**************************************************************************/
/**
- * \file ffvertex_prog.
+ * \file ffvertex_prog.c
*
* Create a vertex program to execute the current fixed function T&L pipeline.
* \author Keith Whitwell
struct state_key {
+ unsigned light_color_material_mask:12;
+ unsigned light_material_mask:12;
unsigned light_global_enabled:1;
unsigned light_local_viewer:1;
unsigned light_twoside:1;
unsigned light_color_material:1;
- unsigned light_color_material_mask:12;
- unsigned light_material_mask:12;
unsigned material_shininess_is_zero:1;
-
unsigned need_eye_coords:1;
unsigned normalize:1;
unsigned rescale_normals:1;
+
unsigned fog_source_is_depth:1;
unsigned tnl_do_vertex_fog:1;
unsigned separate_specular:1;
unsigned texture_enabled_global:1;
unsigned fragprog_inputs_read:12;
+ unsigned varying_vp_inputs;
+
struct {
unsigned light_enabled:1;
unsigned light_eyepos3_is_zero:1;
}
}
+
#define TXG_NONE 0
#define TXG_OBJ_LINEAR 1
#define TXG_EYE_LINEAR 2
return mask;
}
+
/**
* Should fog be computed per-vertex?
*/
#endif
}
+
static GLboolean check_active_shininess( GLcontext *ctx,
const struct state_key *key,
GLuint side )
return GL_FALSE;
}
-
-
-static struct state_key *make_state_key( GLcontext *ctx )
+static void make_state_key( GLcontext *ctx, struct state_key *key )
{
const struct gl_fragment_program *fp;
- struct state_key *key = CALLOC_STRUCT(state_key);
GLuint i;
+ memset(key, 0, sizeof(struct state_key));
fp = ctx->FragmentProgram._Current;
/* This now relies on texenvprogram.c being active:
key->need_eye_coords = ctx->_NeedEyeCoords;
key->fragprog_inputs_read = fp->Base.InputsRead;
+ key->varying_vp_inputs = ctx->varying_vp_inputs;
if (ctx->RenderMode == GL_FEEDBACK) {
/* make sure the vertprog emits color and tex0 */
ctx->Texture._EnabledUnits)
key->texture_enabled_global = 1;
- for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
+ for (i = 0; i < MAX_TEXTURE_COORD_UNITS; i++) {
struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
if (texUnit->_ReallyEnabled)
key->unit[i].texgen_mode0 =
translate_texgen( texUnit->TexGenEnabled & (1<<0),
- texUnit->GenModeS );
+ texUnit->GenS.Mode );
key->unit[i].texgen_mode1 =
translate_texgen( texUnit->TexGenEnabled & (1<<1),
- texUnit->GenModeT );
+ texUnit->GenT.Mode );
key->unit[i].texgen_mode2 =
translate_texgen( texUnit->TexGenEnabled & (1<<2),
- texUnit->GenModeR );
+ texUnit->GenR.Mode );
key->unit[i].texgen_mode3 =
translate_texgen( texUnit->TexGenEnabled & (1<<3),
- texUnit->GenModeQ );
+ texUnit->GenQ.Mode );
}
}
-
- return key;
}
*/
#define DISASSEM 0
-/* Should be tunable by the driver - do we want to do matrix
- * multiplications with DP4's or with MUL/MAD's? SSE works better
- * with the latter, drivers may differ.
- */
-#define PREFER_DP4 0
-
/* Use uregs to represent registers internally, translate to Mesa's
* expected formats on emit.
const struct state_key *state;
struct gl_vertex_program *program;
GLint max_inst; /** number of instructions allocated for program */
+ GLboolean mvp_with_dp4;
GLuint temp_in_use;
GLuint temp_reserved;
};
-static const struct ureg undef = {
+static const struct ureg undef = {
PROGRAM_UNDEFINED,
0,
0,
{
reg.negate ^= 1;
return reg;
-}
+}
static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
GET_SWZ(reg.swz, y),
GET_SWZ(reg.swz, z),
GET_SWZ(reg.swz, w));
-
return reg;
}
+
static struct ureg swizzle1( struct ureg reg, int x )
{
return swizzle(reg, x, x, x, x);
}
+
static struct ureg get_temp( struct tnl_program *p )
{
int bit = _mesa_ffs( ~p->temp_in_use );
return make_ureg(PROGRAM_TEMPORARY, bit-1);
}
+
static struct ureg reserve_temp( struct tnl_program *p )
{
struct ureg temp = get_temp( p );
return temp;
}
+
static void release_temp( struct tnl_program *p, struct ureg reg )
{
if (reg.file == PROGRAM_TEMPORARY) {
}
+static struct ureg register_param5(struct tnl_program *p,
+ GLint s0,
+ GLint s1,
+ GLint s2,
+ GLint s3,
+ GLint s4)
+{
+ gl_state_index tokens[STATE_LENGTH];
+ GLint idx;
+ tokens[0] = s0;
+ tokens[1] = s1;
+ tokens[2] = s2;
+ tokens[3] = s3;
+ tokens[4] = s4;
+ idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
+ return make_ureg(PROGRAM_STATE_VAR, idx);
+}
+
+
+#define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
+#define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
+#define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
+#define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
+
+
/**
* \param input one of VERT_ATTRIB_x tokens.
*/
static struct ureg register_input( struct tnl_program *p, GLuint input )
{
- p->program->Base.InputsRead |= (1<<input);
- return make_ureg(PROGRAM_INPUT, input);
+ /* Material attribs are passed here as inputs >= 32
+ */
+ if (input >= 32 || (p->state->varying_vp_inputs & (1<<input))) {
+ p->program->Base.InputsRead |= (1<<input);
+ return make_ureg(PROGRAM_INPUT, input);
+ }
+ else {
+ return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB, input );
+ }
}
+
/**
* \param input one of VERT_RESULT_x tokens.
*/
return make_ureg(PROGRAM_OUTPUT, output);
}
+
static struct ureg register_const4f( struct tnl_program *p,
GLfloat s0,
GLfloat s1,
return reg.file == PROGRAM_UNDEFINED;
}
+
static struct ureg get_identity_param( struct tnl_program *p )
{
if (is_undef(p->identity))
return p->identity;
}
-static struct ureg register_param5(struct tnl_program *p,
- GLint s0,
- GLint s1,
- GLint s2,
- GLint s3,
- GLint s4)
-{
- gl_state_index tokens[STATE_LENGTH];
- GLint idx;
- tokens[0] = s0;
- tokens[1] = s1;
- tokens[2] = s2;
- tokens[3] = s3;
- tokens[4] = s4;
- idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
- return make_ureg(PROGRAM_STATE_VAR, idx);
-}
-
-
-#define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
-#define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
-#define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
-#define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
-
-
static void register_matrix_param5( struct tnl_program *p,
GLint s0, /* modelview, projection, etc */
GLint s1, /* texture matrix number */
src->File = reg.file;
src->Index = reg.idx;
src->Swizzle = reg.swz;
- src->NegateBase = reg.negate ? NEGATE_XYZW : 0;
+ src->Negate = reg.negate ? NEGATE_XYZW : NEGATE_NONE;
src->Abs = 0;
- src->NegateAbs = 0;
src->RelAddr = 0;
/* Check that bitfield sizes aren't exceeded */
ASSERT(src->Index == reg.idx);
}
+
static void emit_dst( struct prog_dst_register *dst,
struct ureg reg, GLuint mask )
{
ASSERT(dst->Index == reg.idx);
}
+
static void debug_insn( struct prog_instruction *inst, const char *fn,
GLuint line )
{
inst = &p->program->Base.Instructions[nr];
inst->Opcode = (enum prog_opcode) op;
- inst->StringPos = 0;
inst->Data = 0;
emit_arg( &inst->SrcReg[0], src0 );
emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
}
+
/* This version is much easier to implement if writemasks are not
* supported natively on the target or (like SSE), the target doesn't
* have a clean/obvious dotproduct implementation.
release_temp(p, tmp);
}
+
static void emit_matrix_transform_vec3( struct tnl_program *p,
struct ureg dest,
const struct ureg *mat,
struct ureg dest,
struct ureg src )
{
+#if 0
+ /* XXX use this when drivers are ready for NRM3 */
+ emit_op1(p, OPCODE_NRM3, dest, WRITEMASK_XYZ, src);
+#else
struct ureg tmp = get_temp(p);
emit_op2(p, OPCODE_DP3, tmp, WRITEMASK_X, src, src);
emit_op1(p, OPCODE_RSQ, tmp, WRITEMASK_X, tmp);
emit_op2(p, OPCODE_MUL, dest, 0, src, swizzle1(tmp, X));
release_temp(p, tmp);
+#endif
}
+
static void emit_passthrough( struct tnl_program *p,
GLuint input,
GLuint output )
emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
}
+
static struct ureg get_eye_position( struct tnl_program *p )
{
if (is_undef(p->eye_position)) {
p->eye_position = reserve_temp(p);
- if (PREFER_DP4) {
+ if (p->mvp_with_dp4) {
register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
0, modelview );
}
-
static struct ureg get_eye_position_normalized( struct tnl_program *p )
{
if (is_undef(p->eye_position_normalized)) {
}
-
static void build_hpos( struct tnl_program *p )
{
struct ureg pos = register_input( p, VERT_ATTRIB_POS );
struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
struct ureg mvp[4];
- if (PREFER_DP4) {
+ if (p->mvp_with_dp4) {
register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
0, mvp );
emit_matrix_transform_vec4( p, hpos, mvp, pos );
static GLuint material_attrib( GLuint side, GLuint property )
{
- return ((property - STATE_AMBIENT) * 2 +
- side);
+ return (property - STATE_AMBIENT) * 2 + side;
}
-/* Get a bitmask of which material values vary on a per-vertex basis.
+
+/**
+ * Get a bitmask of which material values vary on a per-vertex basis.
*/
static void set_material_flags( struct tnl_program *p )
{
MAT_BIT_FRONT_AMBIENT | \
MAT_BIT_FRONT_DIFFUSE) << (side))
-/* Either return a precalculated constant value or emit code to
+
+/**
+ * Either return a precalculated constant value or emit code to
* calculate these values dynamically in the case where material calls
* are present between begin/end pairs.
*
struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
struct ureg tmp = make_temp(p, material_diffuse);
- emit_op3(p, OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient,
+ emit_op3(p, OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient,
material_ambient, material_emission);
return tmp;
}
register_param3(p, STATE_LIGHT, light, property);
struct ureg material_value = get_material(p, side, property);
struct ureg tmp = get_temp(p);
- emit_op2(p, OPCODE_MUL, tmp, 0, light_value, material_value);
+ emit_op2(p, OPCODE_MUL, tmp, 0, light_value, material_value);
return tmp;
}
else
return register_param4(p, STATE_LIGHTPROD, light, side, property);
}
+
static struct ureg calculate_light_attenuation( struct tnl_program *p,
GLuint i,
struct ureg VPpli,
/* Calculate distance attenuation:
*/
if (p->state->unit[i].light_attenuated) {
-
/* 1/d,d,d,1/d */
emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
/* 1,d,d*d,1/d */
emit_op1(p, OPCODE_RCP, dist, 0, dist);
/* spot-atten * dist-atten */
emit_op2(p, OPCODE_MUL, att, 0, dist, att);
- } else {
+ }
+ else {
/* dist-atten */
emit_op1(p, OPCODE_RCP, att, 0, dist);
}
/*
* NOTE:
- * dot.x = dot(normal, VPpli)
- * dot.y = dot(normal, halfAngle)
- * dot.z = back.shininess
- * dot.w = front.shininess
+ * dots.x = dot(normal, VPpli)
+ * dots.y = dot(normal, halfAngle)
+ * dots.z = back.shininess
+ * dots.w = front.shininess
*/
for (i = 0; i < MAX_LIGHTS; i++)
{
if (!p->state->material_shininess_is_zero) {
struct ureg shininess = get_material(p, 0, STATE_SHININESS);
- emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X));
+ emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X));
release_temp(p, shininess);
}
_col1 = make_temp(p, get_identity_param(p));
else
_col1 = _col0;
-
}
if (twoside) {
if (!p->state->material_shininess_is_zero) {
+ /* Note that we negate the back-face specular exponent here.
+ * The negation will be un-done later in the back-face code below.
+ */
struct ureg shininess = get_material(p, 1, STATE_SHININESS);
emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
negate(swizzle1(shininess,X)));
half = get_temp(p);
emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
emit_normalize_vec3(p, half, half);
- } else {
+ }
+ else {
half = register_param3(p, STATE_INTERNAL,
STATE_LIGHT_HALF_VECTOR, i);
}
}
- }
+ }
else {
struct ureg Ppli = register_param3(p, STATE_INTERNAL,
STATE_LIGHT_POSITION, i);
struct ureg res0, res1;
GLuint mask0, mask1;
-
if (count == nr_lights) {
if (separate) {
mask0 = WRITEMASK_XYZ;
res0 = _col0;
res1 = register_output( p, VERT_RESULT_COL0 );
}
- } else {
+ }
+ else {
mask0 = 0;
mask1 = 0;
res0 = _col0;
res1 = _col1;
}
-
if (!is_undef(att)) {
/* light is attenuated by distance */
emit_op1(p, OPCODE_LIT, lit, 0, dots);
emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
- }
+ }
else if (!p->state->material_shininess_is_zero) {
/* there's a non-zero specular term */
emit_op1(p, OPCODE_LIT, lit, 0, dots);
emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
- }
+ }
else {
/* no attenutation, no specular */
emit_degenerate_lit(p, lit, dots);
res0 = _bfc0;
res1 = register_output( p, VERT_RESULT_BFC0 );
}
- } else {
+ }
+ else {
res0 = _bfc0;
res1 = _bfc1;
mask0 = 0;
mask1 = 0;
}
+ /* For the back face we need to negate the X and Y component
+ * dot products. dots.Z has the negated back-face specular
+ * exponent. We swizzle that into the W position. This
+ * negation makes the back-face specular term positive again.
+ */
dots = negate(swizzle(dots,X,Y,W,Z));
if (!is_undef(att)) {
}
else if (!p->state->material_shininess_is_zero) {
emit_op1(p, OPCODE_LIT, lit, 0, dots);
- emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
- }
+ emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0); /**/
+ }
else {
emit_degenerate_lit(p, lit, dots);
- emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
+ emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0);
}
- emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0);
emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
+ /* restore dots to its original state for subsequent lights
+ * by negating and swizzling again.
+ */
+ dots = negate(swizzle(dots,X,Y,W,Z));
release_temp(p, ambient);
release_temp(p, diffuse);
emit_op1(p, useabs ? OPCODE_ABS : OPCODE_MOV, fog, WRITEMASK_X, input);
}
}
+
static void build_reflect_texgen( struct tnl_program *p,
struct ureg dest,
release_temp(p, tmp);
}
+
static void build_sphere_texgen( struct tnl_program *p,
struct ureg dest,
GLuint writemask )
{
GLuint i, j;
- for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
+ for (i = 0; i < MAX_TEXTURE_COORD_UNITS; i++) {
if (!(p->state->fragprog_inputs_read & FRAG_BIT_TEX(i)))
continue;
case TXG_NONE:
copy_mask |= WRITEMASK_X << j;
}
-
}
-
if (sphere_mask) {
build_sphere_texgen(p, out_texgen, sphere_mask);
}
struct ureg in = (!is_undef(out_texgen) ?
out_texgen :
register_input(p, VERT_ATTRIB_TEX0+i));
- if (PREFER_DP4) {
+ if (p->mvp_with_dp4) {
register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
0, texmat );
emit_matrix_transform_vec4( p, out, texmat, in );
}
release_temps(p);
- }
+ }
else {
emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i);
}
release_temp(p, ut);
}
+
/**
* Emit constant point size.
*/
emit_op1(p, OPCODE_MOV, out, WRITEMASK_X, state_size);
}
+
/**
* Pass-though per-vertex point size, from user's point size array.
*/
static void build_tnl_program( struct tnl_program *p )
-{ /* Emit the program, starting with modelviewproject:
+{
+ /* Emit the program, starting with modelviewproject:
*/
build_hpos(p);
static void
create_new_program( const struct state_key *key,
struct gl_vertex_program *program,
+ GLboolean mvp_with_dp4,
GLuint max_temps)
{
struct tnl_program p;
p.transformed_normal = undef;
p.identity = undef;
p.temp_in_use = 0;
+ p.mvp_with_dp4 = mvp_with_dp4;
if (max_temps >= sizeof(int) * 8)
p.temp_reserved = 0;
_mesa_get_fixed_func_vertex_program(GLcontext *ctx)
{
struct gl_vertex_program *prog;
- struct state_key *key;
+ struct state_key key;
/* Grab all the relevent state and put it in a single structure:
*/
- key = make_state_key(ctx);
+ make_state_key(ctx, &key);
/* Look for an already-prepared program for this state:
*/
prog = (struct gl_vertex_program *)
- _mesa_search_program_cache(ctx->VertexProgram.Cache, key, sizeof(*key));
+ _mesa_search_program_cache(ctx->VertexProgram.Cache, &key, sizeof(key));
if (!prog) {
/* OK, we'll have to build a new one */
if (!prog)
return NULL;
- create_new_program( key, prog,
+ create_new_program( &key, prog,
+ ctx->mvp_with_dp4,
ctx->Const.VertexProgram.MaxTemps );
#if 0
&prog->Base );
#endif
_mesa_program_cache_insert(ctx, ctx->VertexProgram.Cache,
- key, sizeof(*key), &prog->Base);
+ &key, sizeof(key), &prog->Base);
}
- _mesa_free(key);
-
return prog;
}