GLuint i915_emit_texld( struct i915_fragment_program *p,
GLuint dest,
GLuint destmask,
GLuint sampler,
GLuint coord,
GLuint op )
{
if (coord != UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord))) {
/* No real way to work around this in the general case - need to
* allocate and declare a new temporary register (a utemp won't
* do). Will fallback for now.
*/
i915_program_error(p, "Can't (yet) swizzle TEX arguments");
return 0;
}
/* Don't worry about saturate as we only support
*/
if (destmask != A0_DEST_CHANNEL_ALL) {
GLuint tmp = i915_get_utemp(p);
i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op );
i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
return dest;
}
else {
assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
p->nr_tex_indirect++;
}
*(p->csr++) = (op |
T0_DEST( dest ) |
T0_SAMPLER( sampler ));
*(p->csr++) = T1_ADDRESS_REG( coord );
*(p->csr++) = T2_MBZ;
p->nr_tex_insn++;
return dest;
}
}
static void enable_texture_blend_replace( i915ContextPtr i915 )
{
static const GLuint prog[] = {
_3DSTATE_PIXEL_SHADER_PROGRAM,
/* Declare the sampler:
*/
(D0_DCL |
D0_DECL_REG( REG_S(0) ) |
D0_SAMPLE_TYPE_2D |
D0_CHANNEL_NONE),
D1_MBZ,
D2_MBZ,
/* Declare the interpolated texture coordinate:
*/
(D0_DCL |
D0_DECL_REG( REG_T_TEX(0) ) |
D0_CHANNEL_ALL),
D1_MBZ,
D2_MBZ,
/* output-color = texld(sample0, texcoord0)
*/
(T0_TEXLD |
T0_DEST_REG( REG_OC ) |
T0_SAMPLER( 0 )),
T1_ADDRESS_REG(REG_TYPE_T, 0),
T2_MBZ
};
memcpy( i915->meta.Program, prog, sizeof(prog) );
i915->meta.ProgramSize = sizeof(prog) / sizeof(*prog);
i915->meta.Program[0] |= i915->meta.ProgramSize - 2;
i915->meta.emitted &= ~I915_UPLOAD_PROGRAM;
}
GLuint i915_emit_texld( struct i915_fragment_program *p,
GLuint live_regs,
GLuint dest,
GLuint destmask,
GLuint sampler,
GLuint coord,
GLuint op )
{
if (coord != UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord))) {
/* With the help of the "needed registers" table created earlier, pick
* a register we can MOV the swizzled TC to (since TEX doesn't support
* swizzled sources) */
GLuint swizCoord = get_free_rreg(p, live_regs);
if (swizCoord == UREG_BAD)
return 0;
i915_emit_arith( p, A0_MOV, swizCoord, A0_DEST_CHANNEL_ALL, 0, coord, 0, 0 );
coord = swizCoord;
}
/* Don't worry about saturate as we only support texture formats
* that are always in the 0..1 range.
*/
if (destmask != A0_DEST_CHANNEL_ALL) {
GLuint tmp = i915_get_utemp(p);
i915_emit_texld( p, 0, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op );
i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
return dest;
}
else {
assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
/* Can't use unsaved temps for coords, as the phase boundary would result
* in the contents becoming undefined.
*/
assert(GET_UREG_TYPE(coord) != REG_TYPE_U);
if ((GET_UREG_TYPE(coord) != REG_TYPE_R) &&
(GET_UREG_TYPE(coord) != REG_TYPE_OC) &&
(GET_UREG_TYPE(coord) != REG_TYPE_OD) &&
(GET_UREG_TYPE(coord) != REG_TYPE_T)) {
GLuint tmpCoord = get_free_rreg(p, live_regs);
if (tmpCoord == UREG_BAD)
return 0;
i915_emit_arith(p, A0_MOV, tmpCoord, A0_DEST_CHANNEL_ALL, 0, coord, 0, 0);
coord = tmpCoord;
}
/* Output register being oC or oD defines a phase boundary */
if (GET_UREG_TYPE(dest) == REG_TYPE_OC ||
GET_UREG_TYPE(dest) == REG_TYPE_OD)
p->nr_tex_indirect++;
/* Reading from an r# register whose contents depend on output of the
* current phase defines a phase boundary.
*/
if (GET_UREG_TYPE(coord) == REG_TYPE_R &&
p->register_phases[GET_UREG_NR(coord)] == p->nr_tex_indirect)
p->nr_tex_indirect++;
*(p->csr++) = (op |
T0_DEST( dest ) |
T0_SAMPLER( sampler ));
*(p->csr++) = T1_ADDRESS_REG( coord );
*(p->csr++) = T2_MBZ;
if (GET_UREG_TYPE(dest) == REG_TYPE_R)
p->register_phases[GET_UREG_NR(dest)] = p->nr_tex_indirect;
p->nr_tex_insn++;
return dest;
}
}
/**
* Emit a texture load or texkill instruction.
* \param dest the dest i915 register
* \param destmask the dest register writemask
* \param sampler the i915 sampler register
* \param coord the i915 source texcoord operand
* \param opcode the instruction opcode
*/
uint i915_emit_texld( struct i915_fp_compile *p,
uint dest,
uint destmask,
uint sampler,
uint coord,
uint opcode,
uint num_coord )
{
const uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
int temp = -1;
uint ignore = 0;
/* Eliminate the useless texture coordinates. Otherwise we end up generating
* a swizzle for no reason below. */
switch(num_coord) {
case 0:
/* Ignore x */
ignore |= (0xf << UREG_CHANNEL_X_SHIFT);
case 1:
/* Ignore y */
ignore |= (0xf << UREG_CHANNEL_Y_SHIFT);
case 2:
/* Ignore z */
ignore |= (0xf << UREG_CHANNEL_Z_SHIFT);
case 3:
/* Ignore w */
ignore |= (0xf << UREG_CHANNEL_W_SHIFT);
}
if ( (coord &~ignore ) != (k & ~ignore) ) {
/* texcoord is swizzled or negated. Need to allocate a new temporary
* register (a utemp / unpreserved temp) won't do.
*/
uint tempReg;
temp = i915_get_temp(p); /* get temp reg index */
tempReg = UREG(REG_TYPE_R, temp); /* make i915 register */
i915_emit_arith( p, A0_MOV,
tempReg, A0_DEST_CHANNEL_ALL, /* dest reg, writemask */
0, /* saturate */
coord, 0, 0 ); /* src0, src1, src2 */
/* new src texcoord is tempReg */
coord = tempReg;
}
/* Don't worry about saturate as we only support
*/
if (destmask != A0_DEST_CHANNEL_ALL) {
/* if not writing to XYZW... */
uint tmp = i915_get_utemp(p);
i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, opcode, num_coord );
i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
/* XXX release utemp here? */
}
else {
assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
assert(dest == UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
/* Output register being oC or oD defines a phase boundary */
if (GET_UREG_TYPE(dest) == REG_TYPE_OC ||
GET_UREG_TYPE(dest) == REG_TYPE_OD)
p->nr_tex_indirect++;
/* Reading from an r# register whose contents depend on output of the
* current phase defines a phase boundary.
*/
if (GET_UREG_TYPE(coord) == REG_TYPE_R &&
p->register_phases[GET_UREG_NR(coord)] == p->nr_tex_indirect)
p->nr_tex_indirect++;
if (p->csr< p->program + I915_PROGRAM_SIZE) {
*(p->csr++) = (opcode |
T0_DEST( dest ) |
T0_SAMPLER( sampler ));
*(p->csr++) = T1_ADDRESS_REG( coord );
*(p->csr++) = T2_MBZ;
}
else
i915_program_error(p, "Out of instructions");
if (GET_UREG_TYPE(dest) == REG_TYPE_R)
p->register_phases[GET_UREG_NR(dest)] = p->nr_tex_indirect;
p->nr_tex_insn++;
}
if (temp >= 0)
i915_release_temp(p, temp);
return dest;
}