instr *intr,
181
struct ir3_instruction **dst);
182
void (*emit_intrinsic_store_image)(struct ir3_context *ctx,
183
nir_intrinsic_instr *intr);
184
struct ir3_instruction *(*emit_intrinsic_atomic_image)(
185
struct ir3_context *ctx, nir_intrinsic_instr *intr);
186
void (*emit_intrinsic_image_size)(struct ir3_context *ctx,
187
nir_intrinsic_instr *intr,
188
struct ir3_instruction **dst);
189
void (*emit_intrinsic_load_global_ir3)(struct ir3_context *ctx,
190
nir_intrinsic_instr *intr,
191
struct ir3_instruction **dst);
192
void (*emit_intrinsic_store_global_ir3)(struct ir3_context *ctx,
193
nir_intrinsic_instr *intr);
194
struct ir3_instruction *(*emit_intrinsic_atomic_global)(
195
struct ir3_context *ctx, nir_intrinsic_instr *intr);
198
extern const struct ir3_context_funcs ir3_a4xx_funcs;
199
extern const struct ir3_context_funcs ir3_a6xx_funcs;
201
struct ir3_context *ir3_context_init(struct ir3_compiler *compiler,
202
struct ir3_shader_variant *so);
203
void ir3_context_free(struct ir3_context *ctx);
205
struct ir3_instruction **ir3_get_dst_ssa(struct ir3_context *ctx,
206
nir_ssa_def *dst, unsigned n);
207
struct ir3_instruction **ir3_get_dst(struct ir3_context *ctx, nir_dest *dst,
209
struct ir3_instruction *const *ir3_get_src(struct ir3_context *ctx,
211
void ir3_put_dst(struct ir3_context *ctx, nir_dest *dst);
212
struct ir3_instruction *ir3_create_collect(struct ir3_block *block,
213
struct ir3_instruction *const *arr,
215
void ir3_split_dest(struct ir3_block *block, struct ir3_instruction **dst,
216
struct ir3_instruction *src, unsigned base, unsigned n);
217
void ir3_handle_bindless_cat6(struct ir3_instruction *instr, nir_src rsrc);
218
void ir3_handle_nonuniform(struct ir3_instruction *instr,
219
nir_intrinsic_instr *intrin);
220
void emit_intrinsic_image_size_tex(struct ir3_context *ctx,
221
nir_intrinsic_instr *intr,
222
struct ir3_instruction **dst);
224
#define ir3_collect(block, ...) \
226
struct ir3_instruction *__arr[] = {__VA_ARGS__}; \
227
ir3_create_collect(block, __arr, ARRAY_SIZE(__arr)); \
230
NORETURN void ir3_context_error(struct ir3_context *ctx, const char *format,
233
#define compile_assert(ctx, cond) \
236
ir3_context_error((ctx), "failed assert: " #cond "\n"); \
239
struct ir3_instruction *ir3_get_addr0(struct ir3_context *ctx,
240
struct ir3_instruction *src, int align);
241
struct ir3_instruction *ir3_get_addr1(struct ir3_context *ctx,
243
struct ir3_instruction *ir3_get_predicate(struct ir3_context *ctx,
244
struct ir3_instruction *src);
246
void ir3_declare_array(struct ir3_context *ctx, nir_register *reg);
247
struct ir3_array *ir3_get_array(struct ir3_context *ctx, nir_register *reg);
248
struct ir3_instruction *ir3_create_array_load(struct ir3_context *ctx,
249
struct ir3_array *arr, int n,
250
struct ir3_instruction *address);
251
void ir3_create_array_store(struct ir3_context *ctx, struct ir3_array *arr,
252
int n, struct ir3_instruction *src,
253
struct ir3_instruction *address);
256
utype_for_size(unsigned bit_size)
266
unreachable("bad bitsize");
272
utype_src(nir_src src)
274
return utype_for_size(nir_src_bit_size(src));
278
utype_dst(nir_dest dst)
280
return utype_for_size(nir_dest_bit_size(dst));
284
* Convert nir bitsize to ir3 bitsize, handling the special case of 1b bools
285
* which can be 16b or 32b depending on gen.
287
static inline unsigned
288
ir3_bitsize(struct ir3_context *ctx, unsigned nir_bitsize)
290
if (nir_bitsize == 1)
291
return type_size(ctx->compiler->bool_type);
295
#endif /* IR3_CONTEXT_H_ */