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