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 #define I915_CONSTFLAG_PARAM 0x1f
  65
  66 uint
  67 i915_get_temp(struct i915_fp_compile *p)
  68 {
  69    int bit = ffs(~p->temp_flag);
  70    if (!bit) {
  71       i915_program_error(p, "i915_get_temp: out of temporaries\n");
  72       return 0;
  73    }
  74
  75    p->temp_flag |= 1 << (bit - 1);
  76    return UREG(REG_TYPE_R, (bit - 1));
  77 }
  78
  79
  80 uint
  81 i915_get_utemp(struct i915_fp_compile * p)
  82 {
  83    int bit = ffs(~p->utemp_flag);
  84    if (!bit) {
  85       i915_program_error(p, "i915_get_utemp: out of temporaries\n");
  86       return 0;
  87    }
  88
  89    p->utemp_flag |= 1 << (bit - 1);
  90    return UREG(REG_TYPE_U, (bit - 1));
  91 }
  92
  93 void
  94 i915_release_utemps(struct i915_fp_compile *p)
  95 {
  96    p->utemp_flag = ~0x7;
  97 }
  98
  99
 100 uint
 101 i915_emit_decl(struct i915_fp_compile *p,
 102                uint type, uint nr, uint d0_flags)
 103 {
 104    uint reg = UREG(type, nr);
 105
 106    if (type == REG_TYPE_T) {
 107       if (p->decl_t & (1 << nr))
 108          return reg;
 109
 110       p->decl_t |= (1 << nr);
 111    }
 112    else if (type == REG_TYPE_S) {
 113       if (p->decl_s & (1 << nr))
 114          return reg;
 115
 116       p->decl_s |= (1 << nr);
 117    }
 118    else
 119       return reg;
 120
 121    *(p->decl++) = (D0_DCL | D0_DEST(reg) | d0_flags);
 122    *(p->decl++) = D1_MBZ;
 123    *(p->decl++) = D2_MBZ;
 124
 125    p->nr_decl_insn++;
 126    return reg;
 127 }
 128
 129 uint
 130 i915_emit_arith(struct i915_fp_compile * p,
 131                 uint op,
 132                 uint dest,
 133                 uint mask,
 134                 uint saturate, uint src0, uint src1, uint src2)
 135 {
 136    uint c[3];
 137    uint nr_const = 0;
 138
 139    assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
 140    dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest));
 141    assert(dest);
 142
 143    if (GET_UREG_TYPE(src0) == REG_TYPE_CONST)
 144       c[nr_const++] = 0;
 145    if (GET_UREG_TYPE(src1) == REG_TYPE_CONST)
 146       c[nr_const++] = 1;
 147    if (GET_UREG_TYPE(src2) == REG_TYPE_CONST)
 148       c[nr_const++] = 2;
 149
 150    /* Recursively call this function to MOV additional const values
 151     * into temporary registers.  Use utemp registers for this -
 152     * currently shouldn't be possible to run out, but keep an eye on
 153     * this.
 154     */
 155    if (nr_const > 1) {
 156       uint s[3], first, i, old_utemp_flag;
 157
 158       s[0] = src0;
 159       s[1] = src1;
 160       s[2] = src2;
 161       old_utemp_flag = p->utemp_flag;
 162
 163       first = GET_UREG_NR(s[c[0]]);
 164       for (i = 1; i < nr_const; i++) {
 165          if (GET_UREG_NR(s[c[i]]) != first) {
 166             uint tmp = i915_get_utemp(p);
 167
 168             i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
 169                             s[c[i]], 0, 0);
 170             s[c[i]] = tmp;
 171          }
 172       }
 173
 174       src0 = s[0];
 175       src1 = s[1];
 176       src2 = s[2];
 177       p->utemp_flag = old_utemp_flag;   /* restore */
 178    }
 179
 180    *(p->csr++) = (op | A0_DEST(dest) | mask | saturate | A0_SRC0(src0));
 181    *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1));
 182    *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2));
 183
 184    p->nr_alu_insn++;
 185    return dest;
 186 }
 187
 188 uint i915_emit_texld( struct i915_fp_compile *p,
 189                         uint dest,
 190                         uint destmask,
 191                         uint sampler,
 192                         uint coord,
 193                         uint op )
 194 {
 195    uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
 196    if (coord != k) {
 197       /* No real way to work around this in the general case - need to
 198        * allocate and declare a new temporary register (a utemp won't
 199        * do).  Will fallback for now.
 200        */
 201       i915_program_error(p, "Can't (yet) swizzle TEX arguments");
 202       assert(0);
 203       return 0;
 204    }
 205
 206    /* Don't worry about saturate as we only support
 207     */
 208    if (destmask != A0_DEST_CHANNEL_ALL) {
 209       uint tmp = i915_get_utemp(p);
 210       i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op );
 211       i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
 212       return dest;
 213    }
 214    else {
 215       assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
 216       assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
 217
 218       if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
 219          p->nr_tex_indirect++;
 220       }
 221
 222       *(p->csr++) = (op |
 223                      T0_DEST( dest ) |
 224                      T0_SAMPLER( sampler ));
 225
 226       *(p->csr++) = T1_ADDRESS_REG( coord );
 227       *(p->csr++) = T2_MBZ;
 228
 229       p->nr_tex_insn++;
 230       return dest;
 231    }
 232 }
 233
 234
 235 uint
 236 i915_emit_const1f(struct i915_fp_compile * p, float c0)
 237 {
 238    unsigned reg, idx;
 239
 240    if (c0 == 0.0)
 241       return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
 242    if (c0 == 1.0)
 243       return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
 244
 245    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 246       if (p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
 247          continue;
 248       for (idx = 0; idx < 4; idx++) {
 249          if (!(p->constant_flags[reg] & (1 << idx)) ||
 250              p->constants[reg][idx] == c0) {
 251             p->constants[reg][idx] = c0;
 252             p->constant_flags[reg] |= 1 << idx;
 253             if (reg + 1 > p->num_constants)
 254                p->num_constants = reg + 1;
 255             return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
 256          }
 257       }
 258    }
 259
 260    i915_program_error(p, "i915_emit_const1f: out of constants\n");
 261    return 0;
 262 }
 263
 264 uint
 265 i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1)
 266 {
 267    unsigned reg, idx;
 268
 269    if (c0 == 0.0)
 270       return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W);
 271    if (c0 == 1.0)
 272       return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W);
 273
 274    if (c1 == 0.0)
 275       return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W);
 276    if (c1 == 1.0)
 277       return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W);
 278
 279    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 280       if (p->constant_flags[reg] == 0xf ||
 281           p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
 282          continue;
 283       for (idx = 0; idx < 3; idx++) {
 284          if (!(p->constant_flags[reg] & (3 << idx))) {
 285             p->constants[reg][idx + 0] = c0;
 286             p->constants[reg][idx + 1] = c1;
 287             p->constant_flags[reg] |= 3 << idx;
 288             if (reg + 1 > p->num_constants)
 289                p->num_constants = reg + 1;
 290             return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO, ONE);
 291          }
 292       }
 293    }
 294
 295    i915_program_error(p, "i915_emit_const2f: out of constants\n");
 296    return 0;
 297 }
 298
 299
 300
 301 uint
 302 i915_emit_const4f(struct i915_fp_compile * p,
 303                   float c0, float c1, float c2, float c3)
 304 {
 305    unsigned reg;
 306
 307    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 308       if (p->constant_flags[reg] == 0xf &&
 309           p->constants[reg][0] == c0 &&
 310           p->constants[reg][1] == c1 &&
 311           p->constants[reg][2] == c2 &&
 312           p->constants[reg][3] == c3) {
 313          return UREG(REG_TYPE_CONST, reg);
 314       }
 315       else if (p->constant_flags[reg] == 0) {
 316
 317          p->constants[reg][0] = c0;
 318          p->constants[reg][1] = c1;
 319          p->constants[reg][2] = c2;
 320          p->constants[reg][3] = c3;
 321          p->constant_flags[reg] = 0xf;
 322          if (reg + 1 > p->num_constants)
 323             p->num_constants = reg + 1;
 324          return UREG(REG_TYPE_CONST, reg);
 325       }
 326    }
 327
 328    i915_program_error(p, "i915_emit_const4f: out of constants\n");
 329    return 0;
 330 }
 331
 332
 333 uint
 334 i915_emit_const4fv(struct i915_fp_compile * p, const float * c)
 335 {
 336    return i915_emit_const4f(p, c[0], c[1], c[2], c[3]);
 337 }
 338
 339
 340 #if 00000/*UNUSED*/
 341 /* Reserve a slot in the constant file for a Mesa state parameter.
 342  * These will later need to be tracked on statechanges, but that is
 343  * done elsewhere.
 344  */
 345 uint
 346 i915_emit_param4fv(struct i915_fp_compile * p, const float * values)
 347 {
 348    struct i915_fragment_program *fp = p->fp;
 349    int i;
 350
 351    for (i = 0; i < fp->nr_params; i++) {
 352       if (fp->param[i].values == values)
 353          return UREG(REG_TYPE_CONST, fp->param[i].reg);
 354    }
 355
 356    if (p->constants->nr_constants == I915_MAX_CONSTANT ||
 357        fp->nr_params == I915_MAX_CONSTANT) {
 358       i915_program_error(p, "i915_emit_param4fv: out of constants\n");
 359       return 0;
 360    }
 361
 362    {
 363       int reg = p->constants->nr_constants++;
 364       int i = fp->nr_params++;
 365
 366       assert (p->constant_flags[reg] == 0);
 367       p->constant_flags[reg] = I915_CONSTFLAG_PARAM;
 368
 369       fp->param[i].values = values;
 370       fp->param[i].reg = reg;
 371
 372       return UREG(REG_TYPE_CONST, reg);
 373    }
 374 }
 375 #endif