src/gallium/drivers/r600/r600_asm.h

   1 /*
   2  * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
   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  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  */
  23 #ifndef R600_ASM_H
  24 #define R600_ASM_H
  25
  26 #include "r600_pipe.h"
  27 #include "r600_isa.h"
  28 #include "tgsi/tgsi_exec.h"
  29
  30 struct r600_bytecode_alu_src {
  31         unsigned                        sel;
  32         unsigned                        chan;
  33         unsigned                        neg;
  34         unsigned                        abs;
  35         unsigned                        rel;
  36         unsigned                        kc_bank;
  37         unsigned                        kc_rel;
  38         uint32_t                        value;
  39 };
  40
  41 struct r600_bytecode_alu_dst {
  42         unsigned                        sel;
  43         unsigned                        chan;
  44         unsigned                        clamp;
  45         unsigned                        write;
  46         unsigned                        rel;
  47 };
  48
  49 struct r600_bytecode_alu {
  50         struct list_head                list;
  51         struct r600_bytecode_alu_src            src[3];
  52         struct r600_bytecode_alu_dst            dst;
  53         unsigned                        op;
  54         unsigned                        last;
  55         unsigned                        is_op3;
  56         unsigned                        is_lds_idx_op;
  57         unsigned                        execute_mask;
  58         unsigned                        update_pred;
  59         unsigned                        pred_sel;
  60         unsigned                        bank_swizzle;
  61         unsigned                        bank_swizzle_force;
  62         unsigned                        omod;
  63         unsigned                        index_mode;
  64         unsigned                        lds_idx;
  65 };
  66
  67 struct r600_bytecode_tex {
  68         struct list_head                list;
  69         unsigned                        op;
  70         unsigned                        inst_mod;
  71         unsigned                        resource_id;
  72         unsigned                        src_gpr;
  73         unsigned                        src_rel;
  74         unsigned                        dst_gpr;
  75         unsigned                        dst_rel;
  76         unsigned                        dst_sel_x;
  77         unsigned                        dst_sel_y;
  78         unsigned                        dst_sel_z;
  79         unsigned                        dst_sel_w;
  80         unsigned                        lod_bias;
  81         unsigned                        coord_type_x;
  82         unsigned                        coord_type_y;
  83         unsigned                        coord_type_z;
  84         unsigned                        coord_type_w;
  85         int                             offset_x;
  86         int                             offset_y;
  87         int                             offset_z;
  88         unsigned                        sampler_id;
  89         unsigned                        src_sel_x;
  90         unsigned                        src_sel_y;
  91         unsigned                        src_sel_z;
  92         unsigned                        src_sel_w;
  93         /* indexed samplers/resources only on evergreen/cayman */
  94         unsigned                        sampler_index_mode;
  95         unsigned                        resource_index_mode;
  96 };
  97
  98 struct r600_bytecode_vtx {
  99         struct list_head                list;
 100         unsigned                        op;
 101         unsigned                        fetch_type;
 102         unsigned                        buffer_id;
 103         unsigned                        src_gpr;
 104         unsigned                        src_sel_x;
 105         unsigned                        mega_fetch_count;
 106         unsigned                        dst_gpr;
 107         unsigned                        dst_sel_x;
 108         unsigned                        dst_sel_y;
 109         unsigned                        dst_sel_z;
 110         unsigned                        dst_sel_w;
 111         unsigned                        use_const_fields;
 112         unsigned                        data_format;
 113         unsigned                        num_format_all;
 114         unsigned                        format_comp_all;
 115         unsigned                        srf_mode_all;
 116         unsigned                        offset;
 117         unsigned                        endian;
 118         unsigned                        buffer_index_mode;
 119
 120         // READ_SCRATCH fields
 121         unsigned                        uncached;
 122         unsigned                        indexed;
 123         unsigned                        src_sel_y;
 124         unsigned                        src_rel;
 125         unsigned                        elem_size;
 126         unsigned                        array_size;
 127         unsigned                        array_base;
 128         unsigned                        burst_count;
 129         unsigned                        dst_rel;
 130 };
 131
 132 struct r600_bytecode_gds {
 133         struct list_head                list;
 134         unsigned                        op;
 135         unsigned                        src_gpr;
 136         unsigned                        src_rel;
 137         unsigned                        src_sel_x;
 138         unsigned                        src_sel_y;
 139         unsigned                        src_sel_z;
 140         unsigned                        src_gpr2;
 141         unsigned                        dst_gpr;
 142         unsigned                        dst_rel;
 143         unsigned                        dst_sel_x;
 144         unsigned                        dst_sel_y;
 145         unsigned                        dst_sel_z;
 146         unsigned                        dst_sel_w;
 147         unsigned                        uav_index_mode;
 148         unsigned                        uav_id;
 149         unsigned                        alloc_consume;
 150         unsigned                        bcast_first_req;
 151 };
 152
 153 struct r600_bytecode_output {
 154         unsigned                        array_base;
 155         unsigned                        array_size;
 156         unsigned                        comp_mask;
 157         unsigned                        type;
 158
 159         unsigned                        op;
 160
 161         unsigned                        elem_size;
 162         unsigned                        gpr;
 163         unsigned                        swizzle_x;
 164         unsigned                        swizzle_y;
 165         unsigned                        swizzle_z;
 166         unsigned                        swizzle_w;
 167         unsigned                        burst_count;
 168         unsigned                        index_gpr;
 169         unsigned                        mark; /* used by MEM_SCRATCH */
 170 };
 171
 172 struct r600_bytecode_rat {
 173         unsigned                        id;
 174         unsigned                        inst;
 175         unsigned                        index_mode;
 176 };
 177
 178 struct r600_bytecode_kcache {
 179         unsigned                        bank;
 180         unsigned                        mode;
 181         unsigned                        addr;
 182         unsigned                        index_mode;
 183 };
 184
 185 struct r600_bytecode_cf {
 186         struct list_head                list;
 187
 188         unsigned                        op;
 189         unsigned                        addr;
 190         unsigned                        ndw;
 191         unsigned                        id;
 192         unsigned                        cond;
 193         unsigned                        pop_count;
 194         unsigned                        count;
 195         unsigned                        cf_addr; /* control flow addr */
 196         struct r600_bytecode_kcache             kcache[4];
 197         unsigned                        r6xx_uses_waterfall;
 198         unsigned                        eg_alu_extended;
 199         unsigned                        barrier;
 200         unsigned                        end_of_program;
 201         unsigned                        mark;
 202         unsigned                        vpm;
 203         struct list_head                alu;
 204         struct list_head                tex;
 205         struct list_head                vtx;
 206         struct list_head                gds;
 207         struct r600_bytecode_output             output;
 208         struct r600_bytecode_rat                rat;
 209         struct r600_bytecode_alu                *curr_bs_head;
 210         struct r600_bytecode_alu                *prev_bs_head;
 211         struct r600_bytecode_alu                *prev2_bs_head;
 212         unsigned isa[2];
 213 };
 214
 215 #define FC_NONE                         0
 216 #define FC_IF                           1
 217 #define FC_LOOP                         2
 218 #define FC_REP                          3
 219 #define FC_PUSH_VPM                     4
 220 #define FC_PUSH_WQM                     5
 221
 222 struct r600_cf_stack_entry {
 223         int                             type;
 224         struct r600_bytecode_cf         *start;
 225         struct r600_bytecode_cf         **mid; /* used to store the else point */
 226         int                             num_mid;
 227 };
 228
 229 #define SQ_MAX_CALL_DEPTH 0x00000020
 230
 231 #define AR_HANDLE_NORMAL 0
 232 #define AR_HANDLE_RV6XX 1 /* except RV670 */
 233
 234 struct r600_stack_info {
 235         /* current level of non-WQM PUSH operations
 236          * (PUSH, PUSH_ELSE, ALU_PUSH_BEFORE) */
 237         int push;
 238         /* current level of WQM PUSH operations
 239          * (PUSH, PUSH_ELSE, PUSH_WQM) */
 240         int push_wqm;
 241         /* current loop level */
 242         int loop;
 243
 244         /* required depth */
 245         int max_entries;
 246         /* subentries per entry */
 247         int entry_size;
 248 };
 249
 250 struct r600_bytecode {
 251         enum chip_class                 chip_class;
 252         enum radeon_family              family;
 253         bool                            has_compressed_msaa_texturing;
 254         int                             type;
 255         struct list_head                cf;
 256         struct r600_bytecode_cf         *cf_last;
 257         unsigned                        ndw;
 258         unsigned                        ncf;
 259         unsigned                        ngpr;
 260         unsigned                        nstack;
 261         unsigned                        nlds_dw;
 262         unsigned                        nresource;
 263         unsigned                        force_add_cf;
 264         uint32_t                        *bytecode;
 265         uint32_t                        fc_sp;
 266         struct r600_cf_stack_entry      fc_stack[TGSI_EXEC_MAX_NESTING];
 267         struct r600_stack_info          stack;
 268         unsigned        ar_loaded;
 269         unsigned        ar_reg;
 270         unsigned        ar_chan;
 271         unsigned        ar_handling;
 272         unsigned        r6xx_nop_after_rel_dst;
 273         bool            index_loaded[2];
 274         unsigned        index_reg[2]; /* indexing register CF_INDEX_[01] */
 275         unsigned        debug_id;
 276         struct r600_isa* isa;
 277         struct r600_bytecode_output pending_outputs[5];
 278         int n_pending_outputs;
 279         boolean                 need_wait_ack; /* emit a pending WAIT_ACK prior to control flow */
 280         boolean                 precise;
 281 };
 282
 283 /* eg_asm.c */
 284 int eg_bytecode_cf_build(struct r600_bytecode *bc, struct r600_bytecode_cf *cf);
 285 int egcm_load_index_reg(struct r600_bytecode *bc, unsigned id, bool inside_alu_clause);
 286 int eg_bytecode_gds_build(struct r600_bytecode *bc, struct r600_bytecode_gds *gds, unsigned id);
 287 int eg_bytecode_alu_build(struct r600_bytecode *bc,
 288                           struct r600_bytecode_alu *alu, unsigned id);
 289 /* r600_asm.c */
 290 void r600_bytecode_init(struct r600_bytecode *bc,
 291                         enum chip_class chip_class,
 292                         enum radeon_family family,
 293                         bool has_compressed_msaa_texturing);
 294 void r600_bytecode_clear(struct r600_bytecode *bc);
 295 int r600_bytecode_add_alu(struct r600_bytecode *bc,
 296                 const struct r600_bytecode_alu *alu);
 297 int r600_bytecode_add_vtx(struct r600_bytecode *bc,
 298                 const struct r600_bytecode_vtx *vtx);
 299 int r600_bytecode_add_vtx_tc(struct r600_bytecode *bc,
 300                              const struct r600_bytecode_vtx *vtx);
 301 int r600_bytecode_add_tex(struct r600_bytecode *bc,
 302                 const struct r600_bytecode_tex *tex);
 303 int r600_bytecode_add_gds(struct r600_bytecode *bc,
 304                 const struct r600_bytecode_gds *gds);
 305 int r600_bytecode_add_output(struct r600_bytecode *bc,
 306                 const struct r600_bytecode_output *output);
 307 int r600_bytecode_add_pending_output(struct r600_bytecode *bc,
 308                 const struct r600_bytecode_output *output);
 309 void r600_bytecode_need_wait_ack(struct r600_bytecode *bc, boolean needed);
 310 boolean r600_bytecode_get_need_wait_ack(struct r600_bytecode *bc);
 311 int r600_bytecode_build(struct r600_bytecode *bc);
 312 int r600_bytecode_add_cf(struct r600_bytecode *bc);
 313 int r600_bytecode_add_cfinst(struct r600_bytecode *bc,
 314                 unsigned op);
 315 int r600_bytecode_add_alu_type(struct r600_bytecode *bc,
 316                 const struct r600_bytecode_alu *alu, unsigned type);
 317 void r600_bytecode_special_constants(uint32_t value,
 318                 unsigned *sel, unsigned *neg, unsigned abs);
 319 void r600_bytecode_disasm(struct r600_bytecode *bc);
 320 void r600_bytecode_alu_read(struct r600_bytecode *bc,
 321                 struct r600_bytecode_alu *alu, uint32_t word0, uint32_t word1);
 322
 323 int cm_bytecode_add_cf_end(struct r600_bytecode *bc);
 324
 325 void *r600_create_vertex_fetch_shader(struct pipe_context *ctx,
 326                                       unsigned count,
 327                                       const struct pipe_vertex_element *elements);
 328
 329 /* r700_asm.c */
 330 void r700_bytecode_cf_vtx_build(uint32_t *bytecode,
 331                 const struct r600_bytecode_cf *cf);
 332 int r700_bytecode_alu_build(struct r600_bytecode *bc,
 333                 struct r600_bytecode_alu *alu, unsigned id);
 334 void r700_bytecode_alu_read(struct r600_bytecode *bc,
 335                 struct r600_bytecode_alu *alu, uint32_t word0, uint32_t word1);
 336 int r700_bytecode_fetch_mem_build(struct r600_bytecode *bc,
 337                 struct r600_bytecode_vtx *mem, unsigned id);
 338
 339 void r600_bytecode_export_read(struct r600_bytecode *bc,
 340                 struct r600_bytecode_output *output, uint32_t word0, uint32_t word1);
 341 void eg_bytecode_export_read(struct r600_bytecode *bc,
 342                 struct r600_bytecode_output *output, uint32_t word0, uint32_t word1);
 343
 344 void r600_vertex_data_type(enum pipe_format pformat, unsigned *format,
 345                            unsigned *num_format, unsigned *format_comp, unsigned *endian);
 346
 347 static inline int fp64_switch(int i)
 348 {
 349         switch (i) {
 350         case 0:
 351                 return 1;
 352         case 1:
 353                 return 0;
 354         case 2:
 355                 return 3;
 356         case 3:
 357                 return 2;
 358         }
 359         return 0;
 360 }
 361 #endif