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
  35 struct tgsi_token;
  36 struct x86_function;
  37
  38 #include "pipe/p_state.h"
  39 #include "rtasm/rtasm_x86sse.h"
  40
  41
  42
  43
  44
  45 #define X    0
  46 #define Y    1
  47 #define Z    2
  48 #define W    3
  49
  50 #define MAX_INPUTS     PIPE_MAX_ATTRIBS
  51 #define MAX_OUTPUTS    PIPE_MAX_ATTRIBS
  52 #define MAX_TEMPS      PIPE_MAX_ATTRIBS /* say */
  53 #define MAX_CONSTANTS  PIPE_MAX_ATTRIBS /* say */
  54 #define MAX_IMMEDIATES PIPE_MAX_ATTRIBS /* say */
  55 #define MAX_INTERNALS  8
  56
  57 #define AOS_FILE_INTERNAL TGSI_FILE_COUNT
  58
  59 #define FPU_RND_NEG    1
  60 #define FPU_RND_NEAREST 2
  61
  62 struct aos_machine;
  63 typedef void (PIPE_CDECL *lit_func)( struct aos_machine *,
  64                                     float *result,
  65                                     const float *in,
  66                                     unsigned count );
  67
  68 void PIPE_CDECL aos_do_lit( struct aos_machine *machine,
  69                             float *result,
  70                             const float *in,
  71                             unsigned count );
  72
  73 struct shine_tab {
  74    float exponent;
  75    float values[258];
  76    unsigned last_used;
  77 };
  78
  79 struct lit_info {
  80    lit_func func;
  81    struct shine_tab *shine_tab;
  82 };
  83
  84 #define MAX_SHINE_TAB    4
  85 #define MAX_LIT_INFO     16
  86
  87 struct aos_attrib {
  88    const void *input_ptr;
  89    unsigned input_stride;
  90 };
  91
  92
  93
  94
  95 /* This is the temporary storage used by all the aos_sse vs varients.
  96  * Create one per context and reuse by passing a pointer in at
  97  * vs_varient creation??
  98  */
  99 struct aos_machine {
 100    float input    [MAX_INPUTS    ][4];
 101    float output   [MAX_OUTPUTS   ][4];
 102    float temp     [MAX_TEMPS     ][4];
 103    float internal [MAX_INTERNALS ][4];
 104
 105    float scale[4];              /* viewport */
 106    float translate[4];          /* viewport */
 107
 108    float tmp[2][4];             /* scratch space for LIT */
 109
 110    struct shine_tab shine_tab[MAX_SHINE_TAB];
 111    struct lit_info  lit_info[MAX_LIT_INFO];
 112    unsigned now;
 113
 114
 115    ushort fpu_rnd_nearest;
 116    ushort fpu_rnd_neg_inf;
 117    ushort fpu_restore;
 118    ushort fpucntl;              /* one of FPU_* above */
 119
 120    const float (*immediates)[4];     /* points to shader data */
 121    const float (*constants)[4];      /* points to draw data */
 122
 123    const struct aos_attrib *attrib; /* points to ? */
 124 };
 125
 126
 127
 128
 129 struct aos_compilation {
 130    struct x86_function *func;
 131    struct draw_vs_varient_aos_sse *vaos;
 132
 133    unsigned insn_counter;
 134    unsigned num_immediates;
 135    unsigned count;
 136    unsigned lit_count;
 137
 138    struct {
 139       unsigned idx:16;
 140       unsigned file:8;
 141       unsigned dirty:8;
 142       unsigned last_used;
 143    } xmm[8];
 144
 145    unsigned ebp;                /* one of X86_* */
 146
 147    boolean input_fetched[PIPE_MAX_ATTRIBS];
 148    unsigned output_last_write[PIPE_MAX_ATTRIBS];
 149
 150    boolean have_sse2;
 151    boolean error;
 152    short fpucntl;
 153
 154    /* these are actually known values, but putting them in a struct
 155     * like this is helpful to keep them in sync across the file.
 156     */
 157    struct x86_reg tmp_EAX;
 158    struct x86_reg idx_EBX;     /* either start+i or &elt[i] */
 159    struct x86_reg outbuf_ECX;
 160    struct x86_reg machine_EDX;
 161    struct x86_reg count_ESI;    /* decrements to zero */
 162    struct x86_reg temp_EBP;
 163    struct x86_reg stack_ESP;
 164 };
 165
 166 struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp );
 167 void aos_release_xmm_reg( struct aos_compilation *cp, unsigned idx );
 168
 169 void aos_adopt_xmm_reg( struct aos_compilation *cp,
 170                         struct x86_reg reg,
 171                         unsigned file,
 172                         unsigned idx,
 173                         unsigned dirty );
 174
 175 struct x86_reg aos_get_shader_reg( struct aos_compilation *cp,
 176                                    unsigned file,
 177                                    unsigned idx );
 178
 179 boolean aos_fetch_inputs( struct aos_compilation *cp,
 180                           boolean linear );
 181
 182 boolean aos_emit_outputs( struct aos_compilation *cp );
 183
 184
 185 #define IMM_ONES     0              /* 1, 1,1,1 */
 186 #define IMM_SWZ      1              /* 1,-1,0, 0xffffffff */
 187 #define IMM_IDENTITY 2              /* 0, 0,0,1 */
 188 #define IMM_INV_255  3              /* 1/255, 1/255, 1/255, 1/255 */
 189 #define IMM_255      4              /* 255, 255, 255, 255 */
 190 #define IMM_NEGS     5              /* -1,-1,-1,-1 */
 191 #define IMM_RSQ      6              /* -.5,1.5,_,_ */
 192 #define IMM_PSIZE    7              /* not really an immediate - updated each run */
 193
 194 struct x86_reg aos_get_internal( struct aos_compilation *cp,
 195                                  unsigned imm );
 196 struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp,
 197                                      unsigned imm );
 198
 199
 200 #define ERROR(cp, msg)                                                  \
 201 do {                                                                    \
 202    debug_printf("%s: x86 translation failed: %s\n", __FUNCTION__, msg); \
 203    cp->error = 1;                                                       \
 204    assert(0);                                                           \
 205 } while (0)
 206
 207
 208 #define X86_NULL       0
 209 #define X86_IMMEDIATES 1
 210 #define X86_CONSTANTS  2
 211 #define X86_ATTRIBS    3
 212
 213 struct x86_reg aos_get_x86( struct aos_compilation *cp,
 214                             unsigned value );
 215
 216
 217 typedef void (PIPE_CDECL *vaos_run_elts_func)( struct aos_machine *,
 218                                                const unsigned *elts,
 219                                                unsigned count,
 220                                                void *output_buffer);
 221
 222 typedef void (PIPE_CDECL *vaos_run_linear_func)( struct aos_machine *,
 223                                                 unsigned start,
 224                                                 unsigned count,
 225                                                 void *output_buffer);
 226
 227
 228 struct draw_vs_varient_aos_sse {
 229    struct draw_vs_varient base;
 230    struct draw_context *draw;
 231
 232    struct aos_attrib *attrib;
 233
 234    vaos_run_linear_func gen_run_linear;
 235    vaos_run_elts_func gen_run_elts;
 236
 237
 238    struct x86_function func[2];
 239 };
 240
 241
 242
 243 #endif
 244