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 struct shine_tab {
  68    float exponent;
  69    float values[258];
  70    unsigned last_used;
  71 };
  72
  73 struct lit_info {
  74    lit_func func;
  75    struct shine_tab *shine_tab;
  76 };
  77
  78 #define MAX_SHINE_TAB    4
  79 #define MAX_LIT_INFO     16
  80
  81 /* This is the temporary storage used by all the aos_sse vs varients.
  82  * Create one per context and reuse by passing a pointer in at
  83  * vs_varient creation??
  84  */
  85 struct aos_machine {
  86    float input    [MAX_INPUTS    ][4];
  87    float output   [MAX_OUTPUTS   ][4];
  88    float temp     [MAX_TEMPS     ][4];
  89    float constant [MAX_CONSTANTS ][4]; /* fixme -- should just be a pointer */
  90    float immediate[MAX_IMMEDIATES][4]; /* fixme -- should just be a pointer */
  91    float internal [MAX_INTERNALS ][4];
  92
  93    float scale[4];              /* viewport */
  94    float translate[4];          /* viewport */
  95
  96    float tmp[2][4];             /* scratch space for LIT */
  97
  98    struct shine_tab shine_tab[MAX_SHINE_TAB];
  99    struct lit_info  lit_info[MAX_LIT_INFO];
 100    unsigned now;
 101
 102
 103    ushort fpu_rnd_nearest;
 104    ushort fpu_rnd_neg_inf;
 105    ushort fpu_restore;
 106    ushort fpucntl;              /* one of FPU_* above */
 107
 108    struct {
 109       const void *input_ptr;
 110       unsigned input_stride;
 111
 112       unsigned output_offset;
 113    } attrib[PIPE_MAX_ATTRIBS];
 114 };
 115
 116
 117
 118
 119 struct aos_compilation {
 120    struct x86_function *func;
 121    struct draw_vs_varient_aos_sse *vaos;
 122
 123    unsigned insn_counter;
 124    unsigned num_immediates;
 125    unsigned count;
 126    unsigned lit_count;
 127
 128    struct {
 129       unsigned idx:16;
 130       unsigned file:8;
 131       unsigned dirty:8;
 132       unsigned last_used;
 133    } xmm[8];
 134
 135
 136    boolean input_fetched[PIPE_MAX_ATTRIBS];
 137    unsigned output_last_write[PIPE_MAX_ATTRIBS];
 138
 139    boolean have_sse2;
 140    boolean error;
 141    short fpucntl;
 142
 143    /* these are actually known values, but putting them in a struct
 144     * like this is helpful to keep them in sync across the file.
 145     */
 146    struct x86_reg tmp_EAX;
 147    struct x86_reg idx_EBX;     /* either start+i or &elt[i] */
 148    struct x86_reg outbuf_ECX;
 149    struct x86_reg machine_EDX;
 150    struct x86_reg count_ESI;    /* decrements to zero */
 151 };
 152
 153 struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp );
 154 void aos_release_xmm_reg( struct aos_compilation *cp, unsigned idx );
 155
 156 void aos_adopt_xmm_reg( struct aos_compilation *cp,
 157                         struct x86_reg reg,
 158                         unsigned file,
 159                         unsigned idx,
 160                         unsigned dirty );
 161
 162 struct x86_reg aos_get_shader_reg( struct aos_compilation *cp,
 163                                    unsigned file,
 164                                    unsigned idx );
 165
 166 boolean aos_fetch_inputs( struct aos_compilation *cp,
 167                           boolean linear );
 168
 169 boolean aos_emit_outputs( struct aos_compilation *cp );
 170
 171
 172 #define IMM_ONES     0              /* 1, 1,1,1 */
 173 #define IMM_SWZ      1              /* 1,-1,0, 0xffffffff */
 174 #define IMM_IDENTITY 2              /* 0, 0,0,1 */
 175 #define IMM_INV_255  3              /* 1/255, 1/255, 1/255, 1/255 */
 176 #define IMM_255      4              /* 255, 255, 255, 255 */
 177 #define IMM_NEGS     5              /* -1,-1,-1,-1 */
 178 #define IMM_RSQ      6              /* -.5,1.5,_,_ */
 179 #define IMM_PSIZE    7              /* not really an immediate - updated each run */
 180
 181 struct x86_reg aos_get_internal( struct aos_compilation *cp,
 182                                  unsigned imm );
 183 struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp,
 184                                      unsigned imm );
 185
 186
 187 #define ERROR(cp, msg)                                                  \
 188 do {                                                                    \
 189    debug_printf("%s: x86 translation failed: %s\n", __FUNCTION__, msg); \
 190    cp->error = 1;                                                       \
 191    assert(0);                                                           \
 192 } while (0)
 193
 194
 195
 196
 197
 198
 199 struct draw_vs_varient_aos_sse {
 200    struct draw_vs_varient base;
 201    struct draw_context *draw;
 202
 203 #if 0
 204    struct {
 205       const void *ptr;
 206       unsigned stride;
 207    } attrib[PIPE_MAX_ATTRIBS];
 208 #endif
 209
 210    struct aos_machine *machine; /* XXX: temporarily unshared */
 211
 212    vsv_run_linear_func gen_run_linear;
 213    vsv_run_elts_func gen_run_elts;
 214
 215
 216    struct x86_function func[2];
 217 };
 218
 219
 220
 221 #endif
 222