src/gallium/drivers/cell/common.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 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 /**
  29  * Types and tokens which are common to the SPU and PPU code.
  30  */
  31
  32
  33 #ifndef CELL_COMMON_H
  34 #define CELL_COMMON_H
  35
  36 #include "pipe/p_compiler.h"
  37 #include "pipe/p_format.h"
  38 #include "pipe/p_state.h"
  39
  40
  41 /** The standard assert macro doesn't seem to work reliably */
  42 #define ASSERT(x) \
  43    if (!(x)) { \
  44       ubyte *p = NULL; \
  45       fprintf(stderr, "%s:%d: %s(): assertion %s failed.\n", \
  46               __FILE__, __LINE__, __FUNCTION__, #x);             \
  47       *p = 0; \
  48       exit(1); \
  49    }
  50
  51
  52
  53 #define JOIN(x, y) JOIN_AGAIN(x, y)
  54 #define JOIN_AGAIN(x, y) x ## y
  55
  56 #define STATIC_ASSERT(e) \
  57 {typedef char JOIN(assertion_failed_at_line_, __LINE__) [(e) ? 1 : -1];}
  58
  59
  60
  61 /** for sanity checking */
  62 #define ASSERT_ALIGN16(ptr) \
  63   ASSERT((((unsigned long) (ptr)) & 0xf) == 0);
  64
  65
  66 /** round up value to next multiple of 4 */
  67 #define ROUNDUP4(k)  (((k) + 0x3) & ~0x3)
  68
  69 /** round up value to next multiple of 8 */
  70 #define ROUNDUP8(k)  (((k) + 0x7) & ~0x7)
  71
  72 /** round up value to next multiple of 16 */
  73 #define ROUNDUP16(k)  (((k) + 0xf) & ~0xf)
  74
  75
  76 #define CELL_MAX_SPUS 8
  77
  78 #define CELL_MAX_SAMPLERS 4
  79 #define CELL_MAX_TEXTURE_LEVELS 12  /* 2k x 2k */
  80 #define CELL_MAX_CONSTANTS 32  /**< number of float[4] constants */
  81 #define CELL_MAX_WIDTH 1024    /**< max framebuffer width */
  82 #define CELL_MAX_HEIGHT 1024   /**< max framebuffer width */
  83
  84 #define TILE_SIZE 32
  85
  86
  87 /**
  88  * The low byte of a mailbox word contains the command opcode.
  89  * Remaining higher bytes are command specific.
  90  */
  91 #define CELL_CMD_OPCODE_MASK 0xff
  92
  93 #define CELL_CMD_EXIT                 1
  94 #define CELL_CMD_CLEAR_SURFACE        2
  95 #define CELL_CMD_FINISH               3
  96 #define CELL_CMD_RENDER               4
  97 #define CELL_CMD_BATCH                5
  98 #define CELL_CMD_RELEASE_VERTS        6
  99 #define CELL_CMD_STATE_FRAMEBUFFER   10
 100 #define CELL_CMD_STATE_FRAGMENT_OPS  11
 101 #define CELL_CMD_STATE_SAMPLER       12
 102 #define CELL_CMD_STATE_TEXTURE       13
 103 #define CELL_CMD_STATE_VERTEX_INFO   14
 104 #define CELL_CMD_STATE_VIEWPORT      15
 105 #define CELL_CMD_STATE_UNIFORMS      16
 106 #define CELL_CMD_STATE_VS_ARRAY_INFO 17
 107 #define CELL_CMD_STATE_BIND_VS       18
 108 #define CELL_CMD_STATE_FRAGMENT_PROGRAM 19
 109 #define CELL_CMD_STATE_ATTRIB_FETCH  20
 110 #define CELL_CMD_STATE_FS_CONSTANTS  21
 111 #define CELL_CMD_STATE_RASTERIZER    22
 112 #define CELL_CMD_VS_EXECUTE          23
 113 #define CELL_CMD_FLUSH_BUFFER_RANGE  24
 114 #define CELL_CMD_FENCE               25
 115
 116
 117 /** Command/batch buffers */
 118 #define CELL_NUM_BUFFERS 4
 119 #define CELL_BUFFER_SIZE (4*1024)  /**< 16KB would be the max */
 120
 121 #define CELL_BUFFER_STATUS_FREE 10
 122 #define CELL_BUFFER_STATUS_USED 20
 123
 124 /** Debug flags */
 125 #define CELL_DEBUG_CHECKER              (1 << 0)
 126 #define CELL_DEBUG_ASM                  (1 << 1)
 127 #define CELL_DEBUG_SYNC                 (1 << 2)
 128 #define CELL_DEBUG_FRAGMENT_OPS         (1 << 3)
 129 #define CELL_DEBUG_FRAGMENT_OP_FALLBACK (1 << 4)
 130 #define CELL_DEBUG_CMD                  (1 << 5)
 131 #define CELL_DEBUG_CACHE                (1 << 6)
 132
 133 #define CELL_FENCE_IDLE      0
 134 #define CELL_FENCE_EMITTED   1
 135 #define CELL_FENCE_SIGNALLED 2
 136
 137 #define CELL_FACING_FRONT    0
 138 #define CELL_FACING_BACK     1
 139
 140 struct cell_fence
 141 {
 142    /** There's a 16-byte status qword per SPU */
 143    volatile uint status[CELL_MAX_SPUS][4];
 144 };
 145
 146 #ifdef __SPU__
 147 typedef vector unsigned int opcode_t;
 148 #else
 149 typedef unsigned int opcode_t[4];
 150 #endif
 151
 152 /**
 153  * Fence command sent to SPUs.  In response, the SPUs will write
 154  * CELL_FENCE_STATUS_SIGNALLED back to the fence status word in main memory.
 155  */
 156 struct cell_command_fence
 157 {
 158    opcode_t opcode;      /**< CELL_CMD_FENCE */
 159    struct cell_fence *fence;
 160    uint32_t pad_[3];
 161 };
 162
 163
 164 /**
 165  * Command to specify per-fragment operations state and generated code.
 166  * Note that this is a variant-length structure, allocated with as
 167  * much memory as needed to hold the generated code; the "code"
 168  * field *must* be the last field in the structure.  Also, the entire
 169  * length of the structure (including the variant code field) must be
 170  * a multiple of 8 bytes; we require that this structure itself be
 171  * a multiple of 8 bytes, and that the generated code also be a multiple
 172  * of 8 bytes.
 173  *
 174  * Also note that the dsa, blend, blend_color fields are really only needed
 175  * for the fallback/C per-pixel code.  They're not used when we generate
 176  * dynamic SPU fragment code (which is the normal case), and will eventually
 177  * be removed from this structure.
 178  */
 179 struct cell_command_fragment_ops
 180 {
 181    opcode_t opcode;      /**< CELL_CMD_STATE_FRAGMENT_OPS */
 182
 183    /* Fields for the fallback case */
 184    struct pipe_depth_stencil_alpha_state dsa;
 185    struct pipe_blend_state blend;
 186    struct pipe_blend_color blend_color;
 187
 188    /* Fields for the generated SPU code */
 189    unsigned total_code_size;
 190    unsigned front_code_index;
 191    unsigned back_code_index;
 192    /* this field has variant length, and must be the last field in
 193     * the structure
 194     */
 195    unsigned code[0];
 196 };
 197
 198
 199 /** Max instructions for fragment programs */
 200 #define SPU_MAX_FRAGMENT_PROGRAM_INSTS 512
 201
 202 /**
 203  * Command to send a fragment program to SPUs.
 204  */
 205 struct cell_command_fragment_program
 206 {
 207    opcode_t opcode;      /**< CELL_CMD_STATE_FRAGMENT_PROGRAM */
 208    uint num_inst;        /**< Number of instructions */
 209    uint32_t pad[3];
 210    unsigned code[SPU_MAX_FRAGMENT_PROGRAM_INSTS];
 211 };
 212
 213
 214 /**
 215  * Tell SPUs about the framebuffer size, location
 216  */
 217 struct cell_command_framebuffer
 218 {
 219    opcode_t opcode;     /**< CELL_CMD_STATE_FRAMEBUFFER */
 220    int width, height;
 221    void *color_start, *depth_start;
 222    enum pipe_format color_format, depth_format;
 223    uint32_t pad_[2];
 224 };
 225
 226
 227 /**
 228  * Tell SPUs about rasterizer state.
 229  */
 230 struct cell_command_rasterizer
 231 {
 232    opcode_t opcode;    /**< CELL_CMD_STATE_RASTERIZER */
 233    struct pipe_rasterizer_state rasterizer;
 234 };
 235
 236
 237 /**
 238  * Clear framebuffer to the given value/color.
 239  */
 240 struct cell_command_clear_surface
 241 {
 242    opcode_t opcode;     /**< CELL_CMD_CLEAR_SURFACE */
 243    uint surface; /**< Temporary: 0=color, 1=Z */
 244    uint value;
 245    uint32_t pad[2];
 246 };
 247
 248
 249 /**
 250  * Array info used by the vertex shader's vertex puller.
 251  */
 252 struct cell_array_info
 253 {
 254    uint64_t base;      /**< Base address of the 0th element. */
 255    uint attr;          /**< Attribute that this state is for. */
 256    uint pitch;         /**< Byte pitch from one entry to the next. */
 257    uint size;
 258    uint function_offset;
 259 };
 260
 261
 262 struct cell_attribute_fetch_code
 263 {
 264    uint64_t base;
 265    uint size;
 266 };
 267
 268
 269 struct cell_buffer_range
 270 {
 271    uint64_t base;
 272    unsigned size;
 273 };
 274
 275
 276 struct cell_shader_info
 277 {
 278    uint64_t declarations;
 279    uint64_t instructions;
 280    uint64_t  immediates;
 281
 282    unsigned num_outputs;
 283    unsigned num_declarations;
 284    unsigned num_instructions;
 285    unsigned num_immediates;
 286 };
 287
 288
 289 #define SPU_VERTS_PER_BATCH 64
 290 struct cell_command_vs
 291 {
 292    opcode_t opcode;       /**< CELL_CMD_VS_EXECUTE */
 293    uint64_t vOut[SPU_VERTS_PER_BATCH];
 294    unsigned num_elts;
 295    unsigned elts[SPU_VERTS_PER_BATCH];
 296    float plane[12][4];
 297    unsigned nr_planes;
 298    unsigned nr_attrs;
 299 };
 300
 301
 302 struct cell_command_render
 303 {
 304    opcode_t opcode;   /**< CELL_CMD_RENDER */
 305    uint prim_type;    /**< PIPE_PRIM_x */
 306    uint num_verts;
 307    uint vertex_size;  /**< bytes per vertex */
 308    uint num_indexes;
 309    uint vertex_buf;  /**< which cell->buffer[] contains the vertex data */
 310    float xmin, ymin, xmax, ymax;  /* XXX another dummy field */
 311    uint min_index;
 312    boolean inline_verts;
 313    uint32_t pad_[1];
 314 };
 315
 316
 317 struct cell_command_release_verts
 318 {
 319    opcode_t opcode;         /**< CELL_CMD_RELEASE_VERTS */
 320    uint vertex_buf;    /**< in [0, CELL_NUM_BUFFERS-1] */
 321    uint32_t pad_[3];
 322 };
 323
 324
 325 struct cell_command_sampler
 326 {
 327    opcode_t opcode;         /**< CELL_CMD_STATE_SAMPLER */
 328    uint unit;
 329    struct pipe_sampler_state state;
 330    uint32_t pad_[1];
 331 };
 332
 333
 334 struct cell_command_texture
 335 {
 336    opcode_t opcode;     /**< CELL_CMD_STATE_TEXTURE */
 337    uint target;         /**< PIPE_TEXTURE_x */
 338    uint unit;
 339    void *start[CELL_MAX_TEXTURE_LEVELS];   /**< Address in main memory */
 340    ushort width[CELL_MAX_TEXTURE_LEVELS];
 341    ushort height[CELL_MAX_TEXTURE_LEVELS];
 342    ushort depth[CELL_MAX_TEXTURE_LEVELS];
 343 };
 344
 345
 346 #define MAX_SPU_FUNCTIONS 12
 347 /**
 348  * Used to tell the PPU about the address of particular functions in the
 349  * SPU's address space.
 350  */
 351 struct cell_spu_function_info
 352 {
 353    uint num;
 354    char names[MAX_SPU_FUNCTIONS][16];
 355    uint addrs[MAX_SPU_FUNCTIONS];
 356    char pad[12];   /**< Pad struct to multiple of 16 bytes (256 currently) */
 357 };
 358
 359
 360 /** This is the object passed to spe_create_thread() */
 361 struct cell_init_info
 362 {
 363    unsigned id;
 364    unsigned num_spus;
 365    unsigned debug_flags;  /**< mask of CELL_DEBUG_x flags */
 366    float inv_timebase;    /**< 1.0/timebase, for perf measurement */
 367
 368    /** Buffers for command batches, vertex/index data */
 369    ubyte *buffers[CELL_NUM_BUFFERS];
 370    uint *buffer_status;  /**< points at cell_context->buffer_status */
 371
 372    struct cell_spu_function_info *spu_functions;
 373 } ALIGN16_ATTRIB;
 374
 375
 376 #endif /* CELL_COMMON_H */