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
  29 struct r600_bytecode_alu_src {
  30         unsigned                        sel;
  31         unsigned                        chan;
  32         unsigned                        neg;
  33         unsigned                        abs;
  34         unsigned                        rel;
  35         unsigned                        kc_bank;
  36         unsigned                        kc_rel;
  37         uint32_t                        value;
  38 };
  39
  40 struct r600_bytecode_alu_dst {
  41         unsigned                        sel;
  42         unsigned                        chan;
  43         unsigned                        clamp;
  44         unsigned                        write;
  45         unsigned                        rel;
  46 };
  47
  48 struct r600_bytecode_alu {
  49         struct list_head                list;
  50         struct r600_bytecode_alu_src            src[3];
  51         struct r600_bytecode_alu_dst            dst;
  52         unsigned                        op;
  53         unsigned                        last;
  54         unsigned                        is_op3;
  55         unsigned                        execute_mask;
  56         unsigned                        update_pred;
  57         unsigned                        pred_sel;
  58         unsigned                        bank_swizzle;
  59         unsigned                        bank_swizzle_force;
  60         unsigned                        omod;
  61         unsigned                        index_mode;
  62 };
  63
  64 struct r600_bytecode_tex {
  65         struct list_head                list;
  66         unsigned                        op;
  67         unsigned                        inst_mod;
  68         unsigned                        resource_id;
  69         unsigned                        src_gpr;
  70         unsigned                        src_rel;
  71         unsigned                        dst_gpr;
  72         unsigned                        dst_rel;
  73         unsigned                        dst_sel_x;
  74         unsigned                        dst_sel_y;
  75         unsigned                        dst_sel_z;
  76         unsigned                        dst_sel_w;
  77         unsigned                        lod_bias;
  78         unsigned                        coord_type_x;
  79         unsigned                        coord_type_y;
  80         unsigned                        coord_type_z;
  81         unsigned                        coord_type_w;
  82         int                             offset_x;
  83         int                             offset_y;
  84         int                             offset_z;
  85         unsigned                        sampler_id;
  86         unsigned                        src_sel_x;
  87         unsigned                        src_sel_y;
  88         unsigned                        src_sel_z;
  89         unsigned                        src_sel_w;
  90         /* indexed samplers/resources only on evergreen/cayman */
  91         unsigned                        sampler_index_mode;
  92         unsigned                        resource_index_mode;
  93 };
  94
  95 struct r600_bytecode_vtx {
  96         struct list_head                list;
  97         unsigned                        op;
  98         unsigned                        fetch_type;
  99         unsigned                        buffer_id;
 100         unsigned                        src_gpr;
 101         unsigned                        src_sel_x;
 102         unsigned                        mega_fetch_count;
 103         unsigned                        dst_gpr;
 104         unsigned                        dst_sel_x;
 105         unsigned                        dst_sel_y;
 106         unsigned                        dst_sel_z;
 107         unsigned                        dst_sel_w;
 108         unsigned                        use_const_fields;
 109         unsigned                        data_format;
 110         unsigned                        num_format_all;
 111         unsigned                        format_comp_all;
 112         unsigned                        srf_mode_all;
 113         unsigned                        offset;
 114         unsigned                        endian;
 115         unsigned                        buffer_index_mode;
 116 };
 117
 118 struct r600_bytecode_output {
 119         unsigned                        array_base;
 120         unsigned                        array_size;
 121         unsigned                        comp_mask;
 122         unsigned                        type;
 123
 124         unsigned                        op;
 125
 126         unsigned                        elem_size;
 127         unsigned                        gpr;
 128         unsigned                        swizzle_x;
 129         unsigned                        swizzle_y;
 130         unsigned                        swizzle_z;
 131         unsigned                        swizzle_w;
 132         unsigned                        burst_count;
 133         unsigned                        index_gpr;
 134 };
 135
 136 struct r600_bytecode_kcache {
 137         unsigned                        bank;
 138         unsigned                        mode;
 139         unsigned                        addr;
 140         unsigned                        index_mode;
 141 };
 142
 143 struct r600_bytecode_cf {
 144         struct list_head                list;
 145
 146         unsigned                        op;
 147         unsigned                        addr;
 148         unsigned                        ndw;
 149         unsigned                        id;
 150         unsigned                        cond;
 151         unsigned                        pop_count;
 152         unsigned                        count;
 153         unsigned                        cf_addr; /* control flow addr */
 154         struct r600_bytecode_kcache             kcache[4];
 155         unsigned                        r6xx_uses_waterfall;
 156         unsigned                        eg_alu_extended;
 157         unsigned                        barrier;
 158         unsigned                        end_of_program;
 159         struct list_head                alu;
 160         struct list_head                tex;
 161         struct list_head                vtx;
 162         struct r600_bytecode_output             output;
 163         struct r600_bytecode_alu                *curr_bs_head;
 164         struct r600_bytecode_alu                *prev_bs_head;
 165         struct r600_bytecode_alu                *prev2_bs_head;
 166         unsigned isa[2];
 167 };
 168
 169 #define FC_NONE                         0
 170 #define FC_IF                           1
 171 #define FC_LOOP                         2
 172 #define FC_REP                          3
 173 #define FC_PUSH_VPM                     4
 174 #define FC_PUSH_WQM                     5
 175
 176 struct r600_cf_stack_entry {
 177         int                             type;
 178         struct r600_bytecode_cf         *start;
 179         struct r600_bytecode_cf         **mid; /* used to store the else point */
 180         int                             num_mid;
 181 };
 182
 183 #define SQ_MAX_CALL_DEPTH 0x00000020
 184
 185 #define AR_HANDLE_NORMAL 0
 186 #define AR_HANDLE_RV6XX 1 /* except RV670 */
 187
 188 struct r600_stack_info {
 189         /* current level of non-WQM PUSH operations
 190          * (PUSH, PUSH_ELSE, ALU_PUSH_BEFORE) */
 191         int push;
 192         /* current level of WQM PUSH operations
 193          * (PUSH, PUSH_ELSE, PUSH_WQM) */
 194         int push_wqm;
 195         /* current loop level */
 196         int loop;
 197
 198         /* required depth */
 199         int max_entries;
 200         /* subentries per entry */
 201         int entry_size;
 202 };
 203
 204 struct r600_bytecode {
 205         enum chip_class                 chip_class;
 206         enum radeon_family              family;
 207         bool                            has_compressed_msaa_texturing;
 208         int                             type;
 209         struct list_head                cf;
 210         struct r600_bytecode_cf         *cf_last;
 211         unsigned                        ndw;
 212         unsigned                        ncf;
 213         unsigned                        ngpr;
 214         unsigned                        nstack;
 215         unsigned                        nlds_dw;
 216         unsigned                        nresource;
 217         unsigned                        force_add_cf;
 218         uint32_t                        *bytecode;
 219         uint32_t                        fc_sp;
 220         struct r600_cf_stack_entry      fc_stack[32];
 221         struct r600_stack_info          stack;
 222         unsigned        ar_loaded;
 223         unsigned        ar_reg;
 224         unsigned        ar_chan;
 225         unsigned        ar_handling;
 226         unsigned        r6xx_nop_after_rel_dst;
 227         bool            index_loaded[2];
 228         unsigned        index_reg[2]; /* indexing register CF_INDEX_[01] */
 229         unsigned        debug_id;
 230         struct r600_isa* isa;
 231 };
 232
 233 /* eg_asm.c */
 234 int eg_bytecode_cf_build(struct r600_bytecode *bc, struct r600_bytecode_cf *cf);
 235 int egcm_load_index_reg(struct r600_bytecode *bc, unsigned id, bool inside_alu_clause);
 236
 237 /* r600_asm.c */
 238 void r600_bytecode_init(struct r600_bytecode *bc,
 239                         enum chip_class chip_class,
 240                         enum radeon_family family,
 241                         bool has_compressed_msaa_texturing);
 242 void r600_bytecode_clear(struct r600_bytecode *bc);
 243 int r600_bytecode_add_alu(struct r600_bytecode *bc,
 244                 const struct r600_bytecode_alu *alu);
 245 int r600_bytecode_add_vtx(struct r600_bytecode *bc,
 246                 const struct r600_bytecode_vtx *vtx);
 247 int r600_bytecode_add_tex(struct r600_bytecode *bc,
 248                 const struct r600_bytecode_tex *tex);
 249 int r600_bytecode_add_output(struct r600_bytecode *bc,
 250                 const struct r600_bytecode_output *output);
 251 int r600_bytecode_build(struct r600_bytecode *bc);
 252 int r600_bytecode_add_cf(struct r600_bytecode *bc);
 253 int r600_bytecode_add_cfinst(struct r600_bytecode *bc,
 254                 unsigned op);
 255 int r600_bytecode_add_alu_type(struct r600_bytecode *bc,
 256                 const struct r600_bytecode_alu *alu, unsigned type);
 257 void r600_bytecode_special_constants(uint32_t value,
 258                 unsigned *sel, unsigned *neg);
 259 void r600_bytecode_disasm(struct r600_bytecode *bc);
 260 void r600_bytecode_alu_read(struct r600_bytecode *bc,
 261                 struct r600_bytecode_alu *alu, uint32_t word0, uint32_t word1);
 262
 263 int cm_bytecode_add_cf_end(struct r600_bytecode *bc);
 264
 265 void *r600_create_vertex_fetch_shader(struct pipe_context *ctx,
 266                                       unsigned count,
 267                                       const struct pipe_vertex_element *elements);
 268
 269 /* r700_asm.c */
 270 void r700_bytecode_cf_vtx_build(uint32_t *bytecode,
 271                 const struct r600_bytecode_cf *cf);
 272 int r700_bytecode_alu_build(struct r600_bytecode *bc,
 273                 struct r600_bytecode_alu *alu, unsigned id);
 274 void r700_bytecode_alu_read(struct r600_bytecode *bc,
 275                 struct r600_bytecode_alu *alu, uint32_t word0, uint32_t word1);
 276 void r600_bytecode_export_read(struct r600_bytecode *bc,
 277                 struct r600_bytecode_output *output, uint32_t word0, uint32_t word1);
 278 void eg_bytecode_export_read(struct r600_bytecode *bc,
 279                 struct r600_bytecode_output *output, uint32_t word0, uint32_t word1);
 280
 281 void r600_vertex_data_type(enum pipe_format pformat, unsigned *format,
 282                            unsigned *num_format, unsigned *format_comp, unsigned *endian);
 283
 284 static inline int fp64_switch(int i)
 285 {
 286         switch (i) {
 287         case 0:
 288                 return 1;
 289         case 1:
 290                 return 0;
 291         case 2:
 292                 return 3;
 293         case 3:
 294                 return 2;
 295         }
 296         return 0;
 297 }
 298 #endif