src/mesa/drivers/dri/r300/r300_fragprog_swizzle.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 /**
  29  * @file
  30  * Utilities to deal with the somewhat odd restriction on R300 fragment
  31  * program swizzles.
  32  */
  33
  34 #include "r300_fragprog_swizzle.h"
  35
  36 #include "r300_reg.h"
  37 #include "radeon_nqssadce.h"
  38
  39 #define MAKE_SWZ3(x, y, z) (MAKE_SWIZZLE4(SWIZZLE_##x, SWIZZLE_##y, SWIZZLE_##z, SWIZZLE_ZERO))
  40
  41 struct swizzle_data {
  42         GLuint hash; /**< swizzle value this matches */
  43         GLuint base; /**< base value for hw swizzle */
  44         GLuint stride; /**< difference in base between arg0/1/2 */
  45 };
  46
  47 static const struct swizzle_data native_swizzles[] = {
  48         {MAKE_SWZ3(X, Y, Z), R300_ALU_ARGC_SRC0C_XYZ, 4},
  49         {MAKE_SWZ3(X, X, X), R300_ALU_ARGC_SRC0C_XXX, 4},
  50         {MAKE_SWZ3(Y, Y, Y), R300_ALU_ARGC_SRC0C_YYY, 4},
  51         {MAKE_SWZ3(Z, Z, Z), R300_ALU_ARGC_SRC0C_ZZZ, 4},
  52         {MAKE_SWZ3(W, W, W), R300_ALU_ARGC_SRC0A, 1},
  53         {MAKE_SWZ3(Y, Z, X), R300_ALU_ARGC_SRC0C_YZX, 1},
  54         {MAKE_SWZ3(Z, X, Y), R300_ALU_ARGC_SRC0C_ZXY, 1},
  55         {MAKE_SWZ3(W, Z, Y), R300_ALU_ARGC_SRC0CA_WZY, 1},
  56         {MAKE_SWZ3(ONE, ONE, ONE), R300_ALU_ARGC_ONE, 0},
  57         {MAKE_SWZ3(ZERO, ZERO, ZERO), R300_ALU_ARGC_ZERO, 0}
  58 };
  59
  60 static const int num_native_swizzles = sizeof(native_swizzles)/sizeof(native_swizzles[0]);
  61
  62
  63 /**
  64  * Find a native RGB swizzle that matches the given swizzle.
  65  * Returns 0 if none found.
  66  */
  67 static const struct swizzle_data* lookup_native_swizzle(GLuint swizzle)
  68 {
  69         int i, comp;
  70
  71         for(i = 0; i < num_native_swizzles; ++i) {
  72                 const struct swizzle_data* sd = &native_swizzles[i];
  73                 for(comp = 0; comp < 3; ++comp) {
  74                         GLuint swz = GET_SWZ(swizzle, comp);
  75                         if (swz == SWIZZLE_NIL)
  76                                 continue;
  77                         if (swz != GET_SWZ(sd->hash, comp))
  78                                 break;
  79                 }
  80                 if (comp == 3)
  81                         return sd;
  82         }
  83
  84         return 0;
  85 }
  86
  87
  88 /**
  89  * Check whether the given instruction supports the swizzle and negate
  90  * combinations in the given source register.
  91  */
  92 GLboolean r300FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg)
  93 {
  94         if (reg.Abs)
  95                 reg.NegateBase = 0;
  96
  97         if (opcode == OPCODE_KIL ||
  98             opcode == OPCODE_TEX ||
  99             opcode == OPCODE_TXB ||
 100             opcode == OPCODE_TXP) {
 101                 int j;
 102
 103                 if (reg.Abs || reg.NegateBase != (15*reg.NegateAbs))
 104                         return GL_FALSE;
 105
 106                 for(j = 0; j < 4; ++j) {
 107                         GLuint swz = GET_SWZ(reg.Swizzle, j);
 108                         if (swz == SWIZZLE_NIL)
 109                                 continue;
 110                         if (swz != j)
 111                                 return GL_FALSE;
 112                 }
 113
 114                 return GL_TRUE;
 115         }
 116
 117         GLuint relevant = 0;
 118         int j;
 119
 120         for(j = 0; j < 3; ++j)
 121                 if (GET_SWZ(reg.Swizzle, j) != SWIZZLE_NIL)
 122                         relevant |= 1 << j;
 123
 124         if ((reg.NegateBase & relevant) && (reg.NegateBase & relevant) != relevant)
 125                 return GL_FALSE;
 126
 127         if (!lookup_native_swizzle(reg.Swizzle))
 128                 return GL_FALSE;
 129
 130         return GL_TRUE;
 131 }
 132
 133
 134 /**
 135  * Generate MOV dst, src using only native swizzles.
 136  */
 137 void r300FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src)
 138 {
 139         if (src.Abs)
 140                 src.NegateBase = 0;
 141
 142         while(dst.WriteMask) {
 143                 const struct swizzle_data *best_swizzle = 0;
 144                 GLuint best_matchcount = 0;
 145                 GLuint best_matchmask = 0;
 146                 GLboolean rgbnegate;
 147                 int i, comp;
 148
 149                 for(i = 0; i < num_native_swizzles; ++i) {
 150                         const struct swizzle_data *sd = &native_swizzles[i];
 151                         GLuint matchcount = 0;
 152                         GLuint matchmask = 0;
 153                         for(comp = 0; comp < 3; ++comp) {
 154                                 if (!GET_BIT(dst.WriteMask, comp))
 155                                         continue;
 156                                 GLuint swz = GET_SWZ(src.Swizzle, comp);
 157                                 if (swz == SWIZZLE_NIL)
 158                                         continue;
 159                                 if (swz == GET_SWZ(sd->hash, comp)) {
 160                                         matchcount++;
 161                                         matchmask |= 1 << comp;
 162                                 }
 163                         }
 164                         if (matchcount > best_matchcount) {
 165                                 best_swizzle = sd;
 166                                 best_matchcount = matchcount;
 167                                 best_matchmask = matchmask;
 168                                 if (matchmask == (dst.WriteMask & WRITEMASK_XYZ))
 169                                         break;
 170                         }
 171                 }
 172
 173                 if ((src.NegateBase & best_matchmask) != 0) {
 174                         best_matchmask &= src.NegateBase;
 175                         rgbnegate = !src.NegateAbs;
 176                 } else {
 177                         rgbnegate = src.NegateAbs;
 178                 }
 179
 180                 struct prog_instruction *inst;
 181
 182                 _mesa_insert_instructions(s->Program, s->IP, 1);
 183                 inst = s->Program->Instructions + s->IP++;
 184                 inst->Opcode = OPCODE_MOV;
 185                 inst->DstReg = dst;
 186                 inst->DstReg.WriteMask &= (best_matchmask | WRITEMASK_W);
 187                 inst->SrcReg[0] = src;
 188                 /* Note: We rely on NqSSA/DCE to set unused swizzle components to NIL */
 189
 190                 dst.WriteMask &= ~inst->DstReg.WriteMask;
 191         }
 192 }
 193
 194
 195 /**
 196  * Translate an RGB (XYZ) swizzle into the hardware code for the given
 197  * instruction source.
 198  */
 199 GLuint r300FPTranslateRGBSwizzle(GLuint src, GLuint swizzle)
 200 {
 201         const struct swizzle_data* sd = lookup_native_swizzle(swizzle);
 202
 203         if (!sd) {
 204                 _mesa_printf("Not a native swizzle: %08x\n", swizzle);
 205                 return 0;
 206         }
 207
 208         return sd->base + src*sd->stride;
 209 }
 210
 211
 212 /**
 213  * Translate an Alpha (W) swizzle into the hardware code for the given
 214  * instruction source.
 215  */
 216 GLuint r300FPTranslateAlphaSwizzle(GLuint src, GLuint swizzle)
 217 {
 218         if (swizzle < 3)
 219                 return swizzle + 3*src;
 220
 221         switch(swizzle) {
 222         case SWIZZLE_W: return R300_ALU_ARGA_SRC0A + src;
 223         case SWIZZLE_ONE: return R300_ALU_ARGA_ONE;
 224         case SWIZZLE_ZERO: return R300_ALU_ARGA_ZERO;
 225         default: return R300_ALU_ARGA_ONE;
 226         }
 227 }