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 UREG(REG_TYPE_R, (bit - 1));
  75 }
  76
  77
  78 uint
  79 i915_get_utemp(struct i915_fp_compile * p)
  80 {
  81    int bit = ffs(~p->utemp_flag);
  82    if (!bit) {
  83       i915_program_error(p, "i915_get_utemp: out of temporaries\n");
  84       return 0;
  85    }
  86
  87    p->utemp_flag |= 1 << (bit - 1);
  88    return UREG(REG_TYPE_U, (bit - 1));
  89 }
  90
  91 void
  92 i915_release_utemps(struct i915_fp_compile *p)
  93 {
  94    p->utemp_flag = ~0x7;
  95 }
  96
  97
  98 uint
  99 i915_emit_decl(struct i915_fp_compile *p,
 100                uint type, uint nr, uint d0_flags)
 101 {
 102    uint reg = UREG(type, nr);
 103
 104    if (type == REG_TYPE_T) {
 105       if (p->decl_t & (1 << nr))
 106          return reg;
 107
 108       p->decl_t |= (1 << nr);
 109    }
 110    else if (type == REG_TYPE_S) {
 111       if (p->decl_s & (1 << nr))
 112          return reg;
 113
 114       p->decl_s |= (1 << nr);
 115    }
 116    else
 117       return reg;
 118
 119    *(p->decl++) = (D0_DCL | D0_DEST(reg) | d0_flags);
 120    *(p->decl++) = D1_MBZ;
 121    *(p->decl++) = D2_MBZ;
 122
 123    p->nr_decl_insn++;
 124    return reg;
 125 }
 126
 127 uint
 128 i915_emit_arith(struct i915_fp_compile * p,
 129                 uint op,
 130                 uint dest,
 131                 uint mask,
 132                 uint saturate, uint src0, uint src1, uint src2)
 133 {
 134    uint c[3];
 135    uint nr_const = 0;
 136
 137    assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
 138    dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest));
 139    assert(dest);
 140
 141    if (GET_UREG_TYPE(src0) == REG_TYPE_CONST)
 142       c[nr_const++] = 0;
 143    if (GET_UREG_TYPE(src1) == REG_TYPE_CONST)
 144       c[nr_const++] = 1;
 145    if (GET_UREG_TYPE(src2) == REG_TYPE_CONST)
 146       c[nr_const++] = 2;
 147
 148    /* Recursively call this function to MOV additional const values
 149     * into temporary registers.  Use utemp registers for this -
 150     * currently shouldn't be possible to run out, but keep an eye on
 151     * this.
 152     */
 153    if (nr_const > 1) {
 154       uint s[3], first, i, old_utemp_flag;
 155
 156       s[0] = src0;
 157       s[1] = src1;
 158       s[2] = src2;
 159       old_utemp_flag = p->utemp_flag;
 160
 161       first = GET_UREG_NR(s[c[0]]);
 162       for (i = 1; i < nr_const; i++) {
 163          if (GET_UREG_NR(s[c[i]]) != first) {
 164             uint tmp = i915_get_utemp(p);
 165
 166             i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
 167                             s[c[i]], 0, 0);
 168             s[c[i]] = tmp;
 169          }
 170       }
 171
 172       src0 = s[0];
 173       src1 = s[1];
 174       src2 = s[2];
 175       p->utemp_flag = old_utemp_flag;   /* restore */
 176    }
 177
 178    *(p->csr++) = (op | A0_DEST(dest) | mask | saturate | A0_SRC0(src0));
 179    *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1));
 180    *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2));
 181
 182    p->nr_alu_insn++;
 183    return dest;
 184 }
 185
 186 uint i915_emit_texld( struct i915_fp_compile *p,
 187                         uint dest,
 188                         uint destmask,
 189                         uint sampler,
 190                         uint coord,
 191                         uint op )
 192 {
 193    uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
 194    if (coord != k) {
 195       /* No real way to work around this in the general case - need to
 196        * allocate and declare a new temporary register (a utemp won't
 197        * do).  Will fallback for now.
 198        */
 199       i915_program_error(p, "Can't (yet) swizzle TEX arguments");
 200       assert(0);
 201       return 0;
 202    }
 203
 204    /* Don't worry about saturate as we only support
 205     */
 206    if (destmask != A0_DEST_CHANNEL_ALL) {
 207       uint tmp = i915_get_utemp(p);
 208       i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op );
 209       i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
 210       return dest;
 211    }
 212    else {
 213       assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
 214       assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
 215
 216       if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
 217          p->nr_tex_indirect++;
 218       }
 219
 220       *(p->csr++) = (op |
 221                      T0_DEST( dest ) |
 222                      T0_SAMPLER( sampler ));
 223
 224       *(p->csr++) = T1_ADDRESS_REG( coord );
 225       *(p->csr++) = T2_MBZ;
 226
 227       p->nr_tex_insn++;
 228       return dest;
 229    }
 230 }
 231
 232
 233 uint
 234 i915_emit_const1f(struct i915_fp_compile * p, float c0)
 235 {
 236    struct i915_fragment_shader *ifs = p->shader;
 237    unsigned reg, idx;
 238
 239    if (c0 == 0.0)
 240       return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
 241    if (c0 == 1.0)
 242       return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
 243
 244    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 245       if (ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
 246          continue;
 247       for (idx = 0; idx < 4; idx++) {
 248          if (!(ifs->constant_flags[reg] & (1 << idx)) ||
 249              ifs->constants[reg][idx] == c0) {
 250             ifs->constants[reg][idx] = c0;
 251             ifs->constant_flags[reg] |= 1 << idx;
 252             if (reg + 1 > ifs->num_constants)
 253                ifs->num_constants = reg + 1;
 254             return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
 255          }
 256       }
 257    }
 258
 259    i915_program_error(p, "i915_emit_const1f: out of constants\n");
 260    return 0;
 261 }
 262
 263 uint
 264 i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1)
 265 {
 266    struct i915_fragment_shader *ifs = p->shader;
 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 (ifs->constant_flags[reg] == 0xf ||
 281           ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
 282          continue;
 283       for (idx = 0; idx < 3; idx++) {
 284          if (!(ifs->constant_flags[reg] & (3 << idx))) {
 285             ifs->constants[reg][idx + 0] = c0;
 286             ifs->constants[reg][idx + 1] = c1;
 287             ifs->constant_flags[reg] |= 3 << idx;
 288             if (reg + 1 > ifs->num_constants)
 289                ifs->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    struct i915_fragment_shader *ifs = p->shader;
 306    unsigned reg;
 307
 308    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
 309       if (ifs->constant_flags[reg] == 0xf &&
 310           ifs->constants[reg][0] == c0 &&
 311           ifs->constants[reg][1] == c1 &&
 312           ifs->constants[reg][2] == c2 &&
 313           ifs->constants[reg][3] == c3) {
 314          return UREG(REG_TYPE_CONST, reg);
 315       }
 316       else if (ifs->constant_flags[reg] == 0) {
 317
 318          ifs->constants[reg][0] = c0;
 319          ifs->constants[reg][1] = c1;
 320          ifs->constants[reg][2] = c2;
 321          ifs->constants[reg][3] = c3;
 322          ifs->constant_flags[reg] = 0xf;
 323          if (reg + 1 > ifs->num_constants)
 324             ifs->num_constants = reg + 1;
 325          return UREG(REG_TYPE_CONST, reg);
 326       }
 327    }
 328
 329    i915_program_error(p, "i915_emit_const4f: out of constants\n");
 330    return 0;
 331 }
 332
 333
 334 uint
 335 i915_emit_const4fv(struct i915_fp_compile * p, const float * c)
 336 {
 337    return i915_emit_const4f(p, c[0], c[1], c[2], c[3]);
 338 }