src/gallium/drivers/i915simple/i915_fpc_emit.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
   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 TUNGSTEN GRAPHICS 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
  33 #define A0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
  34 #define D0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
  35 #define T0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
  36 #define A0_SRC0( reg ) (((reg)&UREG_MASK)>>UREG_A0_SRC0_SHIFT_LEFT)
  37 #define A1_SRC0( reg ) (((reg)&UREG_MASK)<<UREG_A1_SRC0_SHIFT_RIGHT)
  38 #define A1_SRC1( reg ) (((reg)&UREG_MASK)>>UREG_A1_SRC1_SHIFT_LEFT)
  39 #define A2_SRC1( reg ) (((reg)&UREG_MASK)<<UREG_A2_SRC1_SHIFT_RIGHT)
  40 #define A2_SRC2( reg ) (((reg)&UREG_MASK)>>UREG_A2_SRC2_SHIFT_LEFT)
  41
  42 /* These are special, and don't have swizzle/negate bits.
  43  */
  44 #define T0_SAMPLER( reg )     (GET_UREG_NR(reg)<<T0_SAMPLER_NR_SHIFT)
  45 #define T1_ADDRESS_REG( reg ) ((GET_UREG_NR(reg)<<T1_ADDRESS_REG_NR_SHIFT) | \
  46                                (GET_UREG_TYPE(reg)<<T1_ADDRESS_REG_TYPE_SHIFT))
  47
  48
  49 /* Macros for translating UREG's into the various register fields used
  50  * by the I915 programmable unit.
  51  */
  52 #define UREG_A0_DEST_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_DEST_TYPE_SHIFT)
  53 #define UREG_A0_SRC0_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_SRC0_TYPE_SHIFT)
  54 #define UREG_A1_SRC0_SHIFT_RIGHT (A1_SRC0_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
  55 #define UREG_A1_SRC1_SHIFT_LEFT  (UREG_TYPE_SHIFT - A1_SRC1_TYPE_SHIFT)
  56 #define UREG_A2_SRC1_SHIFT_RIGHT (A2_SRC1_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
  57 #define UREG_A2_SRC2_SHIFT_LEFT  (UREG_TYPE_SHIFT - A2_SRC2_TYPE_SHIFT)
  58
  59 #define UREG_MASK         0xffffff00
  60 #define UREG_TYPE_NR_MASK ((REG_TYPE_MASK << UREG_TYPE_SHIFT) | \
  61                            (REG_NR_MASK << UREG_NR_SHIFT))
  62
  63
  64 uint
  65 i915_get_temp(struct i915_fp_compile *p)
  66 {
  67    int bit = ffs(~p->temp_flag);
  68    if (!bit) {
  69       i915_program_error(p, "i915_get_temp: out of temporaries\n");
  70       return 0;
  71    }
  72
  73    p->temp_flag |= 1 << (bit - 1);
  74    return bit - 1;
  75 }
  76
  77
  78 static void
  79 i915_release_temp(struct i915_fp_compile *p, int reg)
  80 {
  81    p->temp_flag &= ~(1 << reg);
  82 }
  83
  84
  85 /**
  86  * Get unpreserved temporary, a temp whose value is not preserved between
  87  * PS program phases.
  88  */
  89 uint
  90 i915_get_utemp(struct i915_fp_compile * p)
  91 {
  92    int bit = ffs(~p->utemp_flag);
  93    if (!bit) {
  94       i915_program_error(p, "i915_get_utemp: out of temporaries\n");
  95       return 0;
  96    }
  97
  98    p->utemp_flag |= 1 << (bit - 1);
  99    return UREG(REG_TYPE_U, (bit - 1));
 100 }
 101
 102 void
 103 i915_release_utemps(struct i915_fp_compile *p)
 104 {
 105    p->utemp_flag = ~0x7;
 106 }
 107
 108
 109 uint
 110 i915_emit_decl(struct i915_fp_compile *p,
 111                uint type, uint nr, uint d0_flags)
 112 {
 113    uint reg = UREG(type, nr);
 114
 115    if (type == REG_TYPE_T) {
 116       if (p->decl_t & (1 << nr))
 117          return reg;
 118
 119       p->decl_t |= (1 << nr);
 120    }
 121    else if (type == REG_TYPE_S) {
 122       if (p->decl_s & (1 << nr))
 123          return reg;
 124
 125       p->decl_s |= (1 << nr);
 126    }
 127    else
 128       return reg;
 129
 130    *(p->decl++) = (D0_DCL | D0_DEST(reg) | d0_flags);
 131    *(p->decl++) = D1_MBZ;
 132    *(p->decl++) = D2_MBZ;
 133
 134    p->nr_decl_insn++;
 135    return reg;
 136 }
 137
 138 uint
 139 i915_emit_arith(struct i915_fp_compile * p,
 140                 uint op,
 141                 uint dest,
 142                 uint mask,
 143                 uint saturate, uint src0, uint src1, uint src2)
 144 {
 145    uint c[3];
 146    uint nr_const = 0;
 147
 148    assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
 149    dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest));
 150    assert(dest);
 151
 152    if (GET_UREG_TYPE(src0) == REG_TYPE_CONST)
 153       c[nr_const++] = 0;
 154    if (GET_UREG_TYPE(src1) == REG_TYPE_CONST)
 155       c[nr_const++] = 1;
 156    if (GET_UREG_TYPE(src2) == REG_TYPE_CONST)
 157       c[nr_const++] = 2;
 158
 159    /* Recursively call this function to MOV additional const values
 160     * into temporary registers.  Use utemp registers for this -
 161     * currently shouldn't be possible to run out, but keep an eye on
 162     * this.
 163     */
 164    if (nr_const > 1) {
 165       uint s[3], first, i, old_utemp_flag;
 166
 167       s[0] = src0;
 168       s[1] = src1;
 169       s[2] = src2;
 170       old_utemp_flag = p->utemp_flag;
 171
 172       first = GET_UREG_NR(s[c[0]]);
 173       for (i = 1; i < nr_const; i++) {
 174          if (GET_UREG_NR(s[c[i]]) != first) {
 175             uint tmp = i915_get_utemp(p);
 176
 177             i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
 178                             s[c[i]], 0, 0);
 179             s[c[i]] = tmp;
 180          }
 181       }
 182
 183       src0 = s[0];
 184       src1 = s[1];
 185       src2 = s[2];
 186       p->utemp_flag = old_utemp_flag;   /* restore */
 187    }
 188
 189    *(p->csr++) = (op | A0_DEST(dest) | mask | saturate | A0_SRC0(src0));
 190    *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1));
 191    *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2));
 192
 193    p->nr_alu_insn++;
 194    return dest;
 195 }
 196
 197
 198 /**
 199  * Emit a texture load or texkill instruction.
 200  * \param dest  the dest i915 register
 201  * \param destmask  the dest register writemask
 202  * \param sampler  the i915 sampler register
 203  * \param coord  the i915 source texcoord operand
 204  * \param opcode  the instruction opcode
 205  */
 206 uint i915_emit_texld( struct i915_fp_compile *p,
 207                         uint dest,
 208                         uint destmask,
 209                         uint sampler,
 210                         uint coord,
 211                         uint opcode )
 212 {
 213    const uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
 214    int temp = -1;
 215
 216    if (coord != k) {
 217       /* texcoord is swizzled or negated.  Need to allocate a new temporary
 218        * register (a utemp / unpreserved temp) won't do.
 219        */
 220       uint tempReg;
 221
 222       temp = i915_get_temp(p);           /* get temp reg index */
 223       tempReg = UREG(REG_TYPE_R, temp);  /* make i915 register */
 224
 225       i915_emit_arith( p, A0_MOV,
 226                        tempReg, A0_DEST_CHANNEL_ALL, /* dest reg, writemask */
 227                        0,                            /* saturate */
 228                        coord, 0, 0 );                /* src0, src1, src2 */
 229
 230       /* new src texcoord is tempReg */
 231       coord = tempReg;
 232    }
 233
 234    /* Don't worry about saturate as we only support
 235     */
 236    if (destmask != A0_DEST_CHANNEL_ALL) {
 237       /* if not writing to XYZW... */
 238       uint tmp = i915_get_utemp(p);
 239       i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, opcode );
 240       i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
 241       /* XXX release utemp here? */
 242    }
 243    else {
 244       assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
 245       assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
 246
 247       /* is the sampler coord a texcoord input reg? */
 248       if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
 249          p->nr_tex_indirect++;
 250       }
 251
 252       *(p->csr++) = (opcode |
 253                      T0_DEST( dest ) |
 254                      T0_SAMPLER( sampler ));
 255
 256       *(p->csr++) = T1_ADDRESS_REG( coord );
 257       *(p->csr++) = T2_MBZ;
 258
 259       p->nr_tex_insn++;
 260    }
 261
 262    if (temp >= 0)
 263       i915_release_temp(p, temp);
 264
 265    return dest;
 266 }
 267
 268
 269 uint
 270 i915_emit_const1f(struct i915_fp_compile * p, float c0)
 271 {
 272    struct i915_fragment_shader *ifs = p->shader;
 273    unsigned reg, idx;
 274
 275    if (c0 == 0.0)
 276       return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
 277    if (c0 == 1.0)
 278       return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
 279
 280    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 281       if (ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
 282          continue;
 283       for (idx = 0; idx < 4; idx++) {
 284          if (!(ifs->constant_flags[reg] & (1 << idx)) ||
 285              ifs->constants[reg][idx] == c0) {
 286             ifs->constants[reg][idx] = c0;
 287             ifs->constant_flags[reg] |= 1 << idx;
 288             if (reg + 1 > ifs->num_constants)
 289                ifs->num_constants = reg + 1;
 290             return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
 291          }
 292       }
 293    }
 294
 295    i915_program_error(p, "i915_emit_const1f: out of constants\n");
 296    return 0;
 297 }
 298
 299 uint
 300 i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1)
 301 {
 302    struct i915_fragment_shader *ifs = p->shader;
 303    unsigned reg, idx;
 304
 305    if (c0 == 0.0)
 306       return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W);
 307    if (c0 == 1.0)
 308       return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W);
 309
 310    if (c1 == 0.0)
 311       return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W);
 312    if (c1 == 1.0)
 313       return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W);
 314
 315    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 316       if (ifs->constant_flags[reg] == 0xf ||
 317           ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
 318          continue;
 319       for (idx = 0; idx < 3; idx++) {
 320          if (!(ifs->constant_flags[reg] & (3 << idx))) {
 321             ifs->constants[reg][idx + 0] = c0;
 322             ifs->constants[reg][idx + 1] = c1;
 323             ifs->constant_flags[reg] |= 3 << idx;
 324             if (reg + 1 > ifs->num_constants)
 325                ifs->num_constants = reg + 1;
 326             return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO, ONE);
 327          }
 328       }
 329    }
 330
 331    i915_program_error(p, "i915_emit_const2f: out of constants\n");
 332    return 0;
 333 }
 334
 335
 336
 337 uint
 338 i915_emit_const4f(struct i915_fp_compile * p,
 339                   float c0, float c1, float c2, float c3)
 340 {
 341    struct i915_fragment_shader *ifs = p->shader;
 342    unsigned reg;
 343
 344    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 345       if (ifs->constant_flags[reg] == 0xf &&
 346           ifs->constants[reg][0] == c0 &&
 347           ifs->constants[reg][1] == c1 &&
 348           ifs->constants[reg][2] == c2 &&
 349           ifs->constants[reg][3] == c3) {
 350          return UREG(REG_TYPE_CONST, reg);
 351       }
 352       else if (ifs->constant_flags[reg] == 0) {
 353
 354          ifs->constants[reg][0] = c0;
 355          ifs->constants[reg][1] = c1;
 356          ifs->constants[reg][2] = c2;
 357          ifs->constants[reg][3] = c3;
 358          ifs->constant_flags[reg] = 0xf;
 359          if (reg + 1 > ifs->num_constants)
 360             ifs->num_constants = reg + 1;
 361          return UREG(REG_TYPE_CONST, reg);
 362       }
 363    }
 364
 365    i915_program_error(p, "i915_emit_const4f: out of constants\n");
 366    return 0;
 367 }
 368
 369
 370 uint
 371 i915_emit_const4fv(struct i915_fp_compile * p, const float * c)
 372 {
 373    return i915_emit_const4f(p, c[0], c[1], c[2], c[3]);
 374 }