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 fog_mode:2;
unsigned point_attenuated:1;
+ unsigned point_array: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;
-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 */
if (ctx->Point._Attenuated)
key->point_attenuated = 1;
+#if FEATURE_point_size_array
+ if (ctx->Array.ArrayObj->PointSize.Enabled)
+ key->point_array = 1;
+#endif
+
if (ctx->Texture._TexGenEnabled ||
ctx->Texture._TexMatEnabled ||
ctx->Texture._EnabledUnits)
texUnit->GenModeQ );
}
}
-
- return key;
}
* generated program with line/function references for each
* instruction back into this file:
*/
-#define DISASSEM (MESA_VERBOSE&VERBOSE_DISASSEM)
+#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
*/
#define PREFER_DP4 0
-#define MAX_INSN 256
/* Use uregs to represent registers internally, translate to Mesa's
* expected formats on emit.
*/
struct ureg {
GLuint file:4;
- GLint idx:8; /* relative addressing may be negative */
+ GLint idx:9; /* relative addressing may be negative */
+ /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
GLuint negate:1;
GLuint swz:12;
- GLuint pad:7;
+ GLuint pad:6;
};
struct tnl_program {
const struct state_key *state;
struct gl_vertex_program *program;
+ GLint max_inst; /** number of instructions allocated for program */
GLuint temp_in_use;
GLuint temp_reserved;
struct ureg eye_position;
+ struct ureg eye_position_z;
struct ureg eye_position_normalized;
struct ureg transformed_normal;
struct ureg identity;
static const struct ureg undef = {
PROGRAM_UNDEFINED,
- ~0,
+ 0,
0,
0,
0
}
+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.
+ */
static struct ureg register_output( struct tnl_program *p, GLuint output )
{
p->program->Base.OutputsWritten |= (1<<output);
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->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,
dst->CondSwizzle = SWIZZLE_NOOP;
dst->CondSrc = 0;
dst->pad = 0;
+ /* Check that bitfield sizes aren't exceeded */
+ ASSERT(dst->Index == reg.idx);
}
static void debug_insn( struct prog_instruction *inst, const char *fn,
const char *fn,
GLuint line)
{
- GLuint nr = p->program->Base.NumInstructions++;
- struct prog_instruction *inst = &p->program->Base.Instructions[nr];
+ GLuint nr;
+ struct prog_instruction *inst;
- if (p->program->Base.NumInstructions > MAX_INSN) {
- _mesa_problem(0, "Out of instructions in emit_op3fn\n");
- return;
+ assert((GLint) p->program->Base.NumInstructions <= p->max_inst);
+
+ if (p->program->Base.NumInstructions == p->max_inst) {
+ /* need to extend the program's instruction array */
+ struct prog_instruction *newInst;
+
+ /* double the size */
+ p->max_inst *= 2;
+
+ newInst = _mesa_alloc_instructions(p->max_inst);
+ if (!newInst) {
+ _mesa_error(NULL, GL_OUT_OF_MEMORY, "vertex program build");
+ return;
+ }
+
+ _mesa_copy_instructions(newInst,
+ p->program->Base.Instructions,
+ p->program->Base.NumInstructions);
+
+ _mesa_free_instructions(p->program->Base.Instructions,
+ p->program->Base.NumInstructions);
+
+ p->program->Base.Instructions = newInst;
}
+ nr = p->program->Base.NumInstructions++;
+
+ inst = &p->program->Base.Instructions[nr];
inst->Opcode = (enum prog_opcode) op;
inst->StringPos = 0;
inst->Data = 0;
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,
}
+static struct ureg get_eye_position_z( struct tnl_program *p )
+{
+ if (!is_undef(p->eye_position))
+ return swizzle1(p->eye_position, Z);
+
+ if (is_undef(p->eye_position_z)) {
+ struct ureg pos = register_input( p, VERT_ATTRIB_POS );
+ struct ureg modelview[4];
+
+ p->eye_position_z = reserve_temp(p);
+
+ register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
+ 0, modelview );
+
+ emit_op2(p, OPCODE_DP4, p->eye_position_z, 0, pos, modelview[2]);
+ }
+
+ return p->eye_position_z;
+}
+
+
+
static struct ureg get_eye_position_normalized( struct tnl_program *p )
{
if (is_undef(p->eye_position_normalized)) {
}
+/**
+ * Compute:
+ * lit.y = MAX(0, dots.x)
+ * lit.z = SLT(0, dots.x)
+ */
static void emit_degenerate_lit( struct tnl_program *p,
struct ureg lit,
struct ureg dots )
{
- struct ureg id = get_identity_param(p);
-
- /* Note that result.x & result.w will not be examined. Note also that
+ struct ureg id = get_identity_param(p); /* id = {0,0,0,1} */
+
+ /* Note that lit.x & lit.w will not be examined. Note also that
* dots.xyzw == dots.xxxx.
*/
- /* result[1] = MAX2(in, 0)
+ /* MAX lit, id, dots;
*/
- emit_op2(p, OPCODE_MAX, lit, 0, id, dots);
+ emit_op2(p, OPCODE_MAX, lit, WRITEMASK_XYZW, id, dots);
/* result[2] = (in > 0 ? 1 : 0)
+ * SLT lit.z, id.z, dots; # lit.z = (0 < dots.z) ? 1 : 0
*/
- emit_op2(p, OPCODE_SLT, lit, WRITEMASK_Z,
- lit, /* 0 */
- dots); /* in[0] */
+ emit_op2(p, OPCODE_SLT, lit, WRITEMASK_Z, swizzle1(id,Z), dots);
}
struct ureg _bfc0 = undef, _bfc1 = undef;
GLuint i;
+ /*
+ * NOTE:
+ * dot.x = dot(normal, VPpli)
+ * dot.y = dot(normal, halfAngle)
+ * dot.z = back.shininess
+ * dot.w = front.shininess
+ */
+
for (i = 0; i < MAX_LIGHTS; i++)
if (p->state->unit[i].light_enabled)
nr_lights++;
struct ureg res0, res1;
GLuint mask0, mask1;
- if (p->state->material_shininess_is_zero) {
- emit_degenerate_lit(p, lit, dots);
- } else {
- emit_op1(p, OPCODE_LIT, lit, 0, dots);
- }
-
- if (!is_undef(att))
- emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
-
if (count == nr_lights) {
if (separate) {
res1 = _col1;
}
- emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
+
+ 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);
+ emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
+ }
+
emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
struct ureg res0, res1;
GLuint mask0, mask1;
- if (p->state->material_shininess_is_zero) {
- emit_degenerate_lit(p, lit, negate(swizzle(dots,X,Y,W,Z)));
- } else {
- emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z)));
- }
-
- if (!is_undef(att))
- emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
-
if (count == nr_lights) {
if (separate) {
mask0 = WRITEMASK_XYZ;
mask1 = 0;
}
- emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0);
+ dots = negate(swizzle(dots,X,Y,W,Z));
+
+ if (!is_undef(att)) {
+ emit_op1(p, OPCODE_LIT, lit, 0, dots);
+ emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
+ emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
+ }
+ else if (!p->state->material_shininess_is_zero) {
+ emit_op1(p, OPCODE_LIT, lit, 0, dots);
+ emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0);
+ }
+ else {
+ emit_degenerate_lit(p, lit, dots);
+ 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 negate flag for next lighting */
+ dots = negate(dots);
+
release_temp(p, ambient);
release_temp(p, diffuse);
release_temp(p, specular);
struct ureg input;
if (p->state->fog_source_is_depth) {
- input = swizzle1(get_eye_position(p), Z);
+ input = get_eye_position_z(p);
}
else {
input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
}
-static void build_pointsize( struct tnl_program *p )
+/**
+ * Point size attenuation computation.
+ */
+static void build_atten_pointsize( struct tnl_program *p )
{
- struct ureg eye = get_eye_position(p);
+ struct ureg eye = get_eye_position_z(p);
struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
struct ureg out = register_output(p, VERT_RESULT_PSIZ);
/**
* Emit constant point size.
*/
-static void constant_pointsize( struct tnl_program *p )
+static void build_constant_pointsize( struct tnl_program *p )
{
struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
struct ureg out = register_output(p, VERT_RESULT_PSIZ);
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_array_pointsize( struct tnl_program *p )
+{
+ struct ureg in = register_input(p, VERT_ATTRIB_POINT_SIZE);
+ struct ureg out = register_output(p, VERT_RESULT_PSIZ);
+ emit_op1(p, OPCODE_MOV, out, WRITEMASK_X, in);
+}
+
+
static void build_tnl_program( struct tnl_program *p )
{ /* Emit the program, starting with modelviewproject:
*/
build_texture_transform(p);
if (p->state->point_attenuated)
- build_pointsize(p);
+ build_atten_pointsize(p);
+ else if (p->state->point_array)
+ build_array_pointsize(p);
#if 0
else
- constant_pointsize(p);
+ build_constant_pointsize(p);
+#else
+ (void) build_constant_pointsize;
#endif
/* Finish up:
p.state = key;
p.program = program;
p.eye_position = undef;
+ p.eye_position_z = undef;
p.eye_position_normalized = undef;
p.transformed_normal = undef;
p.identity = undef;
else
p.temp_reserved = ~((1<<max_temps)-1);
- p.program->Base.Instructions = _mesa_alloc_instructions(MAX_INSN);
+ /* Start by allocating 32 instructions.
+ * If we need more, we'll grow the instruction array as needed.
+ */
+ p.max_inst = 32;
+ p.program->Base.Instructions = _mesa_alloc_instructions(p.max_inst);
p.program->Base.String = NULL;
p.program->Base.NumInstructions =
p.program->Base.NumTemporaries =
_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->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;
}