brw_wm_state.c revision f75843a517bd188639e6866db2a7b04de3524e16
1/* 2 Copyright (C) Intel Corp. 2006. All Rights Reserved. 3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to 4 develop this 3D driver. 5 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: 13 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. 17 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. 25 26 **********************************************************************/ 27 /* 28 * Authors: 29 * Keith Whitwell <keith@tungstengraphics.com> 30 */ 31 32 33 34#include "brw_context.h" 35#include "brw_state.h" 36#include "brw_defines.h" 37#include "dri_bufmgr.h" 38#include "brw_wm.h" 39 40/*********************************************************************** 41 * WM unit - fragment programs and rasterization 42 */ 43 44struct brw_wm_unit_key { 45 unsigned int total_grf, total_scratch; 46 unsigned int urb_entry_read_length; 47 unsigned int curb_entry_read_length; 48 unsigned int dispatch_grf_start_reg; 49 50 unsigned int curbe_offset; 51 unsigned int urb_size; 52 53 unsigned int max_threads; 54 55 unsigned int nr_surfaces, sampler_count; 56 GLboolean uses_depth, computes_depth, uses_kill, is_glsl; 57 GLboolean polygon_stipple, stats_wm, line_stipple, offset_enable; 58 GLfloat offset_units, offset_factor; 59}; 60 61static void 62wm_unit_populate_key(struct brw_context *brw, struct brw_wm_unit_key *key) 63{ 64 const struct gl_fragment_program *fp = brw->fragment_program; 65 struct intel_context *intel = &brw->intel; 66 67 memset(key, 0, sizeof(*key)); 68 69 if (INTEL_DEBUG & DEBUG_SINGLE_THREAD) 70 key->max_threads = 1; 71 else 72 key->max_threads = 32; 73 74 /* CACHE_NEW_WM_PROG */ 75 key->total_grf = brw->wm.prog_data->total_grf; 76 key->urb_entry_read_length = brw->wm.prog_data->urb_read_length; 77 key->curb_entry_read_length = brw->wm.prog_data->curb_read_length; 78 key->dispatch_grf_start_reg = brw->wm.prog_data->first_curbe_grf; 79 key->total_scratch = ALIGN(brw->wm.prog_data->total_scratch, 1024); 80 81 /* BRW_NEW_URB_FENCE */ 82 key->urb_size = brw->urb.vsize; 83 84 /* BRW_NEW_CURBE_OFFSETS */ 85 key->curbe_offset = brw->curbe.wm_start; 86 87 /* CACHE_NEW_SURFACE */ 88 key->nr_surfaces = brw->wm.nr_surfaces; 89 90 /* CACHE_NEW_SAMPLER */ 91 key->sampler_count = brw->wm.sampler_count; 92 93 /* _NEW_POLYGONSTIPPLE */ 94 key->polygon_stipple = brw->attribs.Polygon->StippleFlag; 95 96 /* BRW_NEW_FRAGMENT_PROGRAM */ 97 key->uses_depth = (fp->Base.InputsRead & (1 << FRAG_ATTRIB_WPOS)) != 0; 98 99 /* as far as we can tell */ 100 key->computes_depth = 101 (fp->Base.OutputsWritten & (1 << FRAG_RESULT_DEPR)) != 0; 102 103 /* _NEW_COLOR */ 104 key->uses_kill = fp->UsesKill || brw->attribs.Color->AlphaEnabled; 105 key->is_glsl = brw_wm_is_glsl(fp); 106 107 /* XXX: This needs a flag to indicate when it changes. */ 108 key->stats_wm = intel->stats_wm; 109 110 /* _NEW_LINE */ 111 key->line_stipple = brw->attribs.Line->StippleFlag; 112 113 /* _NEW_POLYGON */ 114 key->offset_enable = brw->attribs.Polygon->OffsetFill; 115 key->offset_units = brw->attribs.Polygon->OffsetUnits; 116 key->offset_factor = brw->attribs.Polygon->OffsetFactor; 117} 118 119static dri_bo * 120wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key, 121 dri_bo **reloc_bufs) 122{ 123 struct brw_wm_unit_state wm; 124 dri_bo *bo; 125 126 memset(&wm, 0, sizeof(wm)); 127 128 wm.thread0.grf_reg_count = ALIGN(key->total_grf, 16) / 16 - 1; 129 wm.thread0.kernel_start_pointer = brw->wm.prog_bo->offset >> 6; /* reloc */ 130 wm.thread1.depth_coef_urb_read_offset = 1; 131 wm.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754; 132 wm.thread1.binding_table_entry_count = key->nr_surfaces; 133 134 if (key->total_scratch != 0) { 135 wm.thread2.scratch_space_base_pointer = 136 brw->wm.scratch_buffer->offset >> 10; /* reloc */ 137 wm.thread2.per_thread_scratch_space = key->total_scratch / 1024 - 1; 138 } else { 139 wm.thread2.scratch_space_base_pointer = 0; 140 wm.thread2.per_thread_scratch_space = 0; 141 } 142 143 wm.thread3.dispatch_grf_start_reg = key->dispatch_grf_start_reg; 144 wm.thread3.urb_entry_read_length = key->urb_entry_read_length; 145 wm.thread3.const_urb_entry_read_length = key->curb_entry_read_length; 146 wm.thread3.const_urb_entry_read_offset = key->curbe_offset * 2; 147 wm.thread3.urb_entry_read_offset = 0; 148 149 wm.wm4.sampler_count = (key->sampler_count + 1) / 4; 150 if (brw->wm.sampler_bo != NULL) { 151 /* reloc */ 152 wm.wm4.sampler_state_pointer = brw->wm.sampler_bo->offset >> 5; 153 } else { 154 wm.wm4.sampler_state_pointer = 0; 155 } 156 157 wm.wm5.program_uses_depth = key->uses_depth; 158 wm.wm5.program_computes_depth = key->computes_depth; 159 wm.wm5.program_uses_killpixel = key->uses_kill; 160 161 if (key->is_glsl) 162 wm.wm5.enable_8_pix = 1; 163 else 164 wm.wm5.enable_16_pix = 1; 165 166 wm.wm5.max_threads = key->max_threads - 1; 167 wm.wm5.thread_dispatch_enable = 1; /* AKA: color_write */ 168 wm.wm5.legacy_line_rast = 0; 169 wm.wm5.legacy_global_depth_bias = 0; 170 wm.wm5.early_depth_test = 1; /* never need to disable */ 171 wm.wm5.line_aa_region_width = 0; 172 wm.wm5.line_endcap_aa_region_width = 1; 173 174 wm.wm5.polygon_stipple = key->polygon_stipple; 175 176 if (key->offset_enable) { 177 wm.wm5.depth_offset = 1; 178 /* Something wierd going on with legacy_global_depth_bias, 179 * offset_constant, scaling and MRD. This value passes glean 180 * but gives some odd results elsewere (eg. the 181 * quad-offset-units test). 182 */ 183 wm.global_depth_offset_constant = key->offset_units * 2; 184 185 /* This is the only value that passes glean: 186 */ 187 wm.global_depth_offset_scale = key->offset_factor; 188 } 189 190 wm.wm5.line_stipple = key->line_stipple; 191 192 if (INTEL_DEBUG & DEBUG_STATS || key->stats_wm) 193 wm.wm4.stats_enable = 1; 194 195 bo = brw_upload_cache(&brw->cache, BRW_WM_UNIT, 196 key, sizeof(*key), 197 reloc_bufs, 3, 198 &wm, sizeof(wm), 199 NULL, NULL); 200 201 /* Emit WM program relocation */ 202 intel_bo_emit_reloc(bo, 203 I915_GEM_DOMAIN_INSTRUCTION, 0, 204 wm.thread0.grf_reg_count << 1, 205 offsetof(struct brw_wm_unit_state, thread0), 206 brw->wm.prog_bo); 207 208 /* Emit scratch space relocation */ 209 if (key->total_scratch != 0) { 210 intel_bo_emit_reloc(bo, 211 0, 0, 212 wm.thread2.per_thread_scratch_space, 213 offsetof(struct brw_wm_unit_state, thread2), 214 brw->wm.scratch_buffer); 215 } 216 217 /* Emit sampler state relocation */ 218 if (key->sampler_count != 0) { 219 intel_bo_emit_reloc(bo, 220 I915_GEM_DOMAIN_INSTRUCTION, 0, 221 wm.wm4.stats_enable | (wm.wm4.sampler_count << 2), 222 offsetof(struct brw_wm_unit_state, wm4), 223 brw->wm.sampler_bo); 224 } 225 226 return bo; 227} 228 229 230static void upload_wm_unit( struct brw_context *brw ) 231{ 232 struct intel_context *intel = &brw->intel; 233 struct brw_wm_unit_key key; 234 dri_bo *reloc_bufs[3]; 235 wm_unit_populate_key(brw, &key); 236 237 /* Allocate the necessary scratch space if we haven't already. Don't 238 * bother reducing the allocation later, since we use scratch so 239 * rarely. 240 */ 241 assert(key.total_scratch <= 12 * 1024); 242 if (key.total_scratch) { 243 GLuint total = key.total_scratch * key.max_threads; 244 245 if (brw->wm.scratch_buffer && total > brw->wm.scratch_buffer->size) { 246 dri_bo_unreference(brw->wm.scratch_buffer); 247 brw->wm.scratch_buffer = NULL; 248 } 249 if (brw->wm.scratch_buffer == NULL) { 250 brw->wm.scratch_buffer = dri_bo_alloc(intel->bufmgr, 251 "wm scratch", 252 total, 253 4096); 254 } 255 } 256 257 reloc_bufs[0] = brw->wm.prog_bo; 258 reloc_bufs[1] = brw->wm.scratch_buffer; 259 reloc_bufs[2] = brw->wm.sampler_bo; 260 261 dri_bo_unreference(brw->wm.state_bo); 262 brw->wm.state_bo = brw_search_cache(&brw->cache, BRW_WM_UNIT, 263 &key, sizeof(key), 264 reloc_bufs, 3, 265 NULL); 266 if (brw->wm.state_bo == NULL) { 267 brw->wm.state_bo = wm_unit_create_from_key(brw, &key, reloc_bufs); 268 } 269} 270 271const struct brw_tracked_state brw_wm_unit = { 272 .dirty = { 273 .mesa = (_NEW_POLYGON | 274 _NEW_POLYGONSTIPPLE | 275 _NEW_LINE | 276 _NEW_COLOR), 277 278 .brw = (BRW_NEW_FRAGMENT_PROGRAM | 279 BRW_NEW_CURBE_OFFSETS | 280 BRW_NEW_LOCK), 281 282 .cache = (CACHE_NEW_SURFACE | 283 CACHE_NEW_WM_PROG | 284 CACHE_NEW_SAMPLER) 285 }, 286 .prepare = upload_wm_unit, 287}; 288 289