2 * Copyright © 2014 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 "brw_shader.h"
26 #include "glsl/glsl_parser_extras.h"
27 #include "glsl/nir/glsl_to_nir.h"
28 #include "program/prog_to_nir.h"
31 remap_vs_attrs(nir_block
*block
, void *closure
)
33 GLbitfield64 inputs_read
= *((GLbitfield64
*) closure
);
35 nir_foreach_instr(block
, instr
) {
36 if (instr
->type
!= nir_instr_type_intrinsic
)
39 nir_intrinsic_instr
*intrin
= nir_instr_as_intrinsic(instr
);
41 /* We set EmitNoIndirect for VS inputs, so there are no indirects. */
42 assert(intrin
->intrinsic
!= nir_intrinsic_load_input_indirect
);
44 if (intrin
->intrinsic
== nir_intrinsic_load_input
) {
45 /* Attributes come in a contiguous block, ordered by their
46 * gl_vert_attrib value. That means we can compute the slot
47 * number for an attribute by masking out the enabled attributes
48 * before it and counting the bits.
50 int attr
= intrin
->const_index
[0];
51 int slot
= _mesa_bitcount_64(inputs_read
& BITFIELD64_MASK(attr
));
52 intrin
->const_index
[0] = 4 * slot
;
59 brw_nir_lower_inputs(const struct brw_device_info
*devinfo
,
60 nir_shader
*nir
, bool is_scalar
)
63 case MESA_SHADER_VERTEX
:
64 /* For now, leave the vec4 backend doing the old method. */
66 nir_assign_var_locations(&nir
->inputs
, &nir
->num_inputs
,
71 /* Start with the location of the variable's base. */
72 foreach_list_typed(nir_variable
, var
, node
, &nir
->inputs
) {
73 var
->data
.driver_location
= var
->data
.location
;
76 /* Now use nir_lower_io to walk dereference chains. Attribute arrays
77 * are loaded as one vec4 per element (or matrix column), so we use
78 * type_size_vec4 here.
80 nir_lower_io(nir
, nir_var_shader_in
, type_size_vec4
);
82 /* Finally, translate VERT_ATTRIB_* values into the actual registers.
84 * Note that we can use nir->info.inputs_read instead of key->inputs_read
85 * since the two are identical aside from Gen4-5 edge flag differences.
87 GLbitfield64 inputs_read
= nir
->info
.inputs_read
;
88 nir_foreach_overload(nir
, overload
) {
90 nir_foreach_block(overload
->impl
, remap_vs_attrs
, &inputs_read
);
94 case MESA_SHADER_GEOMETRY
: {
96 foreach_list_typed(nir_variable
, var
, node
, &nir
->inputs
) {
97 var
->data
.driver_location
= var
->data
.location
;
100 /* The GLSL linker will have already matched up GS inputs and
101 * the outputs of prior stages. The driver does extend VS outputs
102 * in some cases, but only for legacy OpenGL or Gen4-5 hardware,
103 * neither of which offer geometry shader support. So we can
104 * safely ignore that.
106 * For SSO pipelines, we use a fixed VUE map layout based on variable
107 * locations, so we can rely on rendezvous-by-location to make this
110 * However, we need to ignore VARYING_SLOT_PRIMITIVE_ID, as it's not
111 * written by previous stages and shows up via payload magic.
113 struct brw_vue_map input_vue_map
;
114 GLbitfield64 inputs_read
=
115 nir
->info
.inputs_read
& ~VARYING_BIT_PRIMITIVE_ID
;
116 brw_compute_vue_map(devinfo
, &input_vue_map
, inputs_read
,
117 nir
->info
.separate_shader
);
119 /* Start with the slot for the variable's base. */
120 foreach_list_typed(nir_variable
, var
, node
, &nir
->inputs
) {
121 assert(input_vue_map
.varying_to_slot
[var
->data
.location
] != -1);
122 var
->data
.driver_location
=
123 input_vue_map
.varying_to_slot
[var
->data
.location
];
126 /* Inputs are stored in vec4 slots, so use type_size_vec4(). */
127 nir_lower_io(nir
, nir_var_shader_in
, type_size_vec4
);
131 case MESA_SHADER_FRAGMENT
:
133 nir_assign_var_locations(&nir
->inputs
, &nir
->num_inputs
,
136 case MESA_SHADER_COMPUTE
:
137 /* Compute shaders have no inputs. */
138 assert(exec_list_is_empty(&nir
->inputs
));
141 unreachable("unsupported shader stage");
146 brw_nir_lower_outputs(nir_shader
*nir
, bool is_scalar
)
148 switch (nir
->stage
) {
149 case MESA_SHADER_VERTEX
:
150 case MESA_SHADER_GEOMETRY
:
152 nir_assign_var_locations(&nir
->outputs
, &nir
->num_outputs
,
153 type_size_vec4_times_4
);
154 nir_lower_io(nir
, nir_var_shader_out
, type_size_vec4_times_4
);
156 nir_foreach_variable(var
, &nir
->outputs
)
157 var
->data
.driver_location
= var
->data
.location
;
160 case MESA_SHADER_FRAGMENT
:
161 nir_assign_var_locations(&nir
->outputs
, &nir
->num_outputs
,
164 case MESA_SHADER_COMPUTE
:
165 /* Compute shaders have no outputs. */
166 assert(exec_list_is_empty(&nir
->outputs
));
169 unreachable("unsupported shader stage");
174 nir_optimize(nir_shader
*nir
, bool is_scalar
)
179 nir_lower_vars_to_ssa(nir
);
180 nir_validate_shader(nir
);
183 nir_lower_alu_to_scalar(nir
);
184 nir_validate_shader(nir
);
187 progress
|= nir_copy_prop(nir
);
188 nir_validate_shader(nir
);
191 nir_lower_phis_to_scalar(nir
);
192 nir_validate_shader(nir
);
195 progress
|= nir_copy_prop(nir
);
196 nir_validate_shader(nir
);
197 progress
|= nir_opt_dce(nir
);
198 nir_validate_shader(nir
);
199 progress
|= nir_opt_cse(nir
);
200 nir_validate_shader(nir
);
201 progress
|= nir_opt_peephole_select(nir
);
202 nir_validate_shader(nir
);
203 progress
|= nir_opt_algebraic(nir
);
204 nir_validate_shader(nir
);
205 progress
|= nir_opt_constant_folding(nir
);
206 nir_validate_shader(nir
);
207 progress
|= nir_opt_dead_cf(nir
);
208 nir_validate_shader(nir
);
209 progress
|= nir_opt_remove_phis(nir
);
210 nir_validate_shader(nir
);
211 progress
|= nir_opt_undef(nir
);
212 nir_validate_shader(nir
);
217 brw_create_nir(struct brw_context
*brw
,
218 const struct gl_shader_program
*shader_prog
,
219 const struct gl_program
*prog
,
220 gl_shader_stage stage
,
223 struct gl_context
*ctx
= &brw
->ctx
;
224 const struct brw_device_info
*devinfo
= brw
->intelScreen
->devinfo
;
225 const nir_shader_compiler_options
*options
=
226 ctx
->Const
.ShaderCompilerOptions
[stage
].NirOptions
;
227 static const nir_lower_tex_options tex_options
= {
230 bool debug_enabled
= INTEL_DEBUG
& intel_debug_flag_for_shader_stage(stage
);
233 /* First, lower the GLSL IR or Mesa IR to NIR */
235 nir
= glsl_to_nir(shader_prog
, stage
, options
);
237 nir
= prog_to_nir(prog
, options
);
238 nir_convert_to_ssa(nir
); /* turn registers into SSA */
240 nir_validate_shader(nir
);
242 if (stage
== MESA_SHADER_GEOMETRY
) {
243 nir_lower_gs_intrinsics(nir
);
244 nir_validate_shader(nir
);
247 nir_lower_global_vars_to_local(nir
);
248 nir_validate_shader(nir
);
250 nir_lower_tex(nir
, &tex_options
);
251 nir_validate_shader(nir
);
253 nir_normalize_cubemap_coords(nir
);
254 nir_validate_shader(nir
);
256 nir_split_var_copies(nir
);
257 nir_validate_shader(nir
);
259 nir_optimize(nir
, is_scalar
);
261 /* Lower a bunch of stuff */
262 nir_lower_var_copies(nir
);
263 nir_validate_shader(nir
);
265 /* Get rid of split copies */
266 nir_optimize(nir
, is_scalar
);
268 brw_nir_lower_inputs(devinfo
, nir
, is_scalar
);
269 brw_nir_lower_outputs(nir
, is_scalar
);
270 nir_assign_var_locations(&nir
->uniforms
,
272 is_scalar
? type_size_scalar
: type_size_vec4
);
273 nir_lower_io(nir
, -1, is_scalar
? type_size_scalar
: type_size_vec4
);
274 nir_validate_shader(nir
);
276 nir_remove_dead_variables(nir
);
277 nir_validate_shader(nir
);
280 nir_lower_samplers(nir
, shader_prog
);
281 nir_validate_shader(nir
);
284 nir_lower_system_values(nir
);
285 nir_validate_shader(nir
);
288 nir_lower_atomics(nir
, shader_prog
);
289 nir_validate_shader(nir
);
292 nir_optimize(nir
, is_scalar
);
295 /* Try and fuse multiply-adds */
296 brw_nir_opt_peephole_ffma(nir
);
297 nir_validate_shader(nir
);
300 nir_opt_algebraic_late(nir
);
301 nir_validate_shader(nir
);
303 nir_lower_locals_to_regs(nir
);
304 nir_validate_shader(nir
);
306 nir_lower_to_source_mods(nir
);
307 nir_validate_shader(nir
);
309 nir_validate_shader(nir
);
311 nir_validate_shader(nir
);
313 if (unlikely(debug_enabled
)) {
314 /* Re-index SSA defs so we print more sensible numbers. */
315 nir_foreach_overload(nir
, overload
) {
317 nir_index_ssa_defs(overload
->impl
);
320 fprintf(stderr
, "NIR (SSA form) for %s shader:\n",
321 _mesa_shader_stage_to_string(stage
));
322 nir_print_shader(nir
, stderr
);
325 nir_convert_from_ssa(nir
, true);
326 nir_validate_shader(nir
);
329 nir_move_vec_src_uses_to_dest(nir
);
330 nir_validate_shader(nir
);
332 nir_lower_vec_to_movs(nir
);
333 nir_validate_shader(nir
);
336 /* This is the last pass we run before we start emitting stuff. It
337 * determines when we need to insert boolean resolves on Gen <= 5. We
338 * run it last because it stashes data in instr->pass_flags and we don't
339 * want that to be squashed by other NIR passes.
342 brw_nir_analyze_boolean_resolves(nir
);
346 if (unlikely(debug_enabled
)) {
347 fprintf(stderr
, "NIR (final form) for %s shader:\n",
348 _mesa_shader_stage_to_string(stage
));
349 nir_print_shader(nir
, stderr
);
356 brw_type_for_nir_type(nir_alu_type type
)
359 case nir_type_unsigned
:
360 return BRW_REGISTER_TYPE_UD
;
363 return BRW_REGISTER_TYPE_D
;
365 return BRW_REGISTER_TYPE_F
;
367 unreachable("unknown type");
370 return BRW_REGISTER_TYPE_F
;
373 /* Returns the glsl_base_type corresponding to a nir_alu_type.
374 * This is used by both brw_vec4_nir and brw_fs_nir.
377 brw_glsl_base_type_for_nir_type(nir_alu_type type
)
381 return GLSL_TYPE_FLOAT
;
384 return GLSL_TYPE_INT
;
386 case nir_type_unsigned
:
387 return GLSL_TYPE_UINT
;
390 unreachable("bad type");