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"
41 #include "compiler/brw_nir.h"
42 #include "brw_program.h"
44 #include "util/ralloc.h"
47 * Decide which set of clip planes should be used when clipping via
48 * gl_Position or gl_ClipVertex.
51 brw_select_clip_planes(struct gl_context
*ctx
)
53 if (ctx
->_Shader
->CurrentProgram
[MESA_SHADER_VERTEX
]) {
54 /* There is currently a GLSL vertex shader, so clip according to GLSL
55 * rules, which means compare gl_ClipVertex (or gl_Position, if
56 * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
57 * that were stored in EyeUserPlane at the time the clip planes were
60 return ctx
->Transform
.EyeUserPlane
;
62 /* Either we are using fixed function or an ARB vertex program. In
63 * either case the clip planes are going to be compared against
64 * gl_Position (which is in clip coordinates) so we have to clip using
65 * _ClipUserPlane, which was transformed into clip coordinates by Mesa
68 return ctx
->Transform
._ClipUserPlane
;
73 brw_vs_outputs_written(struct brw_context
*brw
, struct brw_vs_prog_key
*key
,
74 GLbitfield64 user_varyings
)
76 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
77 GLbitfield64 outputs_written
= user_varyings
;
79 if (key
->copy_edgeflag
) {
80 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_EDGE
);
83 if (devinfo
->gen
< 6) {
84 /* Put dummy slots into the VUE for the SF to put the replaced
85 * point sprite coords in. We shouldn't need these dummy slots,
86 * which take up precious URB space, but it would mean that the SF
87 * doesn't get nice aligned pairs of input coords into output
88 * coords, which would be a pain to handle.
90 for (unsigned i
= 0; i
< 8; i
++) {
91 if (key
->point_coord_replace
& (1 << i
))
92 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_TEX0
+ i
);
95 /* if back colors are written, allocate slots for front colors too */
96 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC0
))
97 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL0
);
98 if (outputs_written
& BITFIELD64_BIT(VARYING_SLOT_BFC1
))
99 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_COL1
);
102 /* In order for legacy clipping to work, we need to populate the clip
103 * distance varying slots whenever clipping is enabled, even if the vertex
104 * shader doesn't write to gl_ClipDistance.
106 if (key
->nr_userclip_plane_consts
> 0) {
107 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0
);
108 outputs_written
|= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1
);
111 return outputs_written
;
115 brw_vs_debug_recompile(struct brw_context
*brw
, struct gl_program
*prog
,
116 const struct brw_vs_prog_key
*key
)
118 perf_debug("Recompiling vertex shader for program %d\n", prog
->Id
);
121 const struct brw_vs_prog_key
*old_key
=
122 brw_find_previous_compile(&brw
->cache
, BRW_CACHE_VS_PROG
,
123 key
->program_string_id
);
126 perf_debug(" Didn't find previous compile in the shader cache for "
131 for (unsigned int i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
132 found
|= key_debug(brw
, "Vertex attrib w/a flags",
133 old_key
->gl_attrib_wa_flags
[i
],
134 key
->gl_attrib_wa_flags
[i
]);
137 found
|= key_debug(brw
, "legacy user clipping",
138 old_key
->nr_userclip_plane_consts
,
139 key
->nr_userclip_plane_consts
);
141 found
|= key_debug(brw
, "copy edgeflag",
142 old_key
->copy_edgeflag
, key
->copy_edgeflag
);
143 found
|= key_debug(brw
, "PointCoord replace",
144 old_key
->point_coord_replace
, key
->point_coord_replace
);
145 found
|= key_debug(brw
, "vertex color clamping",
146 old_key
->clamp_vertex_color
, key
->clamp_vertex_color
);
148 found
|= brw_debug_recompile_sampler_key(brw
, &old_key
->tex
, &key
->tex
);
151 perf_debug(" Something else\n");
156 brw_codegen_vs_prog(struct brw_context
*brw
,
157 struct brw_program
*vp
,
158 struct brw_vs_prog_key
*key
)
160 const struct brw_compiler
*compiler
= brw
->screen
->compiler
;
161 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
163 const GLuint
*program
;
164 struct brw_vs_prog_data prog_data
;
165 struct brw_stage_prog_data
*stage_prog_data
= &prog_data
.base
.base
;
167 bool start_busy
= false;
168 double start_time
= 0;
170 memset(&prog_data
, 0, sizeof(prog_data
));
172 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
173 if (vp
->program
.is_arb_asm
)
174 stage_prog_data
->use_alt_mode
= true;
176 mem_ctx
= ralloc_context(NULL
);
178 brw_assign_common_binding_table_offsets(devinfo
, &vp
->program
,
179 &prog_data
.base
.base
, 0);
181 /* Allocate the references to the uniforms that will end up in the
182 * prog_data associated with the compiled program, and which will be freed
183 * by the state cache.
185 int param_count
= vp
->program
.nir
->num_uniforms
/ 4;
187 /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
188 * planes as uniforms.
190 param_count
+= key
->nr_userclip_plane_consts
* 4;
192 stage_prog_data
->param
= rzalloc_array(NULL
, uint32_t, param_count
);
193 stage_prog_data
->pull_param
= rzalloc_array(NULL
, uint32_t, param_count
);
194 stage_prog_data
->nr_params
= param_count
;
196 if (!vp
->program
.is_arb_asm
) {
197 brw_nir_setup_glsl_uniforms(vp
->program
.nir
, &vp
->program
,
198 &prog_data
.base
.base
,
199 compiler
->scalar_stage
[MESA_SHADER_VERTEX
]);
200 brw_nir_analyze_ubo_ranges(compiler
, vp
->program
.nir
,
201 prog_data
.base
.base
.ubo_ranges
);
203 brw_nir_setup_arb_uniforms(vp
->program
.nir
, &vp
->program
,
204 &prog_data
.base
.base
);
207 uint64_t outputs_written
=
208 brw_vs_outputs_written(brw
, key
, vp
->program
.nir
->info
.outputs_written
);
210 brw_compute_vue_map(devinfo
,
211 &prog_data
.base
.vue_map
, outputs_written
,
212 vp
->program
.nir
->info
.separate_shader
);
215 _mesa_fprint_program_opt(stderr
, &vp
->program
, PROG_PRINT_DEBUG
, true);
218 if (unlikely(brw
->perf_debug
)) {
219 start_busy
= (brw
->batch
.last_bo
&&
220 brw_bo_busy(brw
->batch
.last_bo
));
221 start_time
= get_time();
224 if (unlikely(INTEL_DEBUG
& DEBUG_VS
)) {
225 if (vp
->program
.is_arb_asm
)
226 brw_dump_arb_asm("vertex", &vp
->program
);
230 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
) {
231 st_index
= brw_get_shader_time_index(brw
, &vp
->program
, ST_VS
,
232 !vp
->program
.is_arb_asm
);
238 program
= brw_compile_vs(compiler
, brw
, mem_ctx
, key
, &prog_data
,
240 !_mesa_is_gles3(&brw
->ctx
),
241 st_index
, &program_size
, &error_str
);
242 if (program
== NULL
) {
243 if (!vp
->program
.is_arb_asm
) {
244 vp
->program
.sh
.data
->LinkStatus
= linking_failure
;
245 ralloc_strcat(&vp
->program
.sh
.data
->InfoLog
, error_str
);
248 _mesa_problem(NULL
, "Failed to compile vertex shader: %s\n", error_str
);
250 ralloc_free(mem_ctx
);
254 if (unlikely(brw
->perf_debug
)) {
255 if (vp
->compiled_once
) {
256 brw_vs_debug_recompile(brw
, &vp
->program
, key
);
258 if (start_busy
&& !brw_bo_busy(brw
->batch
.last_bo
)) {
259 perf_debug("VS compile took %.03f ms and stalled the GPU\n",
260 (get_time() - start_time
) * 1000);
262 vp
->compiled_once
= true;
265 /* Scratch space is used for register spilling */
266 brw_alloc_stage_scratch(brw
, &brw
->vs
.base
,
267 prog_data
.base
.base
.total_scratch
,
268 devinfo
->max_vs_threads
);
270 brw_upload_cache(&brw
->cache
, BRW_CACHE_VS_PROG
,
271 key
, sizeof(struct brw_vs_prog_key
),
272 program
, program_size
,
273 &prog_data
, sizeof(prog_data
),
274 &brw
->vs
.base
.prog_offset
, &brw
->vs
.base
.prog_data
);
275 ralloc_free(mem_ctx
);
281 brw_vs_state_dirty(const struct brw_context
*brw
)
283 return brw_state_dirty(brw
,
290 BRW_NEW_VERTEX_PROGRAM
|
291 BRW_NEW_VS_ATTRIB_WORKAROUNDS
);
295 brw_vs_populate_key(struct brw_context
*brw
,
296 struct brw_vs_prog_key
*key
)
298 struct gl_context
*ctx
= &brw
->ctx
;
299 /* BRW_NEW_VERTEX_PROGRAM */
300 struct gl_program
*prog
= brw
->programs
[MESA_SHADER_VERTEX
];
301 struct brw_program
*vp
= (struct brw_program
*) prog
;
302 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
304 memset(key
, 0, sizeof(*key
));
306 /* Just upload the program verbatim for now. Always send it all
307 * the inputs it asks for, whether they are varying or not.
309 key
->program_string_id
= vp
->id
;
311 if (ctx
->Transform
.ClipPlanesEnabled
!= 0 &&
312 (ctx
->API
== API_OPENGL_COMPAT
|| ctx
->API
== API_OPENGLES
) &&
313 vp
->program
.info
.clip_distance_array_size
== 0) {
314 key
->nr_userclip_plane_consts
=
315 _mesa_logbase2(ctx
->Transform
.ClipPlanesEnabled
) + 1;
318 if (devinfo
->gen
< 6) {
320 key
->copy_edgeflag
= (ctx
->Polygon
.FrontMode
!= GL_FILL
||
321 ctx
->Polygon
.BackMode
!= GL_FILL
);
324 if (ctx
->Point
.PointSprite
) {
325 key
->point_coord_replace
= ctx
->Point
.CoordReplace
& 0xff;
329 if (prog
->info
.outputs_written
&
330 (VARYING_BIT_COL0
| VARYING_BIT_COL1
| VARYING_BIT_BFC0
|
332 /* _NEW_LIGHT | _NEW_BUFFERS */
333 key
->clamp_vertex_color
= ctx
->Light
._ClampVertexColor
;
337 brw_populate_sampler_prog_key_data(ctx
, prog
, &key
->tex
);
339 /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
340 if (devinfo
->gen
< 8 && !devinfo
->is_haswell
) {
341 memcpy(key
->gl_attrib_wa_flags
, brw
->vb
.attrib_wa_flags
,
342 sizeof(brw
->vb
.attrib_wa_flags
));
347 brw_upload_vs_prog(struct brw_context
*brw
)
349 struct brw_vs_prog_key key
;
350 /* BRW_NEW_VERTEX_PROGRAM */
351 struct brw_program
*vp
=
352 (struct brw_program
*) brw
->programs
[MESA_SHADER_VERTEX
];
354 if (!brw_vs_state_dirty(brw
))
357 brw_vs_populate_key(brw
, &key
);
359 if (!brw_search_cache(&brw
->cache
, BRW_CACHE_VS_PROG
,
361 &brw
->vs
.base
.prog_offset
, &brw
->vs
.base
.prog_data
)) {
362 bool success
= brw_codegen_vs_prog(brw
, vp
, &key
);
369 brw_vs_precompile(struct gl_context
*ctx
, struct gl_program
*prog
)
371 struct brw_context
*brw
= brw_context(ctx
);
372 struct brw_vs_prog_key key
;
373 uint32_t old_prog_offset
= brw
->vs
.base
.prog_offset
;
374 struct brw_stage_prog_data
*old_prog_data
= brw
->vs
.base
.prog_data
;
377 struct brw_program
*bvp
= brw_program(prog
);
379 memset(&key
, 0, sizeof(key
));
381 brw_setup_tex_for_precompile(brw
, &key
.tex
, prog
);
382 key
.program_string_id
= bvp
->id
;
383 key
.clamp_vertex_color
=
384 (prog
->info
.outputs_written
&
385 (VARYING_BIT_COL0
| VARYING_BIT_COL1
| VARYING_BIT_BFC0
|
388 success
= brw_codegen_vs_prog(brw
, bvp
, &key
);
390 brw
->vs
.base
.prog_offset
= old_prog_offset
;
391 brw
->vs
.base
.prog_data
= old_prog_data
;