static GLuint get_dest( struct i915_fragment_program *p, int unit )
{
if (p->ctx->_TriangleCaps & DD_SEPARATE_SPECULAR)
return i915_get_temp( p );
else if (unit != p->last_tex_stage)
return i915_get_temp( p );
else
return UREG(REG_TYPE_OC, 0);
}
static GLuint emit_combine_source( struct i915_fragment_program *p,
GLuint mask,
GLuint unit,
GLenum source,
GLenum operand )
{
GLuint arg, src;
src = get_source(p, source, unit);
switch (operand) {
case GL_ONE_MINUS_SRC_COLOR:
/* Get unused tmp,
* Emit tmp = 1.0 + arg.-x-y-z-w
*/
arg = i915_get_temp( p );
return i915_emit_arith( p, A0_ADD, arg, mask, 0,
swizzle(src, ONE, ONE, ONE, ONE ),
negate(src, 1,1,1,1), 0);
case GL_SRC_ALPHA:
if (mask == A0_DEST_CHANNEL_W)
return src;
else
return swizzle( src, W, W, W, W );
case GL_ONE_MINUS_SRC_ALPHA:
/* Get unused tmp,
* Emit tmp = 1.0 + arg.-w-w-w-w
*/
arg = i915_get_temp( p );
return i915_emit_arith( p, A0_ADD, arg, mask, 0,
swizzle(src, ONE, ONE, ONE, ONE ),
negate( swizzle(src,W,W,W,W), 1,1,1,1), 0);
case GL_SRC_COLOR:
default:
return src;
}
}
static GLuint get_source( struct i915_fragment_program *p,
GLenum src, GLuint unit )
{
switch (src) {
case GL_TEXTURE:
if (p->src_texture == UREG_BAD) {
/* TODO: Use D0_CHANNEL_XY where possible.
*/
GLuint dim = translate_tex_src_bit( p, p->ctx->Texture.Unit[unit]._ReallyEnabled);
GLuint sampler = i915_emit_decl(p, REG_TYPE_S, unit, dim);
GLuint texcoord = i915_emit_decl(p, REG_TYPE_T, unit, D0_CHANNEL_ALL);
GLuint tmp = i915_get_temp( p );
GLuint op = T0_TEXLD;
if (p->VB->TexCoordPtr[unit]->size == 4)
op = T0_TEXLDP;
p->src_texture = i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL,
sampler, texcoord, op );
}
return p->src_texture;
/* Crossbar: */
case GL_TEXTURE0:
case GL_TEXTURE1:
case GL_TEXTURE2:
case GL_TEXTURE3:
case GL_TEXTURE4:
case GL_TEXTURE5:
case GL_TEXTURE6:
case GL_TEXTURE7: {
return UREG_BAD;
}
case GL_CONSTANT:
return i915_emit_const4fv( p, p->ctx->Texture.Unit[unit].EnvColor );
case GL_PRIMARY_COLOR:
return i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
case GL_PREVIOUS:
default:
i915_emit_decl(p,
GET_UREG_TYPE(p->src_previous),
GET_UREG_NR(p->src_previous), D0_CHANNEL_ALL);
return p->src_previous;
}
}
/**
* 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;
}
static GLuint emit_combine( struct i915_fragment_program *p,
GLuint dest,
GLuint mask,
GLuint saturate,
GLuint unit,
GLenum mode,
const GLenum *source,
const GLenum *operand)
{
int tmp, src[3], nr = nr_args(mode);
int i;
for (i = 0; i < nr; i++)
src[i] = emit_combine_source( p, mask, unit, source[i], operand[i] );
switch (mode) {
case GL_REPLACE:
if (mask == A0_DEST_CHANNEL_ALL && !saturate)
return src[0];
else
return i915_emit_arith( p, A0_MOV, dest, mask, saturate, src[0], 0, 0 );
case GL_MODULATE:
return i915_emit_arith( p, A0_MUL, dest, mask, saturate,
src[0], src[1], 0 );
case GL_ADD:
return i915_emit_arith( p, A0_ADD, dest, mask, saturate,
src[0], src[1], 0 );
case GL_ADD_SIGNED:
/* tmp = arg0 + arg1
* result = tmp + -.5
*/
tmp = i915_emit_const1f(p, .5);
tmp = negate(swizzle(tmp,X,X,X,X),1,1,1,1);
i915_emit_arith( p, A0_ADD, dest, mask, 0, src[0], src[1], 0 );
i915_emit_arith( p, A0_ADD, dest, mask, saturate, dest, tmp, 0 );
return dest;
case GL_INTERPOLATE: /* TWO INSTRUCTIONS */
/* Arg0 * (Arg2) + Arg1 * (1-Arg2)
*
* Arg0*Arg2 + Arg1 - Arg1Arg2
*
* tmp = Arg0*Arg2 + Arg1,
* result = (-Arg1)Arg2 + tmp
*/
tmp = i915_get_temp( p );
i915_emit_arith( p, A0_MAD, tmp, mask, 0, src[0], src[2], src[1] );
i915_emit_arith( p, A0_MAD, dest, mask, saturate,
negate(src[1], 1,1,1,1), src[2], tmp );
return dest;
case GL_SUBTRACT:
/* negate src[1] */
return i915_emit_arith( p, A0_ADD, dest, mask, saturate, src[0],
negate(src[1],1,1,1,1), 0 );
case GL_DOT3_RGBA:
case GL_DOT3_RGBA_EXT:
case GL_DOT3_RGB_EXT:
case GL_DOT3_RGB: {
GLuint tmp0 = i915_get_temp( p );
GLuint tmp1 = i915_get_temp( p );
GLuint neg1 = negate(swizzle(i915_emit_const1f(p, 1),X,X,X,X), 1,1,1,1);
GLuint two = swizzle(i915_emit_const1f(p, 2),X,X,X,X);
i915_emit_arith( p, A0_MAD, tmp0, A0_DEST_CHANNEL_ALL, 0,
two, src[0], neg1);
if (src[0] == src[1])
tmp1 = tmp0;
else
i915_emit_arith( p, A0_MAD, tmp1, A0_DEST_CHANNEL_ALL, 0,
two, src[1], neg1);
i915_emit_arith( p, A0_DP3, dest, mask, saturate, tmp0, tmp1, 0);
return dest;
}
default:
return src[0];
}
}