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 )
55 assert(opcode < MAX_OPCODE);
56 return _mesa_num_inst_src_regs(opcode);
61 GLuint brw_wm_is_scalar_result( GLuint opcode )
84 * Do GPU code generation for non-GLSL shader. non-GLSL shaders have
85 * no flow control instructions so we can more readily do SSA-style
89 brw_wm_non_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c)
91 /* Augment fragment program. Add instructions for pre- and
92 * post-fragment-program tasks such as interpolation and fogging.
96 /* Translate to intermediate representation. Build register usage
101 /* Dead code removal.
105 /* Register allocation.
107 c->grf_limit = BRW_WM_MAX_GRF / 2;
111 c->prog_data.total_grf = c->max_wm_grf;
112 if (c->last_scratch) {
113 c->prog_data.total_scratch = c->last_scratch + 0x40;
116 c->prog_data.total_scratch = 0;
126 * All Mesa program -> GPU code generation goes through this function.
127 * Depending on the instructions used (i.e. flow control instructions)
128 * we'll use one of two code generators.
130 static void do_wm_prog( struct brw_context *brw,
131 struct brw_fragment_program *fp,
132 struct brw_wm_prog_key *key)
134 struct brw_wm_compile *c;
135 const GLuint *program;
138 c = brw->wm.compile_data;
140 brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data));
141 c = brw->wm.compile_data;
143 memset(c, 0, sizeof(*brw->wm.compile_data));
145 memcpy(&c->key, key, sizeof(*key));
148 c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
150 brw_init_compile(brw, &c->func);
153 * Shader which use GLSL features such as flow control are handled
154 * differently from "simple" shaders.
156 if (brw_wm_is_glsl(&c->fp->program)) {
157 brw_wm_glsl_emit(brw, c);
160 brw_wm_non_glsl_emit(brw, c);
163 if (INTEL_DEBUG & DEBUG_WM)
164 fprintf(stderr, "\n");
168 program = brw_get_program(&c->func, &program_size);
170 dri_bo_unreference(brw->wm.prog_bo);
171 brw->wm.prog_bo = brw_upload_cache( &brw->cache, BRW_WM_PROG,
172 &c->key, sizeof(c->key),
174 program, program_size,
176 &brw->wm.prog_data );
181 static void brw_wm_populate_key( struct brw_context *brw,
182 struct brw_wm_prog_key *key )
184 GLcontext *ctx = &brw->intel.ctx;
185 /* BRW_NEW_FRAGMENT_PROGRAM */
186 const struct brw_fragment_program *fp =
187 (struct brw_fragment_program *)brw->fragment_program;
192 memset(key, 0, sizeof(*key));
194 /* Build the index for table lookup
197 if (fp->program.UsesKill ||
198 ctx->Color.AlphaEnabled)
199 lookup |= IZ_PS_KILL_ALPHATEST_BIT;
201 if (fp->program.Base.OutputsWritten & (1<<FRAG_RESULT_DEPTH))
202 lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
206 lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
208 if (ctx->Depth.Test &&
209 ctx->Depth.Mask) /* ?? */
210 lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
213 if (ctx->Stencil._Enabled) {
214 lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
216 if (ctx->Stencil.WriteMask[0] ||
217 ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
218 lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
223 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
224 if (ctx->Line.SmoothFlag) {
225 if (brw->intel.reduced_primitive == GL_LINES) {
228 else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
229 if (ctx->Polygon.FrontMode == GL_LINE) {
230 line_aa = AA_SOMETIMES;
232 if (ctx->Polygon.BackMode == GL_LINE ||
233 (ctx->Polygon.CullFlag &&
234 ctx->Polygon.CullFaceMode == GL_BACK))
237 else if (ctx->Polygon.BackMode == GL_LINE) {
238 line_aa = AA_SOMETIMES;
240 if ((ctx->Polygon.CullFlag &&
241 ctx->Polygon.CullFaceMode == GL_FRONT))
247 brw_wm_lookup_iz(line_aa,
252 /* BRW_NEW_WM_INPUT_DIMENSIONS */
253 key->projtex_mask = brw->wm.input_size_masks[4-1] >> (FRAG_ATTRIB_TEX0 - FRAG_ATTRIB_WPOS);
256 key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
259 for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
260 const struct gl_texture_unit *unit = &ctx->Texture.Unit[i];
262 if (unit->_ReallyEnabled) {
263 const struct gl_texture_object *t = unit->_Current;
264 const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
265 if (img->InternalFormat == GL_YCBCR_MESA) {
266 key->yuvtex_mask |= 1 << i;
267 if (img->TexFormat->MesaFormat == MESA_FORMAT_YCBCR)
268 key->yuvtex_swap_mask |= 1 << i;
271 key->tex_swizzles[i] = t->_Swizzle;
274 key->tex_swizzles[i] = SWIZZLE_NOOP;
279 key->shadowtex_mask = fp->program.Base.ShadowSamplers;
283 * Include the draw buffer origin and height so that we can calculate
284 * fragment position values relative to the bottom left of the drawable,
285 * from the incoming screen origin relative position we get as part of our
288 * We could avoid recompiling by including this as a constant referenced by
289 * our program, but if we were to do that it would also be nice to handle
290 * getting that constant updated at batchbuffer submit time (when we
291 * hold the lock and know where the buffer really is) rather than at emit
292 * time when we don't hold the lock and are just guessing. We could also
293 * just avoid using this as key data if the program doesn't use
296 * This pretty much becomes moot with DRI2 and redirected buffers anyway,
297 * as our origins will always be zero then.
299 if (brw->intel.driDrawable != NULL) {
300 key->origin_x = brw->intel.driDrawable->x;
301 key->origin_y = brw->intel.driDrawable->y;
302 key->drawable_height = brw->intel.driDrawable->h;
305 /* The unique fragment program ID */
306 key->program_string_id = fp->id;
310 static void brw_prepare_wm_prog(struct brw_context *brw)
312 struct brw_wm_prog_key key;
313 struct brw_fragment_program *fp = (struct brw_fragment_program *)
314 brw->fragment_program;
316 brw_wm_populate_key(brw, &key);
318 /* Make an early check for the key.
320 dri_bo_unreference(brw->wm.prog_bo);
321 brw->wm.prog_bo = brw_search_cache(&brw->cache, BRW_WM_PROG,
325 if (brw->wm.prog_bo == NULL)
326 do_wm_prog(brw, fp, &key);
330 const struct brw_tracked_state brw_wm_prog = {
332 .mesa = (_NEW_COLOR |
340 .brw = (BRW_NEW_FRAGMENT_PROGRAM |
341 BRW_NEW_WM_INPUT_DIMENSIONS |
342 BRW_NEW_REDUCED_PRIMITIVE),
345 .prepare = brw_prepare_wm_prog