#define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
struct mode_opt {
- GLuint Source:4; /**< SRC_x */
- GLuint Operand:3; /**< OPR_x */
+#ifdef __GNUC__
+ __extension__ GLubyte Source:4; /**< SRC_x */
+ __extension__ GLubyte Operand:3; /**< OPR_x */
+#else
+ GLubyte Source; /**< SRC_x */
+ GLubyte Operand; /**< OPR_x */
+#endif
};
struct state_key {
GLuint fog_mode:2; /**< FOG_x */
GLuint inputs_available:12;
+ /* NOTE: This array of structs must be last! (see "keySize" below) */
struct {
GLuint enabled:1;
GLuint source_index:3; /**< TEXTURE_x_INDEX */
GLuint NumArgsRGB:3; /**< up to MAX_COMBINER_TERMS */
GLuint ModeRGB:5; /**< MODE_x */
- struct mode_opt OptRGB[MAX_COMBINER_TERMS];
GLuint NumArgsA:3; /**< up to MAX_COMBINER_TERMS */
GLuint ModeA:5; /**< MODE_x */
+
+ struct mode_opt OptRGB[MAX_COMBINER_TERMS];
struct mode_opt OptA[MAX_COMBINER_TERMS];
} unit[MAX_TEXTURE_UNITS];
};
}
+/**
+ * Do we need to clamp the results of the given texture env/combine mode?
+ * If the inputs to the mode are in [0,1] we don't always have to clamp
+ * the results.
+ */
+static GLboolean
+need_saturate( GLuint mode )
+{
+ switch (mode) {
+ case MODE_REPLACE:
+ case MODE_MODULATE:
+ case MODE_INTERPOLATE:
+ return GL_FALSE;
+ case MODE_ADD:
+ case MODE_ADD_SIGNED:
+ case MODE_SUBTRACT:
+ case MODE_DOT3_RGB:
+ case MODE_DOT3_RGB_EXT:
+ case MODE_DOT3_RGBA:
+ case MODE_DOT3_RGBA_EXT:
+ case MODE_MODULATE_ADD_ATI:
+ case MODE_MODULATE_SIGNED_ADD_ATI:
+ case MODE_MODULATE_SUBTRACT_ATI:
+ case MODE_ADD_PRODUCTS:
+ case MODE_ADD_PRODUCTS_SIGNED:
+ case MODE_BUMP_ENVMAP_ATI:
+ return GL_TRUE;
+ default:
+ assert(0);
+ return GL_FALSE;
+ }
+}
+
+
+
/**
* Translate TEXTURE_x_BIT to TEXTURE_x_INDEX.
*/
* Examine current texture environment state and generate a unique
* key to identify it.
*/
-static void make_state_key( GLcontext *ctx, struct state_key *key )
+static GLuint make_state_key( GLcontext *ctx, struct state_key *key )
{
GLuint i, j;
GLbitfield inputs_referenced = FRAG_BIT_COL0;
const GLbitfield inputs_available = get_fp_input_mask( ctx );
+ GLuint keySize;
memset(key, 0, sizeof(*key));
key->unit[i].enabled = 1;
key->enabled_units |= (1<<i);
- key->nr_enabled_units = i+1;
+ key->nr_enabled_units = i + 1;
inputs_referenced |= FRAG_BIT_TEX(i);
key->unit[i].source_index =
}
key->inputs_available = (inputs_available & inputs_referenced);
+
+ /* compute size of state key, ignoring unused texture units */
+ keySize = sizeof(*key) - sizeof(key->unit)
+ + key->nr_enabled_units * sizeof(key->unit[0]);
+
+ return keySize;
}
+
/**
* Use uregs to represent registers internally, translate to Mesa's
* expected formats on emit.
{
GLuint i, numArgs = key->unit[unit].NumArgsRGB;
- for (i = 0 ; i < numArgs ; i++) {
+ for (i = 0; i < numArgs; i++) {
if (key->unit[unit].OptA[i].Source != key->unit[unit].OptRGB[i].Source)
return GL_FALSE;
emit_texenv(struct texenv_fragment_program *p, GLuint unit)
{
const struct state_key *key = p->state;
- GLboolean saturate;
+ GLboolean rgb_saturate, alpha_saturate;
GLuint rgb_shift, alpha_shift;
struct ureg out, dest;
/* If we'll do rgb/alpha shifting don't saturate in emit_combine().
* We don't want to clamp twice.
*/
- saturate = !(rgb_shift || alpha_shift);
+ if (rgb_shift)
+ rgb_saturate = GL_FALSE; /* saturate after rgb shift */
+ else if (need_saturate(key->unit[unit].ModeRGB))
+ rgb_saturate = GL_TRUE;
+ else
+ rgb_saturate = GL_FALSE;
+
+ if (alpha_shift)
+ alpha_saturate = GL_FALSE; /* saturate after alpha shift */
+ else if (need_saturate(key->unit[unit].ModeA))
+ alpha_saturate = GL_TRUE;
+ else
+ alpha_saturate = GL_FALSE;
/* If this is the very last calculation, emit direct to output reg:
*/
*/
if (key->unit[unit].ModeRGB == key->unit[unit].ModeA &&
args_match(key, unit)) {
- out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
+ out = emit_combine( p, dest, WRITEMASK_XYZW, rgb_saturate,
unit,
key->unit[unit].NumArgsRGB,
key->unit[unit].ModeRGB,
}
else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT ||
key->unit[unit].ModeRGB == MODE_DOT3_RGBA) {
- out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
+ out = emit_combine( p, dest, WRITEMASK_XYZW, rgb_saturate,
unit,
key->unit[unit].NumArgsRGB,
key->unit[unit].ModeRGB,
/* Need to do something to stop from re-emitting identical
* argument calculations here:
*/
- out = emit_combine( p, dest, WRITEMASK_XYZ, saturate,
+ out = emit_combine( p, dest, WRITEMASK_XYZ, rgb_saturate,
unit,
key->unit[unit].NumArgsRGB,
key->unit[unit].ModeRGB,
key->unit[unit].OptRGB);
- out = emit_combine( p, dest, WRITEMASK_W, saturate,
+ out = emit_combine( p, dest, WRITEMASK_W, alpha_saturate,
unit,
key->unit[unit].NumArgsA,
key->unit[unit].ModeA,
*/
if (alpha_shift || rgb_shift) {
struct ureg shift;
-
- saturate = GL_TRUE; /* always saturate at this point */
+ GLboolean saturate = GL_TRUE; /* always saturate at this point */
if (rgb_shift == alpha_shift) {
shift = register_scalar_const(p, (GLfloat)(1<<rgb_shift));
* Generate instructions for loading all texture source terms.
*/
static GLboolean
-load_texunit_sources( struct texenv_fragment_program *p, int unit )
+load_texunit_sources( struct texenv_fragment_program *p, GLuint unit )
{
const struct state_key *key = p->state;
GLuint i;
* Generate instructions for loading bump map textures.
*/
static GLboolean
-load_texunit_bumpmap( struct texenv_fragment_program *p, int unit )
+load_texunit_bumpmap( struct texenv_fragment_program *p, GLuint unit )
{
const struct state_key *key = p->state;
GLuint bumpedUnitNr = key->unit[unit].OptRGB[1].Source - SRC_TEXTURE0;
texcDst = get_tex_temp( p );
p->texcoord_tex[bumpedUnitNr] = texcDst;
- /* apply rot matrix and add coords to be available in next phase */
- /* dest = (Arg0.xxxx * rotMat0 + Arg1) + (Arg0.yyyy * rotMat1) */
- /* note only 2 coords are affected the rest are left unchanged (mul by 0) */
+ /* Apply rot matrix and add coords to be available in next phase.
+ * dest = (Arg0.xxxx * rotMat0 + Arg1) + (Arg0.yyyy * rotMat1)
+ * note only 2 coords are affected the rest are left unchanged (mul by 0)
+ */
emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0,
swizzle1(bumpMapRes, SWIZZLE_X), rotMat0, texcSrc );
emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0,
swizzle1(bumpMapRes, SWIZZLE_Y), rotMat1, texcDst );
- /* move 0,0,0,1 into bumpmap src if someone (crossbar) is foolish
- enough to access this later, should optimize away */
- emit_arith( p, OPCODE_MOV, bumpMapRes, WRITEMASK_XYZW, 0, constdudvcolor, undef, undef );
+ /* Move 0,0,0,1 into bumpmap src if someone (crossbar) is foolish
+ * enough to access this later, should optimize away.
+ */
+ emit_arith( p, OPCODE_MOV, bumpMapRes, WRITEMASK_XYZW, 0,
+ constdudvcolor, undef, undef );
return GL_TRUE;
}
*/
p.program->Base.Instructions = instBuffer;
p.program->Base.Target = GL_FRAGMENT_PROGRAM_ARB;
- p.program->Base.NumTexIndirections = 1;
+ p.program->Base.String = NULL;
+ p.program->Base.NumTexIndirections = 1; /* is this right? */
p.program->Base.NumTexInstructions = 0;
p.program->Base.NumAluInstructions = 0;
- p.program->Base.String = NULL;
- p.program->Base.NumInstructions =
- p.program->Base.NumTemporaries =
- p.program->Base.NumParameters =
- p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
+ p.program->Base.NumInstructions = 0;
+ p.program->Base.NumTemporaries = 0;
+ p.program->Base.NumParameters = 0;
+ p.program->Base.NumAttributes = 0;
+ p.program->Base.NumAddressRegs = 0;
p.program->Base.Parameters = _mesa_new_parameter_list();
-
- p.program->Base.InputsRead = 0;
+ p.program->Base.InputsRead = 0x0;
p.program->Base.OutputsWritten = 1 << FRAG_RESULT_COLOR;
for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++) {
release_temps(ctx, &p);
if (key->enabled_units) {
- GLboolean needbumpstage = GL_FALSE;
+ GLboolean needbumpstage = GL_FALSE;
+
/* Zeroth pass - bump map textures first */
- for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
- if (key->unit[unit].enabled && key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
+ for (unit = 0; unit < key->nr_enabled_units; unit++)
+ if (key->unit[unit].enabled &&
+ key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
needbumpstage = GL_TRUE;
load_texunit_bumpmap( &p, unit );
}
* all referenced texture sources and emit texld instructions
* for each:
*/
- for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
+ for (unit = 0; unit < key->nr_enabled_units; unit++)
if (key->unit[unit].enabled) {
load_texunit_sources( &p, unit );
p.last_tex_stage = unit;
/* Second pass - emit combine instructions to build final color:
*/
- for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++)
- if (key->enabled_units & (1<<unit)) {
+ for (unit = 0; unit < key->nr_enabled_units; unit++)
+ if (key->unit[unit].enabled) {
p.src_previous = emit_texenv( &p, unit );
reserve_temp(&p, p.src_previous); /* don't re-use this temp reg */
release_temps(ctx, &p); /* release all temps */
* a reduced value and not what is expected in FogOption
*/
p.program->FogOption = ctx->Fog.Mode;
- p.program->Base.InputsRead |= FRAG_BIT_FOGC; /* XXX new */
- } else
+ p.program->Base.InputsRead |= FRAG_BIT_FOGC;
+ }
+ else {
p.program->FogOption = GL_NONE;
+ }
if (p.program->Base.NumTexIndirections > ctx->Const.FragmentProgram.MaxTexIndirections)
program_error(&p, "Exceeded max nr indirect texture lookups");
{
struct gl_fragment_program *prog;
struct state_key key;
+ GLuint keySize;
- make_state_key(ctx, &key);
+ keySize = make_state_key(ctx, &key);
prog = (struct gl_fragment_program *)
_mesa_search_program_cache(ctx->FragmentProgram.Cache,
- &key, sizeof(key));
+ &key, keySize);
if (!prog) {
prog = (struct gl_fragment_program *)
create_new_program(ctx, &key, prog);
_mesa_program_cache_insert(ctx, ctx->FragmentProgram.Cache,
- &key, sizeof(key), &prog->Base);
+ &key, keySize, &prog->Base);
}
return prog;
projects / mesa.git / blobdiff