/**
* Point size attenuation computation.
*/
static void build_atten_pointsize( struct tnl_program *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);
struct ureg ut = get_temp(p);
/* dist = |eyez| */
emit_op1(p, OPCODE_ABS, ut, WRITEMASK_Y, swizzle1(eye, Z));
/* p1 + dist * (p2 + dist * p3); */
emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
ut, swizzle1(state_attenuation, X));
/* 1 / sqrt(factor) */
emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
#if 0
/* out = pointSize / sqrt(factor) */
emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
#else
/* this is a good place to clamp the point size since there's likely
* no hardware registers to clamp point size at rasterization time.
*/
emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
#endif
release_temp(p, ut);
}
static void build_pointsize( struct tnl_program *p )
{
struct ureg eye = get_eye_position(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);
struct ureg ut = get_temp(p);
/* dist = |eyez| */
emit_op1(p, OPCODE_ABS, ut, WRITEMASK_Y, swizzle1(eye, Z));
/* p1 + dist * (p2 + dist * p3); */
emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
ut, swizzle1(state_attenuation, X));
/* 1 / sqrt(factor) */
emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
#if 1
/* out = pointSize / sqrt(factor) */
emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
#else
/* not sure, might make sense to do clamping here,
but it's not done in t_vb_points neither */
emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
#endif
release_temp(p, ut);
}
static struct brw_fp_src get_pixel_w( struct brw_wm_compile *c )
{
if (src_is_undef(c->fp_pixel_w)) {
struct brw_fp_dst pixel_w = get_temp(c);
struct brw_fp_src deltas = get_delta_xy(c);
/* XXX: assuming position is always first -- valid?
*/
struct brw_fp_src interp_wpos = src_reg(BRW_FILE_PAYLOAD, 0);
/* deltas.xyw = DELTAS2 deltas.xy, payload.interp_wpos.x
*/
emit_op3(c,
WM_PIXELW,
dst_mask(pixel_w, BRW_WRITEMASK_W),
interp_wpos,
deltas,
src_undef());
c->fp_pixel_w = src_reg_from_dst(pixel_w);
}
return c->fp_pixel_w;
}
static void build_fog( struct tnl_program *p )
{
struct ureg fog = register_output(p, VERT_RESULT_FOGC);
struct ureg input;
GLuint useabs = p->state->fog_source_is_depth && p->state->fog_option &&
(p->state->fog_option != FOG_EXP2);
if (p->state->fog_source_is_depth) {
input = swizzle1(get_eye_position(p), Z);
}
else {
input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
}
if (p->state->fog_option &&
p->state->tnl_do_vertex_fog) {
struct ureg params = register_param2(p, STATE_INTERNAL,
STATE_FOG_PARAMS_OPTIMIZED);
struct ureg tmp = get_temp(p);
struct ureg id = get_identity_param(p);
emit_op1(p, OPCODE_MOV, fog, 0, id);
if (useabs) {
emit_op1(p, OPCODE_ABS, tmp, 0, input);
}
switch (p->state->fog_option) {
case FOG_LINEAR: {
emit_op3(p, OPCODE_MAD, tmp, 0, useabs ? tmp : input,
swizzle1(params,X), swizzle1(params,Y));
emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
break;
}
case FOG_EXP:
emit_op2(p, OPCODE_MUL, tmp, 0, useabs ? tmp : input,
swizzle1(params,Z));
emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, ureg_negate(tmp));
break;
case FOG_EXP2:
emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W));
emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp);
emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, ureg_negate(tmp));
break;
}
release_temp(p, tmp);
}
else {
/* results = incoming fog coords (compute fog per-fragment later)
*
* KW: Is it really necessary to do anything in this case?
*/
emit_op1(p, useabs ? OPCODE_ABS : OPCODE_MOV, fog, 0, input);
}
}
static void build_fog( struct tnl_program *p )
{
struct ureg fog = register_output(p, VERT_RESULT_FOGC);
struct ureg input;
if (p->state->fog_source_is_depth) {
input = swizzle1(get_eye_position(p), Z);
}
else {
input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
}
if (p->state->fog_mode && p->state->tnl_do_vertex_fog) {
struct ureg params = register_param2(p, STATE_INTERNAL,
STATE_FOG_PARAMS_OPTIMIZED);
struct ureg tmp = get_temp(p);
GLboolean useabs = (p->state->fog_mode != FOG_EXP2);
if (useabs) {
emit_op1(p, OPCODE_ABS, tmp, 0, input);
}
switch (p->state->fog_mode) {
case FOG_LINEAR: {
struct ureg id = get_identity_param(p);
emit_op3(p, OPCODE_MAD, tmp, 0, useabs ? tmp : input,
swizzle1(params,X), swizzle1(params,Y));
emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
break;
}
case FOG_EXP:
emit_op2(p, OPCODE_MUL, tmp, 0, useabs ? tmp : input,
swizzle1(params,Z));
emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
break;
case FOG_EXP2:
emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W));
emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp);
emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
break;
}
release_temp(p, tmp);
}
else {
/* results = incoming fog coords (compute fog per-fragment later)
*
* KW: Is it really necessary to do anything in this case?
* BP: Yes, we always need to compute the absolute value, unless
* we want to push that down into the fragment program...
*/
GLboolean useabs = GL_TRUE;
emit_op1(p, useabs ? OPCODE_ABS : OPCODE_MOV, fog, WRITEMASK_X, input);
}
}
/* 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.
*/
static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
struct ureg dest,
const struct ureg *mat,
struct ureg src)
{
struct ureg tmp;
if (dest.file != PROGRAM_TEMPORARY)
tmp = get_temp(p);
else
tmp = dest;
emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
if (dest.file != PROGRAM_TEMPORARY)
release_temp(p, tmp);
}
static void build_sphere_texgen( struct tnl_program *p,
struct ureg dest,
GLuint writemask )
{
struct ureg normal = get_transformed_normal(p);
struct ureg eye_hat = get_eye_position_normalized(p);
struct ureg tmp = get_temp(p);
struct ureg half = register_scalar_const(p, .5);
struct ureg r = get_temp(p);
struct ureg inv_m = get_temp(p);
struct ureg id = get_identity_param(p);
/* Could share the above calculations, but it would be
* a fairly odd state for someone to set (both sphere and
* reflection active for different texture coordinate
* components. Of course - if two texture units enable
* reflect and/or sphere, things start to tilt in favour
* of seperating this out:
*/
/* n.u */
emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
/* 2n.u */
emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
/* (-2n.u)n + u */
emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
/* r + 0,0,1 */
emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
/* rx^2 + ry^2 + (rz+1)^2 */
emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
/* 2/m */
emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
/* 1/m */
emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
/* r/m + 1/2 */
emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
release_temp(p, tmp);
release_temp(p, r);
release_temp(p, inv_m);
}
/**
* 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.
*
* Probably want to shift this to the program compilation phase - if
* we always emitted the calculation here, a smart compiler could
* detect that it was constant (given a certain set of inputs), and
* lift it out of the main loop. That way the programs created here
* would be independent of the vertex_buffer details.
*/
static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
{
if (p->materials & SCENE_COLOR_BITS(side)) {
struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
struct ureg material_emission = get_material(p, side, STATE_EMISSION);
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,
material_ambient, material_emission);
return tmp;
}
else
return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
}
static void build_reflect_texgen( struct tnl_program *p,
struct ureg dest,
GLuint writemask )
{
struct ureg normal = get_transformed_normal(p);
struct ureg eye_hat = get_eye_position_normalized(p);
struct ureg tmp = get_temp(p);
/* n.u */
emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
/* 2n.u */
emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
/* (-2n.u)n + u */
emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);
release_temp(p, tmp);
}
static struct brw_fp_src get_delta_xy( struct brw_wm_compile *c )
{
if (src_is_undef(c->fp_delta_xy)) {
struct brw_fp_dst delta_xy = get_temp(c);
struct brw_fp_src pixel_xy = get_pixel_xy(c);
struct brw_fp_src payload_r0_depth = src_reg(BRW_FILE_PAYLOAD, PAYLOAD_DEPTH);
/* deltas.xy = DELTAXY pixel_xy, payload[0]
*/
emit_op3(c,
WM_DELTAXY,
dst_mask(delta_xy, BRW_WRITEMASK_XY),
pixel_xy,
payload_r0_depth,
src_undef());
c->fp_delta_xy = src_reg_from_dst(delta_xy);
}
return c->fp_delta_xy;
}
/* Many opcodes produce the same value across all the result channels.
* We'd rather not have to support that splatting in the opcode implementations,
* and brw_wm_pass*.c wants to optimize them out by shuffling references around
* anyway. We can easily get both by emitting the opcode to one channel, and
* then MOVing it to the others, which brw_wm_pass*.c already understands.
*/
static void emit_scalar_insn(struct brw_wm_compile *c,
unsigned opcode,
struct brw_fp_dst dst,
struct brw_fp_src src0,
struct brw_fp_src src1,
struct brw_fp_src src2 )
{
unsigned first_chan = ffs(dst.writemask) - 1;
unsigned first_mask = 1 << first_chan;
if (dst.writemask == 0)
return;
emit_op3( c, opcode,
dst_mask(dst, first_mask),
src0, src1, src2 );
if (dst.writemask != first_mask) {
emit_op1(c, TGSI_OPCODE_MOV,
dst_mask(dst, ~first_mask),
src_scalar(src_reg_from_dst(dst), first_chan));
}
}
/* Need to add some addtional parameters to allow lighting in object
* space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
* space lighting.
*/
static void build_lighting( struct tnl_program *p )
{
const GLboolean twoside = p->state->light_twoside;
const GLboolean separate = p->state->separate_specular;
GLuint nr_lights = 0, count = 0;
struct ureg normal = get_transformed_normal(p);
struct ureg lit = get_temp(p);
struct ureg dots = get_temp(p);
struct ureg _col0 = undef, _col1 = undef;
struct ureg _bfc0 = undef, _bfc1 = undef;
GLuint i;
/*
* NOTE:
* 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->unit[i].light_enabled)
nr_lights++;
set_material_flags(p);
{
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));
release_temp(p, shininess);
}
_col0 = make_temp(p, get_scenecolor(p, 0));
if (separate)
_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)));
release_temp(p, shininess);
}
_bfc0 = make_temp(p, get_scenecolor(p, 1));
if (separate)
_bfc1 = make_temp(p, get_identity_param(p));
else
_bfc1 = _bfc0;
}
/* If no lights, still need to emit the scenecolor.
*/
{
struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
emit_op1(p, OPCODE_MOV, res0, 0, _col0);
}
if (separate) {
struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
emit_op1(p, OPCODE_MOV, res1, 0, _col1);
}
if (twoside) {
struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
}
if (twoside && separate) {
struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
}
if (nr_lights == 0) {
release_temps(p);
return;
}
for (i = 0; i < MAX_LIGHTS; i++) {
if (p->state->unit[i].light_enabled) {
struct ureg half = undef;
struct ureg att = undef, VPpli = undef;
count++;
if (p->state->unit[i].light_eyepos3_is_zero) {
/* Can used precomputed constants in this case.
* Attenuation never applies to infinite lights.
*/
VPpli = register_param3(p, STATE_INTERNAL,
STATE_LIGHT_POSITION_NORMALIZED, i);
if (!p->state->material_shininess_is_zero) {
if (p->state->light_local_viewer) {
struct ureg eye_hat = get_eye_position_normalized(p);
half = get_temp(p);
emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
emit_normalize_vec3(p, half, half);
}
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 V = get_eye_position(p);
struct ureg dist = get_temp(p);
VPpli = get_temp(p);
/* Calculate VPpli vector
*/
emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
/* Normalize VPpli. The dist value also used in
* attenuation below.
*/
emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
emit_op1(p, OPCODE_RSQ, dist, 0, dist);
emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
/* Calculate attenuation:
*/
if (!p->state->unit[i].light_spotcutoff_is_180 ||
p->state->unit[i].light_attenuated) {
att = calculate_light_attenuation(p, i, VPpli, dist);
}
/* Calculate viewer direction, or use infinite viewer:
*/
if (!p->state->material_shininess_is_zero) {
half = get_temp(p);
if (p->state->light_local_viewer) {
struct ureg eye_hat = get_eye_position_normalized(p);
emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
}
else {
struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
}
emit_normalize_vec3(p, half, half);
}
release_temp(p, dist);
}
/* Calculate dot products:
*/
if (p->state->material_shininess_is_zero) {
emit_op2(p, OPCODE_DP3, dots, 0, normal, VPpli);
}
else {
emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
}
/* Front face lighting:
*/
{
struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
struct ureg res0, res1;
GLuint mask0, mask1;
if (count == nr_lights) {
if (separate) {
mask0 = WRITEMASK_XYZ;
mask1 = WRITEMASK_XYZ;
res0 = register_output( p, VERT_RESULT_COL0 );
res1 = register_output( p, VERT_RESULT_COL1 );
}
else {
mask0 = 0;
mask1 = WRITEMASK_XYZ;
res0 = _col0;
res1 = register_output( p, VERT_RESULT_COL0 );
}
}
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);
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);
release_temp(p, ambient);
release_temp(p, diffuse);
release_temp(p, specular);
}
/* Back face lighting:
*/
if (twoside) {
struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
struct ureg res0, res1;
GLuint mask0, mask1;
if (count == nr_lights) {
if (separate) {
mask0 = WRITEMASK_XYZ;
mask1 = WRITEMASK_XYZ;
res0 = register_output( p, VERT_RESULT_BFC0 );
res1 = register_output( p, VERT_RESULT_BFC1 );
}
else {
mask0 = 0;
mask1 = WRITEMASK_XYZ;
res0 = _bfc0;
res1 = register_output( p, VERT_RESULT_BFC0 );
}
}
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)) {
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 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);
release_temp(p, specular);
}
release_temp(p, half);
release_temp(p, VPpli);
release_temp(p, att);
}
}
release_temps( p );
}
/**
* Some TEX instructions require extra code, cube map coordinate
* normalization, or coordinate scaling for RECT textures, etc.
* This function emits those extra instructions and the TEX
* instruction itself.
*/
static void precalc_tex( struct brw_wm_compile *c,
struct brw_fp_dst dst,
unsigned target,
unsigned unit,
struct brw_fp_src src0,
struct brw_fp_src sampler )
{
struct brw_fp_src coord;
struct brw_fp_dst tmp = dst_undef();
assert(unit < BRW_MAX_TEX_UNIT);
/* Cubemap: find longest component of coord vector and normalize
* it.
*/
if (target == TGSI_TEXTURE_CUBE) {
struct brw_fp_src tmpsrc;
tmp = get_temp(c);
tmpsrc = src_reg_from_dst(tmp);
/* tmp = abs(src0) */
emit_op1(c,
TGSI_OPCODE_MOV,
tmp,
src_abs(src0));
/* tmp.X = MAX(tmp.X, tmp.Y) */
emit_op2(c, TGSI_OPCODE_MAX,
dst_mask(tmp, BRW_WRITEMASK_X),
src_scalar(tmpsrc, X),
src_scalar(tmpsrc, Y));
/* tmp.X = MAX(tmp.X, tmp.Z) */
emit_op2(c, TGSI_OPCODE_MAX,
dst_mask(tmp, BRW_WRITEMASK_X),
tmpsrc,
src_scalar(tmpsrc, Z));
/* tmp.X = 1 / tmp.X */
emit_op1(c, TGSI_OPCODE_RCP,
dst_mask(tmp, BRW_WRITEMASK_X),
tmpsrc);
/* tmp = src0 * tmp.xxxx */
emit_op2(c, TGSI_OPCODE_MUL,
tmp,
src0,
src_scalar(tmpsrc, X));
coord = tmpsrc;
}
else if (target == TGSI_TEXTURE_RECT ||
target == TGSI_TEXTURE_SHADOWRECT) {
/* XXX: need a mechanism for internally generated constants.
*/
coord = src0;
}
else {
coord = src0;
}
/* Need to emit YUV texture conversions by hand. Probably need to
* do this here - the alternative is in brw_wm_emit.c, but the
* conversion requires allocating a temporary variable which we
* don't have the facility to do that late in the compilation.
*/
if (c->key.yuvtex_mask & (1 << unit)) {
/* convert ycbcr to RGBA */
GLboolean swap_uv = c->key.yuvtex_swap_mask & (1<<unit);
struct brw_fp_dst tmp = get_temp(c);
struct brw_fp_src tmpsrc = src_reg_from_dst(tmp);
struct brw_fp_src C0 = src_imm4f( c, -.5, -.0625, -.5, 1.164 );
struct brw_fp_src C1 = src_imm4f( c, 1.596, -0.813, 2.018, -.391 );
/* tmp = TEX ...
*/
emit_tex_op(c,
TGSI_OPCODE_TEX,
dst_saturate(tmp, dst.saturate),
unit,
target,
sampler.index,
coord,
src_undef(),
src_undef());
/* tmp.xyz = ADD TMP, C0
*/
emit_op2(c, TGSI_OPCODE_ADD,
dst_mask(tmp, BRW_WRITEMASK_XYZ),
tmpsrc,
C0);
/* YUV.y = MUL YUV.y, C0.w
*/
emit_op2(c, TGSI_OPCODE_MUL,
dst_mask(tmp, BRW_WRITEMASK_Y),
tmpsrc,
src_scalar(C0, W));
/*
* if (UV swaped)
* RGB.xyz = MAD YUV.zzx, C1, YUV.y
* else
* RGB.xyz = MAD YUV.xxz, C1, YUV.y
*/
emit_op3(c, TGSI_OPCODE_MAD,
dst_mask(dst, BRW_WRITEMASK_XYZ),
( swap_uv ?
src_swizzle(tmpsrc, Z,Z,X,X) :
src_swizzle(tmpsrc, X,X,Z,Z)),
C1,
src_scalar(tmpsrc, Y));
/* RGB.y = MAD YUV.z, C1.w, RGB.y
*/
emit_op3(c,
TGSI_OPCODE_MAD,
dst_mask(dst, BRW_WRITEMASK_Y),
src_scalar(tmpsrc, Z),
src_scalar(C1, W),
src_scalar(src_reg_from_dst(dst), Y));
release_temp(c, tmp);
}
else {
/* ordinary RGBA tex instruction */
emit_tex_op(c,
TGSI_OPCODE_TEX,
dst,
unit,
target,
sampler.index,
coord,
src_undef(),
src_undef());
}
/* XXX: add GL_EXT_texture_swizzle support to gallium -- by
* generating shader variants in mesa state tracker.
*/
/* Release this temp if we ended up allocating it:
*/
if (!dst_is_undef(tmp))
release_temp(c, tmp);
}
static void emit_interp( struct brw_wm_compile *c,
GLuint idx,
GLuint semantic,
GLuint interp_mode )
{
struct brw_fp_dst dst = dst_reg(TGSI_FILE_INPUT, idx);
struct brw_fp_src interp = src_reg(BRW_FILE_PAYLOAD, idx);
struct brw_fp_src deltas = get_delta_xy(c);
/* Need to use PINTERP on attributes which have been
* multiplied by 1/W in the SF program, and LINTERP on those
* which have not:
*/
switch (semantic) {
case TGSI_SEMANTIC_POSITION:
/* Have to treat wpos.xy specially:
*/
emit_op1(c,
WM_WPOSXY,
dst_mask(dst, BRW_WRITEMASK_XY),
get_pixel_xy(c));
/* TGSI_FILE_INPUT.attr.xyzw = INTERP payload.interp[attr].x, deltas.xyw
*/
emit_op2(c,
WM_LINTERP,
dst_mask(dst, BRW_WRITEMASK_ZW),
interp,
deltas);
break;
case TGSI_SEMANTIC_COLOR:
if (c->key.flat_shade) {
emit_op1(c,
WM_CINTERP,
dst,
interp);
}
else if (interp_mode == TGSI_INTERPOLATE_LINEAR) {
emit_op2(c,
WM_LINTERP,
dst,
interp,
deltas);
}
else {
emit_op3(c,
WM_PINTERP,
dst,
interp,
deltas,
get_pixel_w(c));
}
break;
case TGSI_SEMANTIC_FOG:
/* Interpolate the fog coordinate */
emit_op3(c,
WM_PINTERP,
dst_mask(dst, BRW_WRITEMASK_X),
interp,
deltas,
get_pixel_w(c));
emit_op1(c,
TGSI_OPCODE_MOV,
dst_mask(dst, BRW_WRITEMASK_YZ),
src_imm1f(c, 0.0));
emit_op1(c,
TGSI_OPCODE_MOV,
dst_mask(dst, BRW_WRITEMASK_W),
src_imm1f(c, 1.0));
break;
case TGSI_SEMANTIC_FACE:
/* XXX review/test this case */
emit_op0(c,
WM_FRONTFACING,
dst_mask(dst, BRW_WRITEMASK_X));
emit_op1(c,
TGSI_OPCODE_MOV,
dst_mask(dst, BRW_WRITEMASK_YZ),
src_imm1f(c, 0.0));
emit_op1(c,
TGSI_OPCODE_MOV,
dst_mask(dst, BRW_WRITEMASK_W),
src_imm1f(c, 1.0));
break;
case TGSI_SEMANTIC_PSIZE:
/* XXX review/test this case */
emit_op3(c,
WM_PINTERP,
dst_mask(dst, BRW_WRITEMASK_XY),
interp,
deltas,
get_pixel_w(c));
emit_op1(c,
TGSI_OPCODE_MOV,
dst_mask(dst, BRW_WRITEMASK_Z),
src_imm1f(c, 0.0f));
emit_op1(c,
TGSI_OPCODE_MOV,
dst_mask(dst, BRW_WRITEMASK_W),
src_imm1f(c, 1.0f));
break;
default:
switch (interp_mode) {
case TGSI_INTERPOLATE_CONSTANT:
emit_op1(c,
WM_CINTERP,
dst,
interp);
break;
case TGSI_INTERPOLATE_LINEAR:
emit_op2(c,
WM_LINTERP,
dst,
interp,
deltas);
break;
case TGSI_INTERPOLATE_PERSPECTIVE:
emit_op3(c,
WM_PINTERP,
dst,
interp,
deltas,
get_pixel_w(c));
break;
}
break;
}
}
static void emit_insn( struct brw_wm_compile *c,
const struct tgsi_full_instruction *inst )
{
unsigned opcode = inst->Instruction.Opcode;
struct brw_fp_dst dst;
struct brw_fp_src src[3];
int i;
dst = translate_dst( c, &inst->Dst[0],
inst->Instruction.Saturate );
for (i = 0; i < inst->Instruction.NumSrcRegs; i++)
src[i] = translate_src( c, &inst->Src[i] );
switch (opcode) {
case TGSI_OPCODE_ABS:
emit_op1(c, TGSI_OPCODE_MOV,
dst,
src_abs(src[0]));
break;
case TGSI_OPCODE_SUB:
emit_op2(c, TGSI_OPCODE_ADD,
dst,
src[0],
src_negate(src[1]));
break;
case TGSI_OPCODE_SCS:
emit_op1(c, TGSI_OPCODE_SCS,
dst_mask(dst, BRW_WRITEMASK_XY),
src[0]);
break;
case TGSI_OPCODE_DST:
precalc_dst(c, dst, src[0], src[1]);
break;
case TGSI_OPCODE_LIT:
precalc_lit(c, dst, src[0]);
break;
case TGSI_OPCODE_TEX:
precalc_tex(c, dst,
inst->Texture.Texture,
src[1].index, /* use sampler unit for tex idx */
src[0], /* coord */
src[1]); /* sampler */
break;
case TGSI_OPCODE_TXP:
precalc_txp(c, dst,
inst->Texture.Texture,
src[1].index, /* use sampler unit for tex idx */
src[0], /* coord */
src[1]); /* sampler */
break;
case TGSI_OPCODE_TXB:
/* XXX: TXB not done
*/
precalc_tex(c, dst,
inst->Texture.Texture,
src[1].index, /* use sampler unit for tex idx*/
src[0],
src[1]);
break;
case TGSI_OPCODE_XPD:
emit_op2(c, TGSI_OPCODE_XPD,
dst_mask(dst, BRW_WRITEMASK_XYZ),
src[0],
src[1]);
break;
case TGSI_OPCODE_KIL:
emit_op1(c, TGSI_OPCODE_KIL,
dst_mask(dst_undef(), 0),
src[0]);
break;
case TGSI_OPCODE_END:
emit_fb_write(c);
break;
default:
if (!c->key.has_flow_control &&
brw_wm_is_scalar_result(opcode))
emit_scalar_insn(c, opcode, dst, src[0], src[1], src[2]);
else
emit_op3(c, opcode, dst, src[0], src[1], src[2]);
break;
}
}
/* Need to add some addtional parameters to allow lighting in object
* space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
* space lighting.
*/
static void build_lighting( struct tnl_program *p )
{
const GLboolean twoside = p->state->light_twoside;
const GLboolean separate = p->state->separate_specular;
GLuint nr_lights = 0, count = 0;
struct ureg normal = get_eye_normal(p);
struct ureg lit = get_temp(p);
struct ureg dots = get_temp(p);
struct ureg _col0 = undef, _col1 = undef;
struct ureg _bfc0 = undef, _bfc1 = undef;
GLuint i;
for (i = 0; i < MAX_LIGHTS; i++)
if (p->state->unit[i].light_enabled)
nr_lights++;
set_material_flags(p);
{
struct ureg shininess = get_material(p, 0, STATE_SHININESS);
emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X));
release_temp(p, shininess);
_col0 = make_temp(p, get_scenecolor(p, 0));
if (separate)
_col1 = make_temp(p, get_identity_param(p));
else
_col1 = _col0;
}
if (twoside) {
struct ureg shininess = get_material(p, 1, STATE_SHININESS);
emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
ureg_negate(swizzle1(shininess,X)));
release_temp(p, shininess);
_bfc0 = make_temp(p, get_scenecolor(p, 1));
if (separate)
_bfc1 = make_temp(p, get_identity_param(p));
else
_bfc1 = _bfc0;
}
/* If no lights, still need to emit the scenecolor.
*/
/* KW: changed to do this always - v1.17 "Fix lighting alpha result"?
*/
if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
{
struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
emit_op1(p, OPCODE_MOV, res0, 0, _col0);
if (twoside) {
struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
}
}
if (separate && (p->state->fragprog_inputs_read & FRAG_BIT_COL1)) {
struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
emit_op1(p, OPCODE_MOV, res1, 0, _col1);
if (twoside) {
struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
}
}
if (nr_lights == 0) {
release_temps(p);
return;
}
for (i = 0; i < MAX_LIGHTS; i++) {
if (p->state->unit[i].light_enabled) {
struct ureg half = undef;
struct ureg att = undef, VPpli = undef;
count++;
if (p->state->unit[i].light_eyepos3_is_zero) {
/* Can used precomputed constants in this case.
* Attenuation never applies to infinite lights.
*/
VPpli = register_param3(p, STATE_LIGHT, i,
STATE_POSITION_NORMALIZED);
if (p->state->light_local_viewer) {
struct ureg eye_hat = get_eye_position_normalized(p);
half = get_temp(p);
emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
emit_normalize_vec3(p, half, half);
} else {
half = register_param3(p, STATE_LIGHT, i, STATE_HALF_VECTOR);
}
}
else {
struct ureg Ppli = register_param3(p, STATE_LIGHT, i,
STATE_POSITION);
struct ureg V = get_eye_position(p);
struct ureg dist = get_temp(p);
VPpli = get_temp(p);
half = get_temp(p);
/* Calulate VPpli vector
*/
emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
/* Normalize VPpli. The dist value also used in
* attenuation below.
*/
emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
emit_op1(p, OPCODE_RSQ, dist, 0, dist);
emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
/* Calculate attenuation:
*/
if (!p->state->unit[i].light_spotcutoff_is_180 ||
p->state->unit[i].light_attenuated) {
att = calculate_light_attenuation(p, i, VPpli, dist);
}
/* Calculate viewer direction, or use infinite viewer:
*/
if (p->state->light_local_viewer) {
struct ureg eye_hat = get_eye_position_normalized(p);
emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
}
else {
struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
}
emit_normalize_vec3(p, half, half);
release_temp(p, dist);
}
/* Calculate dot products:
*/
emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
/* Front face lighting:
*/
{
struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
struct ureg res0, res1;
GLuint mask0, mask1;
emit_op1(p, OPCODE_LIT, lit, 0, dots);
if (!is_undef(att))
emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
mask0 = 0;
mask1 = 0;
res0 = _col0;
res1 = _col1;
if (count == nr_lights) {
if (separate) {
mask0 = WRITEMASK_XYZ;
mask1 = WRITEMASK_XYZ;
if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
res0 = register_output( p, VERT_RESULT_COL0 );
if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
res1 = register_output( p, VERT_RESULT_COL1 );
}
else {
mask1 = WRITEMASK_XYZ;
if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
res1 = register_output( p, VERT_RESULT_COL0 );
}
}
emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), 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);
release_temp(p, ambient);
release_temp(p, diffuse);
release_temp(p, specular);
}
/* Back face lighting:
*/
if (twoside) {
struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
struct ureg res0, res1;
GLuint mask0, mask1;
emit_op1(p, OPCODE_LIT, lit, 0, ureg_negate(swizzle(dots,X,Y,W,Z)));
if (!is_undef(att))
emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
mask0 = 0;
mask1 = 0;
res0 = _bfc0;
res1 = _bfc1;
if (count == nr_lights) {
if (separate) {
mask0 = WRITEMASK_XYZ;
mask1 = WRITEMASK_XYZ;
if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
res0 = register_output( p, VERT_RESULT_BFC0 );
if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
res1 = register_output( p, VERT_RESULT_BFC1 );
}
else {
mask1 = WRITEMASK_XYZ;
if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
res1 = register_output( p, VERT_RESULT_BFC0 );
}
}
emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), 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);
release_temp(p, ambient);
release_temp(p, diffuse);
release_temp(p, specular);
}
release_temp(p, half);
release_temp(p, VPpli);
release_temp(p, att);
}
}
release_temps( p );
}
