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 (at) tungstengraphics.com> 30 */ 31 32 33 #include "brw_context.h" 34 #include "brw_state.h" 35 #include "brw_defines.h" 36 37 #include "main/macros.h" 38 #include "main/samplerobj.h" 39 40 41 /* Samplers aren't strictly wm state from the hardware's perspective, 42 * but that is the only situation in which we use them in this driver. 43 */ 44 45 46 47 uint32_t 48 translate_wrap_mode(GLenum wrap, bool using_nearest) 49 { 50 switch( wrap ) { 51 case GL_REPEAT: 52 return BRW_TEXCOORDMODE_WRAP; 53 case GL_CLAMP: 54 /* GL_CLAMP is the weird mode where coordinates are clamped to 55 * [0.0, 1.0], so linear filtering of coordinates outside of 56 * [0.0, 1.0] give you half edge texel value and half border 57 * color. The fragment shader will clamp the coordinates, and 58 * we set clamp_border here, which gets the result desired. We 59 * just use clamp(_to_edge) for nearest, because for nearest 60 * clamping to 1.0 gives border color instead of the desired 61 * edge texels. 62 */ 63 if (using_nearest) 64 return BRW_TEXCOORDMODE_CLAMP; 65 else 66 return BRW_TEXCOORDMODE_CLAMP_BORDER; 67 case GL_CLAMP_TO_EDGE: 68 return BRW_TEXCOORDMODE_CLAMP; 69 case GL_CLAMP_TO_BORDER: 70 return BRW_TEXCOORDMODE_CLAMP_BORDER; 71 case GL_MIRRORED_REPEAT: 72 return BRW_TEXCOORDMODE_MIRROR; 73 default: 74 return BRW_TEXCOORDMODE_WRAP; 75 } 76 } 77 78 /** 79 * Upload SAMPLER_BORDER_COLOR_STATE. 80 */ 81 void 82 upload_default_color(struct brw_context *brw, struct gl_sampler_object *sampler, 83 int unit, int ss_index) 84 { 85 struct intel_context *intel = &brw->intel; 86 struct gl_context *ctx = &intel->ctx; 87 struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; 88 struct gl_texture_object *texObj = texUnit->_Current; 89 struct gl_texture_image *firstImage = texObj->Image[0][texObj->BaseLevel]; 90 float color[4]; 91 92 if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) { 93 /* GL specs that border color for depth textures is taken from the 94 * R channel, while the hardware uses A. Spam R into all the 95 * channels for safety. 96 */ 97 color[0] = sampler->BorderColor.f[0]; 98 color[1] = sampler->BorderColor.f[0]; 99 color[2] = sampler->BorderColor.f[0]; 100 color[3] = sampler->BorderColor.f[0]; 101 } else { 102 color[0] = sampler->BorderColor.f[0]; 103 color[1] = sampler->BorderColor.f[1]; 104 color[2] = sampler->BorderColor.f[2]; 105 color[3] = sampler->BorderColor.f[3]; 106 } 107 108 if (intel->gen == 5 || intel->gen == 6) { 109 struct gen5_sampler_default_color *sdc; 110 111 sdc = brw_state_batch(brw, AUB_TRACE_SAMPLER_DEFAULT_COLOR, 112 sizeof(*sdc), 32, &brw->wm.sdc_offset[ss_index]); 113 114 memset(sdc, 0, sizeof(*sdc)); 115 116 UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[0], color[0]); 117 UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[1], color[1]); 118 UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[2], color[2]); 119 UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[3], color[3]); 120 121 UNCLAMPED_FLOAT_TO_USHORT(sdc->us[0], color[0]); 122 UNCLAMPED_FLOAT_TO_USHORT(sdc->us[1], color[1]); 123 UNCLAMPED_FLOAT_TO_USHORT(sdc->us[2], color[2]); 124 UNCLAMPED_FLOAT_TO_USHORT(sdc->us[3], color[3]); 125 126 UNCLAMPED_FLOAT_TO_SHORT(sdc->s[0], color[0]); 127 UNCLAMPED_FLOAT_TO_SHORT(sdc->s[1], color[1]); 128 UNCLAMPED_FLOAT_TO_SHORT(sdc->s[2], color[2]); 129 UNCLAMPED_FLOAT_TO_SHORT(sdc->s[3], color[3]); 130 131 sdc->hf[0] = _mesa_float_to_half(color[0]); 132 sdc->hf[1] = _mesa_float_to_half(color[1]); 133 sdc->hf[2] = _mesa_float_to_half(color[2]); 134 sdc->hf[3] = _mesa_float_to_half(color[3]); 135 136 sdc->b[0] = sdc->s[0] >> 8; 137 sdc->b[1] = sdc->s[1] >> 8; 138 sdc->b[2] = sdc->s[2] >> 8; 139 sdc->b[3] = sdc->s[3] >> 8; 140 141 sdc->f[0] = color[0]; 142 sdc->f[1] = color[1]; 143 sdc->f[2] = color[2]; 144 sdc->f[3] = color[3]; 145 } else { 146 struct brw_sampler_default_color *sdc; 147 148 sdc = brw_state_batch(brw, AUB_TRACE_SAMPLER_DEFAULT_COLOR, 149 sizeof(*sdc), 32, &brw->wm.sdc_offset[ss_index]); 150 151 COPY_4V(sdc->color, color); 152 } 153 } 154 155 /** 156 * Sets the sampler state for a single unit based off of the sampler key 157 * entry. 158 */ 159 static void brw_update_sampler_state(struct brw_context *brw, 160 int unit, 161 int ss_index, 162 struct brw_sampler_state *sampler) 163 { 164 struct intel_context *intel = &brw->intel; 165 struct gl_context *ctx = &intel->ctx; 166 struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; 167 struct gl_texture_object *texObj = texUnit->_Current; 168 struct gl_sampler_object *gl_sampler = _mesa_get_samplerobj(ctx, unit); 169 bool using_nearest = false; 170 171 /* These don't use samplers at all. */ 172 if (texObj->Target == GL_TEXTURE_BUFFER) 173 return; 174 175 switch (gl_sampler->MinFilter) { 176 case GL_NEAREST: 177 sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST; 178 sampler->ss0.mip_filter = BRW_MIPFILTER_NONE; 179 using_nearest = true; 180 break; 181 case GL_LINEAR: 182 sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR; 183 sampler->ss0.mip_filter = BRW_MIPFILTER_NONE; 184 break; 185 case GL_NEAREST_MIPMAP_NEAREST: 186 sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST; 187 sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST; 188 break; 189 case GL_LINEAR_MIPMAP_NEAREST: 190 sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR; 191 sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST; 192 break; 193 case GL_NEAREST_MIPMAP_LINEAR: 194 sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST; 195 sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR; 196 break; 197 case GL_LINEAR_MIPMAP_LINEAR: 198 sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR; 199 sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR; 200 break; 201 default: 202 break; 203 } 204 205 /* Set Anisotropy: 206 */ 207 if (gl_sampler->MaxAnisotropy > 1.0) { 208 sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC; 209 sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC; 210 211 if (gl_sampler->MaxAnisotropy > 2.0) { 212 sampler->ss3.max_aniso = MIN2((gl_sampler->MaxAnisotropy - 2) / 2, 213 BRW_ANISORATIO_16); 214 } 215 } 216 else { 217 switch (gl_sampler->MagFilter) { 218 case GL_NEAREST: 219 sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST; 220 using_nearest = true; 221 break; 222 case GL_LINEAR: 223 sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR; 224 break; 225 default: 226 break; 227 } 228 } 229 230 sampler->ss1.r_wrap_mode = translate_wrap_mode(gl_sampler->WrapR, 231 using_nearest); 232 sampler->ss1.s_wrap_mode = translate_wrap_mode(gl_sampler->WrapS, 233 using_nearest); 234 sampler->ss1.t_wrap_mode = translate_wrap_mode(gl_sampler->WrapT, 235 using_nearest); 236 237 if (intel->gen >= 6 && 238 sampler->ss0.min_filter != sampler->ss0.mag_filter) 239 sampler->ss0.min_mag_neq = 1; 240 241 /* Cube-maps on 965 and later must use the same wrap mode for all 3 242 * coordinate dimensions. Futher, only CUBE and CLAMP are valid. 243 */ 244 if (texObj->Target == GL_TEXTURE_CUBE_MAP) { 245 if (ctx->Texture.CubeMapSeamless && 246 (gl_sampler->MinFilter != GL_NEAREST || 247 gl_sampler->MagFilter != GL_NEAREST)) { 248 sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE; 249 sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE; 250 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE; 251 } else { 252 sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP; 253 sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP; 254 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP; 255 } 256 } else if (texObj->Target == GL_TEXTURE_1D) { 257 /* There's a bug in 1D texture sampling - it actually pays 258 * attention to the wrap_t value, though it should not. 259 * Override the wrap_t value here to GL_REPEAT to keep 260 * any nonexistent border pixels from floating in. 261 */ 262 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP; 263 } 264 265 266 /* Set shadow function: 267 */ 268 if (gl_sampler->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB) { 269 /* Shadowing is "enabled" by emitting a particular sampler 270 * message (sample_c). So need to recompile WM program when 271 * shadow comparison is enabled on each/any texture unit. 272 */ 273 sampler->ss0.shadow_function = 274 intel_translate_shadow_compare_func(gl_sampler->CompareFunc); 275 } 276 277 /* Set LOD bias: 278 */ 279 sampler->ss0.lod_bias = S_FIXED(CLAMP(texUnit->LodBias + 280 gl_sampler->LodBias, -16, 15), 6); 281 282 sampler->ss0.lod_preclamp = 1; /* OpenGL mode */ 283 sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */ 284 285 /* Set BaseMipLevel, MaxLOD, MinLOD: 286 * 287 * XXX: I don't think that using firstLevel, lastLevel works, 288 * because we always setup the surface state as if firstLevel == 289 * level zero. Probably have to subtract firstLevel from each of 290 * these: 291 */ 292 sampler->ss0.base_level = U_FIXED(0, 1); 293 294 sampler->ss1.max_lod = U_FIXED(CLAMP(gl_sampler->MaxLod, 0, 13), 6); 295 sampler->ss1.min_lod = U_FIXED(CLAMP(gl_sampler->MinLod, 0, 13), 6); 296 297 /* On Gen6+, the sampler can handle non-normalized texture 298 * rectangle coordinates natively 299 */ 300 if (intel->gen >= 6 && texObj->Target == GL_TEXTURE_RECTANGLE) { 301 sampler->ss3.non_normalized_coord = 1; 302 } 303 304 upload_default_color(brw, gl_sampler, unit, ss_index); 305 306 if (intel->gen >= 6) { 307 sampler->ss2.default_color_pointer = brw->wm.sdc_offset[ss_index] >> 5; 308 } else { 309 /* reloc */ 310 sampler->ss2.default_color_pointer = (intel->batch.bo->offset + 311 brw->wm.sdc_offset[ss_index]) >> 5; 312 313 drm_intel_bo_emit_reloc(intel->batch.bo, 314 brw->sampler.offset + 315 ss_index * sizeof(struct brw_sampler_state) + 316 offsetof(struct brw_sampler_state, ss2), 317 intel->batch.bo, brw->wm.sdc_offset[ss_index], 318 I915_GEM_DOMAIN_SAMPLER, 0); 319 } 320 321 if (sampler->ss0.min_filter != BRW_MAPFILTER_NEAREST) 322 sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MIN | 323 BRW_ADDRESS_ROUNDING_ENABLE_V_MIN | 324 BRW_ADDRESS_ROUNDING_ENABLE_R_MIN; 325 if (sampler->ss0.mag_filter != BRW_MAPFILTER_NEAREST) 326 sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MAG | 327 BRW_ADDRESS_ROUNDING_ENABLE_V_MAG | 328 BRW_ADDRESS_ROUNDING_ENABLE_R_MAG; 329 } 330 331 332 static void 333 brw_upload_samplers(struct brw_context *brw) 334 { 335 struct gl_context *ctx = &brw->intel.ctx; 336 struct brw_sampler_state *samplers; 337 338 /* BRW_NEW_VERTEX_PROGRAM and BRW_NEW_FRAGMENT_PROGRAM */ 339 struct gl_program *vs = (struct gl_program *) brw->vertex_program; 340 struct gl_program *fs = (struct gl_program *) brw->fragment_program; 341 342 GLbitfield SamplersUsed = vs->SamplersUsed | fs->SamplersUsed; 343 344 /* ARB programs use the texture unit number as the sampler index, so we 345 * need to find the highest unit used. A bit-count will not work. 346 */ 347 brw->sampler.count = _mesa_fls(SamplersUsed); 348 349 if (brw->sampler.count == 0) 350 return; 351 352 samplers = brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE, 353 brw->sampler.count * sizeof(*samplers), 354 32, &brw->sampler.offset); 355 memset(samplers, 0, brw->sampler.count * sizeof(*samplers)); 356 357 for (unsigned s = 0; s < brw->sampler.count; s++) { 358 if (SamplersUsed & (1 << s)) { 359 const unsigned unit = (fs->SamplersUsed & (1 << s)) ? 360 fs->SamplerUnits[s] : vs->SamplerUnits[s]; 361 if (ctx->Texture.Unit[unit]._ReallyEnabled) 362 brw_update_sampler_state(brw, unit, s, &samplers[s]); 363 } 364 } 365 366 brw->state.dirty.cache |= CACHE_NEW_SAMPLER; 367 } 368 369 const struct brw_tracked_state brw_samplers = { 370 .dirty = { 371 .mesa = _NEW_TEXTURE, 372 .brw = BRW_NEW_BATCH | 373 BRW_NEW_VERTEX_PROGRAM | 374 BRW_NEW_FRAGMENT_PROGRAM, 375 .cache = 0 376 }, 377 .emit = brw_upload_samplers, 378 }; 379 380 381