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 #include "radeon_dataflow.h"
  34
  35
  36 /**
  37  * Transform the given clause in the following way:
  38  *  1. Replace it with an empty clause
  39  *  2. For every instruction in the original clause, try the given
  40  *     transformations in order.
  41  *  3. If one of the transformations returns GL_TRUE, assume that it
  42  *     has emitted the appropriate instruction(s) into the new clause;
  43  *     otherwise, copy the instruction verbatim.
  44  *
  45  * \note The transformation is currently not recursive; in other words,
  46  * instructions emitted by transformations are not transformed.
  47  *
  48  * \note The transform is called 'local' because it can only look at
  49  * one instruction at a time.
  50  */
  51 void rc_local_transform(
  52         struct radeon_compiler * c,
  53         void *user)
  54 {
  55         struct radeon_program_transformation *transformations =
  56                 (struct radeon_program_transformation*)user;
  57         struct rc_instruction * inst = c->Program.Instructions.Next;
  58
  59         while(inst != &c->Program.Instructions) {
  60                 struct rc_instruction * current = inst;
  61                 int i;
  62
  63                 inst = inst->Next;
  64
  65                 for(i = 0; transformations[i].function; ++i) {
  66                         struct radeon_program_transformation* t = transformations + i;
  67
  68                         if (t->function(c, current, t->userData))
  69                                 break;
  70                 }
  71         }
  72 }
  73
  74 /**
  75  * Left multiplication of a register with a swizzle
  76  */
  77 struct rc_src_register lmul_swizzle(unsigned int swizzle, struct rc_src_register srcreg)
  78 {
  79         struct rc_src_register tmp = srcreg;
  80         int i;
  81         tmp.Swizzle = 0;
  82         tmp.Negate = 0;
  83         for(i = 0; i < 4; ++i) {
  84                 rc_swizzle swz = GET_SWZ(swizzle, i);
  85                 if (swz < 4) {
  86                         tmp.Swizzle |= GET_SWZ(srcreg.Swizzle, swz) << (i*3);
  87                         tmp.Negate |= GET_BIT(srcreg.Negate, swz) << i;
  88                 } else {
  89                         tmp.Swizzle |= swz << (i*3);
  90                 }
  91         }
  92         return tmp;
  93 }
  94
  95 struct get_used_temporaries_data {
  96         unsigned char * Used;
  97         unsigned int UsedLength;
  98 };
  99
 100 static void get_used_temporaries_cb(
 101         void * userdata,
 102         struct rc_instruction * inst,
 103         rc_register_file file,
 104         unsigned int index,
 105         unsigned int mask)
 106 {
 107         struct get_used_temporaries_data * d = userdata;
 108
 109         if (file != RC_FILE_TEMPORARY)
 110                 return;
 111
 112         if (index >= d->UsedLength)
 113                 return;
 114
 115         d->Used[index] |= mask;
 116 }
 117
 118 /**
 119  * This function fills in the parameter 'used' with a writemask that
 120  * represent which components of each temporary register are used by the
 121  * program.  This is meant to be combined with rc_find_free_temporary_list as a
 122  * more efficient version of rc_find_free_temporary.
 123  * @param used The function does not initialize this parameter.
 124  */
 125 void rc_get_used_temporaries(
 126         struct radeon_compiler * c,
 127         unsigned char * used,
 128         unsigned int used_length)
 129 {
 130         struct rc_instruction * inst;
 131         struct get_used_temporaries_data d;
 132         d.Used = used;
 133         d.UsedLength = used_length;
 134
 135         for(inst = c->Program.Instructions.Next;
 136                         inst != &c->Program.Instructions; inst = inst->Next) {
 137
 138                 rc_for_all_reads_mask(inst, get_used_temporaries_cb, &d);
 139                 rc_for_all_writes_mask(inst, get_used_temporaries_cb, &d);
 140         }
 141 }
 142
 143 /* Search a list of used temporaries for a free one
 144  * \sa rc_get_used_temporaries
 145  * @note If this functions finds a free temporary, it will mark it as used
 146  * in the used temporary list (param 'used')
 147  * @param used list of used temporaries
 148  * @param used_length number of items in param 'used'
 149  * @param mask which components must be free in the temporary index that is
 150  * returned.
 151  * @return -1 If there are no more free temporaries, otherwise the index of
 152  * a temporary register where the components specified in param 'mask' are
 153  * not being used.
 154  */
 155 int rc_find_free_temporary_list(
 156         struct radeon_compiler * c,
 157         unsigned char * used,
 158         unsigned int used_length,
 159         unsigned int mask)
 160 {
 161         int i;
 162         for(i = 0; i < used_length; i++) {
 163                 if ((~used[i] & mask) == mask) {
 164                         used[i] |= mask;
 165                         return i;
 166                 }
 167         }
 168         return -1;
 169 }
 170
 171 unsigned int rc_find_free_temporary(struct radeon_compiler * c)
 172 {
 173         unsigned char used[RC_REGISTER_MAX_INDEX];
 174         int free;
 175
 176         memset(used, 0, sizeof(used));
 177
 178         rc_get_used_temporaries(c, used, RC_REGISTER_MAX_INDEX);
 179
 180         free = rc_find_free_temporary_list(c, used, RC_REGISTER_MAX_INDEX,
 181                                                                 RC_MASK_XYZW);
 182         if (free < 0) {
 183                 rc_error(c, "Ran out of temporary registers\n");
 184                 return 0;
 185         }
 186         return free;
 187 }
 188
 189
 190 struct rc_instruction *rc_alloc_instruction(struct radeon_compiler * c)
 191 {
 192         struct rc_instruction * inst = memory_pool_malloc(&c->Pool, sizeof(struct rc_instruction));
 193
 194         memset(inst, 0, sizeof(struct rc_instruction));
 195
 196         inst->U.I.Opcode = RC_OPCODE_ILLEGAL_OPCODE;
 197         inst->U.I.DstReg.WriteMask = RC_MASK_XYZW;
 198         inst->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZW;
 199         inst->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_XYZW;
 200         inst->U.I.SrcReg[2].Swizzle = RC_SWIZZLE_XYZW;
 201
 202         return inst;
 203 }
 204
 205 void rc_insert_instruction(struct rc_instruction * after, struct rc_instruction * inst)
 206 {
 207         inst->Prev = after;
 208         inst->Next = after->Next;
 209
 210         inst->Prev->Next = inst;
 211         inst->Next->Prev = inst;
 212 }
 213
 214 struct rc_instruction *rc_insert_new_instruction(struct radeon_compiler * c, struct rc_instruction * after)
 215 {
 216         struct rc_instruction * inst = rc_alloc_instruction(c);
 217
 218         rc_insert_instruction(after, inst);
 219
 220         return inst;
 221 }
 222
 223 void rc_remove_instruction(struct rc_instruction * inst)
 224 {
 225         inst->Prev->Next = inst->Next;
 226         inst->Next->Prev = inst->Prev;
 227 }
 228
 229 /**
 230  * Return the number of instructions in the program.
 231  */
 232 unsigned int rc_recompute_ips(struct radeon_compiler * c)
 233 {
 234         unsigned int ip = 0;
 235         struct rc_instruction * inst;
 236
 237         for(inst = c->Program.Instructions.Next;
 238             inst != &c->Program.Instructions;
 239             inst = inst->Next) {
 240                 inst->IP = ip++;
 241         }
 242
 243         c->Program.Instructions.IP = 0xcafedead;
 244
 245         return ip;
 246 }
 247
 248 rc_swizzle rc_mask_to_swizzle(unsigned int mask)
 249 {
 250         switch(mask) {
 251                 case RC_MASK_X: return RC_SWIZZLE_X;
 252                 case RC_MASK_Y: return RC_SWIZZLE_Y;
 253                 case RC_MASK_Z: return RC_SWIZZLE_Z;
 254                 case RC_MASK_W: return RC_SWIZZLE_W;
 255                 default: return RC_SWIZZLE_UNUSED;
 256         }
 257 }