src/gallium/drivers/i915/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 #include "util/u_math.h"
  32
  33
  34 #define A0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
  35 #define D0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
  36 #define T0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
  37 #define A0_SRC0( reg ) (((reg)&UREG_MASK)>>UREG_A0_SRC0_SHIFT_LEFT)
  38 #define A1_SRC0( reg ) (((reg)&UREG_MASK)<<UREG_A1_SRC0_SHIFT_RIGHT)
  39 #define A1_SRC1( reg ) (((reg)&UREG_MASK)>>UREG_A1_SRC1_SHIFT_LEFT)
  40 #define A2_SRC1( reg ) (((reg)&UREG_MASK)<<UREG_A2_SRC1_SHIFT_RIGHT)
  41 #define A2_SRC2( reg ) (((reg)&UREG_MASK)>>UREG_A2_SRC2_SHIFT_LEFT)
  42
  43 /* These are special, and don't have swizzle/negate bits.
  44  */
  45 #define T0_SAMPLER( reg )     (GET_UREG_NR(reg)<<T0_SAMPLER_NR_SHIFT)
  46 #define T1_ADDRESS_REG( reg ) ((GET_UREG_NR(reg)<<T1_ADDRESS_REG_NR_SHIFT) | \
  47                                (GET_UREG_TYPE(reg)<<T1_ADDRESS_REG_TYPE_SHIFT))
  48
  49
  50 /* Macros for translating UREG's into the various register fields used
  51  * by the I915 programmable unit.
  52  */
  53 #define UREG_A0_DEST_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_DEST_TYPE_SHIFT)
  54 #define UREG_A0_SRC0_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_SRC0_TYPE_SHIFT)
  55 #define UREG_A1_SRC0_SHIFT_RIGHT (A1_SRC0_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
  56 #define UREG_A1_SRC1_SHIFT_LEFT  (UREG_TYPE_SHIFT - A1_SRC1_TYPE_SHIFT)
  57 #define UREG_A2_SRC1_SHIFT_RIGHT (A2_SRC1_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
  58 #define UREG_A2_SRC2_SHIFT_LEFT  (UREG_TYPE_SHIFT - A2_SRC2_TYPE_SHIFT)
  59
  60 #define UREG_MASK         0xffffff00
  61 #define UREG_TYPE_NR_MASK ((REG_TYPE_MASK << UREG_TYPE_SHIFT) | \
  62                            (REG_NR_MASK << UREG_NR_SHIFT))
  63
  64
  65 uint
  66 i915_get_temp(struct i915_fp_compile *p)
  67 {
  68    int bit = ffs(~p->temp_flag);
  69    if (!bit) {
  70       i915_program_error(p, "i915_get_temp: out of temporaries\n");
  71       return 0;
  72    }
  73
  74    p->temp_flag |= 1 << (bit - 1);
  75    return bit - 1;
  76 }
  77
  78
  79 static void
  80 i915_release_temp(struct i915_fp_compile *p, int reg)
  81 {
  82    p->temp_flag &= ~(1 << reg);
  83 }
  84
  85
  86 /**
  87  * Get unpreserved temporary, a temp whose value is not preserved between
  88  * PS program phases.
  89  */
  90 uint
  91 i915_get_utemp(struct i915_fp_compile * p)
  92 {
  93    int bit = ffs(~p->utemp_flag);
  94    if (!bit) {
  95       i915_program_error(p, "i915_get_utemp: out of temporaries\n");
  96       return 0;
  97    }
  98
  99    p->utemp_flag |= 1 << (bit - 1);
 100    return UREG(REG_TYPE_U, (bit - 1));
 101 }
 102
 103 void
 104 i915_release_utemps(struct i915_fp_compile *p)
 105 {
 106    p->utemp_flag = ~0x7;
 107 }
 108
 109
 110 uint
 111 i915_emit_decl(struct i915_fp_compile *p,
 112                uint type, uint nr, uint d0_flags)
 113 {
 114    uint reg = UREG(type, nr);
 115
 116    if (type == REG_TYPE_T) {
 117       if (p->decl_t & (1 << nr))
 118          return reg;
 119
 120       p->decl_t |= (1 << nr);
 121    }
 122    else if (type == REG_TYPE_S) {
 123       if (p->decl_s & (1 << nr))
 124          return reg;
 125
 126       p->decl_s |= (1 << nr);
 127    }
 128    else
 129       return reg;
 130
 131    if (p->decl< p->declarations + I915_PROGRAM_SIZE) {
 132       *(p->decl++) = (D0_DCL | D0_DEST(reg) | d0_flags);
 133       *(p->decl++) = D1_MBZ;
 134       *(p->decl++) = D2_MBZ;
 135    }
 136    else
 137       i915_program_error(p, "Out of declarations\n");
 138
 139    p->nr_decl_insn++;
 140    return reg;
 141 }
 142
 143 uint
 144 i915_emit_arith(struct i915_fp_compile * p,
 145                 uint op,
 146                 uint dest,
 147                 uint mask,
 148                 uint saturate, uint src0, uint src1, uint src2)
 149 {
 150    uint c[3];
 151    uint nr_const = 0;
 152
 153    assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
 154    dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest));
 155    assert(dest);
 156
 157    if (GET_UREG_TYPE(src0) == REG_TYPE_CONST)
 158       c[nr_const++] = 0;
 159    if (GET_UREG_TYPE(src1) == REG_TYPE_CONST)
 160       c[nr_const++] = 1;
 161    if (GET_UREG_TYPE(src2) == REG_TYPE_CONST)
 162       c[nr_const++] = 2;
 163
 164    /* Recursively call this function to MOV additional const values
 165     * into temporary registers.  Use utemp registers for this -
 166     * currently shouldn't be possible to run out, but keep an eye on
 167     * this.
 168     */
 169    if (nr_const > 1) {
 170       uint s[3], first, i, old_utemp_flag;
 171
 172       s[0] = src0;
 173       s[1] = src1;
 174       s[2] = src2;
 175       old_utemp_flag = p->utemp_flag;
 176
 177       first = GET_UREG_NR(s[c[0]]);
 178       for (i = 1; i < nr_const; i++) {
 179          if (GET_UREG_NR(s[c[i]]) != first) {
 180             uint tmp = i915_get_utemp(p);
 181
 182             i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
 183                             s[c[i]], 0, 0);
 184             s[c[i]] = tmp;
 185          }
 186       }
 187
 188       src0 = s[0];
 189       src1 = s[1];
 190       src2 = s[2];
 191       p->utemp_flag = old_utemp_flag;   /* restore */
 192    }
 193
 194    if (p->csr< p->program + I915_PROGRAM_SIZE) {
 195       *(p->csr++) = (op | A0_DEST(dest) | mask | saturate | A0_SRC0(src0));
 196       *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1));
 197       *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2));
 198    }
 199    else
 200       i915_program_error(p, "Out of instructions\n");
 201
 202    p->nr_alu_insn++;
 203    return dest;
 204 }
 205
 206
 207 /**
 208  * Emit a texture load or texkill instruction.
 209  * \param dest  the dest i915 register
 210  * \param destmask  the dest register writemask
 211  * \param sampler  the i915 sampler register
 212  * \param coord  the i915 source texcoord operand
 213  * \param opcode  the instruction opcode
 214  */
 215 uint i915_emit_texld( struct i915_fp_compile *p,
 216                         uint dest,
 217                         uint destmask,
 218                         uint sampler,
 219                         uint coord,
 220                         uint opcode )
 221 {
 222    const uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
 223    int temp = -1;
 224
 225    if (coord != k) {
 226       /* texcoord is swizzled or negated.  Need to allocate a new temporary
 227        * register (a utemp / unpreserved temp) won't do.
 228        */
 229       uint tempReg;
 230
 231       temp = i915_get_temp(p);           /* get temp reg index */
 232       tempReg = UREG(REG_TYPE_R, temp);  /* make i915 register */
 233
 234       i915_emit_arith( p, A0_MOV,
 235                        tempReg, A0_DEST_CHANNEL_ALL, /* dest reg, writemask */
 236                        0,                            /* saturate */
 237                        coord, 0, 0 );                /* src0, src1, src2 */
 238
 239       /* new src texcoord is tempReg */
 240       coord = tempReg;
 241    }
 242
 243    /* Don't worry about saturate as we only support
 244     */
 245    if (destmask != A0_DEST_CHANNEL_ALL) {
 246       /* if not writing to XYZW... */
 247       uint tmp = i915_get_utemp(p);
 248       i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, opcode );
 249       i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
 250       /* XXX release utemp here? */
 251    }
 252    else {
 253       assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
 254       assert(dest == UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
 255
 256       /* is the sampler coord a texcoord input reg? */
 257       if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
 258          p->nr_tex_indirect++;
 259       }
 260
 261       if (p->csr< p->program + I915_PROGRAM_SIZE) {
 262          *(p->csr++) = (opcode |
 263                         T0_DEST( dest ) |
 264                         T0_SAMPLER( sampler ));
 265
 266          *(p->csr++) = T1_ADDRESS_REG( coord );
 267          *(p->csr++) = T2_MBZ;
 268       }
 269    else
 270       i915_program_error(p, "Out of instructions\n");
 271
 272       p->nr_tex_insn++;
 273    }
 274
 275    if (temp >= 0)
 276       i915_release_temp(p, temp);
 277
 278    return dest;
 279 }
 280
 281
 282 uint
 283 i915_emit_const1f(struct i915_fp_compile * p, float c0)
 284 {
 285    struct i915_fragment_shader *ifs = p->shader;
 286    unsigned reg, idx;
 287
 288    if (c0 == 0.0)
 289       return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
 290    if (c0 == 1.0)
 291       return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
 292
 293    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 294       if (ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
 295          continue;
 296       for (idx = 0; idx < 4; idx++) {
 297          if (!(ifs->constant_flags[reg] & (1 << idx)) ||
 298              ifs->constants[reg][idx] == c0) {
 299             ifs->constants[reg][idx] = c0;
 300             ifs->constant_flags[reg] |= 1 << idx;
 301             if (reg + 1 > ifs->num_constants)
 302                ifs->num_constants = reg + 1;
 303             return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
 304          }
 305       }
 306    }
 307
 308    i915_program_error(p, "i915_emit_const1f: out of constants\n");
 309    return 0;
 310 }
 311
 312 uint
 313 i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1)
 314 {
 315    struct i915_fragment_shader *ifs = p->shader;
 316    unsigned reg, idx;
 317
 318    if (c0 == 0.0)
 319       return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W);
 320    if (c0 == 1.0)
 321       return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W);
 322
 323    if (c1 == 0.0)
 324       return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W);
 325    if (c1 == 1.0)
 326       return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W);
 327
 328    // XXX emit swizzle here for 0, 1, -1 and any combination thereof
 329    // we can use swizzle + neg for that
 330    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 331       if (ifs->constant_flags[reg] == 0xf ||
 332           ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
 333          continue;
 334       for (idx = 0; idx < 3; idx++) {
 335          if (!(ifs->constant_flags[reg] & (3 << idx))) {
 336             ifs->constants[reg][idx + 0] = c0;
 337             ifs->constants[reg][idx + 1] = c1;
 338             ifs->constant_flags[reg] |= 3 << idx;
 339             if (reg + 1 > ifs->num_constants)
 340                ifs->num_constants = reg + 1;
 341             return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO, ONE);
 342          }
 343       }
 344    }
 345
 346    i915_program_error(p, "i915_emit_const2f: out of constants\n");
 347    return 0;
 348 }
 349
 350 uint
 351 i915_emit_const4f(struct i915_fp_compile * p,
 352                   float c0, float c1, float c2, float c3)
 353 {
 354    struct i915_fragment_shader *ifs = p->shader;
 355    unsigned reg;
 356
 357    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 358       if (ifs->constant_flags[reg] == 0xf &&
 359           ifs->constants[reg][0] == c0 &&
 360           ifs->constants[reg][1] == c1 &&
 361           ifs->constants[reg][2] == c2 &&
 362           ifs->constants[reg][3] == c3) {
 363          return UREG(REG_TYPE_CONST, reg);
 364       }
 365       else if (ifs->constant_flags[reg] == 0) {
 366
 367          ifs->constants[reg][0] = c0;
 368          ifs->constants[reg][1] = c1;
 369          ifs->constants[reg][2] = c2;
 370          ifs->constants[reg][3] = c3;
 371          ifs->constant_flags[reg] = 0xf;
 372          if (reg + 1 > ifs->num_constants)
 373             ifs->num_constants = reg + 1;
 374          return UREG(REG_TYPE_CONST, reg);
 375       }
 376    }
 377
 378    i915_program_error(p, "i915_emit_const4f: out of constants\n");
 379    return 0;
 380 }
 381
 382
 383 uint
 384 i915_emit_const4fv(struct i915_fp_compile * p, const float * c)
 385 {
 386    return i915_emit_const4f(p, c[0], c[1], c[2], c[3]);
 387 }