src/mesa/drivers/dri/r300/compiler/radeon_program.c

   1 /*
   2  * Copyright (C) 2008 Nicolai Haehnle.
   3  *
   4  * All Rights Reserved.
   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 #include "radeon_program.h"
  29
  30 #include <stdio.h>
  31
  32 #include "radeon_compiler.h"
  33
  34
  35 /**
  36  * Transform the given clause in the following way:
  37  *  1. Replace it with an empty clause
  38  *  2. For every instruction in the original clause, try the given
  39  *     transformations in order.
  40  *  3. If one of the transformations returns GL_TRUE, assume that it
  41  *     has emitted the appropriate instruction(s) into the new clause;
  42  *     otherwise, copy the instruction verbatim.
  43  *
  44  * \note The transformation is currently not recursive; in other words,
  45  * instructions emitted by transformations are not transformed.
  46  *
  47  * \note The transform is called 'local' because it can only look at
  48  * one instruction at a time.
  49  */
  50 void radeonLocalTransform(
  51         struct radeon_compiler * c,
  52         int num_transformations,
  53         struct radeon_program_transformation* transformations)
  54 {
  55         struct rc_instruction * inst = c->Program.Instructions.Next;
  56
  57         while(inst != &c->Program.Instructions) {
  58                 struct rc_instruction * current = inst;
  59                 int i;
  60
  61                 inst = inst->Next;
  62
  63                 for(i = 0; i < num_transformations; ++i) {
  64                         struct radeon_program_transformation* t = transformations + i;
  65
  66                         if (t->function(c, current, t->userData))
  67                                 break;
  68                 }
  69         }
  70 }
  71
  72 /**
  73  * Left multiplication of a register with a swizzle
  74  */
  75 struct rc_src_register lmul_swizzle(unsigned int swizzle, struct rc_src_register srcreg)
  76 {
  77         struct rc_src_register tmp = srcreg;
  78         int i;
  79         tmp.Swizzle = 0;
  80         tmp.Negate = 0;
  81         for(i = 0; i < 4; ++i) {
  82                 rc_swizzle swz = GET_SWZ(swizzle, i);
  83                 if (swz < 4) {
  84                         tmp.Swizzle |= GET_SWZ(srcreg.Swizzle, swz) << (i*3);
  85                         tmp.Negate |= GET_BIT(srcreg.Negate, swz) << i;
  86                 } else {
  87                         tmp.Swizzle |= swz << (i*3);
  88                 }
  89         }
  90         return tmp;
  91 }
  92
  93 unsigned int rc_find_free_temporary(struct radeon_compiler * c)
  94 {
  95         char used[RC_REGISTER_MAX_INDEX];
  96         unsigned int i;
  97
  98         memset(used, 0, sizeof(used));
  99
 100         for (struct rc_instruction * rcinst = c->Program.Instructions.Next; rcinst != &c->Program.Instructions; rcinst = rcinst->Next) {
 101                 const struct rc_sub_instruction *inst = &rcinst->I;
 102                 const struct rc_opcode_info *opcode = rc_get_opcode_info(inst->Opcode);
 103                 unsigned int k;
 104
 105                 for (k = 0; k < opcode->NumSrcRegs; k++) {
 106                         if (inst->SrcReg[k].File == RC_FILE_TEMPORARY)
 107                                 used[inst->SrcReg[k].Index] = 1;
 108                 }
 109
 110                 if (opcode->HasDstReg) {
 111                         if (inst->DstReg.File == RC_FILE_TEMPORARY)
 112                                 used[inst->DstReg.Index] = 1;
 113                 }
 114         }
 115
 116         for (i = 0; i < RC_REGISTER_MAX_INDEX; i++) {
 117                 if (!used[i])
 118                         return i;
 119         }
 120
 121         rc_error(c, "Ran out of temporary registers\n");
 122         return 0;
 123 }
 124
 125
 126 struct rc_instruction *rc_alloc_instruction(struct radeon_compiler * c)
 127 {
 128         struct rc_instruction * inst = memory_pool_malloc(&c->Pool, sizeof(struct rc_instruction));
 129
 130         memset(inst, 0, sizeof(struct rc_instruction));
 131
 132         inst->I.Opcode = RC_OPCODE_ILLEGAL_OPCODE;
 133         inst->I.DstReg.WriteMask = RC_MASK_XYZW;
 134         inst->I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZW;
 135         inst->I.SrcReg[1].Swizzle = RC_SWIZZLE_XYZW;
 136         inst->I.SrcReg[2].Swizzle = RC_SWIZZLE_XYZW;
 137
 138         return inst;
 139 }
 140
 141
 142 struct rc_instruction *rc_insert_new_instruction(struct radeon_compiler * c, struct rc_instruction * after)
 143 {
 144         struct rc_instruction * inst = rc_alloc_instruction(c);
 145
 146         inst->Prev = after;
 147         inst->Next = after->Next;
 148
 149         inst->Prev->Next = inst;
 150         inst->Next->Prev = inst;
 151
 152         return inst;
 153 }
 154
 155 void rc_remove_instruction(struct rc_instruction * inst)
 156 {
 157         inst->Prev->Next = inst->Next;
 158         inst->Next->Prev = inst->Prev;
 159 }
 160
 161 static const char * textarget_to_string(rc_texture_target target)
 162 {
 163         switch(target) {
 164         case RC_TEXTURE_2D_ARRAY: return "2D_ARRAY";
 165         case RC_TEXTURE_1D_ARRAY: return "1D_ARRAY";
 166         case RC_TEXTURE_CUBE: return "CUBE";
 167         case RC_TEXTURE_3D: return "3D";
 168         case RC_TEXTURE_RECT: return "RECT";
 169         case RC_TEXTURE_2D: return "2D";
 170         case RC_TEXTURE_1D: return "1D";
 171         default: return "BAD_TEXTURE_TARGET";
 172         }
 173 }
 174
 175 static void rc_print_register(FILE * f, rc_register_file file, int index, unsigned int reladdr)
 176 {
 177         if (file == RC_FILE_NONE) {
 178                 fprintf(f, "none");
 179         } else {
 180                 const char * filename;
 181                 switch(file) {
 182                 case RC_FILE_TEMPORARY: filename = "temp"; break;
 183                 case RC_FILE_INPUT: filename = "input"; break;
 184                 case RC_FILE_OUTPUT: filename = "output"; break;
 185                 case RC_FILE_ADDRESS: filename = "addr"; break;
 186                 case RC_FILE_CONSTANT: filename = "const"; break;
 187                 default: filename = "BAD FILE"; break;
 188                 }
 189                 fprintf(f, "%s[%i%s]", filename, index, reladdr ? " + addr[0]" : "");
 190         }
 191 }
 192
 193 static void rc_print_mask(FILE * f, unsigned int mask)
 194 {
 195         if (mask & RC_MASK_X) fprintf(f, "x");
 196         if (mask & RC_MASK_Y) fprintf(f, "y");
 197         if (mask & RC_MASK_Z) fprintf(f, "z");
 198         if (mask & RC_MASK_W) fprintf(f, "w");
 199 }
 200
 201 static void rc_print_dst_register(FILE * f, struct rc_dst_register dst)
 202 {
 203         rc_print_register(f, dst.File, dst.Index, dst.RelAddr);
 204         if (dst.WriteMask != RC_MASK_XYZW) {
 205                 fprintf(f, ".");
 206                 rc_print_mask(f, dst.WriteMask);
 207         }
 208 }
 209
 210 static void rc_print_swizzle(FILE * f, unsigned int swizzle, unsigned int negate)
 211 {
 212         unsigned int comp;
 213         for(comp = 0; comp < 4; ++comp) {
 214                 rc_swizzle swz = GET_SWZ(swizzle, comp);
 215                 if (GET_BIT(negate, comp))
 216                         fprintf(f, "-");
 217                 switch(swz) {
 218                 case RC_SWIZZLE_X: fprintf(f, "x"); break;
 219                 case RC_SWIZZLE_Y: fprintf(f, "y"); break;
 220                 case RC_SWIZZLE_Z: fprintf(f, "z"); break;
 221                 case RC_SWIZZLE_W: fprintf(f, "w"); break;
 222                 case RC_SWIZZLE_ZERO: fprintf(f, "0"); break;
 223                 case RC_SWIZZLE_ONE: fprintf(f, "1"); break;
 224                 case RC_SWIZZLE_HALF: fprintf(f, "H"); break;
 225                 case RC_SWIZZLE_UNUSED: fprintf(f, "_"); break;
 226                 }
 227         }
 228 }
 229
 230 static void rc_print_src_register(FILE * f, struct rc_src_register src)
 231 {
 232         int trivial_negate = (src.Negate == RC_MASK_NONE || src.Negate == RC_MASK_XYZW);
 233
 234         if (src.Negate == RC_MASK_XYZW)
 235                 fprintf(f, "-");
 236         if (src.Abs)
 237                 fprintf(f, "|");
 238
 239         rc_print_register(f, src.File, src.Index, src.RelAddr);
 240
 241         if (src.Abs && !trivial_negate)
 242                 fprintf(f, "|");
 243
 244         if (src.Swizzle != RC_SWIZZLE_XYZW || !trivial_negate) {
 245                 fprintf(f, ".");
 246                 rc_print_swizzle(f, src.Swizzle, trivial_negate ? 0 : src.Negate);
 247         }
 248
 249         if (src.Abs && trivial_negate)
 250                 fprintf(f, "|");
 251 }
 252
 253 static void rc_print_instruction(FILE * f, struct rc_instruction * inst)
 254 {
 255         const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode);
 256         unsigned int reg;
 257
 258         fprintf(f, "%s", opcode->Name);
 259
 260         switch(inst->I.SaturateMode) {
 261         case RC_SATURATE_NONE: break;
 262         case RC_SATURATE_ZERO_ONE: fprintf(f, "_SAT"); break;
 263         case RC_SATURATE_MINUS_PLUS_ONE: fprintf(f, "_SAT2"); break;
 264         default: fprintf(f, "_BAD_SAT"); break;
 265         }
 266
 267         if (opcode->HasDstReg) {
 268                 fprintf(f, " ");
 269                 rc_print_dst_register(f, inst->I.DstReg);
 270                 if (opcode->NumSrcRegs)
 271                         fprintf(f, ",");
 272         }
 273
 274         for(reg = 0; reg < opcode->NumSrcRegs; ++reg) {
 275                 if (reg > 0)
 276                         fprintf(f, ",");
 277                 fprintf(f, " ");
 278                 rc_print_src_register(f, inst->I.SrcReg[reg]);
 279         }
 280
 281         if (opcode->HasTexture) {
 282                 fprintf(f, ", %s%s[%u]",
 283                         textarget_to_string(inst->I.TexSrcTarget),
 284                         inst->I.TexShadow ? "SHADOW" : "",
 285                         inst->I.TexSrcUnit);
 286         }
 287
 288         fprintf(f, ";\n");
 289 }
 290
 291 /**
 292  * Print program to stderr, default options.
 293  */
 294 void rc_print_program(const struct rc_program *prog)
 295 {
 296         unsigned int linenum = 0;
 297         struct rc_instruction *inst;
 298
 299         fprintf(stderr, "# Radeon Compiler Program\n");
 300
 301         for(inst = prog->Instructions.Next; inst != &prog->Instructions; inst = inst->Next) {
 302                 fprintf(stderr, "%3d: ", linenum);
 303
 304                 rc_print_instruction(stderr, inst);
 305
 306                 linenum++;
 307         }
 308 }