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         rc_dataflow_remove_instruction(inst);
 158         inst->Prev->Next = inst->Next;
 159         inst->Next->Prev = inst->Prev;
 160 }