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"
43 #include "util/ralloc.h"
46 brw_codegen_vs_prog(struct brw_context
*brw
,
47 struct gl_shader_program
*prog
,
48 struct brw_vertex_program
*vp
,
49 struct brw_vs_prog_key
*key
)
52 const GLuint
*program
;
53 struct brw_vs_prog_data prog_data
;
54 struct brw_stage_prog_data
*stage_prog_data
= &prog_data
.base
.base
;
57 struct brw_shader
*vs
= NULL
;
58 bool start_busy
= false;
59 double start_time
= 0;
62 vs
= (struct brw_shader
*) prog
->_LinkedShaders
[MESA_SHADER_VERTEX
];
64 memset(&prog_data
, 0, sizeof(prog_data
));
66 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
68 stage_prog_data
->use_alt_mode
= true;
70 mem_ctx
= ralloc_context(NULL
);
72 brw_assign_common_binding_table_offsets(MESA_SHADER_VERTEX
,
73 brw
->intelScreen
->devinfo
,
74 prog
, &vp
->program
.Base
,
75 &prog_data
.base
.base
, 0);
77 /* Allocate the references to the uniforms that will end up in the
78 * prog_data associated with the compiled program, and which will be freed
81 int param_count
= vp
->program
.Base
.nir
->num_uniforms
;
82 if (!brw
->intelScreen
->compiler
->scalar_vs
)
86 prog_data
.base
.base
.nr_image_params
= vs
->base
.NumImages
;
88 /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
91 param_count
+= key
->nr_userclip_plane_consts
* 4;
93 stage_prog_data
->param
=
94 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
95 stage_prog_data
->pull_param
=
96 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
97 stage_prog_data
->image_param
=
98 rzalloc_array(NULL
, struct brw_image_param
,
99 stage_prog_data
->nr_image_params
);
100 stage_prog_data
->nr_params
= param_count
;
103 brw_nir_setup_glsl_uniforms(vp
->program
.Base
.nir
, prog
, &vp
->program
.Base
,
104 &prog_data
.base
.base
,
105 brw
->intelScreen
->compiler
->scalar_vs
);
107 brw_nir_setup_arb_uniforms(vp
->program
.Base
.nir
, &vp
->program
.Base
,
108 &prog_data
.base
.base
);
111 GLbitfield64 outputs_written
= vp
->program
.Base
.OutputsWritten
;
112 prog_data
.inputs_read
= vp
->program
.Base
.InputsRead
;
114 if (key
->copy_edgeflag
) {
115 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_EDGE
);
116 prog_data
.inputs_read
|= VERT_BIT_EDGEFLAG
;
120 /* Put dummy slots into the VUE for the SF to put the replaced
121 * point sprite coords in. We shouldn't need these dummy slots,
122 * which take up precious URB space, but it would mean that the SF
123 * doesn't get nice aligned pairs of input coords into output
124 * coords, which would be a pain to handle.
126 for (i
= 0; i
< 8; i
++) {
127 if (key
->point_coord_replace
& (1 << i
))
128 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_TEX0
+ i
);
131 /* if back colors are written, allocate slots for front colors too */
132 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC0
))
133 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL0
);
134 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC1
))
135 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL1
);
138 /* In order for legacy clipping to work, we need to populate the clip
139 * distance varying slots whenever clipping is enabled, even if the vertex
140 * shader doesn't write to gl_ClipDistance.
142 if (key
->nr_userclip_plane_consts
> 0) {
143 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0
);
144 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1
);
147 brw_compute_vue_map(brw
->intelScreen
->devinfo
,
148 &prog_data
.base
.vue_map
, outputs_written
,
149 prog
? prog
->SeparateShader
: false);
152 _mesa_fprint_program_opt(stderr
, &vp
->program
.Base
, PROG_PRINT_DEBUG
,
156 if (unlikely(brw
->perf_debug
)) {
157 start_busy
= (brw
->batch
.last_bo
&&
158 drm_intel_bo_busy(brw
->batch
.last_bo
));
159 start_time
= get_time();
162 if (unlikely(INTEL_DEBUG
& DEBUG_VS
)) {
163 brw_dump_ir("vertex", prog
, vs
? &vs
->base
: NULL
, &vp
->program
.Base
);
165 fprintf(stderr
, "VS Output ");
166 brw_print_vue_map(stderr
, &prog_data
.base
.vue_map
);
170 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
171 st_index
= brw_get_shader_time_index(brw
, prog
, &vp
->program
.Base
, ST_VS
);
176 program
= brw_compile_vs(brw
->intelScreen
->compiler
, brw
, mem_ctx
, key
,
177 &prog_data
, vp
->program
.Base
.nir
,
178 brw_select_clip_planes(&brw
->ctx
),
179 !_mesa_is_gles3(&brw
->ctx
),
180 st_index
, &program_size
, &error_str
);
181 if (program
== NULL
) {
183 prog
->LinkStatus
= false;
184 ralloc_strcat(&prog
->InfoLog
, error_str
);
187 _mesa_problem(NULL
, "Failed to compile vertex shader: %s\n", error_str
);
189 ralloc_free(mem_ctx
);
193 if (unlikely(brw
->perf_debug
) && vs
) {
194 if (vs
->compiled_once
) {
195 brw_vs_debug_recompile(brw
, prog
, key
);
197 if (start_busy
&& !drm_intel_bo_busy(brw
->batch
.last_bo
)) {
198 perf_debug("VS compile took %.03f ms and stalled the GPU\n",
199 (get_time() - start_time
) * 1000);
201 vs
->compiled_once
= true;
204 /* Scratch space is used for register spilling */
205 if (prog_data
.base
.base
.total_scratch
) {
206 brw_get_scratch_bo(brw
, &brw
->vs
.base
.scratch_bo
,
207 prog_data
.base
.base
.total_scratch
*
208 brw
->max_vs_threads
);
211 brw_upload_cache(&brw
->cache
, BRW_CACHE_VS_PROG
,
212 key
, sizeof(struct brw_vs_prog_key
),
213 program
, program_size
,
214 &prog_data
, sizeof(prog_data
),
215 &brw
->vs
.base
.prog_offset
, &brw
->vs
.prog_data
);
216 ralloc_free(mem_ctx
);
222 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
225 perf_debug(" %s %d->%d\n", name
, a
, b
);
232 brw_vs_debug_recompile(struct brw_context
*brw
,
233 struct gl_shader_program
*prog
,
234 const struct brw_vs_prog_key
*key
)
236 struct brw_cache_item
*c
= NULL
;
237 const struct brw_vs_prog_key
*old_key
= NULL
;
240 perf_debug("Recompiling vertex shader for program %d\n", prog
->Name
);
242 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
243 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
244 if (c
->cache_id
== BRW_CACHE_VS_PROG
) {
247 if (old_key
->program_string_id
== key
->program_string_id
)
256 perf_debug(" Didn't find previous compile in the shader cache for "
261 for (unsigned int i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
262 found
|= key_debug(brw
, "Vertex attrib w/a flags",
263 old_key
->gl_attrib_wa_flags
[i
],
264 key
->gl_attrib_wa_flags
[i
]);
267 found
|= key_debug(brw
, "legacy user clipping",
268 old_key
->nr_userclip_plane_consts
,
269 key
->nr_userclip_plane_consts
);
271 found
|= key_debug(brw
, "copy edgeflag",
272 old_key
->copy_edgeflag
, key
->copy_edgeflag
);
273 found
|= key_debug(brw
, "PointCoord replace",
274 old_key
->point_coord_replace
, key
->point_coord_replace
);
275 found
|= key_debug(brw
, "vertex color clamping",
276 old_key
->clamp_vertex_color
, key
->clamp_vertex_color
);
278 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
281 perf_debug(" Something else\n");
286 brw_vs_state_dirty(struct brw_context
*brw
)
288 return brw_state_dirty(brw
,
295 BRW_NEW_VERTEX_PROGRAM
|
296 BRW_NEW_VS_ATTRIB_WORKAROUNDS
);
300 brw_vs_populate_key(struct brw_context
*brw
,
301 struct brw_vs_prog_key
*key
)
303 struct gl_context
*ctx
= &brw
->ctx
;
304 /* BRW_NEW_VERTEX_PROGRAM */
305 struct brw_vertex_program
*vp
=
306 (struct brw_vertex_program
*)brw
->vertex_program
;
307 struct gl_program
*prog
= (struct gl_program
*) brw
->vertex_program
;
310 memset(key
, 0, sizeof(*key
));
312 /* Just upload the program verbatim for now. Always send it all
313 * the inputs it asks for, whether they are varying or not.
315 key
->program_string_id
= vp
->id
;
317 if (ctx
->Transform
.ClipPlanesEnabled
!= 0 &&
318 (ctx
->API
== API_OPENGL_COMPAT
||
319 ctx
->API
== API_OPENGLES
) &&
320 vp
->program
.Base
.ClipDistanceArraySize
== 0) {
321 key
->nr_userclip_plane_consts
=
322 _mesa_logbase2(ctx
->Transform
.ClipPlanesEnabled
) + 1;
327 key
->copy_edgeflag
= (ctx
->Polygon
.FrontMode
!= GL_FILL
||
328 ctx
->Polygon
.BackMode
!= GL_FILL
);
331 if (prog
->OutputsWritten
& (VARYING_BIT_COL0
| VARYING_BIT_COL1
|
332 VARYING_BIT_BFC0
| VARYING_BIT_BFC1
)) {
333 /* _NEW_LIGHT | _NEW_BUFFERS */
334 key
->clamp_vertex_color
= ctx
->Light
._ClampVertexColor
;
338 if (brw
->gen
< 6 && ctx
->Point
.PointSprite
) {
339 for (i
= 0; i
< 8; i
++) {
340 if (ctx
->Point
.CoordReplace
[i
])
341 key
->point_coord_replace
|= (1 << i
);
346 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->vs
.base
.sampler_count
,
349 /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
350 memcpy(key
->gl_attrib_wa_flags
, brw
->vb
.attrib_wa_flags
,
351 sizeof(brw
->vb
.attrib_wa_flags
));
355 brw_upload_vs_prog(struct brw_context
*brw
)
357 struct gl_context
*ctx
= &brw
->ctx
;
358 struct gl_shader_program
**current
= ctx
->_Shader
->CurrentProgram
;
359 struct brw_vs_prog_key key
;
360 /* BRW_NEW_VERTEX_PROGRAM */
361 struct brw_vertex_program
*vp
=
362 (struct brw_vertex_program
*)brw
->vertex_program
;
364 if (!brw_vs_state_dirty(brw
))
367 brw_vs_populate_key(brw
, &key
);
369 if (!brw_search_cache(&brw
->cache
, BRW_CACHE_VS_PROG
,
371 &brw
->vs
.base
.prog_offset
, &brw
->vs
.prog_data
)) {
372 bool success
= brw_codegen_vs_prog(brw
, current
[MESA_SHADER_VERTEX
],
377 brw
->vs
.base
.prog_data
= &brw
->vs
.prog_data
->base
.base
;
381 brw_vs_precompile(struct gl_context
*ctx
,
382 struct gl_shader_program
*shader_prog
,
383 struct gl_program
*prog
)
385 struct brw_context
*brw
= brw_context(ctx
);
386 struct brw_vs_prog_key key
;
387 uint32_t old_prog_offset
= brw
->vs
.base
.prog_offset
;
388 struct brw_vs_prog_data
*old_prog_data
= brw
->vs
.prog_data
;
391 struct gl_vertex_program
*vp
= (struct gl_vertex_program
*) prog
;
392 struct brw_vertex_program
*bvp
= brw_vertex_program(vp
);
394 memset(&key
, 0, sizeof(key
));
396 brw_setup_tex_for_precompile(brw
, &key
.tex
, prog
);
397 key
.program_string_id
= bvp
->id
;
398 key
.clamp_vertex_color
=
399 (prog
->OutputsWritten
& (VARYING_BIT_COL0
| VARYING_BIT_COL1
|
400 VARYING_BIT_BFC0
| VARYING_BIT_BFC1
));
402 success
= brw_codegen_vs_prog(brw
, shader_prog
, bvp
, &key
);
404 brw
->vs
.base
.prog_offset
= old_prog_offset
;
405 brw
->vs
.prog_data
= old_prog_data
;