2 * Copyright © 2015 Intel Corporation
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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
25 #include "nir_builder.h"
28 * \file nir_lower_gs_intrinsics.c
30 * Geometry Shaders can call EmitVertex()/EmitStreamVertex() to output an
31 * arbitrary number of vertices. However, the shader must declare the maximum
32 * number of vertices that it will ever output - further attempts to emit
33 * vertices result in undefined behavior according to the GLSL specification.
35 * Drivers might use this maximum number of vertices to allocate enough space
36 * to hold the geometry shader's output. Some drivers (such as i965) need to
37 * implement "safety checks" which ensure that the shader hasn't emitted too
38 * many vertices, to avoid overflowing that space and trashing other memory.
40 * The count of emitted vertices can also be useful in buffer offset
41 * calculations, so drivers know where to write the GS output.
43 * However, for simple geometry shaders that emit a statically determinable
44 * number of vertices, this extra bookkeeping is unnecessary and inefficient.
45 * By tracking the vertex count in NIR, we allow constant folding/propagation
46 * and dead control flow optimizations to eliminate most of it where possible.
48 * This pass introduces a new global variable which stores the current vertex
49 * count (initialized to 0), and converts emit_vertex/end_primitive intrinsics
50 * to their *_with_counter variants. emit_vertex is also wrapped in a safety
51 * check to avoid buffer overflows. Finally, it adds a set_vertex_count
52 * intrinsic at the end of the program, informing the driver of the final
58 nir_variable
*vertex_count_var
;
63 * Replace emit_vertex intrinsics with:
65 * if (vertex_count < max_vertices) {
66 * emit_vertex_with_counter vertex_count ...
71 rewrite_emit_vertex(nir_intrinsic_instr
*intrin
, struct state
*state
)
73 nir_builder
*b
= state
->builder
;
75 /* Load the vertex count */
76 b
->cursor
= nir_before_instr(&intrin
->instr
);
77 nir_ssa_def
*count
= nir_load_var(b
, state
->vertex_count_var
);
79 nir_ssa_def
*max_vertices
= nir_imm_int(b
, b
->shader
->info
.gs
.vertices_out
);
81 /* Create: if (vertex_count < max_vertices) and insert it.
83 * The new if statement needs to be hooked up to the control flow graph
84 * before we start inserting instructions into it.
86 nir_if
*if_stmt
= nir_if_create(b
->shader
);
87 if_stmt
->condition
= nir_src_for_ssa(nir_ilt(b
, count
, max_vertices
));
88 nir_builder_cf_insert(b
, &if_stmt
->cf_node
);
90 /* Fill out the new then-block */
91 b
->cursor
= nir_after_cf_list(&if_stmt
->then_list
);
93 nir_intrinsic_instr
*lowered
=
94 nir_intrinsic_instr_create(b
->shader
,
95 nir_intrinsic_emit_vertex_with_counter
);
96 nir_intrinsic_set_stream_id(lowered
, nir_intrinsic_stream_id(intrin
));
97 lowered
->src
[0] = nir_src_for_ssa(count
);
98 nir_builder_instr_insert(b
, &lowered
->instr
);
100 /* Increment the vertex count by 1 */
101 nir_store_var(b
, state
->vertex_count_var
,
102 nir_iadd(b
, count
, nir_imm_int(b
, 1)),
105 nir_instr_remove(&intrin
->instr
);
107 state
->progress
= true;
111 * Replace end_primitive with end_primitive_with_counter.
114 rewrite_end_primitive(nir_intrinsic_instr
*intrin
, struct state
*state
)
116 nir_builder
*b
= state
->builder
;
118 b
->cursor
= nir_before_instr(&intrin
->instr
);
119 nir_ssa_def
*count
= nir_load_var(b
, state
->vertex_count_var
);
121 nir_intrinsic_instr
*lowered
=
122 nir_intrinsic_instr_create(b
->shader
,
123 nir_intrinsic_end_primitive_with_counter
);
124 nir_intrinsic_set_stream_id(lowered
, nir_intrinsic_stream_id(intrin
));
125 lowered
->src
[0] = nir_src_for_ssa(count
);
126 nir_builder_instr_insert(b
, &lowered
->instr
);
128 nir_instr_remove(&intrin
->instr
);
130 state
->progress
= true;
134 rewrite_intrinsics(nir_block
*block
, void *closure
)
136 struct state
*state
= closure
;
138 nir_foreach_instr_safe(block
, instr
) {
139 if (instr
->type
!= nir_instr_type_intrinsic
)
142 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
143 switch (intrin
->intrinsic
) {
144 case nir_intrinsic_emit_vertex
:
145 rewrite_emit_vertex(intrin
, state
);
147 case nir_intrinsic_end_primitive
:
148 rewrite_end_primitive(intrin
, state
);
151 /* not interesting; skip this */
160 * Add a set_vertex_count intrinsic at the end of the program
161 * (representing the final vertex count).
164 append_set_vertex_count(nir_block
*end_block
, struct state
*state
)
166 nir_builder
*b
= state
->builder
;
167 nir_shader
*shader
= state
->builder
->shader
;
169 /* Insert the new intrinsic in all of the predecessors of the end block,
170 * but before any jump instructions (return).
172 struct set_entry
*entry
;
173 set_foreach(end_block
->predecessors
, entry
) {
174 nir_block
*pred
= (nir_block
*) entry
->key
;
175 b
->cursor
= nir_after_block_before_jump(pred
);
177 nir_ssa_def
*count
= nir_load_var(b
, state
->vertex_count_var
);
179 nir_intrinsic_instr
*set_vertex_count
=
180 nir_intrinsic_instr_create(shader
, nir_intrinsic_set_vertex_count
);
181 set_vertex_count
->src
[0] = nir_src_for_ssa(count
);
183 nir_builder_instr_insert(b
, &set_vertex_count
->instr
);
188 nir_lower_gs_intrinsics(nir_shader
*shader
)
191 state
.progress
= false;
193 /* Create the counter variable */
194 nir_variable
*var
= rzalloc(shader
, nir_variable
);
195 var
->data
.mode
= nir_var_global
;
196 var
->type
= glsl_uint_type();
197 var
->name
= "vertex_count";
198 var
->constant_initializer
= rzalloc(shader
, nir_constant
); /* initialize to 0 */
200 exec_list_push_tail(&shader
->globals
, &var
->node
);
201 state
.vertex_count_var
= var
;
203 nir_foreach_function(shader
, function
) {
204 if (function
->impl
) {
206 nir_builder_init(&b
, function
->impl
);
209 nir_foreach_block(function
->impl
, rewrite_intrinsics
, &state
);
211 /* This only works because we have a single main() function. */
212 append_set_vertex_count(function
->impl
->end_block
, &state
);
214 nir_metadata_preserve(function
->impl
, 0);
218 return state
.progress
;