2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keithw@vmware.com>
33 #include "main/compiler.h"
34 #include "main/context.h"
35 #include "brw_context.h"
38 #include "brw_state.h"
39 #include "program/prog_print.h"
40 #include "program/prog_parameter.h"
42 #include "brw_program.h"
44 #include "util/ralloc.h"
47 brw_vs_outputs_written(struct brw_context
*brw
, struct brw_vs_prog_key
*key
,
48 GLbitfield64 user_varyings
)
50 GLbitfield64 outputs_written
= user_varyings
;
52 if (key
->copy_edgeflag
) {
53 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_EDGE
);
57 /* Put dummy slots into the VUE for the SF to put the replaced
58 * point sprite coords in. We shouldn't need these dummy slots,
59 * which take up precious URB space, but it would mean that the SF
60 * doesn't get nice aligned pairs of input coords into output
61 * coords, which would be a pain to handle.
63 for (unsigned i
= 0; i
< 8; i
++) {
64 if (key
->point_coord_replace
& (1 << i
))
65 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_TEX0
+ i
);
68 /* if back colors are written, allocate slots for front colors too */
69 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC0
))
70 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL0
);
71 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC1
))
72 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL1
);
75 /* In order for legacy clipping to work, we need to populate the clip
76 * distance varying slots whenever clipping is enabled, even if the vertex
77 * shader doesn't write to gl_ClipDistance.
79 if (key
->nr_userclip_plane_consts
> 0) {
80 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0
);
81 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1
);
84 return outputs_written
;
88 brw_codegen_vs_prog(struct brw_context
*brw
,
89 struct gl_shader_program
*prog
,
90 struct brw_program
*vp
,
91 struct brw_vs_prog_key
*key
)
93 const struct brw_compiler
*compiler
= brw
->screen
->compiler
;
94 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
96 const GLuint
*program
;
97 struct brw_vs_prog_data prog_data
;
98 struct brw_stage_prog_data
*stage_prog_data
= &prog_data
.base
.base
;
100 bool start_busy
= false;
101 double start_time
= 0;
103 memset(&prog_data
, 0, sizeof(prog_data
));
105 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
107 stage_prog_data
->use_alt_mode
= true;
109 mem_ctx
= ralloc_context(NULL
);
111 brw_assign_common_binding_table_offsets(MESA_SHADER_VERTEX
, devinfo
, prog
,
112 &vp
->program
, &prog_data
.base
.base
,
115 /* Allocate the references to the uniforms that will end up in the
116 * prog_data associated with the compiled program, and which will be freed
117 * by the state cache.
119 int param_count
= vp
->program
.nir
->num_uniforms
/ 4;
121 prog_data
.base
.base
.nr_image_params
= vp
->program
.info
.num_images
;
123 /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
124 * planes as uniforms.
126 param_count
+= key
->nr_userclip_plane_consts
* 4;
128 stage_prog_data
->param
=
129 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
130 stage_prog_data
->pull_param
=
131 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
132 stage_prog_data
->image_param
=
133 rzalloc_array(NULL
, struct brw_image_param
,
134 stage_prog_data
->nr_image_params
);
135 stage_prog_data
->nr_params
= param_count
;
138 brw_nir_setup_glsl_uniforms(vp
->program
.nir
, prog
, &vp
->program
,
139 &prog_data
.base
.base
,
140 compiler
->scalar_stage
[MESA_SHADER_VERTEX
]);
142 brw_nir_setup_arb_uniforms(vp
->program
.nir
, &vp
->program
,
143 &prog_data
.base
.base
);
146 uint64_t outputs_written
=
147 brw_vs_outputs_written(brw
, key
, vp
->program
.info
.outputs_written
);
148 prog_data
.inputs_read
= vp
->program
.info
.inputs_read
;
149 prog_data
.double_inputs_read
= vp
->program
.info
.double_inputs_read
;
151 if (key
->copy_edgeflag
) {
152 prog_data
.inputs_read
|= VERT_BIT_EDGEFLAG
;
155 brw_compute_vue_map(devinfo
,
156 &prog_data
.base
.vue_map
, outputs_written
,
157 vp
->program
.nir
->info
->separate_shader
);
160 _mesa_fprint_program_opt(stderr
, &vp
->program
, PROG_PRINT_DEBUG
, true);
163 if (unlikely(brw
->perf_debug
)) {
164 start_busy
= (brw
->batch
.last_bo
&&
165 drm_intel_bo_busy(brw
->batch
.last_bo
));
166 start_time
= get_time();
169 if (unlikely(INTEL_DEBUG
& DEBUG_VS
)) {
171 brw_dump_arb_asm("vertex", &vp
->program
);
173 fprintf(stderr
, "VS Output ");
174 brw_print_vue_map(stderr
, &prog_data
.base
.vue_map
);
178 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
179 bool is_glsl_sh
= prog
!= NULL
;
180 st_index
= brw_get_shader_time_index(brw
, &vp
->program
, ST_VS
,
187 program
= brw_compile_vs(compiler
, brw
, mem_ctx
, key
, &prog_data
,
189 brw_select_clip_planes(&brw
->ctx
),
190 !_mesa_is_gles3(&brw
->ctx
),
191 st_index
, &program_size
, &error_str
);
192 if (program
== NULL
) {
194 vp
->program
.sh
.data
->LinkStatus
= false;
195 ralloc_strcat(&vp
->program
.sh
.data
->InfoLog
, error_str
);
198 _mesa_problem(NULL
, "Failed to compile vertex shader: %s\n", error_str
);
200 ralloc_free(mem_ctx
);
204 if (unlikely(brw
->perf_debug
)) {
205 if (vp
->compiled_once
) {
206 brw_vs_debug_recompile(brw
, prog
, key
);
208 if (start_busy
&& !drm_intel_bo_busy(brw
->batch
.last_bo
)) {
209 perf_debug("VS compile took %.03f ms and stalled the GPU\n",
210 (get_time() - start_time
) * 1000);
212 vp
->compiled_once
= true;
215 /* Scratch space is used for register spilling */
216 brw_alloc_stage_scratch(brw
, &brw
->vs
.base
,
217 prog_data
.base
.base
.total_scratch
,
218 devinfo
->max_vs_threads
);
220 brw_upload_cache(&brw
->cache
, BRW_CACHE_VS_PROG
,
221 key
, sizeof(struct brw_vs_prog_key
),
222 program
, program_size
,
223 &prog_data
, sizeof(prog_data
),
224 &brw
->vs
.base
.prog_offset
, &brw
->vs
.base
.prog_data
);
225 ralloc_free(mem_ctx
);
231 brw_vs_debug_recompile(struct brw_context
*brw
,
232 struct gl_shader_program
*prog
,
233 const struct brw_vs_prog_key
*key
)
235 struct brw_cache_item
*c
= NULL
;
236 const struct brw_vs_prog_key
*old_key
= NULL
;
239 perf_debug("Recompiling vertex shader for program %d\n", prog
->Name
);
241 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
242 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
243 if (c
->cache_id
== BRW_CACHE_VS_PROG
) {
246 if (old_key
->program_string_id
== key
->program_string_id
)
255 perf_debug(" Didn't find previous compile in the shader cache for "
260 for (unsigned int i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
261 found
|= key_debug(brw
, "Vertex attrib w/a flags",
262 old_key
->gl_attrib_wa_flags
[i
],
263 key
->gl_attrib_wa_flags
[i
]);
266 found
|= key_debug(brw
, "legacy user clipping",
267 old_key
->nr_userclip_plane_consts
,
268 key
->nr_userclip_plane_consts
);
270 found
|= key_debug(brw
, "copy edgeflag",
271 old_key
->copy_edgeflag
, key
->copy_edgeflag
);
272 found
|= key_debug(brw
, "PointCoord replace",
273 old_key
->point_coord_replace
, key
->point_coord_replace
);
274 found
|= key_debug(brw
, "vertex color clamping",
275 old_key
->clamp_vertex_color
, key
->clamp_vertex_color
);
277 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
280 perf_debug(" Something else\n");
285 brw_vs_state_dirty(const struct brw_context
*brw
)
287 return brw_state_dirty(brw
,
294 BRW_NEW_VERTEX_PROGRAM
|
295 BRW_NEW_VS_ATTRIB_WORKAROUNDS
);
299 brw_vs_populate_key(struct brw_context
*brw
,
300 struct brw_vs_prog_key
*key
)
302 struct gl_context
*ctx
= &brw
->ctx
;
303 /* BRW_NEW_VERTEX_PROGRAM */
304 struct brw_program
*vp
= (struct brw_program
*)brw
->vertex_program
;
305 struct gl_program
*prog
= (struct gl_program
*) brw
->vertex_program
;
307 memset(key
, 0, sizeof(*key
));
309 /* Just upload the program verbatim for now. Always send it all
310 * the inputs it asks for, whether they are varying or not.
312 key
->program_string_id
= vp
->id
;
314 if (ctx
->Transform
.ClipPlanesEnabled
!= 0 &&
315 (ctx
->API
== API_OPENGL_COMPAT
|| ctx
->API
== API_OPENGLES
) &&
316 vp
->program
.ClipDistanceArraySize
== 0) {
317 key
->nr_userclip_plane_consts
=
318 _mesa_logbase2(ctx
->Transform
.ClipPlanesEnabled
) + 1;
323 key
->copy_edgeflag
= (ctx
->Polygon
.FrontMode
!= GL_FILL
||
324 ctx
->Polygon
.BackMode
!= GL_FILL
);
327 if (ctx
->Point
.PointSprite
) {
328 key
->point_coord_replace
= ctx
->Point
.CoordReplace
& 0xff;
332 if (prog
->nir
->info
->outputs_written
&
333 (VARYING_BIT_COL0
| VARYING_BIT_COL1
| VARYING_BIT_BFC0
|
335 /* _NEW_LIGHT | _NEW_BUFFERS */
336 key
->clamp_vertex_color
= ctx
->Light
._ClampVertexColor
;
340 brw_populate_sampler_prog_key_data(ctx
, prog
, &key
->tex
);
342 /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
343 if (brw
->gen
< 8 && !brw
->is_haswell
) {
344 memcpy(key
->gl_attrib_wa_flags
, brw
->vb
.attrib_wa_flags
,
345 sizeof(brw
->vb
.attrib_wa_flags
));
350 brw_upload_vs_prog(struct brw_context
*brw
)
352 struct gl_context
*ctx
= &brw
->ctx
;
353 struct gl_shader_program
**current
= ctx
->_Shader
->CurrentProgram
;
354 struct brw_vs_prog_key key
;
355 /* BRW_NEW_VERTEX_PROGRAM */
356 struct brw_program
*vp
= (struct brw_program
*)brw
->vertex_program
;
358 if (!brw_vs_state_dirty(brw
))
361 brw_vs_populate_key(brw
, &key
);
363 if (!brw_search_cache(&brw
->cache
, BRW_CACHE_VS_PROG
,
365 &brw
->vs
.base
.prog_offset
, &brw
->vs
.base
.prog_data
)) {
366 bool success
= brw_codegen_vs_prog(brw
, current
[MESA_SHADER_VERTEX
],
374 brw_vs_precompile(struct gl_context
*ctx
,
375 struct gl_shader_program
*shader_prog
,
376 struct gl_program
*prog
)
378 struct brw_context
*brw
= brw_context(ctx
);
379 struct brw_vs_prog_key key
;
380 uint32_t old_prog_offset
= brw
->vs
.base
.prog_offset
;
381 struct brw_stage_prog_data
*old_prog_data
= brw
->vs
.base
.prog_data
;
384 struct brw_program
*bvp
= brw_program(prog
);
386 memset(&key
, 0, sizeof(key
));
388 brw_setup_tex_for_precompile(brw
, &key
.tex
, prog
);
389 key
.program_string_id
= bvp
->id
;
390 key
.clamp_vertex_color
=
391 (prog
->nir
->info
->outputs_written
&
392 (VARYING_BIT_COL0
| VARYING_BIT_COL1
| VARYING_BIT_BFC0
|
395 success
= brw_codegen_vs_prog(brw
, shader_prog
, bvp
, &key
);
397 brw
->vs
.base
.prog_offset
= old_prog_offset
;
398 brw
->vs
.base
.prog_data
= old_prog_data
;