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_wm.h" 35 36 37 static GLuint get_tracked_mask(struct brw_wm_compile *c, 38 struct brw_wm_instruction *inst) 39 { 40 GLuint i; 41 for (i = 0; i < 4; i++) { 42 if (inst->writemask & (1<<i)) { 43 if (!inst->dst[i]->contributes_to_output) { 44 inst->writemask &= ~(1<<i); 45 inst->dst[i] = 0; 46 } 47 } 48 } 49 50 return inst->writemask; 51 } 52 53 /* Remove a reference from a value's usage chain. 54 */ 55 static void unlink_ref(struct brw_wm_ref *ref) 56 { 57 struct brw_wm_value *value = ref->value; 58 59 if (ref == value->lastuse) { 60 value->lastuse = ref->prevuse; 61 } 62 else { 63 struct brw_wm_ref *i = value->lastuse; 64 while (i->prevuse != ref) i = i->prevuse; 65 i->prevuse = ref->prevuse; 66 } 67 } 68 69 static void track_arg(struct brw_wm_compile *c, 70 struct brw_wm_instruction *inst, 71 GLuint arg, 72 GLuint readmask) 73 { 74 GLuint i; 75 76 for (i = 0; i < 4; i++) { 77 struct brw_wm_ref *ref = inst->src[arg][i]; 78 if (ref) { 79 if (readmask & (1<<i)) { 80 ref->value->contributes_to_output = 1; 81 } 82 else { 83 unlink_ref(ref); 84 inst->src[arg][i] = NULL; 85 } 86 } 87 } 88 } 89 90 static GLuint get_texcoord_mask( GLuint tex_idx ) 91 { 92 switch (tex_idx) { 93 case TEXTURE_1D_INDEX: 94 return WRITEMASK_X; 95 case TEXTURE_2D_INDEX: 96 case TEXTURE_1D_ARRAY_INDEX: 97 case TEXTURE_EXTERNAL_INDEX: 98 return WRITEMASK_XY; 99 case TEXTURE_3D_INDEX: 100 case TEXTURE_2D_ARRAY_INDEX: 101 return WRITEMASK_XYZ; 102 case TEXTURE_CUBE_INDEX: 103 return WRITEMASK_XYZ; 104 case TEXTURE_RECT_INDEX: 105 return WRITEMASK_XY; 106 default: return 0; 107 } 108 } 109 110 111 /* Step two: Basically this is dead code elimination. 112 * 113 * Iterate backwards over instructions, noting which values 114 * contribute to the final result. Adjust writemasks to only 115 * calculate these values. 116 */ 117 void brw_wm_pass1( struct brw_wm_compile *c ) 118 { 119 GLint insn; 120 121 for (insn = c->nr_insns-1; insn >= 0; insn--) { 122 struct brw_wm_instruction *inst = &c->instruction[insn]; 123 GLuint writemask; 124 GLuint read0, read1, read2; 125 126 if (inst->opcode == OPCODE_KIL) { 127 track_arg(c, inst, 0, WRITEMASK_XYZW); /* All args contribute to final */ 128 continue; 129 } 130 131 if (inst->opcode == WM_FB_WRITE) { 132 track_arg(c, inst, 0, WRITEMASK_XYZW); 133 track_arg(c, inst, 1, WRITEMASK_XYZW); 134 if (c->source_depth_to_render_target && c->computes_depth) 135 track_arg(c, inst, 2, WRITEMASK_Z); 136 else 137 track_arg(c, inst, 2, 0); 138 continue; 139 } 140 141 /* Lookup all the registers which were written by this 142 * instruction and get a mask of those that contribute to the output: 143 */ 144 writemask = get_tracked_mask(c, inst); 145 if (!writemask) { 146 GLuint arg; 147 for (arg = 0; arg < 3; arg++) 148 track_arg(c, inst, arg, 0); 149 continue; 150 } 151 152 read0 = 0; 153 read1 = 0; 154 read2 = 0; 155 156 /* Mark all inputs which contribute to the marked outputs: 157 */ 158 switch (inst->opcode) { 159 case OPCODE_ABS: 160 case OPCODE_FLR: 161 case OPCODE_FRC: 162 case OPCODE_MOV: 163 case OPCODE_SSG: 164 case OPCODE_SWZ: 165 case OPCODE_TRUNC: 166 read0 = writemask; 167 break; 168 169 case OPCODE_SUB: 170 case OPCODE_SLT: 171 case OPCODE_SLE: 172 case OPCODE_SGE: 173 case OPCODE_SGT: 174 case OPCODE_SEQ: 175 case OPCODE_SNE: 176 case OPCODE_ADD: 177 case OPCODE_MAX: 178 case OPCODE_MIN: 179 case OPCODE_MUL: 180 read0 = writemask; 181 read1 = writemask; 182 break; 183 184 case OPCODE_DDX: 185 case OPCODE_DDY: 186 read0 = writemask; 187 break; 188 189 case OPCODE_MAD: 190 case OPCODE_CMP: 191 case OPCODE_LRP: 192 read0 = writemask; 193 read1 = writemask; 194 read2 = writemask; 195 break; 196 197 case OPCODE_XPD: 198 if (writemask & WRITEMASK_X) read0 |= WRITEMASK_YZ; 199 if (writemask & WRITEMASK_Y) read0 |= WRITEMASK_XZ; 200 if (writemask & WRITEMASK_Z) read0 |= WRITEMASK_XY; 201 read1 = read0; 202 break; 203 204 case OPCODE_COS: 205 case OPCODE_EX2: 206 case OPCODE_LG2: 207 case OPCODE_RCP: 208 case OPCODE_RSQ: 209 case OPCODE_SIN: 210 case OPCODE_SCS: 211 case WM_CINTERP: 212 case WM_PIXELXY: 213 read0 = WRITEMASK_X; 214 break; 215 216 case OPCODE_POW: 217 read0 = WRITEMASK_X; 218 read1 = WRITEMASK_X; 219 break; 220 221 case OPCODE_TEX: 222 case OPCODE_TXP: 223 read0 = get_texcoord_mask(inst->tex_idx); 224 225 if (inst->tex_shadow) 226 read0 |= WRITEMASK_Z; 227 break; 228 229 case OPCODE_TXB: 230 /* Shadow ignored for txb. 231 */ 232 read0 = get_texcoord_mask(inst->tex_idx) | WRITEMASK_W; 233 break; 234 235 case WM_WPOSXY: 236 read0 = writemask & WRITEMASK_XY; 237 break; 238 239 case WM_DELTAXY: 240 read0 = writemask & WRITEMASK_XY; 241 read1 = WRITEMASK_X; 242 break; 243 244 case WM_PIXELW: 245 read0 = WRITEMASK_X; 246 read1 = WRITEMASK_XY; 247 break; 248 249 case WM_LINTERP: 250 read0 = WRITEMASK_X; 251 read1 = WRITEMASK_XY; 252 break; 253 254 case WM_PINTERP: 255 read0 = WRITEMASK_X; /* interpolant */ 256 read1 = WRITEMASK_XY; /* deltas */ 257 read2 = WRITEMASK_W; /* pixel w */ 258 break; 259 260 case OPCODE_DP2: 261 read0 = WRITEMASK_XY; 262 read1 = WRITEMASK_XY; 263 break; 264 265 case OPCODE_DP3: 266 read0 = WRITEMASK_XYZ; 267 read1 = WRITEMASK_XYZ; 268 break; 269 270 case OPCODE_DPH: 271 read0 = WRITEMASK_XYZ; 272 read1 = WRITEMASK_XYZW; 273 break; 274 275 case OPCODE_DP4: 276 read0 = WRITEMASK_XYZW; 277 read1 = WRITEMASK_XYZW; 278 break; 279 280 case OPCODE_LIT: 281 read0 = WRITEMASK_XYW; 282 break; 283 284 case OPCODE_DST: 285 case WM_FRONTFACING: 286 default: 287 break; 288 } 289 290 track_arg(c, inst, 0, read0); 291 track_arg(c, inst, 1, read1); 292 track_arg(c, inst, 2, read2); 293 } 294 295 if (unlikely(INTEL_DEBUG & DEBUG_WM)) { 296 brw_wm_print_program(c, "pass1"); 297 } 298 } 299
