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 "main/texformat.h"
33 #include "brw_context.h"
36 #include "brw_state.h"
39 /** Return number of src args for given instruction */
40 GLuint brw_wm_nr_args( GLuint opcode )
57 assert(opcode < MAX_OPCODE);
58 return _mesa_num_inst_src_regs(opcode);
63 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;
118 /* Scratch space is used for register spilling */
119 if (c->last_scratch) {
120 c->prog_data.total_scratch = c->last_scratch + 0x40;
123 c->prog_data.total_scratch = 0;
133 * All Mesa program -> GPU code generation goes through this function.
134 * Depending on the instructions used (i.e. flow control instructions)
135 * we'll use one of two code generators.
137 static void do_wm_prog( struct brw_context *brw,
138 struct brw_fragment_program *fp,
139 struct brw_wm_prog_key *key)
141 struct brw_wm_compile *c;
142 const GLuint *program;
145 c = brw->wm.compile_data;
147 brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data));
148 c = brw->wm.compile_data;
150 /* Ouch - big out of memory problem. Can't continue
151 * without triggering a segfault, no way to signal,
157 memset(c, 0, sizeof(*brw->wm.compile_data));
159 memcpy(&c->key, key, sizeof(*key));
162 c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
164 brw_init_compile(brw, &c->func);
166 /* temporary sanity check assertion */
167 ASSERT(fp->isGLSL == brw_wm_is_glsl(&c->fp->program));
170 * Shader which use GLSL features such as flow control are handled
171 * differently from "simple" shaders.
174 brw_wm_glsl_emit(brw, c);
177 brw_wm_non_glsl_emit(brw, c);
180 if (INTEL_DEBUG & DEBUG_WM)
181 fprintf(stderr, "\n");
185 program = brw_get_program(&c->func, &program_size);
187 dri_bo_unreference(brw->wm.prog_bo);
188 brw->wm.prog_bo = brw_upload_cache( &brw->cache, BRW_WM_PROG,
189 &c->key, sizeof(c->key),
191 program, program_size,
193 &brw->wm.prog_data );
198 static void brw_wm_populate_key( struct brw_context *brw,
199 struct brw_wm_prog_key *key )
201 GLcontext *ctx = &brw->intel.ctx;
202 /* BRW_NEW_FRAGMENT_PROGRAM */
203 const struct brw_fragment_program *fp =
204 (struct brw_fragment_program *)brw->fragment_program;
209 memset(key, 0, sizeof(*key));
211 /* Build the index for table lookup
214 if (fp->program.UsesKill ||
215 ctx->Color.AlphaEnabled)
216 lookup |= IZ_PS_KILL_ALPHATEST_BIT;
218 if (fp->program.Base.OutputsWritten & (1<<FRAG_RESULT_DEPTH))
219 lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
223 lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
225 if (ctx->Depth.Test &&
226 ctx->Depth.Mask) /* ?? */
227 lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
230 if (ctx->Stencil._Enabled) {
231 lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
233 if (ctx->Stencil.WriteMask[0] ||
234 ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
235 lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
240 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
241 if (ctx->Line.SmoothFlag) {
242 if (brw->intel.reduced_primitive == GL_LINES) {
245 else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
246 if (ctx->Polygon.FrontMode == GL_LINE) {
247 line_aa = AA_SOMETIMES;
249 if (ctx->Polygon.BackMode == GL_LINE ||
250 (ctx->Polygon.CullFlag &&
251 ctx->Polygon.CullFaceMode == GL_BACK))
254 else if (ctx->Polygon.BackMode == GL_LINE) {
255 line_aa = AA_SOMETIMES;
257 if ((ctx->Polygon.CullFlag &&
258 ctx->Polygon.CullFaceMode == GL_FRONT))
264 brw_wm_lookup_iz(line_aa,
269 /* BRW_NEW_WM_INPUT_DIMENSIONS */
270 key->proj_attrib_mask = brw->wm.input_size_masks[4-1];
273 key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
276 key->linear_color = (ctx->Hint.PerspectiveCorrection == GL_FASTEST);
279 for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
280 const struct gl_texture_unit *unit = &ctx->Texture.Unit[i];
282 if (unit->_ReallyEnabled) {
283 const struct gl_texture_object *t = unit->_Current;
284 const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
285 if (img->InternalFormat == GL_YCBCR_MESA) {
286 key->yuvtex_mask |= 1 << i;
287 if (img->TexFormat->MesaFormat == MESA_FORMAT_YCBCR)
288 key->yuvtex_swap_mask |= 1 << i;
291 key->tex_swizzles[i] = t->_Swizzle;
294 key->tex_swizzles[i] = SWIZZLE_NOOP;
299 key->shadowtex_mask = fp->program.Base.ShadowSamplers;
303 * Include the draw buffer origin and height so that we can calculate
304 * fragment position values relative to the bottom left of the drawable,
305 * from the incoming screen origin relative position we get as part of our
308 * We could avoid recompiling by including this as a constant referenced by
309 * our program, but if we were to do that it would also be nice to handle
310 * getting that constant updated at batchbuffer submit time (when we
311 * hold the lock and know where the buffer really is) rather than at emit
312 * time when we don't hold the lock and are just guessing. We could also
313 * just avoid using this as key data if the program doesn't use
316 * This pretty much becomes moot with DRI2 and redirected buffers anyway,
317 * as our origins will always be zero then.
319 if (brw->intel.driDrawable != NULL) {
320 key->origin_x = brw->intel.driDrawable->x;
321 key->origin_y = brw->intel.driDrawable->y;
322 key->drawable_height = brw->intel.driDrawable->h;
325 /* CACHE_NEW_VS_PROG */
326 key->vp_outputs_written = brw->vs.prog_data->outputs_written & DO_SETUP_BITS;
328 /* The unique fragment program ID */
329 key->program_string_id = fp->id;
333 static void brw_prepare_wm_prog(struct brw_context *brw)
335 struct brw_wm_prog_key key;
336 struct brw_fragment_program *fp = (struct brw_fragment_program *)
337 brw->fragment_program;
339 brw_wm_populate_key(brw, &key);
341 /* Make an early check for the key.
343 dri_bo_unreference(brw->wm.prog_bo);
344 brw->wm.prog_bo = brw_search_cache(&brw->cache, BRW_WM_PROG,
348 if (brw->wm.prog_bo == NULL)
349 do_wm_prog(brw, fp, &key);
353 const struct brw_tracked_state brw_wm_prog = {
355 .mesa = (_NEW_COLOR |
364 .brw = (BRW_NEW_FRAGMENT_PROGRAM |
365 BRW_NEW_WM_INPUT_DIMENSIONS |
366 BRW_NEW_REDUCED_PRIMITIVE),
367 .cache = CACHE_NEW_VS_PROG,
369 .prepare = brw_prepare_wm_prog