src/gallium/auxiliary/draw/draw_vs_aos.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 /* Authors:  Keith Whitwell <keith@tungstengraphics.com>
  29  */
  30
  31 #ifndef DRAW_VS_AOS_H
  32 #define DRAW_VS_AOS_H
  33
  34 #include "pipe/p_config.h"
  35 #include "tgsi/tgsi_exec.h"
  36 #include "draw_vs.h"
  37
  38 #ifdef PIPE_ARCH_X86
  39
  40 struct tgsi_token;
  41 struct x86_function;
  42
  43 #include "pipe/p_state.h"
  44 #include "rtasm/rtasm_x86sse.h"
  45
  46
  47
  48
  49
  50 #define X    0
  51 #define Y    1
  52 #define Z    2
  53 #define W    3
  54
  55 #define MAX_INPUTS     PIPE_MAX_ATTRIBS
  56 #define MAX_OUTPUTS    PIPE_MAX_SHADER_OUTPUTS
  57 #define MAX_TEMPS      TGSI_EXEC_NUM_TEMPS
  58 #define MAX_CONSTANTS  1024  /** only used for sanity checking */
  59 #define MAX_IMMEDIATES 1024  /** only used for sanity checking */
  60 #define MAX_INTERNALS  8     /** see IMM_x values below */
  61
  62 #define AOS_FILE_INTERNAL TGSI_FILE_COUNT
  63
  64 #define FPU_RND_NEG    1
  65 #define FPU_RND_NEAREST 2
  66
  67 struct aos_machine;
  68 typedef void (PIPE_CDECL *lit_func)( struct aos_machine *,
  69                                     float *result,
  70                                     const float *in,
  71                                     unsigned count );
  72
  73 void PIPE_CDECL aos_do_lit( struct aos_machine *machine,
  74                             float *result,
  75                             const float *in,
  76                             unsigned count );
  77
  78 struct shine_tab {
  79    float exponent;
  80    float values[258];
  81    unsigned last_used;
  82 };
  83
  84 struct lit_info {
  85    lit_func func;
  86    struct shine_tab *shine_tab;
  87 };
  88
  89 #define MAX_SHINE_TAB    4
  90 #define MAX_LIT_INFO     16
  91
  92 struct aos_buffer {
  93    const void *base_ptr;
  94    unsigned stride;
  95    void *ptr;                   /* updated per vertex */
  96 };
  97
  98
  99
 100
 101 /* This is the temporary storage used by all the aos_sse vs varients.
 102  * Create one per context and reuse by passing a pointer in at
 103  * vs_varient creation??
 104  */
 105 struct aos_machine {
 106    float input    [MAX_INPUTS    ][4];
 107    float output   [MAX_OUTPUTS   ][4];
 108    float temp     [MAX_TEMPS     ][4];
 109    float internal [MAX_INTERNALS ][4];
 110
 111    float scale[4];              /* viewport */
 112    float translate[4];          /* viewport */
 113
 114    float tmp[2][4];             /* scratch space for LIT */
 115
 116    struct shine_tab shine_tab[MAX_SHINE_TAB];
 117    struct lit_info  lit_info[MAX_LIT_INFO];
 118    unsigned now;
 119
 120
 121    ushort fpu_rnd_nearest;
 122    ushort fpu_rnd_neg_inf;
 123    ushort fpu_restore;
 124    ushort fpucntl;              /* one of FPU_* above */
 125
 126    const float (*immediates)[4];     /* points to shader data */
 127    const void *constants[PIPE_MAX_CONSTANT_BUFFERS]; /* points to draw data */
 128
 129    const struct aos_buffer *buffer; /* points to ? */
 130 };
 131
 132
 133
 134
 135 struct aos_compilation {
 136    struct x86_function *func;
 137    struct draw_vs_varient_aos_sse *vaos;
 138
 139    unsigned insn_counter;
 140    unsigned num_immediates;
 141    unsigned count;
 142    unsigned lit_count;
 143
 144    struct {
 145       unsigned idx:16;
 146       unsigned file:8;
 147       unsigned dirty:8;
 148       unsigned last_used;
 149    } xmm[8];
 150
 151    unsigned x86_reg[2];                /* one of X86_* */
 152
 153    boolean input_fetched[PIPE_MAX_ATTRIBS];
 154    unsigned output_last_write[PIPE_MAX_ATTRIBS];
 155
 156    boolean have_sse2;
 157    boolean error;
 158    short fpucntl;
 159
 160    /* these are actually known values, but putting them in a struct
 161     * like this is helpful to keep them in sync across the file.
 162     */
 163    struct x86_reg tmp_EAX;
 164    struct x86_reg idx_EBX;     /* either start+i or &elt[i] */
 165    struct x86_reg outbuf_ECX;
 166    struct x86_reg machine_EDX;
 167    struct x86_reg count_ESI;    /* decrements to zero */
 168    struct x86_reg temp_EBP;
 169    struct x86_reg stack_ESP;
 170 };
 171
 172 struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp );
 173 void aos_release_xmm_reg( struct aos_compilation *cp, unsigned idx );
 174
 175 void aos_adopt_xmm_reg( struct aos_compilation *cp,
 176                         struct x86_reg reg,
 177                         unsigned file,
 178                         unsigned idx,
 179                         unsigned dirty );
 180
 181 void aos_spill_all( struct aos_compilation *cp );
 182
 183 struct x86_reg aos_get_shader_reg( struct aos_compilation *cp,
 184                                    unsigned file,
 185                                    unsigned idx );
 186
 187 boolean aos_init_inputs( struct aos_compilation *cp, boolean linear );
 188 boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear );
 189 boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear );
 190
 191 boolean aos_emit_outputs( struct aos_compilation *cp );
 192
 193
 194 #define IMM_ONES     0              /* 1, 1,1,1 */
 195 #define IMM_SWZ      1              /* 1,-1,0, 0xffffffff */
 196 #define IMM_IDENTITY 2              /* 0, 0,0,1 */
 197 #define IMM_INV_255  3              /* 1/255, 1/255, 1/255, 1/255 */
 198 #define IMM_255      4              /* 255, 255, 255, 255 */
 199 #define IMM_NEGS     5              /* -1,-1,-1,-1 */
 200 #define IMM_RSQ      6              /* -.5,1.5,_,_ */
 201 #define IMM_PSIZE    7              /* not really an immediate - updated each run */
 202
 203 struct x86_reg aos_get_internal( struct aos_compilation *cp,
 204                                  unsigned imm );
 205 struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp,
 206                                      unsigned imm );
 207
 208
 209 #define AOS_ERROR(cp, msg)                                                  \
 210 do {                                                                    \
 211    if (0) debug_printf("%s: x86 translation failed: %s\n", __FUNCTION__, msg); \
 212    cp->error = 1;                                                       \
 213 } while (0)
 214
 215
 216 #define X86_NULL       0
 217 #define X86_IMMEDIATES 1
 218 #define X86_CONSTANTS  2
 219 #define X86_BUFFERS    3
 220
 221 struct x86_reg aos_get_x86( struct aos_compilation *cp,
 222                             unsigned which_reg,
 223                             unsigned value );
 224
 225
 226 typedef void (PIPE_CDECL *vaos_run_elts_func)( struct aos_machine *,
 227                                                const unsigned *elts,
 228                                                unsigned count,
 229                                                void *output_buffer);
 230
 231 typedef void (PIPE_CDECL *vaos_run_linear_func)( struct aos_machine *,
 232                                                 unsigned start,
 233                                                 unsigned count,
 234                                                 void *output_buffer);
 235
 236
 237 struct draw_vs_varient_aos_sse {
 238    struct draw_vs_varient base;
 239    struct draw_context *draw;
 240
 241    struct aos_buffer *buffer;
 242    unsigned nr_vb;
 243
 244    vaos_run_linear_func gen_run_linear;
 245    vaos_run_elts_func gen_run_elts;
 246
 247
 248    struct x86_function func[2];
 249 };
 250
 251
 252 #endif
 253
 254 #endif
 255