src/gallium/drivers/i915/i915_fpc_optimize.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2011 The Chromium OS authors.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * 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, sub license, 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 portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL GOOGLE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include "i915_reg.h"
  29 #include "i915_context.h"
  30 #include "i915_fpc.h"
  31
  32 #include "pipe/p_shader_tokens.h"
  33 #include "util/u_math.h"
  34 #include "util/u_memory.h"
  35 #include "util/u_string.h"
  36 #include "tgsi/tgsi_parse.h"
  37 #include "tgsi/tgsi_dump.h"
  38
  39 static boolean same_dst_reg(struct i915_full_dst_register* d1, struct i915_full_dst_register* d2)
  40 {
  41    return (d1->Register.File == d2->Register.File &&
  42            d1->Register.Indirect == d2->Register.Indirect &&
  43            d1->Register.Dimension == d2->Register.Dimension &&
  44            d1->Register.Index == d2->Register.Index);
  45 }
  46
  47 static boolean same_src_reg(struct i915_full_src_register* d1, struct i915_full_src_register* d2)
  48 {
  49    return (d1->Register.File == d2->Register.File &&
  50            d1->Register.Indirect == d2->Register.Indirect &&
  51            d1->Register.Dimension == d2->Register.Dimension &&
  52            d1->Register.Index == d2->Register.Index &&
  53            d1->Register.Absolute == d2->Register.Absolute &&
  54            d1->Register.Negate == d2->Register.Negate);
  55 }
  56
  57 static boolean is_unswizzled(struct i915_full_src_register* r,
  58                              unsigned write_mask)
  59 {
  60    if ( write_mask & TGSI_WRITEMASK_X && r->Register.SwizzleX != TGSI_SWIZZLE_X)
  61       return FALSE;
  62    if ( write_mask & TGSI_WRITEMASK_Y && r->Register.SwizzleY != TGSI_SWIZZLE_Y)
  63       return FALSE;
  64    if ( write_mask & TGSI_WRITEMASK_Z && r->Register.SwizzleZ != TGSI_SWIZZLE_Z)
  65       return FALSE;
  66    if ( write_mask & TGSI_WRITEMASK_W && r->Register.SwizzleW != TGSI_SWIZZLE_W)
  67       return FALSE;
  68    return TRUE;
  69 }
  70
  71 static boolean op_commutes(unsigned opcode)
  72 {
  73    if (opcode == TGSI_OPCODE_ADD) return TRUE;
  74    if (opcode == TGSI_OPCODE_MUL) return TRUE;
  75    return FALSE;
  76 }
  77
  78 static unsigned op_neutral_element(unsigned opcode)
  79 {
  80    if (opcode == TGSI_OPCODE_ADD)
  81       return TGSI_SWIZZLE_ZERO;
  82    if (opcode == TGSI_OPCODE_MUL)
  83       return TGSI_SWIZZLE_ONE;
  84
  85    debug_printf("Unknown opcode %d\n",opcode);
  86    return TGSI_SWIZZLE_ZERO;
  87 }
  88
  89 /*
  90  * Sets the swizzle to the neutral element for the operation for the bits
  91  * of writemask which are set, swizzle to identity otherwise.
  92  */
  93 static void set_neutral_element_swizzle(struct i915_full_src_register* r,
  94                                         unsigned write_mask,
  95                                         unsigned neutral)
  96 {
  97    if ( write_mask & TGSI_WRITEMASK_X )
  98       r->Register.SwizzleX = neutral;
  99    else
 100       r->Register.SwizzleX = TGSI_SWIZZLE_X;
 101
 102    if ( write_mask & TGSI_WRITEMASK_Y )
 103       r->Register.SwizzleY = neutral;
 104    else
 105       r->Register.SwizzleY = TGSI_SWIZZLE_Y;
 106
 107    if ( write_mask & TGSI_WRITEMASK_Z )
 108       r->Register.SwizzleZ = neutral;
 109    else
 110       r->Register.SwizzleZ = TGSI_SWIZZLE_Z;
 111
 112    if ( write_mask & TGSI_WRITEMASK_W )
 113       r->Register.SwizzleW = neutral;
 114    else
 115       r->Register.SwizzleW = TGSI_SWIZZLE_W;
 116 }
 117
 118 /*
 119  * Optimize away things like:
 120  *    MUL OUT[0].xyz, TEMP[1], TEMP[2]
 121  *    MOV OUT[0].w, TEMP[2]
 122  * into:
 123  *    MUL OUT[0].xyzw, TEMP[1].xyz1, TEMP[2]
 124  * This is useful for optimizing texenv.
 125  */
 126 static void i915_fpc_optimize_mov_after_alu(union i915_full_token* current, union i915_full_token* next)
 127 {
 128    if ( current->Token.Type == TGSI_TOKEN_TYPE_INSTRUCTION  &&
 129         next->Token.Type == TGSI_TOKEN_TYPE_INSTRUCTION  &&
 130         op_commutes(current->FullInstruction.Instruction.Opcode) &&
 131         current->FullInstruction.Instruction.Saturate == next->FullInstruction.Instruction.Saturate &&
 132         next->FullInstruction.Instruction.Opcode == TGSI_OPCODE_MOV &&
 133         same_dst_reg(&next->FullInstruction.Dst[0], &current->FullInstruction.Dst[0]) &&
 134         same_src_reg(&next->FullInstruction.Src[0], &current->FullInstruction.Src[1]) &&
 135         is_unswizzled(&current->FullInstruction.Src[0], current->FullInstruction.Dst[0].Register.WriteMask) &&
 136         is_unswizzled(&current->FullInstruction.Src[1], current->FullInstruction.Dst[0].Register.WriteMask) &&
 137         is_unswizzled(&next->FullInstruction.Src[0], next->FullInstruction.Dst[0].Register.WriteMask) )
 138    {
 139       next->FullInstruction.Instruction.Opcode = TGSI_OPCODE_NOP;
 140
 141       set_neutral_element_swizzle(&current->FullInstruction.Src[1], 0, 0);
 142       set_neutral_element_swizzle(&current->FullInstruction.Src[0],
 143                                   next->FullInstruction.Dst[0].Register.WriteMask,
 144                                   op_neutral_element(current->FullInstruction.Instruction.Opcode));
 145
 146       current->FullInstruction.Dst[0].Register.WriteMask = current->FullInstruction.Dst[0].Register.WriteMask |
 147                                                            next->FullInstruction.Dst[0].Register.WriteMask;
 148       return;
 149    }
 150
 151    if ( current->Token.Type == TGSI_TOKEN_TYPE_INSTRUCTION  &&
 152         next->Token.Type == TGSI_TOKEN_TYPE_INSTRUCTION  &&
 153         op_commutes(current->FullInstruction.Instruction.Opcode) &&
 154         current->FullInstruction.Instruction.Saturate == next->FullInstruction.Instruction.Saturate &&
 155         next->FullInstruction.Instruction.Opcode == TGSI_OPCODE_MOV &&
 156         same_dst_reg(&next->FullInstruction.Dst[0], &current->FullInstruction.Dst[0]) &&
 157         same_src_reg(&next->FullInstruction.Src[0], &current->FullInstruction.Src[0]) &&
 158         is_unswizzled(&current->FullInstruction.Src[0], current->FullInstruction.Dst[0].Register.WriteMask) &&
 159         is_unswizzled(&current->FullInstruction.Src[1], current->FullInstruction.Dst[0].Register.WriteMask) &&
 160         is_unswizzled(&next->FullInstruction.Src[0], next->FullInstruction.Dst[0].Register.WriteMask) )
 161    {
 162       next->FullInstruction.Instruction.Opcode = TGSI_OPCODE_NOP;
 163
 164       set_neutral_element_swizzle(&current->FullInstruction.Src[0], 0, 0);
 165       set_neutral_element_swizzle(&current->FullInstruction.Src[1],
 166                                   next->FullInstruction.Dst[0].Register.WriteMask,
 167                                   op_neutral_element(current->FullInstruction.Instruction.Opcode));
 168
 169       current->FullInstruction.Dst[0].Register.WriteMask = current->FullInstruction.Dst[0].Register.WriteMask |
 170                                                            next->FullInstruction.Dst[0].Register.WriteMask;
 171       return;
 172    }
 173 }
 174
 175 static void copy_src_reg(struct i915_src_register* o, const struct tgsi_src_register* i)
 176 {
 177    o->File      = i->File;
 178    o->Indirect  = i->Indirect;
 179    o->Dimension = i->Dimension;
 180    o->Index     = i->Index;
 181    o->SwizzleX  = i->SwizzleX;
 182    o->SwizzleY  = i->SwizzleY;
 183    o->SwizzleZ  = i->SwizzleZ;
 184    o->SwizzleW  = i->SwizzleW;
 185    o->Absolute  = i->Absolute;
 186    o->Negate    = i->Negate;
 187 }
 188
 189 static void copy_dst_reg(struct i915_dst_register* o, const struct tgsi_dst_register* i)
 190 {
 191    o->File      = i->File;
 192    o->WriteMask = i->WriteMask;
 193    o->Indirect  = i->Indirect;
 194    o->Dimension = i->Dimension;
 195    o->Index     = i->Index;
 196 }
 197
 198 static void copy_instruction(struct i915_full_instruction* o, const struct tgsi_full_instruction* i)
 199 {
 200    memcpy(&o->Instruction, &i->Instruction, sizeof(o->Instruction));
 201    memcpy(&o->Texture, &i->Texture, sizeof(o->Texture));
 202
 203    copy_dst_reg(&o->Dst[0].Register, &i->Dst[0].Register);
 204
 205    copy_src_reg(&o->Src[0].Register, &i->Src[0].Register);
 206    copy_src_reg(&o->Src[1].Register, &i->Src[1].Register);
 207    copy_src_reg(&o->Src[2].Register, &i->Src[2].Register);
 208 }
 209
 210 static void copy_token(union i915_full_token* o, union tgsi_full_token* i)
 211 {
 212    if (i->Token.Type != TGSI_TOKEN_TYPE_INSTRUCTION)
 213       memcpy(o, i, sizeof(*o));
 214    else
 215       copy_instruction(&o->FullInstruction, &i->FullInstruction);
 216
 217 }
 218
 219 struct i915_token_list* i915_optimize(const struct tgsi_token *tokens)
 220 {
 221    struct i915_token_list *out_tokens = MALLOC(sizeof(struct i915_token_list));
 222    struct tgsi_parse_context parse;
 223    int i = 0;
 224
 225    out_tokens->NumTokens = 0;
 226
 227    /* Count the tokens */
 228    tgsi_parse_init( &parse, tokens );
 229    while( !tgsi_parse_end_of_tokens( &parse ) ) {
 230       tgsi_parse_token( &parse );
 231       out_tokens->NumTokens++;
 232    }
 233    tgsi_parse_free (&parse);
 234
 235    /* Allocate our tokens */
 236    out_tokens->Tokens = MALLOC(sizeof(union i915_full_token) * out_tokens->NumTokens);
 237
 238    tgsi_parse_init( &parse, tokens );
 239    while( !tgsi_parse_end_of_tokens( &parse ) ) {
 240       tgsi_parse_token( &parse );
 241       copy_token(&out_tokens->Tokens[i] , &parse.FullToken);
 242
 243       if (i > 0)
 244          i915_fpc_optimize_mov_after_alu(&out_tokens->Tokens[i-1], &out_tokens->Tokens[i]);
 245
 246       i++;
 247    }
 248    tgsi_parse_free (&parse);
 249
 250    return out_tokens;
 251 }
 252
 253 void i915_optimize_free(struct i915_token_list* tokens)
 254 {
 255    free(tokens->Tokens);
 256    free(tokens);
 257 }
 258
 259