| 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 nvvertexec.c |
| 27 |
* Code to execute vertex programs. |
| 28 |
* \author Brian Paul |
| 29 |
*/ |
| 30 |
|
| 31 |
#include "glheader.h" |
| 32 |
#include "context.h" |
| 33 |
#include "imports.h" |
| 34 |
#include "macros.h" |
| 35 |
#include "mtypes.h" |
| 36 |
#include "nvvertexec.h" |
| 37 |
#include "nvvertprog.h" |
| 38 |
#include "program.h" |
| 39 |
#include "math/m_matrix.h" |
| 40 |
|
| 41 |
|
| 42 |
static const GLfloat ZeroVec[4] = { 0.0F, 0.0F, 0.0F, 0.0F }; |
| 43 |
|
| 44 |
|
| 45 |
/** |
| 46 |
* Load/initialize the vertex program registers which need to be set |
| 47 |
* per-vertex. |
| 48 |
*/ |
| 49 |
void |
| 50 |
_mesa_init_vp_per_vertex_registers(GLcontext *ctx) |
| 51 |
{ |
| 52 |
/* Input registers get initialized from the current vertex attribs */ |
| 53 |
MEMCPY(ctx->VertexProgram.Inputs, ctx->Current.Attrib, |
| 54 |
VERT_ATTRIB_MAX * 4 * sizeof(GLfloat)); |
| 55 |
|
| 56 |
if (ctx->VertexProgram.Current->IsNVProgram) { |
| 57 |
GLuint i; |
| 58 |
/* Output/result regs are initialized to [0,0,0,1] */ |
| 59 |
for (i = 0; i < MAX_NV_VERTEX_PROGRAM_OUTPUTS; i++) { |
| 60 |
ASSIGN_4V(ctx->VertexProgram.Outputs[i], 0.0F, 0.0F, 0.0F, 1.0F); |
| 61 |
} |
| 62 |
/* Temp regs are initialized to [0,0,0,0] */ |
| 63 |
for (i = 0; i < MAX_NV_VERTEX_PROGRAM_TEMPS; i++) { |
| 64 |
ASSIGN_4V(ctx->VertexProgram.Temporaries[i], 0.0F, 0.0F, 0.0F, 0.0F); |
| 65 |
} |
| 66 |
ASSIGN_4V(ctx->VertexProgram.AddressReg, 0, 0, 0, 0); |
| 67 |
} |
| 68 |
} |
| 69 |
|
| 70 |
|
| 71 |
|
| 72 |
/** |
| 73 |
* Copy the 16 elements of a matrix into four consecutive program |
| 74 |
* registers starting at 'pos'. |
| 75 |
*/ |
| 76 |
static void |
| 77 |
load_matrix(GLfloat registers[][4], GLuint pos, const GLfloat mat[16]) |
| 78 |
{ |
| 79 |
GLuint i; |
| 80 |
for (i = 0; i < 4; i++) { |
| 81 |
registers[pos + i][0] = mat[0 + i]; |
| 82 |
registers[pos + i][1] = mat[4 + i]; |
| 83 |
registers[pos + i][2] = mat[8 + i]; |
| 84 |
registers[pos + i][3] = mat[12 + i]; |
| 85 |
} |
| 86 |
} |
| 87 |
|
| 88 |
|
| 89 |
/** |
| 90 |
* As above, but transpose the matrix. |
| 91 |
*/ |
| 92 |
static void |
| 93 |
load_transpose_matrix(GLfloat registers[][4], GLuint pos, |
| 94 |
const GLfloat mat[16]) |
| 95 |
{ |
| 96 |
MEMCPY(registers[pos], mat, 16 * sizeof(GLfloat)); |
| 97 |
} |
| 98 |
|
| 99 |
|
| 100 |
/** |
| 101 |
* Load program parameter registers with tracked matrices (if NV program) |
| 102 |
* or GL state values (if ARB program). |
| 103 |
* This needs to be done per glBegin/glEnd, not per-vertex. |
| 104 |
*/ |
| 105 |
void |
| 106 |
_mesa_init_vp_per_primitive_registers(GLcontext *ctx) |
| 107 |
{ |
| 108 |
if (ctx->VertexProgram.Current->IsNVProgram) { |
| 109 |
GLuint i; |
| 110 |
|
| 111 |
for (i = 0; i < MAX_NV_VERTEX_PROGRAM_PARAMS / 4; i++) { |
| 112 |
/* point 'mat' at source matrix */ |
| 113 |
GLmatrix *mat; |
| 114 |
if (ctx->VertexProgram.TrackMatrix[i] == GL_MODELVIEW) { |
| 115 |
mat = ctx->ModelviewMatrixStack.Top; |
| 116 |
} |
| 117 |
else if (ctx->VertexProgram.TrackMatrix[i] == GL_PROJECTION) { |
| 118 |
mat = ctx->ProjectionMatrixStack.Top; |
| 119 |
} |
| 120 |
else if (ctx->VertexProgram.TrackMatrix[i] == GL_TEXTURE) { |
| 121 |
mat = ctx->TextureMatrixStack[ctx->Texture.CurrentUnit].Top; |
| 122 |
} |
| 123 |
else if (ctx->VertexProgram.TrackMatrix[i] == GL_COLOR) { |
| 124 |
mat = ctx->ColorMatrixStack.Top; |
| 125 |
} |
| 126 |
else if (ctx->VertexProgram.TrackMatrix[i]==GL_MODELVIEW_PROJECTION_NV) { |
| 127 |
/* XXX verify the combined matrix is up to date */ |
| 128 |
mat = &ctx->_ModelProjectMatrix; |
| 129 |
} |
| 130 |
else if (ctx->VertexProgram.TrackMatrix[i] >= GL_MATRIX0_NV && |
| 131 |
ctx->VertexProgram.TrackMatrix[i] <= GL_MATRIX7_NV) { |
| 132 |
GLuint n = ctx->VertexProgram.TrackMatrix[i] - GL_MATRIX0_NV; |
| 133 |
ASSERT(n < MAX_PROGRAM_MATRICES); |
| 134 |
mat = ctx->ProgramMatrixStack[n].Top; |
| 135 |
} |
| 136 |
else { |
| 137 |
/* no matrix is tracked, but we leave the register values as-is */ |
| 138 |
assert(ctx->VertexProgram.TrackMatrix[i] == GL_NONE); |
| 139 |
continue; |
| 140 |
} |
| 141 |
|
| 142 |
/* load the matrix */ |
| 143 |
if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_IDENTITY_NV) { |
| 144 |
load_matrix(ctx->VertexProgram.Parameters, i*4, mat->m); |
| 145 |
} |
| 146 |
else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_INVERSE_NV) { |
| 147 |
_math_matrix_analyse(mat); /* update the inverse */ |
| 148 |
ASSERT(!_math_matrix_is_dirty(mat)); |
| 149 |
load_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv); |
| 150 |
} |
| 151 |
else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_TRANSPOSE_NV) { |
| 152 |
load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->m); |
| 153 |
} |
| 154 |
else { |
| 155 |
assert(ctx->VertexProgram.TrackMatrixTransform[i] |
| 156 |
== GL_INVERSE_TRANSPOSE_NV); |
| 157 |
_math_matrix_analyse(mat); /* update the inverse */ |
| 158 |
ASSERT(!_math_matrix_is_dirty(mat)); |
| 159 |
load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv); |
| 160 |
} |
| 161 |
} |
| 162 |
} |
| 163 |
else { |
| 164 |
/* Using and ARB vertex program */ |
| 165 |
if (ctx->VertexProgram.Current->Parameters) { |
| 166 |
/* Grab the state GL state and put into registers */ |
| 167 |
_mesa_load_state_parameters(ctx, |
| 168 |
ctx->VertexProgram.Current->Parameters); |
| 169 |
} |
| 170 |
} |
| 171 |
} |
| 172 |
|
| 173 |
|
| 174 |
|
| 175 |
/** |
| 176 |
* For debugging. Dump the current vertex program machine registers. |
| 177 |
*/ |
| 178 |
void |
| 179 |
_mesa_dump_vp_state( const struct gl_vertex_program_state *state ) |
| 180 |
{ |
| 181 |
int i; |
| 182 |
_mesa_printf("VertexIn:\n"); |
| 183 |
for (i = 0; i < MAX_NV_VERTEX_PROGRAM_INPUTS; i++) { |
| 184 |
_mesa_printf("%d: %f %f %f %f ", i, |
| 185 |
state->Inputs[i][0], |
| 186 |
state->Inputs[i][1], |
| 187 |
state->Inputs[i][2], |
| 188 |
state->Inputs[i][3]); |
| 189 |
} |
| 190 |
_mesa_printf("\n"); |
| 191 |
|
| 192 |
_mesa_printf("VertexOut:\n"); |
| 193 |
for (i = 0; i < MAX_NV_VERTEX_PROGRAM_OUTPUTS; i++) { |
| 194 |
_mesa_printf("%d: %f %f %f %f ", i, |
| 195 |
state->Outputs[i][0], |
| 196 |
state->Outputs[i][1], |
| 197 |
state->Outputs[i][2], |
| 198 |
state->Outputs[i][3]); |
| 199 |
} |
| 200 |
_mesa_printf("\n"); |
| 201 |
|
| 202 |
_mesa_printf("Registers:\n"); |
| 203 |
for (i = 0; i < MAX_NV_VERTEX_PROGRAM_TEMPS; i++) { |
| 204 |
_mesa_printf("%d: %f %f %f %f ", i, |
| 205 |
state->Temporaries[i][0], |
| 206 |
state->Temporaries[i][1], |
| 207 |
state->Temporaries[i][2], |
| 208 |
state->Temporaries[i][3]); |
| 209 |
} |
| 210 |
_mesa_printf("\n"); |
| 211 |
|
| 212 |
_mesa_printf("Parameters:\n"); |
| 213 |
for (i = 0; i < MAX_NV_VERTEX_PROGRAM_PARAMS; i++) { |
| 214 |
_mesa_printf("%d: %f %f %f %f ", i, |
| 215 |
state->Parameters[i][0], |
| 216 |
state->Parameters[i][1], |
| 217 |
state->Parameters[i][2], |
| 218 |
state->Parameters[i][3]); |
| 219 |
} |
| 220 |
_mesa_printf("\n"); |
| 221 |
} |
| 222 |
|
| 223 |
|
| 224 |
|
| 225 |
/** |
| 226 |
* Return a pointer to the 4-element float vector specified by the given |
| 227 |
* source register. |
| 228 |
*/ |
| 229 |
static INLINE const GLfloat * |
| 230 |
get_register_pointer( const struct vp_src_register *source, |
| 231 |
const struct gl_vertex_program_state *state ) |
| 232 |
{ |
| 233 |
if (source->RelAddr) { |
| 234 |
const GLint reg = source->Index + state->AddressReg[0]; |
| 235 |
ASSERT( (source->File == PROGRAM_ENV_PARAM) || |
| 236 |
(source->File == PROGRAM_STATE_VAR) ); |
| 237 |
if (reg < 0 || reg > MAX_NV_VERTEX_PROGRAM_PARAMS) |
| 238 |
return ZeroVec; |
| 239 |
else if (source->File == PROGRAM_ENV_PARAM) |
| 240 |
return state->Parameters[reg]; |
| 241 |
else |
| 242 |
return state->Current->Parameters->ParameterValues[reg]; |
| 243 |
} |
| 244 |
else { |
| 245 |
switch (source->File) { |
| 246 |
case PROGRAM_TEMPORARY: |
| 247 |
ASSERT(source->Index < MAX_NV_VERTEX_PROGRAM_TEMPS); |
| 248 |
return state->Temporaries[source->Index]; |
| 249 |
case PROGRAM_INPUT: |
| 250 |
ASSERT(source->Index < MAX_NV_VERTEX_PROGRAM_INPUTS); |
| 251 |
return state->Inputs[source->Index]; |
| 252 |
case PROGRAM_OUTPUT: |
| 253 |
/* This is only needed for the PRINT instruction */ |
| 254 |
ASSERT(source->Index < MAX_NV_VERTEX_PROGRAM_OUTPUTS); |
| 255 |
return state->Outputs[source->Index]; |
| 256 |
case PROGRAM_LOCAL_PARAM: |
| 257 |
ASSERT(source->Index < MAX_PROGRAM_LOCAL_PARAMS); |
| 258 |
return state->Current->Base.LocalParams[source->Index]; |
| 259 |
case PROGRAM_ENV_PARAM: |
| 260 |
ASSERT(source->Index < MAX_NV_VERTEX_PROGRAM_PARAMS); |
| 261 |
return state->Parameters[source->Index]; |
| 262 |
case PROGRAM_STATE_VAR: |
| 263 |
ASSERT(source->Index < state->Current->Parameters->NumParameters); |
| 264 |
return state->Current->Parameters->ParameterValues[source->Index]; |
| 265 |
default: |
| 266 |
_mesa_problem(NULL, |
| 267 |
"Bad source register file in get_register_pointer"); |
| 268 |
return NULL; |
| 269 |
} |
| 270 |
} |
| 271 |
return NULL; |
| 272 |
} |
| 273 |
|
| 274 |
|
| 275 |
/** |
| 276 |
* Fetch a 4-element float vector from the given source register. |
| 277 |
* Apply swizzling and negating as needed. |
| 278 |
*/ |
| 279 |
static INLINE void |
| 280 |
fetch_vector4( const struct vp_src_register *source, |
| 281 |
const struct gl_vertex_program_state *state, |
| 282 |
GLfloat result[4] ) |
| 283 |
{ |
| 284 |
const GLfloat *src = get_register_pointer(source, state); |
| 285 |
|
| 286 |
if (source->Negate) { |
| 287 |
result[0] = -src[GET_SWZ(source->Swizzle, 0)]; |
| 288 |
result[1] = -src[GET_SWZ(source->Swizzle, 1)]; |
| 289 |
result[2] = -src[GET_SWZ(source->Swizzle, 2)]; |
| 290 |
result[3] = -src[GET_SWZ(source->Swizzle, 3)]; |
| 291 |
} |
| 292 |
else { |
| 293 |
result[0] = src[GET_SWZ(source->Swizzle, 0)]; |
| 294 |
result[1] = src[GET_SWZ(source->Swizzle, 1)]; |
| 295 |
result[2] = src[GET_SWZ(source->Swizzle, 2)]; |
| 296 |
result[3] = src[GET_SWZ(source->Swizzle, 3)]; |
| 297 |
} |
| 298 |
} |
| 299 |
|
| 300 |
|
| 301 |
|
| 302 |
/** |
| 303 |
* As above, but only return result[0] element. |
| 304 |
*/ |
| 305 |
static INLINE void |
| 306 |
fetch_vector1( const struct vp_src_register *source, |
| 307 |
const struct gl_vertex_program_state *state, |
| 308 |
GLfloat result[4] ) |
| 309 |
{ |
| 310 |
const GLfloat *src = get_register_pointer(source, state); |
| 311 |
|
| 312 |
if (source->Negate) { |
| 313 |
result[0] = -src[GET_SWZ(source->Swizzle, 0)]; |
| 314 |
} |
| 315 |
else { |
| 316 |
result[0] = src[GET_SWZ(source->Swizzle, 0)]; |
| 317 |
} |
| 318 |
} |
| 319 |
|
| 320 |
|
| 321 |
/** |
| 322 |
* Store 4 floats into a register. |
| 323 |
*/ |
| 324 |
static void |
| 325 |
store_vector4( const struct vp_dst_register *dest, |
| 326 |
struct gl_vertex_program_state *state, |
| 327 |
const GLfloat value[4] ) |
| 328 |
{ |
| 329 |
GLfloat *dst; |
| 330 |
switch (dest->File) { |
| 331 |
case PROGRAM_TEMPORARY: |
| 332 |
dst = state->Temporaries[dest->Index]; |
| 333 |
break; |
| 334 |
case PROGRAM_OUTPUT: |
| 335 |
dst = state->Outputs[dest->Index]; |
| 336 |
break; |
| 337 |
case PROGRAM_ENV_PARAM: |
| 338 |
{ |
| 339 |
/* a slight hack */ |
| 340 |
GET_CURRENT_CONTEXT(ctx); |
| 341 |
dst = ctx->VertexProgram.Parameters[dest->Index]; |
| 342 |
} |
| 343 |
break; |
| 344 |
default: |
| 345 |
_mesa_problem(NULL, "Invalid register file in store_vector4(file=%d)", |
| 346 |
dest->File); |
| 347 |
return; |
| 348 |
} |
| 349 |
|
| 350 |
if (dest->WriteMask & WRITEMASK_X) |
| 351 |
dst[0] = value[0]; |
| 352 |
if (dest->WriteMask & WRITEMASK_Y) |
| 353 |
dst[1] = value[1]; |
| 354 |
if (dest->WriteMask & WRITEMASK_Z) |
| 355 |
dst[2] = value[2]; |
| 356 |
if (dest->WriteMask & WRITEMASK_W) |
| 357 |
dst[3] = value[3]; |
| 358 |
} |
| 359 |
|
| 360 |
|
| 361 |
/** |
| 362 |
* Set x to positive or negative infinity. |
| 363 |
*/ |
| 364 |
#if defined(USE_IEEE) || defined(_WIN32) |
| 365 |
#define SET_POS_INFINITY(x) ( *((GLuint *) (void *)&x) = 0x7F800000 ) |
| 366 |
#define SET_NEG_INFINITY(x) ( *((GLuint *) (void *)&x) = 0xFF800000 ) |
| 367 |
#elif defined(VMS) |
| 368 |
#define SET_POS_INFINITY(x) x = __MAXFLOAT |
| 369 |
#define SET_NEG_INFINITY(x) x = -__MAXFLOAT |
| 370 |
#else |
| 371 |
#define SET_POS_INFINITY(x) x = (GLfloat) HUGE_VAL |
| 372 |
#define SET_NEG_INFINITY(x) x = (GLfloat) -HUGE_VAL |
| 373 |
#endif |
| 374 |
|
| 375 |
#define SET_FLOAT_BITS(x, bits) ((fi_type *) (void *) &(x))->i = bits |
| 376 |
|
| 377 |
|
| 378 |
/** |
| 379 |
* Execute the given vertex program |
| 380 |
*/ |
| 381 |
void |
| 382 |
_mesa_exec_vertex_program(GLcontext *ctx, const struct vertex_program *program) |
| 383 |
{ |
| 384 |
struct gl_vertex_program_state *state = &ctx->VertexProgram; |
| 385 |
const struct vp_instruction *inst; |
| 386 |
|
| 387 |
ctx->_CurrentProgram = GL_VERTEX_PROGRAM_ARB; /* or NV, doesn't matter */ |
| 388 |
|
| 389 |
/* If the program is position invariant, multiply the input |
| 390 |
* position and the MVP matrix and stick it into the output pos slot |
| 391 |
*/ |
| 392 |
if (ctx->VertexProgram.Current->IsPositionInvariant) { |
| 393 |
TRANSFORM_POINT( ctx->VertexProgram.Outputs[0], |
| 394 |
ctx->_ModelProjectMatrix.m, |
| 395 |
ctx->VertexProgram.Inputs[0]); |
| 396 |
|
| 397 |
/* XXX: This could go elsewhere */ |
| 398 |
ctx->VertexProgram.Current->OutputsWritten |= 0x1; |
| 399 |
} |
| 400 |
for (inst = program->Instructions; ; inst++) { |
| 401 |
|
| 402 |
if (ctx->VertexProgram.CallbackEnabled && |
| 403 |
ctx->VertexProgram.Callback) { |
| 404 |
ctx->VertexProgram.CurrentPosition = inst->StringPos; |
| 405 |
ctx->VertexProgram.Callback(program->Base.Target, |
| 406 |
ctx->VertexProgram.CallbackData); |
| 407 |
} |
| 408 |
|
| 409 |
switch (inst->Opcode) { |
| 410 |
case VP_OPCODE_MOV: |
| 411 |
{ |
| 412 |
GLfloat t[4]; |
| 413 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 414 |
store_vector4( &inst->DstReg, state, t ); |
| 415 |
} |
| 416 |
break; |
| 417 |
case VP_OPCODE_LIT: |
| 418 |
{ |
| 419 |
const GLfloat epsilon = 1.0F / 256.0F; /* per NV spec */ |
| 420 |
GLfloat t[4], lit[4]; |
| 421 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 422 |
t[0] = MAX2(t[0], 0.0F); |
| 423 |
t[1] = MAX2(t[1], 0.0F); |
| 424 |
t[3] = CLAMP(t[3], -(128.0F - epsilon), (128.0F - epsilon)); |
| 425 |
lit[0] = 1.0; |
| 426 |
lit[1] = t[0]; |
| 427 |
lit[2] = (t[0] > 0.0) ? (GLfloat) _mesa_pow(t[1], t[3]) : 0.0F; |
| 428 |
lit[3] = 1.0; |
| 429 |
store_vector4( &inst->DstReg, state, lit ); |
| 430 |
} |
| 431 |
break; |
| 432 |
case VP_OPCODE_RCP: |
| 433 |
{ |
| 434 |
GLfloat t[4]; |
| 435 |
fetch_vector1( &inst->SrcReg[0], state, t ); |
| 436 |
if (t[0] != 1.0F) |
| 437 |
t[0] = 1.0F / t[0]; /* div by zero is infinity! */ |
| 438 |
t[1] = t[2] = t[3] = t[0]; |
| 439 |
store_vector4( &inst->DstReg, state, t ); |
| 440 |
} |
| 441 |
break; |
| 442 |
case VP_OPCODE_RSQ: |
| 443 |
{ |
| 444 |
GLfloat t[4]; |
| 445 |
fetch_vector1( &inst->SrcReg[0], state, t ); |
| 446 |
t[0] = INV_SQRTF(FABSF(t[0])); |
| 447 |
t[1] = t[2] = t[3] = t[0]; |
| 448 |
store_vector4( &inst->DstReg, state, t ); |
| 449 |
} |
| 450 |
break; |
| 451 |
case VP_OPCODE_EXP: |
| 452 |
{ |
| 453 |
GLfloat t[4], q[4], floor_t0; |
| 454 |
fetch_vector1( &inst->SrcReg[0], state, t ); |
| 455 |
floor_t0 = (float) floor(t[0]); |
| 456 |
if (floor_t0 > FLT_MAX_EXP) { |
| 457 |
SET_POS_INFINITY(q[0]); |
| 458 |
SET_POS_INFINITY(q[2]); |
| 459 |
} |
| 460 |
else if (floor_t0 < FLT_MIN_EXP) { |
| 461 |
q[0] = 0.0F; |
| 462 |
q[2] = 0.0F; |
| 463 |
} |
| 464 |
else { |
| 465 |
#ifdef USE_IEEE |
| 466 |
GLint ii = (GLint) floor_t0; |
| 467 |
ii = (ii < 23) + 0x3f800000; |
| 468 |
SET_FLOAT_BITS(q[0], ii); |
| 469 |
q[0] = *((GLfloat *) (void *)&ii); |
| 470 |
#else |
| 471 |
q[0] = (GLfloat) pow(2.0, floor_t0); |
| 472 |
#endif |
| 473 |
q[2] = (GLfloat) (q[0] * LOG2(q[1])); |
| 474 |
} |
| 475 |
q[1] = t[0] - floor_t0; |
| 476 |
q[3] = 1.0F; |
| 477 |
store_vector4( &inst->DstReg, state, q ); |
| 478 |
} |
| 479 |
break; |
| 480 |
case VP_OPCODE_LOG: |
| 481 |
{ |
| 482 |
GLfloat t[4], q[4], abs_t0; |
| 483 |
fetch_vector1( &inst->SrcReg[0], state, t ); |
| 484 |
abs_t0 = (GLfloat) fabs(t[0]); |
| 485 |
if (abs_t0 != 0.0F) { |
| 486 |
/* Since we really can't handle infinite values on VMS |
| 487 |
* like other OSes we'll use __MAXFLOAT to represent |
| 488 |
* infinity. This may need some tweaking. |
| 489 |
*/ |
| 490 |
#ifdef VMS |
| 491 |
if (abs_t0 == __MAXFLOAT) |
| 492 |
#else |
| 493 |
if (IS_INF_OR_NAN(abs_t0)) |
| 494 |
#endif |
| 495 |
{ |
| 496 |
SET_POS_INFINITY(q[0]); |
| 497 |
q[1] = 1.0F; |
| 498 |
SET_POS_INFINITY(q[2]); |
| 499 |
} |
| 500 |
else { |
| 501 |
int exponent; |
| 502 |
double mantissa = frexp(t[0], &exponent); |
| 503 |
q[0] = (GLfloat) (exponent - 1); |
| 504 |
q[1] = (GLfloat) (2.0 * mantissa); /* map [.5, 1) -> [1, 2) */ |
| 505 |
q[2] = (GLfloat) (q[0] + LOG2(q[1])); |
| 506 |
} |
| 507 |
} |
| 508 |
else { |
| 509 |
SET_NEG_INFINITY(q[0]); |
| 510 |
q[1] = 1.0F; |
| 511 |
SET_NEG_INFINITY(q[2]); |
| 512 |
} |
| 513 |
q[3] = 1.0; |
| 514 |
store_vector4( &inst->DstReg, state, q ); |
| 515 |
} |
| 516 |
break; |
| 517 |
case VP_OPCODE_MUL: |
| 518 |
{ |
| 519 |
GLfloat t[4], u[4], prod[4]; |
| 520 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 521 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 522 |
prod[0] = t[0] * u[0]; |
| 523 |
prod[1] = t[1] * u[1]; |
| 524 |
prod[2] = t[2] * u[2]; |
| 525 |
prod[3] = t[3] * u[3]; |
| 526 |
store_vector4( &inst->DstReg, state, prod ); |
| 527 |
} |
| 528 |
break; |
| 529 |
case VP_OPCODE_ADD: |
| 530 |
{ |
| 531 |
GLfloat t[4], u[4], sum[4]; |
| 532 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 533 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 534 |
sum[0] = t[0] + u[0]; |
| 535 |
sum[1] = t[1] + u[1]; |
| 536 |
sum[2] = t[2] + u[2]; |
| 537 |
sum[3] = t[3] + u[3]; |
| 538 |
store_vector4( &inst->DstReg, state, sum ); |
| 539 |
} |
| 540 |
break; |
| 541 |
case VP_OPCODE_DP3: |
| 542 |
{ |
| 543 |
GLfloat t[4], u[4], dot[4]; |
| 544 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 545 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 546 |
dot[0] = t[0] * u[0] + t[1] * u[1] + t[2] * u[2]; |
| 547 |
dot[1] = dot[2] = dot[3] = dot[0]; |
| 548 |
store_vector4( &inst->DstReg, state, dot ); |
| 549 |
} |
| 550 |
break; |
| 551 |
case VP_OPCODE_DP4: |
| 552 |
{ |
| 553 |
GLfloat t[4], u[4], dot[4]; |
| 554 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 555 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 556 |
dot[0] = t[0] * u[0] + t[1] * u[1] + t[2] * u[2] + t[3] * u[3]; |
| 557 |
dot[1] = dot[2] = dot[3] = dot[0]; |
| 558 |
store_vector4( &inst->DstReg, state, dot ); |
| 559 |
} |
| 560 |
break; |
| 561 |
case VP_OPCODE_DST: |
| 562 |
{ |
| 563 |
GLfloat t[4], u[4], dst[4]; |
| 564 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 565 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 566 |
dst[0] = 1.0F; |
| 567 |
dst[1] = t[1] * u[1]; |
| 568 |
dst[2] = t[2]; |
| 569 |
dst[3] = u[3]; |
| 570 |
store_vector4( &inst->DstReg, state, dst ); |
| 571 |
} |
| 572 |
break; |
| 573 |
case VP_OPCODE_MIN: |
| 574 |
{ |
| 575 |
GLfloat t[4], u[4], min[4]; |
| 576 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 577 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 578 |
min[0] = (t[0] < u[0]) ? t[0] : u[0]; |
| 579 |
min[1] = (t[1] < u[1]) ? t[1] : u[1]; |
| 580 |
min[2] = (t[2] < u[2]) ? t[2] : u[2]; |
| 581 |
min[3] = (t[3] < u[3]) ? t[3] : u[3]; |
| 582 |
store_vector4( &inst->DstReg, state, min ); |
| 583 |
} |
| 584 |
break; |
| 585 |
case VP_OPCODE_MAX: |
| 586 |
{ |
| 587 |
GLfloat t[4], u[4], max[4]; |
| 588 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 589 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 590 |
max[0] = (t[0] > u[0]) ? t[0] : u[0]; |
| 591 |
max[1] = (t[1] > u[1]) ? t[1] : u[1]; |
| 592 |
max[2] = (t[2] > u[2]) ? t[2] : u[2]; |
| 593 |
max[3] = (t[3] > u[3]) ? t[3] : u[3]; |
| 594 |
store_vector4( &inst->DstReg, state, max ); |
| 595 |
} |
| 596 |
break; |
| 597 |
case VP_OPCODE_SLT: |
| 598 |
{ |
| 599 |
GLfloat t[4], u[4], slt[4]; |
| 600 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 601 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 602 |
slt[0] = (t[0] < u[0]) ? 1.0F : 0.0F; |
| 603 |
slt[1] = (t[1] < u[1]) ? 1.0F : 0.0F; |
| 604 |
slt[2] = (t[2] < u[2]) ? 1.0F : 0.0F; |
| 605 |
slt[3] = (t[3] < u[3]) ? 1.0F : 0.0F; |
| 606 |
store_vector4( &inst->DstReg, state, slt ); |
| 607 |
} |
| 608 |
break; |
| 609 |
case VP_OPCODE_SGE: |
| 610 |
{ |
| 611 |
GLfloat t[4], u[4], sge[4]; |
| 612 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 613 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 614 |
sge[0] = (t[0] >= u[0]) ? 1.0F : 0.0F; |
| 615 |
sge[1] = (t[1] >= u[1]) ? 1.0F : 0.0F; |
| 616 |
sge[2] = (t[2] >= u[2]) ? 1.0F : 0.0F; |
| 617 |
sge[3] = (t[3] >= u[3]) ? 1.0F : 0.0F; |
| 618 |
store_vector4( &inst->DstReg, state, sge ); |
| 619 |
} |
| 620 |
break; |
| 621 |
case VP_OPCODE_MAD: |
| 622 |
{ |
| 623 |
GLfloat t[4], u[4], v[4], sum[4]; |
| 624 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 625 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 626 |
fetch_vector4( &inst->SrcReg[2], state, v ); |
| 627 |
sum[0] = t[0] * u[0] + v[0]; |
| 628 |
sum[1] = t[1] * u[1] + v[1]; |
| 629 |
sum[2] = t[2] * u[2] + v[2]; |
| 630 |
sum[3] = t[3] * u[3] + v[3]; |
| 631 |
store_vector4( &inst->DstReg, state, sum ); |
| 632 |
} |
| 633 |
break; |
| 634 |
case VP_OPCODE_ARL: |
| 635 |
{ |
| 636 |
GLfloat t[4]; |
| 637 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 638 |
state->AddressReg[0] = (GLint) floor(t[0]); |
| 639 |
} |
| 640 |
break; |
| 641 |
case VP_OPCODE_DPH: |
| 642 |
{ |
| 643 |
GLfloat t[4], u[4], dot[4]; |
| 644 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 645 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 646 |
dot[0] = t[0] * u[0] + t[1] * u[1] + t[2] * u[2] + u[3]; |
| 647 |
dot[1] = dot[2] = dot[3] = dot[0]; |
| 648 |
store_vector4( &inst->DstReg, state, dot ); |
| 649 |
} |
| 650 |
break; |
| 651 |
case VP_OPCODE_RCC: |
| 652 |
{ |
| 653 |
GLfloat t[4], u; |
| 654 |
fetch_vector1( &inst->SrcReg[0], state, t ); |
| 655 |
if (t[0] == 1.0F) |
| 656 |
u = 1.0F; |
| 657 |
else |
| 658 |
u = 1.0F / t[0]; |
| 659 |
if (u > 0.0F) { |
| 660 |
if (u > 1.884467e+019F) { |
| 661 |
u = 1.884467e+019F; /* IEEE 32-bit binary value 0x5F800000 */ |
| 662 |
} |
| 663 |
else if (u < 5.42101e-020F) { |
| 664 |
u = 5.42101e-020F; /* IEEE 32-bit binary value 0x1F800000 */ |
| 665 |
} |
| 666 |
} |
| 667 |
else { |
| 668 |
if (u < -1.884467e+019F) { |
| 669 |
u = -1.884467e+019F; /* IEEE 32-bit binary value 0xDF800000 */ |
| 670 |
} |
| 671 |
else if (u > -5.42101e-020F) { |
| 672 |
u = -5.42101e-020F; /* IEEE 32-bit binary value 0x9F800000 */ |
| 673 |
} |
| 674 |
} |
| 675 |
t[0] = t[1] = t[2] = t[3] = u; |
| 676 |
store_vector4( &inst->DstReg, state, t ); |
| 677 |
} |
| 678 |
break; |
| 679 |
case VP_OPCODE_SUB: /* GL_NV_vertex_program1_1 */ |
| 680 |
{ |
| 681 |
GLfloat t[4], u[4], sum[4]; |
| 682 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 683 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 684 |
sum[0] = t[0] - u[0]; |
| 685 |
sum[1] = t[1] - u[1]; |
| 686 |
sum[2] = t[2] - u[2]; |
| 687 |
sum[3] = t[3] - u[3]; |
| 688 |
store_vector4( &inst->DstReg, state, sum ); |
| 689 |
} |
| 690 |
break; |
| 691 |
case VP_OPCODE_ABS: /* GL_NV_vertex_program1_1 */ |
| 692 |
{ |
| 693 |
GLfloat t[4]; |
| 694 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 695 |
if (t[0] < 0.0) t[0] = -t[0]; |
| 696 |
if (t[1] < 0.0) t[1] = -t[1]; |
| 697 |
if (t[2] < 0.0) t[2] = -t[2]; |
| 698 |
if (t[3] < 0.0) t[3] = -t[3]; |
| 699 |
store_vector4( &inst->DstReg, state, t ); |
| 700 |
} |
| 701 |
break; |
| 702 |
case VP_OPCODE_FLR: /* GL_ARB_vertex_program */ |
| 703 |
{ |
| 704 |
GLfloat t[4]; |
| 705 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 706 |
t[0] = FLOORF(t[0]); |
| 707 |
t[1] = FLOORF(t[1]); |
| 708 |
t[2] = FLOORF(t[2]); |
| 709 |
t[3] = FLOORF(t[3]); |
| 710 |
store_vector4( &inst->DstReg, state, t ); |
| 711 |
} |
| 712 |
break; |
| 713 |
case VP_OPCODE_FRC: /* GL_ARB_vertex_program */ |
| 714 |
{ |
| 715 |
GLfloat t[4]; |
| 716 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 717 |
t[0] = t[0] - FLOORF(t[0]); |
| 718 |
t[1] = t[1] - FLOORF(t[1]); |
| 719 |
t[2] = t[2] - FLOORF(t[2]); |
| 720 |
t[3] = t[3] - FLOORF(t[3]); |
| 721 |
store_vector4( &inst->DstReg, state, t ); |
| 722 |
} |
| 723 |
break; |
| 724 |
case VP_OPCODE_EX2: /* GL_ARB_vertex_program */ |
| 725 |
{ |
| 726 |
GLfloat t[4]; |
| 727 |
fetch_vector1( &inst->SrcReg[0], state, t ); |
| 728 |
t[0] = t[1] = t[2] = t[3] = (GLfloat)_mesa_pow(2.0, t[0]); |
| 729 |
store_vector4( &inst->DstReg, state, t ); |
| 730 |
} |
| 731 |
break; |
| 732 |
case VP_OPCODE_LG2: /* GL_ARB_vertex_program */ |
| 733 |
{ |
| 734 |
GLfloat t[4]; |
| 735 |
fetch_vector1( &inst->SrcReg[0], state, t ); |
| 736 |
t[0] = t[1] = t[2] = t[3] = LOG2(t[0]); |
| 737 |
store_vector4( &inst->DstReg, state, t ); |
| 738 |
} |
| 739 |
break; |
| 740 |
case VP_OPCODE_POW: /* GL_ARB_vertex_program */ |
| 741 |
{ |
| 742 |
GLfloat t[4], u[4]; |
| 743 |
fetch_vector1( &inst->SrcReg[0], state, t ); |
| 744 |
fetch_vector1( &inst->SrcReg[1], state, u ); |
| 745 |
t[0] = t[1] = t[2] = t[3] = (GLfloat)_mesa_pow(t[0], u[0]); |
| 746 |
store_vector4( &inst->DstReg, state, t ); |
| 747 |
} |
| 748 |
break; |
| 749 |
case VP_OPCODE_XPD: /* GL_ARB_vertex_program */ |
| 750 |
{ |
| 751 |
GLfloat t[4], u[4], cross[4]; |
| 752 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 753 |
fetch_vector4( &inst->SrcReg[1], state, u ); |
| 754 |
cross[0] = t[1] * u[2] - t[2] * u[1]; |
| 755 |
cross[1] = t[2] * u[0] - t[0] * u[2]; |
| 756 |
cross[2] = t[0] * u[1] - t[1] * u[0]; |
| 757 |
store_vector4( &inst->DstReg, state, cross ); |
| 758 |
} |
| 759 |
break; |
| 760 |
case VP_OPCODE_SWZ: /* GL_ARB_vertex_program */ |
| 761 |
{ |
| 762 |
const struct vp_src_register *source = &inst->SrcReg[0]; |
| 763 |
const GLfloat *src = get_register_pointer(source, state); |
| 764 |
GLfloat result[4]; |
| 765 |
GLuint i; |
| 766 |
|
| 767 |
/* do extended swizzling here */ |
| 768 |
for (i = 0; i < 4; i++) { |
| 769 |
if (GET_SWZ(source->Swizzle, i) == SWIZZLE_ZERO) |
| 770 |
result[i] = 0.0; |
| 771 |
else if (GET_SWZ(source->Swizzle, i) == SWIZZLE_ONE) |
| 772 |
result[i] = 1.0; |
| 773 |
else |
| 774 |
result[i] = src[GET_SWZ(source->Swizzle, i)]; |
| 775 |
if (source->Negate & (1 << i)) |
| 776 |
result[i] = -result[i]; |
| 777 |
} |
| 778 |
store_vector4( &inst->DstReg, state, result ); |
| 779 |
} |
| 780 |
break; |
| 781 |
case VP_OPCODE_PRINT: |
| 782 |
if (inst->SrcReg[0].File) { |
| 783 |
GLfloat t[4]; |
| 784 |
fetch_vector4( &inst->SrcReg[0], state, t ); |
| 785 |
_mesa_printf("%s%g, %g, %g, %g\n", |
| 786 |
(char *) inst->Data, t[0], t[1], t[2], t[3]); |
| 787 |
} |
| 788 |
else { |
| 789 |
_mesa_printf("%s\n", (char *) inst->Data); |
| 790 |
} |
| 791 |
break; |
| 792 |
case VP_OPCODE_END: |
| 793 |
ctx->_CurrentProgram = 0; |
| 794 |
return; |
| 795 |
default: |
| 796 |
/* bad instruction opcode */ |
| 797 |
_mesa_problem(ctx, "Bad VP Opcode in _mesa_exec_vertex_program"); |
| 798 |
ctx->_CurrentProgram = 0; |
| 799 |
return; |
| 800 |
} /* switch */ |
| 801 |
} /* for */ |
| 802 |
|
| 803 |
ctx->_CurrentProgram = 0; |
| 804 |
} |
| 805 |
|
| 806 |
|
| 807 |
|
| 808 |
/** |
| 809 |
Thoughts on vertex program optimization: |
| 810 |
|
| 811 |
The obvious thing to do is to compile the vertex program into X86/SSE/3DNow! |
| 812 |
assembly code. That will probably be a lot of work. |
| 813 |
|
| 814 |
Another approach might be to replace the vp_instruction->Opcode field with |
| 815 |
a pointer to a specialized C function which executes the instruction. |
| 816 |
In particular we can write functions which skip swizzling, negating, |
| 817 |
masking, relative addressing, etc. when they're not needed. |
| 818 |
|
| 819 |
For example: |
| 820 |
|
| 821 |
void simple_add( struct vp_instruction *inst ) |
| 822 |
{ |
| 823 |
GLfloat *sum = machine->Registers[inst->DstReg.Register]; |
| 824 |
GLfloat *a = machine->Registers[inst->SrcReg[0].Register]; |
| 825 |
GLfloat *b = machine->Registers[inst->SrcReg[1].Register]; |
| 826 |
sum[0] = a[0] + b[0]; |
| 827 |
sum[1] = a[1] + b[1]; |
| 828 |
sum[2] = a[2] + b[2]; |
| 829 |
sum[3] = a[3] + b[3]; |
| 830 |
} |
| 831 |
|
| 832 |
*/ |
| 833 |
|
| 834 |
/* |
| 835 |
|
| 836 |
KW: |
| 837 |
|
| 838 |
A first step would be to 'vectorize' the programs in the same way as |
| 839 |
the normal transformation code in the tnl module. Thus each opcode |
| 840 |
takes zero or more input vectors (registers) and produces one or more |
| 841 |
output vectors. |
| 842 |
|
| 843 |
These operations would intially be coded in C, with machine-specific |
| 844 |
assembly following, as is currently the case for matrix |
| 845 |
transformations in the math/ directory. The preprocessing scheme for |
| 846 |
selecting simpler operations Brian describes above would also work |
| 847 |
here. |
| 848 |
|
| 849 |
This should give reasonable performance without excessive effort. |
| 850 |
|
| 851 |
*/ |