2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) 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 <keith@tungstengraphics.com>
32 #include "brw_context.h"
34 #include "brw_state.h"
35 #include "main/formats.h"
36 #include "main/samplerobj.h"
38 /** Return number of src args for given instruction */
39 GLuint
brw_wm_nr_args( GLuint opcode
)
56 assert(opcode
< MAX_OPCODE
);
57 return _mesa_num_inst_src_regs(opcode
);
62 GLuint
brw_wm_is_scalar_result( GLuint opcode
)
86 * Do GPU code generation for non-GLSL shader. non-GLSL shaders have
87 * no flow control instructions so we can more readily do SSA-style
91 brw_wm_non_glsl_emit(struct brw_context
*brw
, struct brw_wm_compile
*c
)
93 /* Augment fragment program. Add instructions for pre- and
94 * post-fragment-program tasks such as interpolation and fogging.
98 /* Translate to intermediate representation. Build register usage
103 /* Dead code removal.
107 /* Register allocation.
108 * Divide by two because we operate on 16 pixels at a time and require
109 * two GRF entries for each logical shader register.
111 c
->grf_limit
= BRW_WM_MAX_GRF
/ 2;
115 /* how many general-purpose registers are used */
116 c
->prog_data
.total_grf
= c
->max_wm_grf
;
124 brw_wm_payload_setup(struct brw_context
*brw
,
125 struct brw_wm_compile
*c
)
127 struct intel_context
*intel
= &brw
->intel
;
128 bool uses_depth
= (c
->fp
->program
.Base
.InputsRead
&
129 (1 << FRAG_ATTRIB_WPOS
)) != 0;
131 if (intel
->gen
>= 6) {
132 /* R0-1: masks, pixel X/Y coordinates. */
133 c
->nr_payload_regs
= 2;
134 /* R2: only for 32-pixel dispatch.*/
135 /* R3-4: perspective pixel location barycentric */
136 c
->nr_payload_regs
+= 2;
137 /* R5-6: perspective pixel location bary for dispatch width != 8 */
138 if (c
->dispatch_width
== 16) {
139 c
->nr_payload_regs
+= 2;
141 /* R7-10: perspective centroid barycentric */
142 /* R11-14: perspective sample barycentric */
143 /* R15-18: linear pixel location barycentric */
144 /* R19-22: linear centroid barycentric */
145 /* R23-26: linear sample barycentric */
147 /* R27: interpolated depth if uses source depth */
149 c
->source_depth_reg
= c
->nr_payload_regs
;
150 c
->nr_payload_regs
++;
151 if (c
->dispatch_width
== 16) {
152 /* R28: interpolated depth if not 8-wide. */
153 c
->nr_payload_regs
++;
156 /* R29: interpolated W set if GEN6_WM_USES_SOURCE_W.
159 c
->source_w_reg
= c
->nr_payload_regs
;
160 c
->nr_payload_regs
++;
161 if (c
->dispatch_width
== 16) {
162 /* R30: interpolated W if not 8-wide. */
163 c
->nr_payload_regs
++;
166 /* R31: MSAA position offsets. */
167 /* R32-: bary for 32-pixel. */
168 /* R58-59: interp W for 32-pixel. */
170 if (c
->fp
->program
.Base
.OutputsWritten
&
171 BITFIELD64_BIT(FRAG_RESULT_DEPTH
)) {
172 c
->source_depth_to_render_target
= GL_TRUE
;
173 c
->computes_depth
= GL_TRUE
;
176 brw_wm_lookup_iz(intel
, c
);
181 * All Mesa program -> GPU code generation goes through this function.
182 * Depending on the instructions used (i.e. flow control instructions)
183 * we'll use one of two code generators.
185 static void do_wm_prog( struct brw_context
*brw
,
186 struct brw_fragment_program
*fp
,
187 struct brw_wm_prog_key
*key
)
189 struct intel_context
*intel
= &brw
->intel
;
190 struct brw_wm_compile
*c
;
191 const GLuint
*program
;
194 c
= brw
->wm
.compile_data
;
196 brw
->wm
.compile_data
= calloc(1, sizeof(*brw
->wm
.compile_data
));
197 c
= brw
->wm
.compile_data
;
199 /* Ouch - big out of memory problem. Can't continue
200 * without triggering a segfault, no way to signal,
205 c
->instruction
= calloc(1, BRW_WM_MAX_INSN
* sizeof(*c
->instruction
));
206 c
->prog_instructions
= calloc(1, BRW_WM_MAX_INSN
*
207 sizeof(*c
->prog_instructions
));
208 c
->vreg
= calloc(1, BRW_WM_MAX_VREG
* sizeof(*c
->vreg
));
209 c
->refs
= calloc(1, BRW_WM_MAX_REF
* sizeof(*c
->refs
));
211 void *instruction
= c
->instruction
;
212 void *prog_instructions
= c
->prog_instructions
;
213 void *vreg
= c
->vreg
;
214 void *refs
= c
->refs
;
215 memset(c
, 0, sizeof(*brw
->wm
.compile_data
));
216 c
->instruction
= instruction
;
217 c
->prog_instructions
= prog_instructions
;
221 memcpy(&c
->key
, key
, sizeof(*key
));
224 c
->env_param
= brw
->intel
.ctx
.FragmentProgram
.Parameters
;
226 brw_init_compile(brw
, &c
->func
);
228 if (!brw_wm_fs_emit(brw
, c
)) {
229 /* Fallback for fixed function and ARB_fp shaders. */
230 c
->dispatch_width
= 16;
231 brw_wm_payload_setup(brw
, c
);
232 brw_wm_non_glsl_emit(brw
, c
);
233 c
->prog_data
.dispatch_width
= 16;
236 /* Scratch space is used for register spilling */
237 if (c
->last_scratch
) {
238 uint32_t total_scratch
;
240 /* Per-thread scratch space is power-of-two sized. */
241 for (c
->prog_data
.total_scratch
= 1024;
242 c
->prog_data
.total_scratch
<= c
->last_scratch
;
243 c
->prog_data
.total_scratch
*= 2) {
246 total_scratch
= c
->prog_data
.total_scratch
* brw
->wm_max_threads
;
248 if (brw
->wm
.scratch_bo
&& total_scratch
> brw
->wm
.scratch_bo
->size
) {
249 drm_intel_bo_unreference(brw
->wm
.scratch_bo
);
250 brw
->wm
.scratch_bo
= NULL
;
252 if (brw
->wm
.scratch_bo
== NULL
) {
253 brw
->wm
.scratch_bo
= drm_intel_bo_alloc(intel
->bufmgr
,
260 c
->prog_data
.total_scratch
= 0;
263 if (unlikely(INTEL_DEBUG
& DEBUG_WM
))
264 fprintf(stderr
, "\n");
268 program
= brw_get_program(&c
->func
, &program_size
);
270 drm_intel_bo_unreference(brw
->wm
.prog_bo
);
271 brw
->wm
.prog_bo
= brw_upload_cache(&brw
->cache
, BRW_WM_PROG
,
272 &c
->key
, sizeof(c
->key
),
273 program
, program_size
,
274 &c
->prog_data
, sizeof(c
->prog_data
),
280 static void brw_wm_populate_key( struct brw_context
*brw
,
281 struct brw_wm_prog_key
*key
)
283 struct gl_context
*ctx
= &brw
->intel
.ctx
;
284 /* BRW_NEW_FRAGMENT_PROGRAM */
285 const struct brw_fragment_program
*fp
=
286 (struct brw_fragment_program
*)brw
->fragment_program
;
291 memset(key
, 0, sizeof(*key
));
293 /* Build the index for table lookup
296 key
->alpha_test
= ctx
->Color
.AlphaEnabled
;
297 if (fp
->program
.UsesKill
||
298 ctx
->Color
.AlphaEnabled
)
299 lookup
|= IZ_PS_KILL_ALPHATEST_BIT
;
301 if (fp
->program
.Base
.OutputsWritten
& BITFIELD64_BIT(FRAG_RESULT_DEPTH
))
302 lookup
|= IZ_PS_COMPUTES_DEPTH_BIT
;
306 lookup
|= IZ_DEPTH_TEST_ENABLE_BIT
;
308 if (ctx
->Depth
.Test
&&
309 ctx
->Depth
.Mask
) /* ?? */
310 lookup
|= IZ_DEPTH_WRITE_ENABLE_BIT
;
313 if (ctx
->Stencil
._Enabled
) {
314 lookup
|= IZ_STENCIL_TEST_ENABLE_BIT
;
316 if (ctx
->Stencil
.WriteMask
[0] ||
317 ctx
->Stencil
.WriteMask
[ctx
->Stencil
._BackFace
])
318 lookup
|= IZ_STENCIL_WRITE_ENABLE_BIT
;
323 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
324 if (ctx
->Line
.SmoothFlag
) {
325 if (brw
->intel
.reduced_primitive
== GL_LINES
) {
328 else if (brw
->intel
.reduced_primitive
== GL_TRIANGLES
) {
329 if (ctx
->Polygon
.FrontMode
== GL_LINE
) {
330 line_aa
= AA_SOMETIMES
;
332 if (ctx
->Polygon
.BackMode
== GL_LINE
||
333 (ctx
->Polygon
.CullFlag
&&
334 ctx
->Polygon
.CullFaceMode
== GL_BACK
))
337 else if (ctx
->Polygon
.BackMode
== GL_LINE
) {
338 line_aa
= AA_SOMETIMES
;
340 if ((ctx
->Polygon
.CullFlag
&&
341 ctx
->Polygon
.CullFaceMode
== GL_FRONT
))
347 key
->iz_lookup
= lookup
;
348 key
->line_aa
= line_aa
;
349 key
->stats_wm
= brw
->intel
.stats_wm
;
351 /* BRW_NEW_WM_INPUT_DIMENSIONS */
352 key
->proj_attrib_mask
= brw
->wm
.input_size_masks
[4-1];
355 key
->flat_shade
= (ctx
->Light
.ShadeModel
== GL_FLAT
);
358 key
->linear_color
= (ctx
->Hint
.PerspectiveCorrection
== GL_FASTEST
);
360 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
361 key
->clamp_fragment_color
= ctx
->Color
._ClampFragmentColor
;
364 for (i
= 0; i
< BRW_MAX_TEX_UNIT
; i
++) {
365 const struct gl_texture_unit
*unit
= &ctx
->Texture
.Unit
[i
];
367 if (unit
->_ReallyEnabled
) {
368 const struct gl_texture_object
*t
= unit
->_Current
;
369 const struct gl_texture_image
*img
= t
->Image
[0][t
->BaseLevel
];
370 struct gl_sampler_object
*sampler
= _mesa_get_samplerobj(ctx
, i
);
371 int swizzles
[SWIZZLE_NIL
+ 1] = {
381 /* GL_DEPTH_TEXTURE_MODE is normally handled through
382 * brw_wm_surface_state, but it applies to shadow compares as
383 * well and our shadow compares always return the result in
386 if (sampler
->CompareMode
== GL_COMPARE_R_TO_TEXTURE_ARB
) {
387 if (sampler
->DepthMode
== GL_ALPHA
) {
388 swizzles
[0] = SWIZZLE_ZERO
;
389 swizzles
[1] = SWIZZLE_ZERO
;
390 swizzles
[2] = SWIZZLE_ZERO
;
391 } else if (sampler
->DepthMode
== GL_LUMINANCE
) {
392 swizzles
[3] = SWIZZLE_ONE
;
393 } else if (sampler
->DepthMode
== GL_RED
) {
394 /* See table 3.23 of the GL 3.0 spec. */
395 swizzles
[1] = SWIZZLE_ZERO
;
396 swizzles
[2] = SWIZZLE_ZERO
;
397 swizzles
[3] = SWIZZLE_ONE
;
401 if (img
->InternalFormat
== GL_YCBCR_MESA
) {
402 key
->yuvtex_mask
|= 1 << i
;
403 if (img
->TexFormat
== MESA_FORMAT_YCBCR
)
404 key
->yuvtex_swap_mask
|= 1 << i
;
407 key
->tex_swizzles
[i
] =
408 MAKE_SWIZZLE4(swizzles
[GET_SWZ(t
->_Swizzle
, 0)],
409 swizzles
[GET_SWZ(t
->_Swizzle
, 1)],
410 swizzles
[GET_SWZ(t
->_Swizzle
, 2)],
411 swizzles
[GET_SWZ(t
->_Swizzle
, 3)]);
414 key
->tex_swizzles
[i
] = SWIZZLE_NOOP
;
419 key
->shadowtex_mask
= fp
->program
.Base
.ShadowSamplers
;
423 * Include the draw buffer origin and height so that we can calculate
424 * fragment position values relative to the bottom left of the drawable,
425 * from the incoming screen origin relative position we get as part of our
428 * This is only needed for the WM_WPOSXY opcode when the fragment program
429 * uses the gl_FragCoord input.
431 * We could avoid recompiling by including this as a constant referenced by
432 * our program, but if we were to do that it would also be nice to handle
433 * getting that constant updated at batchbuffer submit time (when we
434 * hold the lock and know where the buffer really is) rather than at emit
435 * time when we don't hold the lock and are just guessing. We could also
436 * just avoid using this as key data if the program doesn't use
439 * For DRI2 the origin_x/y will always be (0,0) but we still need the
440 * drawable height in order to invert the Y axis.
442 if (fp
->program
.Base
.InputsRead
& FRAG_BIT_WPOS
) {
443 key
->drawable_height
= ctx
->DrawBuffer
->Height
;
444 key
->render_to_fbo
= ctx
->DrawBuffer
->Name
!= 0;
448 key
->nr_color_regions
= ctx
->DrawBuffer
->_NumColorDrawBuffers
;
450 /* CACHE_NEW_VS_PROG */
451 key
->vp_outputs_written
= brw
->vs
.prog_data
->outputs_written
;
453 /* The unique fragment program ID */
454 key
->program_string_id
= fp
->id
;
458 static void brw_prepare_wm_prog(struct brw_context
*brw
)
460 struct brw_wm_prog_key key
;
461 struct brw_fragment_program
*fp
= (struct brw_fragment_program
*)
462 brw
->fragment_program
;
464 brw_wm_populate_key(brw
, &key
);
466 /* Make an early check for the key.
468 drm_intel_bo_unreference(brw
->wm
.prog_bo
);
469 brw
->wm
.prog_bo
= brw_search_cache(&brw
->cache
, BRW_WM_PROG
,
472 if (brw
->wm
.prog_bo
== NULL
)
473 do_wm_prog(brw
, fp
, &key
);
477 const struct brw_tracked_state brw_wm_prog
= {
479 .mesa
= (_NEW_COLOR
|
489 .brw
= (BRW_NEW_FRAGMENT_PROGRAM
|
490 BRW_NEW_WM_INPUT_DIMENSIONS
|
491 BRW_NEW_REDUCED_PRIMITIVE
),
492 .cache
= CACHE_NEW_VS_PROG
,
494 .prepare
= brw_prepare_wm_prog