src/mesa/tnl/t_vb_arbprogram.h

   1 /*
   2  * Mesa 3-D graphics library
   3  * Version:  6.5
   4  *
   5  * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the "Software"),
   9  * to deal in the Software without restriction, including without limitation
  10  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  11  * and/or sell copies of the Software, and to permit persons to whom the
  12  * Software is furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included
  15  * in all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  21  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  22  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  */
  24
  25 /**
  26  * \file t_arb_program.c
  27  * Compile vertex programs to an intermediate representation.
  28  * Execute vertex programs over a buffer of vertices.
  29  * \author Keith Whitwell, Brian Paul
  30  */
  31
  32
  33 #ifndef _T_VB_ARBPROGRAM_H_
  34 #define _T_VB_ARBPROGRAM_H_
  35
  36
  37 /* New, internal instructions:
  38  */
  39 #define RSW        (MAX_OPCODE)
  40 #define MSK        (MAX_OPCODE+1)
  41 #define REL        (MAX_OPCODE+2)
  42
  43 /**
  44  * Register files for vertex programs
  45  */
  46 #define FILE_REG         0  /* temporaries */
  47 #define FILE_LOCAL_PARAM 1  /* local parameters */
  48 #define FILE_ENV_PARAM   2  /* global parameters */
  49 #define FILE_STATE_PARAM 3  /* GL state references */
  50
  51 #define REG_ARG0   0
  52 #define REG_ARG1   1
  53 #define REG_ARG2   2
  54 #define REG_RES    3
  55 #define REG_ADDR   4
  56 #define REG_TMP0   5
  57 #define REG_TMP11  16
  58 #define REG_OUT0   17
  59 #define REG_OUT14  31
  60 #define REG_IN0    32
  61 #define REG_IN31   63
  62 #define REG_ID     64           /* 0,0,0,1 */
  63 #define REG_ONES   65           /* 1,1,1,1 */
  64 #define REG_SWZ    66           /* -1,1,0,0 */
  65 #define REG_NEG    67           /* -1,-1,-1,-1 */
  66 #define REG_LIT    68           /* 1,0,0,1 */
  67 #define REG_LIT2    69           /* 1,0,0,1 */
  68 #define REG_SCRATCH 70          /* internal temporary. XXX we can't actually use this because 70 doesn't fit in the 5-bit 'dst' instruction field! */
  69 #define REG_UNDEF  127          /* special case - never used */
  70 #define REG_MAX    128
  71 #define REG_INVALID ~0
  72
  73 /* ARB_vp instructions are broken down into one or more of the
  74  * following micro-instructions, each representable in a 64 bit packed
  75  * structure.
  76  */
  77 struct reg {
  78    GLuint file:2;
  79    GLuint idx:7;
  80 };
  81
  82
  83 union instruction {
  84    struct {
  85       GLuint opcode:7;
  86       GLuint dst:5;
  87       GLuint file0:2;
  88       GLuint idx0:7;
  89       GLuint file1:2;
  90       GLuint idx1:7;
  91       GLuint pad:2;
  92       GLuint pad2;
  93    } alu;
  94
  95    struct {
  96       GLuint opcode:7;
  97       GLuint dst:5;
  98       GLuint file0:2;
  99       GLuint idx0:7;
 100       GLuint neg:4;
 101       GLuint swz:8;             /* xyzw only */
 102    } rsw;
 103
 104    struct {
 105       GLuint opcode:7;
 106       GLuint dst:5;
 107       GLuint file:2;
 108       GLuint idx:7;
 109       GLuint mask:4;
 110       GLuint pad:7;
 111       GLuint pad2;
 112    } msk;
 113 };
 114
 115
 116 /**
 117  * Reduced swizzle is a 2-bit field; only X/Y/Z/W are allowed, not 0/1.
 118  */
 119 #define RSW_NOOP ((0<<0) | (1<<2) | (2<<4) | (3<<6))
 120 #define GET_RSW(swz, idx)      (((swz) >> ((idx)*2)) & 0x3)
 121
 122
 123 struct input {
 124    GLuint idx;
 125    GLfloat *data;
 126    GLuint stride;
 127    GLuint size;
 128 };
 129
 130 struct output {
 131    GLuint idx;
 132    GLfloat *data;
 133 };
 134
 135
 136
 137 /*--------------------------------------------------------------------------- */
 138 #if defined(USE_SSE_ASM)
 139 #ifdef NO_FAST_MATH
 140 #define RESTORE_FPU (DEFAULT_X86_FPU)
 141 #define RND_NEG_FPU (DEFAULT_X86_FPU | 0x400)
 142 #else
 143 #define RESTORE_FPU (FAST_X86_FPU)
 144 #define RND_NEG_FPU (FAST_X86_FPU | 0x400)
 145 #endif
 146 #else
 147 #define RESTORE_FPU 0
 148 #define RND_NEG_FPU 0
 149 #endif
 150
 151
 152 /**
 153  * Private storage for the vertex program pipeline stage.
 154  */
 155 struct arb_vp_machine {
 156    GLfloat (*File[4])[4];       /* All values referencable from the program. */
 157
 158    struct input input[_TNL_ATTRIB_MAX];
 159    GLuint nr_inputs;
 160
 161    struct output output[VERT_RESULT_MAX];
 162    GLuint nr_outputs;
 163
 164    GLvector4f attribs[VERT_RESULT_MAX]; /**< result vectors. */
 165    GLvector4f ndcCoords;              /**< normalized device coords */
 166    GLubyte *clipmask;                 /**< clip flags */
 167    GLubyte ormask, andmask;           /**< for clipping */
 168
 169    GLuint vtx_nr;               /**< loop counter */
 170
 171    struct vertex_buffer *VB;
 172
 173    GLshort fpucntl_rnd_neg;     /* constant value */
 174    GLshort fpucntl_restore;     /* constant value */
 175
 176    GLboolean try_codegen;
 177 };
 178
 179 struct tnl_compiled_program {
 180    union instruction instructions[1024];
 181    GLint nr_instructions;
 182    void (*compiled_func)( struct arb_vp_machine * ); /**< codegen'd program */
 183 };
 184
 185 void _tnl_program_string_change( struct vertex_program * );
 186 void _tnl_program_destroy( struct vertex_program * );
 187
 188 void _tnl_disassem_vba_insn( union instruction op );
 189
 190 GLboolean _tnl_sse_codegen_vertex_program(struct tnl_compiled_program *p);
 191
 192 #endif