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_codegen_vs_prog(struct brw_context
*brw
,
48 struct gl_shader_program
*prog
,
49 struct brw_vertex_program
*vp
,
50 struct brw_vs_prog_key
*key
)
52 const struct brw_compiler
*compiler
= brw
->intelScreen
->compiler
;
54 const GLuint
*program
;
55 struct brw_vs_prog_data prog_data
;
56 struct brw_stage_prog_data
*stage_prog_data
= &prog_data
.base
.base
;
59 struct brw_shader
*vs
= NULL
;
60 bool start_busy
= false;
61 double start_time
= 0;
64 vs
= (struct brw_shader
*) prog
->_LinkedShaders
[MESA_SHADER_VERTEX
];
66 memset(&prog_data
, 0, sizeof(prog_data
));
68 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
70 stage_prog_data
->use_alt_mode
= true;
72 mem_ctx
= ralloc_context(NULL
);
74 brw_assign_common_binding_table_offsets(MESA_SHADER_VERTEX
,
75 brw
->intelScreen
->devinfo
,
76 prog
, &vp
->program
.Base
,
77 &prog_data
.base
.base
, 0);
79 /* Allocate the references to the uniforms that will end up in the
80 * prog_data associated with the compiled program, and which will be freed
83 int param_count
= vp
->program
.Base
.nir
->num_uniforms
;
84 if (!compiler
->scalar_stage
[MESA_SHADER_VERTEX
])
88 prog_data
.base
.base
.nr_image_params
= vs
->base
.NumImages
;
90 /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
93 param_count
+= key
->nr_userclip_plane_consts
* 4;
95 stage_prog_data
->param
=
96 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
97 stage_prog_data
->pull_param
=
98 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
99 stage_prog_data
->image_param
=
100 rzalloc_array(NULL
, struct brw_image_param
,
101 stage_prog_data
->nr_image_params
);
102 stage_prog_data
->nr_params
= param_count
;
105 brw_nir_setup_glsl_uniforms(vp
->program
.Base
.nir
, prog
, &vp
->program
.Base
,
106 &prog_data
.base
.base
,
107 compiler
->scalar_stage
[MESA_SHADER_VERTEX
]);
109 brw_nir_setup_arb_uniforms(vp
->program
.Base
.nir
, &vp
->program
.Base
,
110 &prog_data
.base
.base
);
113 GLbitfield64 outputs_written
= vp
->program
.Base
.OutputsWritten
;
114 prog_data
.inputs_read
= vp
->program
.Base
.InputsRead
;
116 if (key
->copy_edgeflag
) {
117 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_EDGE
);
118 prog_data
.inputs_read
|= VERT_BIT_EDGEFLAG
;
122 /* Put dummy slots into the VUE for the SF to put the replaced
123 * point sprite coords in. We shouldn't need these dummy slots,
124 * which take up precious URB space, but it would mean that the SF
125 * doesn't get nice aligned pairs of input coords into output
126 * coords, which would be a pain to handle.
128 for (i
= 0; i
< 8; i
++) {
129 if (key
->point_coord_replace
& (1 << i
))
130 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_TEX0
+ i
);
133 /* if back colors are written, allocate slots for front colors too */
134 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC0
))
135 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL0
);
136 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC1
))
137 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL1
);
140 /* In order for legacy clipping to work, we need to populate the clip
141 * distance varying slots whenever clipping is enabled, even if the vertex
142 * shader doesn't write to gl_ClipDistance.
144 if (key
->nr_userclip_plane_consts
> 0) {
145 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0
);
146 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1
);
149 brw_compute_vue_map(brw
->intelScreen
->devinfo
,
150 &prog_data
.base
.vue_map
, outputs_written
,
151 prog
? prog
->SeparateShader
: false);
154 _mesa_fprint_program_opt(stderr
, &vp
->program
.Base
, PROG_PRINT_DEBUG
,
158 if (unlikely(brw
->perf_debug
)) {
159 start_busy
= (brw
->batch
.last_bo
&&
160 drm_intel_bo_busy(brw
->batch
.last_bo
));
161 start_time
= get_time();
164 if (unlikely(INTEL_DEBUG
& DEBUG_VS
)) {
165 brw_dump_ir("vertex", prog
, vs
? &vs
->base
: NULL
, &vp
->program
.Base
);
167 fprintf(stderr
, "VS Output ");
168 brw_print_vue_map(stderr
, &prog_data
.base
.vue_map
);
172 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
173 st_index
= brw_get_shader_time_index(brw
, prog
, &vp
->program
.Base
, ST_VS
);
178 program
= brw_compile_vs(compiler
, brw
, mem_ctx
, key
,
179 &prog_data
, vp
->program
.Base
.nir
,
180 brw_select_clip_planes(&brw
->ctx
),
181 !_mesa_is_gles3(&brw
->ctx
),
182 st_index
, &program_size
, &error_str
);
183 if (program
== NULL
) {
185 prog
->LinkStatus
= false;
186 ralloc_strcat(&prog
->InfoLog
, error_str
);
189 _mesa_problem(NULL
, "Failed to compile vertex shader: %s\n", error_str
);
191 ralloc_free(mem_ctx
);
195 if (unlikely(brw
->perf_debug
) && vs
) {
196 if (vs
->compiled_once
) {
197 brw_vs_debug_recompile(brw
, prog
, key
);
199 if (start_busy
&& !drm_intel_bo_busy(brw
->batch
.last_bo
)) {
200 perf_debug("VS compile took %.03f ms and stalled the GPU\n",
201 (get_time() - start_time
) * 1000);
203 vs
->compiled_once
= true;
206 /* Scratch space is used for register spilling */
207 if (prog_data
.base
.base
.total_scratch
) {
208 brw_get_scratch_bo(brw
, &brw
->vs
.base
.scratch_bo
,
209 prog_data
.base
.base
.total_scratch
*
210 brw
->max_vs_threads
);
213 brw_upload_cache(&brw
->cache
, BRW_CACHE_VS_PROG
,
214 key
, sizeof(struct brw_vs_prog_key
),
215 program
, program_size
,
216 &prog_data
, sizeof(prog_data
),
217 &brw
->vs
.base
.prog_offset
, &brw
->vs
.prog_data
);
218 ralloc_free(mem_ctx
);
224 key_debug(struct brw_context
*brw
, const char *name
, int a
, int b
)
227 perf_debug(" %s %d->%d\n", name
, a
, b
);
234 brw_vs_debug_recompile(struct brw_context
*brw
,
235 struct gl_shader_program
*prog
,
236 const struct brw_vs_prog_key
*key
)
238 struct brw_cache_item
*c
= NULL
;
239 const struct brw_vs_prog_key
*old_key
= NULL
;
242 perf_debug("Recompiling vertex shader for program %d\n", prog
->Name
);
244 for (unsigned int i
= 0; i
< brw
->cache
.size
; i
++) {
245 for (c
= brw
->cache
.items
[i
]; c
; c
= c
->next
) {
246 if (c
->cache_id
== BRW_CACHE_VS_PROG
) {
249 if (old_key
->program_string_id
== key
->program_string_id
)
258 perf_debug(" Didn't find previous compile in the shader cache for "
263 for (unsigned int i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
264 found
|= key_debug(brw
, "Vertex attrib w/a flags",
265 old_key
->gl_attrib_wa_flags
[i
],
266 key
->gl_attrib_wa_flags
[i
]);
269 found
|= key_debug(brw
, "legacy user clipping",
270 old_key
->nr_userclip_plane_consts
,
271 key
->nr_userclip_plane_consts
);
273 found
|= key_debug(brw
, "copy edgeflag",
274 old_key
->copy_edgeflag
, key
->copy_edgeflag
);
275 found
|= key_debug(brw
, "PointCoord replace",
276 old_key
->point_coord_replace
, key
->point_coord_replace
);
277 found
|= key_debug(brw
, "vertex color clamping",
278 old_key
->clamp_vertex_color
, key
->clamp_vertex_color
);
280 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
283 perf_debug(" Something else\n");
288 brw_vs_state_dirty(struct brw_context
*brw
)
290 return brw_state_dirty(brw
,
297 BRW_NEW_VERTEX_PROGRAM
|
298 BRW_NEW_VS_ATTRIB_WORKAROUNDS
);
302 brw_vs_populate_key(struct brw_context
*brw
,
303 struct brw_vs_prog_key
*key
)
305 struct gl_context
*ctx
= &brw
->ctx
;
306 /* BRW_NEW_VERTEX_PROGRAM */
307 struct brw_vertex_program
*vp
=
308 (struct brw_vertex_program
*)brw
->vertex_program
;
309 struct gl_program
*prog
= (struct gl_program
*) brw
->vertex_program
;
312 memset(key
, 0, sizeof(*key
));
314 /* Just upload the program verbatim for now. Always send it all
315 * the inputs it asks for, whether they are varying or not.
317 key
->program_string_id
= vp
->id
;
319 if (ctx
->Transform
.ClipPlanesEnabled
!= 0 &&
320 (ctx
->API
== API_OPENGL_COMPAT
||
321 ctx
->API
== API_OPENGLES
) &&
322 vp
->program
.Base
.ClipDistanceArraySize
== 0) {
323 key
->nr_userclip_plane_consts
=
324 _mesa_logbase2(ctx
->Transform
.ClipPlanesEnabled
) + 1;
329 key
->copy_edgeflag
= (ctx
->Polygon
.FrontMode
!= GL_FILL
||
330 ctx
->Polygon
.BackMode
!= GL_FILL
);
333 if (prog
->OutputsWritten
& (VARYING_BIT_COL0
| VARYING_BIT_COL1
|
334 VARYING_BIT_BFC0
| VARYING_BIT_BFC1
)) {
335 /* _NEW_LIGHT | _NEW_BUFFERS */
336 key
->clamp_vertex_color
= ctx
->Light
._ClampVertexColor
;
340 if (brw
->gen
< 6 && ctx
->Point
.PointSprite
) {
341 for (i
= 0; i
< 8; i
++) {
342 if (ctx
->Point
.CoordReplace
[i
])
343 key
->point_coord_replace
|= (1 << i
);
348 brw_populate_sampler_prog_key_data(ctx
, prog
, brw
->vs
.base
.sampler_count
,
351 /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
352 memcpy(key
->gl_attrib_wa_flags
, brw
->vb
.attrib_wa_flags
,
353 sizeof(brw
->vb
.attrib_wa_flags
));
357 brw_upload_vs_prog(struct brw_context
*brw
)
359 struct gl_context
*ctx
= &brw
->ctx
;
360 struct gl_shader_program
**current
= ctx
->_Shader
->CurrentProgram
;
361 struct brw_vs_prog_key key
;
362 /* BRW_NEW_VERTEX_PROGRAM */
363 struct brw_vertex_program
*vp
=
364 (struct brw_vertex_program
*)brw
->vertex_program
;
366 if (!brw_vs_state_dirty(brw
))
369 brw_vs_populate_key(brw
, &key
);
371 if (!brw_search_cache(&brw
->cache
, BRW_CACHE_VS_PROG
,
373 &brw
->vs
.base
.prog_offset
, &brw
->vs
.prog_data
)) {
374 bool success
= brw_codegen_vs_prog(brw
, current
[MESA_SHADER_VERTEX
],
379 brw
->vs
.base
.prog_data
= &brw
->vs
.prog_data
->base
.base
;
383 brw_vs_precompile(struct gl_context
*ctx
,
384 struct gl_shader_program
*shader_prog
,
385 struct gl_program
*prog
)
387 struct brw_context
*brw
= brw_context(ctx
);
388 struct brw_vs_prog_key key
;
389 uint32_t old_prog_offset
= brw
->vs
.base
.prog_offset
;
390 struct brw_vs_prog_data
*old_prog_data
= brw
->vs
.prog_data
;
393 struct gl_vertex_program
*vp
= (struct gl_vertex_program
*) prog
;
394 struct brw_vertex_program
*bvp
= brw_vertex_program(vp
);
396 memset(&key
, 0, sizeof(key
));
398 brw_setup_tex_for_precompile(brw
, &key
.tex
, prog
);
399 key
.program_string_id
= bvp
->id
;
400 key
.clamp_vertex_color
=
401 (prog
->OutputsWritten
& (VARYING_BIT_COL0
| VARYING_BIT_COL1
|
402 VARYING_BIT_BFC0
| VARYING_BIT_BFC1
));
404 success
= brw_codegen_vs_prog(brw
, shader_prog
, bvp
, &key
);
406 brw
->vs
.base
.prog_offset
= old_prog_offset
;
407 brw
->vs
.prog_data
= old_prog_data
;