src/gallium/drivers/freedreno/a2xx/instr-a2xx.h

   1 /*
   2  * Copyright (c) 2012 Rob Clark <robdclark@gmail.com>
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  */
  23
  24 #ifndef INSTR_A2XX_H_
  25 #define INSTR_A2XX_H_
  26
  27 #define PACKED __attribute__((__packed__))
  28
  29 #include "util/u_math.h"
  30 #include "adreno_common.xml.h"
  31 #include "adreno_pm4.xml.h"
  32 #include "a2xx.xml.h"
  33
  34
  35 /*
  36  * ALU instructions:
  37  */
  38
  39 typedef enum {
  40         ADDs = 0,
  41         ADD_PREVs = 1,
  42         MULs = 2,
  43         MUL_PREVs = 3,
  44         MUL_PREV2s = 4,
  45         MAXs = 5,
  46         MINs = 6,
  47         SETEs = 7,
  48         SETGTs = 8,
  49         SETGTEs = 9,
  50         SETNEs = 10,
  51         FRACs = 11,
  52         TRUNCs = 12,
  53         FLOORs = 13,
  54         EXP_IEEE = 14,
  55         LOG_CLAMP = 15,
  56         LOG_IEEE = 16,
  57         RECIP_CLAMP = 17,
  58         RECIP_FF = 18,
  59         RECIP_IEEE = 19,
  60         RECIPSQ_CLAMP = 20,
  61         RECIPSQ_FF = 21,
  62         RECIPSQ_IEEE = 22,
  63         MOVAs = 23,
  64         MOVA_FLOORs = 24,
  65         SUBs = 25,
  66         SUB_PREVs = 26,
  67         PRED_SETEs = 27,
  68         PRED_SETNEs = 28,
  69         PRED_SETGTs = 29,
  70         PRED_SETGTEs = 30,
  71         PRED_SET_INVs = 31,
  72         PRED_SET_POPs = 32,
  73         PRED_SET_CLRs = 33,
  74         PRED_SET_RESTOREs = 34,
  75         KILLEs = 35,
  76         KILLGTs = 36,
  77         KILLGTEs = 37,
  78         KILLNEs = 38,
  79         KILLONEs = 39,
  80         SQRT_IEEE = 40,
  81         MUL_CONST_0 = 42,
  82         MUL_CONST_1 = 43,
  83         ADD_CONST_0 = 44,
  84         ADD_CONST_1 = 45,
  85         SUB_CONST_0 = 46,
  86         SUB_CONST_1 = 47,
  87         SIN = 48,
  88         COS = 49,
  89         RETAIN_PREV = 50,
  90 } instr_scalar_opc_t;
  91
  92 typedef enum {
  93         ADDv = 0,
  94         MULv = 1,
  95         MAXv = 2,
  96         MINv = 3,
  97         SETEv = 4,
  98         SETGTv = 5,
  99         SETGTEv = 6,
 100         SETNEv = 7,
 101         FRACv = 8,
 102         TRUNCv = 9,
 103         FLOORv = 10,
 104         MULADDv = 11,
 105         CNDEv = 12,
 106         CNDGTEv = 13,
 107         CNDGTv = 14,
 108         DOT4v = 15,
 109         DOT3v = 16,
 110         DOT2ADDv = 17,
 111         CUBEv = 18,
 112         MAX4v = 19,
 113         PRED_SETE_PUSHv = 20,
 114         PRED_SETNE_PUSHv = 21,
 115         PRED_SETGT_PUSHv = 22,
 116         PRED_SETGTE_PUSHv = 23,
 117         KILLEv = 24,
 118         KILLGTv = 25,
 119         KILLGTEv = 26,
 120         KILLNEv = 27,
 121         DSTv = 28,
 122         MOVAv = 29,
 123 } instr_vector_opc_t;
 124
 125 typedef struct PACKED {
 126         /* dword0: */
 127         uint8_t             vector_dest              : 6;
 128         uint8_t             vector_dest_rel          : 1;
 129         uint8_t             low_precision_16b_fp     : 1;
 130         uint8_t             scalar_dest              : 6;
 131         uint8_t             scalar_dest_rel          : 1;
 132         uint8_t             export_data              : 1;
 133         uint8_t             vector_write_mask        : 4;
 134         uint8_t             scalar_write_mask        : 4;
 135         uint8_t             vector_clamp             : 1;
 136         uint8_t             scalar_clamp             : 1;
 137         instr_scalar_opc_t  scalar_opc               : 6;
 138         /* dword1: */
 139         uint8_t             src3_swiz                : 8;
 140         uint8_t             src2_swiz                : 8;
 141         uint8_t             src1_swiz                : 8;
 142         uint8_t             src3_reg_negate          : 1;
 143         uint8_t             src2_reg_negate          : 1;
 144         uint8_t             src1_reg_negate          : 1;
 145         uint8_t             pred_select              : 2;
 146         uint8_t             relative_addr            : 1;
 147         uint8_t             const_1_rel_abs          : 1;
 148         uint8_t             const_0_rel_abs          : 1;
 149         /* dword2: */
 150         uint8_t             src3_reg                 : 6;
 151         uint8_t             src3_reg_select          : 1;
 152         uint8_t             src3_reg_abs             : 1;
 153         uint8_t             src2_reg                 : 6;
 154         uint8_t             src2_reg_select          : 1;
 155         uint8_t             src2_reg_abs             : 1;
 156         uint8_t             src1_reg                 : 6;
 157         uint8_t             src1_reg_select          : 1;
 158         uint8_t             src1_reg_abs             : 1;
 159         instr_vector_opc_t  vector_opc               : 5;
 160         uint8_t             src3_sel                 : 1;
 161         uint8_t             src2_sel                 : 1;
 162         uint8_t             src1_sel                 : 1;
 163 } instr_alu_t;
 164
 165
 166
 167 /*
 168  * CF instructions:
 169  */
 170
 171 typedef enum {
 172         NOP = 0,
 173         EXEC = 1,
 174         EXEC_END = 2,
 175         COND_EXEC = 3,
 176         COND_EXEC_END = 4,
 177         COND_PRED_EXEC = 5,
 178         COND_PRED_EXEC_END = 6,
 179         LOOP_START = 7,
 180         LOOP_END = 8,
 181         COND_CALL = 9,
 182         RETURN = 10,
 183         COND_JMP = 11,
 184         ALLOC = 12,
 185         COND_EXEC_PRED_CLEAN = 13,
 186         COND_EXEC_PRED_CLEAN_END = 14,
 187         MARK_VS_FETCH_DONE = 15,
 188 } instr_cf_opc_t;
 189
 190 typedef enum {
 191         RELATIVE_ADDR = 0,
 192         ABSOLUTE_ADDR = 1,
 193 } instr_addr_mode_t;
 194
 195 typedef enum {
 196         SQ_NO_ALLOC = 0,
 197         SQ_POSITION = 1,
 198         SQ_PARAMETER_PIXEL = 2,
 199         SQ_MEMORY = 3,
 200 } instr_alloc_type_t;
 201
 202 typedef struct PACKED {
 203         uint16_t            address                  : 9;
 204         uint8_t             reserved0                : 3;
 205         uint8_t             count                    : 3;
 206         uint8_t             yeild                    : 1;
 207         uint16_t            serialize                : 12;
 208         uint8_t             vc                       : 6;   /* vertex cache? */
 209         uint8_t             bool_addr                : 8;
 210         uint8_t             condition                : 1;
 211         instr_addr_mode_t   address_mode             : 1;
 212         instr_cf_opc_t      opc                      : 4;
 213 } instr_cf_exec_t;
 214
 215 typedef struct PACKED {
 216         uint16_t            address                  : 10;
 217         uint8_t             reserved0                : 6;
 218         uint8_t             loop_id                  : 5;
 219         uint32_t            reserved1                : 22;
 220         instr_addr_mode_t   address_mode             : 1;
 221         instr_cf_opc_t      opc                      : 4;
 222 } instr_cf_loop_t;
 223
 224 typedef struct PACKED {
 225         uint16_t            address                  : 10;
 226         uint8_t             reserved0                : 3;
 227         uint8_t             force_call               : 1;
 228         uint8_t             predicated_jmp           : 1;
 229         uint32_t            reserved1                : 18;
 230         uint8_t             direction                : 1;
 231         uint8_t             bool_addr                : 8;
 232         uint8_t             condition                : 1;
 233         instr_addr_mode_t   address_mode             : 1;
 234         instr_cf_opc_t      opc                      : 4;
 235 } instr_cf_jmp_call_t;
 236
 237 typedef struct PACKED {
 238         uint8_t             size                     : 4;
 239         uint64_t            reserved0                : 36;
 240         uint8_t             no_serial                : 1;
 241         instr_alloc_type_t  buffer_select            : 2;
 242         uint8_t             alloc_mode               : 1;
 243         instr_cf_opc_t      opc                      : 4;
 244 } instr_cf_alloc_t;
 245
 246 typedef union PACKED {
 247         instr_cf_exec_t     exec;
 248         instr_cf_loop_t     loop;
 249         instr_cf_jmp_call_t jmp_call;
 250         instr_cf_alloc_t    alloc;
 251         struct PACKED {
 252                 uint64_t        dummy                    : 44;
 253                 instr_cf_opc_t  opc                      : 4;
 254         };
 255 } instr_cf_t;
 256
 257
 258
 259 /*
 260  * FETCH instructions:
 261  */
 262
 263 typedef enum {
 264         VTX_FETCH = 0,
 265         TEX_FETCH = 1,
 266         TEX_GET_BORDER_COLOR_FRAC = 16,
 267         TEX_GET_COMP_TEX_LOD = 17,
 268         TEX_GET_GRADIENTS = 18,
 269         TEX_GET_WEIGHTS = 19,
 270         TEX_SET_TEX_LOD = 24,
 271         TEX_SET_GRADIENTS_H = 25,
 272         TEX_SET_GRADIENTS_V = 26,
 273         TEX_RESERVED_4 = 27,
 274 } instr_fetch_opc_t;
 275
 276 typedef enum {
 277         TEX_FILTER_POINT = 0,
 278         TEX_FILTER_LINEAR = 1,
 279         TEX_FILTER_BASEMAP = 2,            /* only applicable for mip-filter */
 280         TEX_FILTER_USE_FETCH_CONST = 3,
 281 } instr_tex_filter_t;
 282
 283 typedef enum {
 284         ANISO_FILTER_DISABLED = 0,
 285         ANISO_FILTER_MAX_1_1 = 1,
 286         ANISO_FILTER_MAX_2_1 = 2,
 287         ANISO_FILTER_MAX_4_1 = 3,
 288         ANISO_FILTER_MAX_8_1 = 4,
 289         ANISO_FILTER_MAX_16_1 = 5,
 290         ANISO_FILTER_USE_FETCH_CONST = 7,
 291 } instr_aniso_filter_t;
 292
 293 typedef enum {
 294         ARBITRARY_FILTER_2X4_SYM = 0,
 295         ARBITRARY_FILTER_2X4_ASYM = 1,
 296         ARBITRARY_FILTER_4X2_SYM = 2,
 297         ARBITRARY_FILTER_4X2_ASYM = 3,
 298         ARBITRARY_FILTER_4X4_SYM = 4,
 299         ARBITRARY_FILTER_4X4_ASYM = 5,
 300         ARBITRARY_FILTER_USE_FETCH_CONST = 7,
 301 } instr_arbitrary_filter_t;
 302
 303 typedef enum {
 304         SAMPLE_CENTROID = 0,
 305         SAMPLE_CENTER = 1,
 306 } instr_sample_loc_t;
 307
 308 typedef enum a2xx_sq_surfaceformat instr_surf_fmt_t;
 309
 310 typedef struct PACKED {
 311         /* dword0: */
 312         instr_fetch_opc_t   opc                      : 5;
 313         uint8_t             src_reg                  : 6;
 314         uint8_t             src_reg_am               : 1;
 315         uint8_t             dst_reg                  : 6;
 316         uint8_t             dst_reg_am               : 1;
 317         uint8_t             fetch_valid_only         : 1;
 318         uint8_t             const_idx                : 5;
 319         uint8_t             tx_coord_denorm          : 1;
 320         uint8_t             src_swiz                 : 6;
 321         /* dword1: */
 322         uint16_t            dst_swiz                 : 12;
 323         instr_tex_filter_t  mag_filter               : 2;
 324         instr_tex_filter_t  min_filter               : 2;
 325         instr_tex_filter_t  mip_filter               : 2;
 326         instr_aniso_filter_t aniso_filter            : 3;
 327         instr_arbitrary_filter_t arbitrary_filter    : 3;
 328         instr_tex_filter_t  vol_mag_filter           : 2;
 329         instr_tex_filter_t  vol_min_filter           : 2;
 330         uint8_t             use_comp_lod             : 1;
 331         uint8_t             use_reg_lod              : 2;  /* 0 for cube, 1 for 2d */
 332         uint8_t             pred_select              : 1;
 333         /* dword2: */
 334         uint8_t             use_reg_gradients        : 1;
 335         instr_sample_loc_t  sample_location          : 1;
 336         uint8_t             lod_bias                 : 7;
 337         uint8_t             unused                   : 7;
 338         uint8_t             offset_x                 : 5;
 339         uint8_t             offset_y                 : 5;
 340         uint8_t             offset_z                 : 5;
 341         uint8_t             pred_condition           : 1;
 342 } instr_fetch_tex_t;
 343
 344 typedef struct PACKED {
 345         /* dword0: */
 346         instr_fetch_opc_t   opc                      : 5;
 347         uint8_t             src_reg                  : 6;
 348         uint8_t             src_reg_am               : 1;
 349         uint8_t             dst_reg                  : 6;
 350         uint8_t             dst_reg_am               : 1;
 351         uint8_t             must_be_one              : 1;
 352         uint8_t             const_index              : 5;
 353         uint8_t             const_index_sel          : 2;
 354         uint8_t             reserved0                : 3;
 355         uint8_t             src_swiz                 : 2;
 356         /* dword1: */
 357         uint16_t            dst_swiz                 : 12;
 358         uint8_t             format_comp_all          : 1;   /* '1' for signed, '0' for unsigned? */
 359         uint8_t             num_format_all           : 1;   /* '0' for normalized, '1' for unnormalized */
 360         uint8_t             signed_rf_mode_all       : 1;
 361         uint8_t             reserved1                : 1;
 362         instr_surf_fmt_t    format                   : 6;
 363         uint8_t             reserved2                : 1;
 364         uint8_t             exp_adjust_all           : 7;
 365         uint8_t             reserved3                : 1;
 366         uint8_t             pred_select              : 1;
 367         /* dword2: */
 368         uint8_t             stride                   : 8;
 369         /* possibly offset and reserved4 are swapped on a200? */
 370         uint8_t             offset                   : 8;
 371         uint8_t             reserved4                : 8;
 372         uint8_t             reserved5                : 7;
 373         uint8_t             pred_condition           : 1;
 374 } instr_fetch_vtx_t;
 375
 376 typedef union PACKED {
 377         instr_fetch_tex_t   tex;
 378         instr_fetch_vtx_t   vtx;
 379         struct PACKED {
 380                 /* dword0: */
 381                 instr_fetch_opc_t opc                    : 5;
 382                 uint32_t        dummy0                   : 27;
 383                 /* dword1: */
 384                 uint32_t        dummy1                   : 32;
 385                 /* dword2: */
 386                 uint32_t        dummy2                   : 32;
 387         };
 388 } instr_fetch_t;
 389
 390 #endif /* INSTR_H_ */