static void emit_sop(struct brw_wm_compile *c,
struct prog_instruction *inst, GLuint cond)
{
struct brw_compile *p = &c->func;
GLuint mask = inst->DstReg.WriteMask;
struct brw_reg dst, src0, src1;
int i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
dst = get_dst_reg(c, inst, i, 1);
src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
brw_push_insn_state(p);
brw_CMP(p, brw_null_reg(), cond, src0, src1);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_MOV(p, dst, brw_imm_f(0.0));
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
brw_MOV(p, dst, brw_imm_f(1.0));
brw_pop_insn_state(p);
}
}
}
/**
* Sets VERT_RESULT_FOGC.Y for gl_FrontFacing
*
* This is currently executed if the fragment program uses VERT_RESULT_FOGC
* at all, but this could be eliminated with a scan of the FP contents.
*/
static void
do_front_facing( struct brw_sf_compile *c )
{
struct brw_compile *p = &c->func;
int i;
if (!have_attr(c, VERT_RESULT_FOGC))
return;
brw_push_insn_state(p);
brw_CMP(p, brw_null_reg(),
c->key.frontface_ccw ? BRW_CONDITIONAL_G : BRW_CONDITIONAL_L,
c->det, brw_imm_f(0));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
for (i = 0; i < 3; i++) {
struct brw_reg fogc = get_vert_attr(c, c->vert[i],FRAG_ATTRIB_FOGC);
brw_MOV(p, get_element(fogc, 1), brw_imm_f(0));
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
brw_MOV(p, get_element(fogc, 1), brw_imm_f(1));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
brw_pop_insn_state(p);
}
void brw_clip_tri_init_vertices( struct brw_clip_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */
/* Initial list of indices for incoming vertexes:
*/
brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK));
brw_CMP(p,
vec1(brw_null_reg()),
BRW_CONDITIONAL_EQ,
tmp0,
brw_imm_ud(_3DPRIM_TRISTRIP_REVERSE));
/* XXX: Is there an easier way to do this? Need to reverse every
* second tristrip element: Can ignore sometimes?
*/
brw_IF(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_element(c->reg.inlist, 0), brw_address(c->reg.vertex[1]) );
brw_MOV(p, get_element(c->reg.inlist, 1), brw_address(c->reg.vertex[0]) );
if (c->need_direction)
brw_MOV(p, c->reg.dir, brw_imm_f(-1));
}
brw_ELSE(p);
{
brw_MOV(p, get_element(c->reg.inlist, 0), brw_address(c->reg.vertex[0]) );
brw_MOV(p, get_element(c->reg.inlist, 1), brw_address(c->reg.vertex[1]) );
if (c->need_direction)
brw_MOV(p, c->reg.dir, brw_imm_f(1));
}
brw_ENDIF(p);
brw_MOV(p, get_element(c->reg.inlist, 2), brw_address(c->reg.vertex[2]) );
brw_MOV(p, brw_vec8_grf(c->reg.outlist.nr, 0), brw_imm_f(0));
brw_MOV(p, c->reg.nr_verts, brw_imm_ud(3));
}
static void emit_lit(struct brw_wm_compile *c,
const struct brw_reg *dst,
GLuint mask,
const struct brw_reg *arg0)
{
struct brw_compile *p = &c->func;
assert((mask & WRITEMASK_XW) == 0);
if (mask & WRITEMASK_Y) {
brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
brw_MOV(p, dst[1], arg0[0]);
brw_set_saturate(p, 0);
}
if (mask & WRITEMASK_Z) {
emit_math2(c, BRW_MATH_FUNCTION_POW,
&dst[2],
WRITEMASK_X | (mask & SATURATE),
&arg0[1],
&arg0[3]);
}
/* Ordinarily you'd use an iff statement to skip or shortcircuit
* some of the POW calculations above, but 16-wide iff statements
* seem to lock c1 hardware, so this is a nasty workaround:
*/
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_LE, arg0[0], brw_imm_f(0));
{
if (mask & WRITEMASK_Y)
brw_MOV(p, dst[1], brw_imm_f(0));
if (mask & WRITEMASK_Z)
brw_MOV(p, dst[2], brw_imm_f(0));
}
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
static void merge_edgeflags( struct brw_clip_compile *c )
{
struct brw_compile *p = &c->func;
const struct brw_context *brw = p->brw;
struct brw_reg tmp0 = get_element_ud(c->reg.tmp0, 0);
brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK));
brw_CMP(p,
vec1(brw_null_reg()),
BRW_CONDITIONAL_EQ,
tmp0,
brw_imm_ud(_3DPRIM_POLYGON));
/* Get away with using reg.vertex because we know that this is not
* a _3DPRIM_TRISTRIP_REVERSE:
*/
brw_IF(p, BRW_EXECUTE_1);
{
brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<8));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_EQ);
brw_MOV(p, byte_offset(c->reg.vertex[0],
brw_varying_to_offset(&c->vue_map,
VARYING_SLOT_EDGE)),
brw_imm_f(0));
brw_inst_set_pred_control(brw, brw_last_inst, BRW_PREDICATE_NORMAL);
brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<9));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_EQ);
brw_MOV(p, byte_offset(c->reg.vertex[2],
brw_varying_to_offset(&c->vue_map,
VARYING_SLOT_EDGE)),
brw_imm_f(0));
brw_inst_set_pred_control(brw, brw_last_inst, BRW_PREDICATE_NORMAL);
}
brw_ENDIF(p);
}
static void copy_bfc( struct brw_clip_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_instruction *ccw;
GLuint conditional;
/* Do we have any colors to copy?
*/
if (!(c->offset[VERT_RESULT_COL0] && c->offset[VERT_RESULT_BFC0]) &&
!(c->offset[VERT_RESULT_COL1] && c->offset[VERT_RESULT_BFC1]))
return;
/* In some wierd degnerate cases we can end up testing the
* direction twice, once for culling and once for bfc copying. Oh
* well, that's what you get for setting wierd GL state.
*/
if (c->key.copy_bfc_ccw)
conditional = BRW_CONDITIONAL_GE;
else
conditional = BRW_CONDITIONAL_L;
brw_CMP(p,
vec1(brw_null_reg()),
conditional,
get_element(c->reg.dir, 2),
brw_imm_f(0));
ccw = brw_IF(p, BRW_EXECUTE_1);
{
GLuint i;
for (i = 0; i < 3; i++) {
if (c->offset[VERT_RESULT_COL0] && c->offset[VERT_RESULT_BFC0])
brw_MOV(p,
byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_COL0]),
byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_BFC0]));
if (c->offset[VERT_RESULT_COL1] && c->offset[VERT_RESULT_BFC1])
brw_MOV(p,
byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_COL1]),
byte_offset(c->reg.vertex[i], c->offset[VERT_RESULT_BFC1]));
}
}
brw_ENDIF(p, ccw);
}
/* Kill pixel - set execution mask to zero for those pixels which
* fail.
*/
static void emit_kil( struct brw_wm_compile *c,
struct brw_reg *arg0)
{
struct brw_compile *p = &c->func;
struct brw_reg r0uw = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
GLuint i;
/* XXX - usually won't need 4 compares!
*/
for (i = 0; i < 4; i++) {
brw_push_insn_state(p);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_GE, arg0[i], brw_imm_f(0));
brw_set_predicate_control_flag_value(p, 0xff);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_AND(p, r0uw, brw_flag_reg(), r0uw);
brw_pop_insn_state(p);
}
}
/**
* Load a GPR from scratch memory.
*/
static void emit_unspill( struct brw_wm_compile *c,
struct brw_reg reg,
GLuint slot )
{
struct brw_compile *p = &c->func;
/* Slot 0 is the undef value.
*/
if (slot == 0) {
brw_MOV(p, reg, brw_imm_f(0));
return;
}
/*
mov (1) r0.2<1>:d 0x000000c0:d { Align1 NoMask }
send (16) r110.0<1>:uw m1 r0.0<8;8,1>:uw 0x041243ff:ud { Align1 }
*/
brw_oword_block_read(p, vec16(reg), brw_message_reg(1), 2, slot);
}
void emit_cmp(struct brw_compile *p,
const struct brw_reg *dst,
GLuint mask,
const struct brw_reg *arg0,
const struct brw_reg *arg1,
const struct brw_reg *arg2)
{
GLuint i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0[i], brw_imm_f(0));
brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
brw_SEL(p, dst[i], arg1[i], arg2[i]);
brw_set_saturate(p, 0);
brw_set_predicate_control_flag_value(p, 0xff);
}
}
}
static void emit_points(struct brw_clip_compile *c,
bool do_offset )
{
struct brw_compile *p = &c->func;
const struct brw_context *brw = p->brw;
struct brw_indirect v0 = brw_indirect(0, 0);
struct brw_indirect v0ptr = brw_indirect(2, 0);
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
/* draw if edgeflag != 0
*/
brw_CMP(p,
vec1(brw_null_reg()), BRW_CONDITIONAL_NZ,
deref_1f(v0, brw_varying_to_offset(&c->vue_map,
VARYING_SLOT_EDGE)),
brw_imm_f(0));
brw_IF(p, BRW_EXECUTE_1);
{
if (do_offset)
apply_one_offset(c, v0);
brw_clip_emit_vue(c, v0, BRW_URB_WRITE_ALLOCATE_COMPLETE,
(_3DPRIM_POINTLIST << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_START | URB_WRITE_PRIM_END);
}
brw_ENDIF(p);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_NZ);
}
brw_WHILE(p);
brw_inst_set_pred_control(brw, brw_last_inst, BRW_PREDICATE_NORMAL);
}
static void do_twoside_color( struct brw_sf_compile *c )
{
struct brw_compile *p = &c->func;
GLuint backface_conditional = c->key.frontface_ccw ? BRW_CONDITIONAL_G : BRW_CONDITIONAL_L;
/* Already done in clip program:
*/
if (c->key.primitive == SF_UNFILLED_TRIS)
return;
/* XXX: What happens if BFC isn't present? This could only happen
* for user-supplied vertex programs, as t_vp_build.c always does
* the right thing.
*/
if (!(have_attr(c, VERT_RESULT_COL0) && have_attr(c, VERT_RESULT_BFC0)) &&
!(have_attr(c, VERT_RESULT_COL1) && have_attr(c, VERT_RESULT_BFC1)))
return;
/* Need to use BRW_EXECUTE_4 and also do an 4-wide compare in order
* to get all channels active inside the IF. In the clipping code
* we run with NoMask, so it's not an option and we can use
* BRW_EXECUTE_1 for all comparisions.
*/
brw_push_insn_state(p);
brw_CMP(p, vec4(brw_null_reg()), backface_conditional, c->det, brw_imm_f(0));
brw_IF(p, BRW_EXECUTE_4);
{
switch (c->nr_verts) {
case 3:
copy_bfc(c, c->vert[2]);
case 2:
copy_bfc(c, c->vert[1]);
case 1:
copy_bfc(c, c->vert[0]);
}
}
brw_ENDIF(p);
brw_pop_insn_state(p);
}
static void emit_points(struct brw_clip_compile *c,
bool do_offset )
{
struct brw_compile *p = &c->func;
struct brw_indirect v0 = brw_indirect(0, 0);
struct brw_indirect v0ptr = brw_indirect(2, 0);
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
/* draw if edgeflag != 0
*/
brw_CMP(p,
vec1(brw_null_reg()), BRW_CONDITIONAL_NZ,
deref_1f(v0, brw_vert_result_to_offset(&c->vue_map,
VARYING_SLOT_EDGE)),
brw_imm_f(0));
brw_IF(p, BRW_EXECUTE_1);
{
if (do_offset)
apply_one_offset(c, v0);
brw_clip_emit_vue(c, v0, 1, 0,
(_3DPRIM_POINTLIST << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_START | URB_WRITE_PRIM_END);
}
brw_ENDIF(p);
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p);
}
static void emit_points(struct brw_clip_compile *c,
GLboolean do_offset )
{
struct brw_compile *p = &c->func;
struct brw_instruction *loop;
struct brw_instruction *draw_point;
struct brw_indirect v0 = brw_indirect(0, 0);
struct brw_indirect v0ptr = brw_indirect(2, 0);
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
loop = brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
/* draw if edgeflag != 0
*/
brw_CMP(p,
vec1(brw_null_reg()), BRW_CONDITIONAL_NZ,
deref_1f(v0, c->offset[VERT_RESULT_EDGE]),
brw_imm_f(0));
draw_point = brw_IF(p, BRW_EXECUTE_1);
{
if (do_offset)
apply_one_offset(c, v0);
brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_POINTLIST << 2) | R02_PRIM_START | R02_PRIM_END);
}
brw_ENDIF(p, draw_point);
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p, loop);
}
static void emit_unspill( struct brw_wm_compile *c,
struct brw_reg reg,
GLuint slot )
{
struct brw_compile *p = &c->func;
/* Slot 0 is the undef value.
*/
if (slot == 0) {
brw_MOV(p, reg, brw_imm_f(0));
return;
}
/*
mov (1) r0.2<1>:d 0x000000c0:d { Align1 NoMask }
send (16) r110.0<1>:uw m1 r0.0<8;8,1>:uw 0x041243ff:ud { Align1 }
*/
brw_dp_READ_16(p,
retype(vec16(reg), BRW_REGISTER_TYPE_UW),
1,
slot);
}
/**
* Loads the clip distance for a vertex into `dst`, and ends with
* a comparison of it to zero with the condition `cond`.
*
* - If using a fixed plane, the distance is dot(hpos, plane).
* - If using a user clip plane, the distance is directly available in the vertex.
*/
static inline void
load_clip_distance(struct brw_clip_compile *c, struct brw_indirect vtx,
struct brw_reg dst, GLuint hpos_offset, int cond)
{
struct brw_codegen *p = &c->func;
dst = vec4(dst);
brw_AND(p, vec1(brw_null_reg()), c->reg.vertex_src_mask, brw_imm_ud(1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_IF(p, BRW_EXECUTE_1);
{
struct brw_indirect temp_ptr = brw_indirect(7, 0);
brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx), c->reg.clipdistance_offset);
brw_MOV(p, vec1(dst), deref_1f(temp_ptr, 0));
}
brw_ELSE(p);
{
brw_MOV(p, dst, deref_4f(vtx, hpos_offset));
brw_DP4(p, dst, dst, c->reg.plane_equation);
}
brw_ENDIF(p);
brw_CMP(p, brw_null_reg(), cond, vec1(dst), brw_imm_f(0.0f));
}
static void emit_lrp(struct brw_wm_compile *c,
struct prog_instruction *inst)
{
struct brw_compile *p = &c->func;
GLuint mask = inst->DstReg.WriteMask;
struct brw_reg dst, tmp1, tmp2, src0, src1, src2;
int i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
dst = get_dst_reg(c, inst, i, 1);
src0 = get_src_reg(c, &inst->SrcReg[0], i, 1);
src1 = get_src_reg(c, &inst->SrcReg[1], i, 1);
if (src1.nr == dst.nr) {
tmp1 = alloc_tmp(c);
brw_MOV(p, tmp1, src1);
} else
tmp1 = src1;
src2 = get_src_reg(c, &inst->SrcReg[2], i, 1);
if (src2.nr == dst.nr) {
tmp2 = alloc_tmp(c);
brw_MOV(p, tmp2, src2);
} else
tmp2 = src2;
brw_ADD(p, dst, negate(src0), brw_imm_f(1.0));
brw_MUL(p, brw_null_reg(), dst, tmp2);
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
brw_MAC(p, dst, src0, tmp1);
brw_set_saturate(p, 0);
}
release_tmps(c);
}
}
void emit_lrp(struct brw_compile *p,
const struct brw_reg *dst,
GLuint mask,
const struct brw_reg *arg0,
const struct brw_reg *arg1,
const struct brw_reg *arg2)
{
GLuint i;
/* Uses dst as a temporary:
*/
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
/* Can I use the LINE instruction for this?
*/
brw_ADD(p, dst[i], negate(arg0[i]), brw_imm_f(1.0));
brw_MUL(p, brw_null_reg(), dst[i], arg2[i]);
brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
brw_MAC(p, dst[i], arg0[i], arg1[i]);
brw_set_saturate(p, 0);
}
}
}
/* Line clipping, more or less following the following algorithm:
*
* for (p=0;p<MAX_PLANES;p++) {
* if (clipmask & (1 << p)) {
* GLfloat dp0 = DOTPROD( vtx0, plane[p] );
* GLfloat dp1 = DOTPROD( vtx1, plane[p] );
*
* if (dp1 < 0.0f) {
* GLfloat t = dp1 / (dp1 - dp0);
* if (t > t1) t1 = t;
* } else {
* GLfloat t = dp0 / (dp0 - dp1);
* if (t > t0) t0 = t;
* }
*
* if (t0 + t1 >= 1.0)
* return;
* }
* }
*
* interp( ctx, newvtx0, vtx0, vtx1, t0 );
* interp( ctx, newvtx1, vtx1, vtx0, t1 );
*
*/
static void clip_and_emit_line( struct brw_clip_compile *c )
{
struct brw_codegen *p = &c->func;
struct brw_indirect vtx0 = brw_indirect(0, 0);
struct brw_indirect vtx1 = brw_indirect(1, 0);
struct brw_indirect newvtx0 = brw_indirect(2, 0);
struct brw_indirect newvtx1 = brw_indirect(3, 0);
struct brw_indirect plane_ptr = brw_indirect(4, 0);
struct brw_reg v1_null_ud = retype(vec1(brw_null_reg()), BRW_REGISTER_TYPE_UD);
GLuint hpos_offset = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS);
GLint clipdist0_offset = c->key.nr_userclip
? brw_varying_to_offset(&c->vue_map, VARYING_SLOT_CLIP_DIST0)
: 0;
brw_MOV(p, get_addr_reg(vtx0), brw_address(c->reg.vertex[0]));
brw_MOV(p, get_addr_reg(vtx1), brw_address(c->reg.vertex[1]));
brw_MOV(p, get_addr_reg(newvtx0), brw_address(c->reg.vertex[2]));
brw_MOV(p, get_addr_reg(newvtx1), brw_address(c->reg.vertex[3]));
brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c));
/* Note: init t0, t1 together:
*/
brw_MOV(p, vec2(c->reg.t0), brw_imm_f(0));
brw_clip_init_planes(c);
brw_clip_init_clipmask(c);
/* -ve rhw workaround */
if (p->devinfo->has_negative_rhw_bug) {
brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2),
brw_imm_ud(1<<20));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(0x3f));
brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
}
/* Set the initial vertex source mask: The first 6 planes are the bounds
* of the view volume; the next 8 planes are the user clipping planes.
*/
brw_MOV(p, c->reg.vertex_src_mask, brw_imm_ud(0x3fc0));
/* Set the initial clipdistance offset to be 6 floats before gl_ClipDistance[0].
* We'll increment 6 times before we start hitting actual user clipping. */
brw_MOV(p, c->reg.clipdistance_offset, brw_imm_d(clipdist0_offset - 6*sizeof(float)));
brw_DO(p, BRW_EXECUTE_1);
{
/* if (planemask & 1)
*/
brw_AND(p, v1_null_ud, c->reg.planemask, brw_imm_ud(1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_IF(p, BRW_EXECUTE_1);
{
brw_AND(p, v1_null_ud, c->reg.vertex_src_mask, brw_imm_ud(1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_IF(p, BRW_EXECUTE_1);
{
/* user clip distance: just fetch the correct float from each vertex */
struct brw_indirect temp_ptr = brw_indirect(7, 0);
brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx0), c->reg.clipdistance_offset);
brw_MOV(p, c->reg.dp0, deref_1f(temp_ptr, 0));
brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx1), c->reg.clipdistance_offset);
brw_MOV(p, c->reg.dp1, deref_1f(temp_ptr, 0));
}
brw_ELSE(p);
{
/* fixed plane: fetch the hpos, dp4 against the plane. */
if (c->key.nr_userclip)
brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0));
else
brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0));
brw_DP4(p, vec4(c->reg.dp0), deref_4f(vtx0, hpos_offset), c->reg.plane_equation);
brw_DP4(p, vec4(c->reg.dp1), deref_4f(vtx1, hpos_offset), c->reg.plane_equation);
}
brw_ENDIF(p);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, vec1(c->reg.dp1), brw_imm_f(0.0f));
brw_IF(p, BRW_EXECUTE_1);
{
/*
* Both can be negative on GM965/G965 due to RHW workaround
* if so, this object should be rejected.
*/
if (p->devinfo->has_negative_rhw_bug) {
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_LE, c->reg.dp0, brw_imm_f(0.0));
brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_kill_thread(c);
}
brw_ENDIF(p);
}
brw_ADD(p, c->reg.t, c->reg.dp1, negate(c->reg.dp0));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp1);
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t1 );
brw_MOV(p, c->reg.t1, c->reg.t);
brw_inst_set_pred_control(p->devinfo, brw_last_inst,
BRW_PREDICATE_NORMAL);
}
brw_ELSE(p);
{
/* Coming back in. We know that both cannot be negative
* because the line would have been culled in that case.
*/
/* If both are positive, do nothing */
/* Only on GM965/G965 */
if (p->devinfo->has_negative_rhw_bug) {
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0));
brw_IF(p, BRW_EXECUTE_1);
}
{
brw_ADD(p, c->reg.t, c->reg.dp0, negate(c->reg.dp1));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp0);
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t0 );
brw_MOV(p, c->reg.t0, c->reg.t);
brw_inst_set_pred_control(p->devinfo, brw_last_inst,
BRW_PREDICATE_NORMAL);
}
if (p->devinfo->has_negative_rhw_bug) {
brw_ENDIF(p);
}
}
brw_ENDIF(p);
}
brw_ENDIF(p);
/* plane_ptr++;
*/
brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c));
/* while (planemask>>=1) != 0
*/
brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
brw_SHR(p, c->reg.vertex_src_mask, c->reg.vertex_src_mask, brw_imm_ud(1));
brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
brw_ADD(p, c->reg.clipdistance_offset, c->reg.clipdistance_offset, brw_imm_w(sizeof(float)));
brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
}
brw_WHILE(p);
brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
brw_ADD(p, c->reg.t, c->reg.t0, c->reg.t1);
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.t, brw_imm_f(1.0));
brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_interp_vertex(c, newvtx0, vtx0, vtx1, c->reg.t0, false);
brw_clip_interp_vertex(c, newvtx1, vtx1, vtx0, c->reg.t1, false);
brw_clip_emit_vue(c, newvtx0, BRW_URB_WRITE_ALLOCATE_COMPLETE,
(_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_START);
brw_clip_emit_vue(c, newvtx1, BRW_URB_WRITE_EOT_COMPLETE,
(_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_END);
}
brw_ENDIF(p);
brw_clip_kill_thread(c);
}
/***********************************************************************
* Output clipped polygon as an unfilled primitive:
*/
static void emit_lines(struct brw_clip_compile *c,
bool do_offset)
{
struct brw_compile *p = &c->func;
const struct brw_context *brw = p->brw;
struct brw_indirect v0 = brw_indirect(0, 0);
struct brw_indirect v1 = brw_indirect(1, 0);
struct brw_indirect v0ptr = brw_indirect(2, 0);
struct brw_indirect v1ptr = brw_indirect(3, 0);
/* Need a seperate loop for offset:
*/
if (do_offset) {
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
apply_one_offset(c, v0);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_G);
}
brw_WHILE(p);
brw_inst_set_pred_control(brw, brw_last_inst, BRW_PREDICATE_NORMAL);
}
/* v1ptr = &inlist[nr_verts]
* *v1ptr = v0
*/
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v0ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v1ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
brw_MOV(p, deref_1uw(v1ptr, 0), deref_1uw(v0ptr, 0));
brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_MOV(p, get_addr_reg(v1), deref_1uw(v0ptr, 2));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
/* draw edge if edgeflag != 0 */
brw_CMP(p,
vec1(brw_null_reg()), BRW_CONDITIONAL_NZ,
deref_1f(v0, brw_varying_to_offset(&c->vue_map,
VARYING_SLOT_EDGE)),
brw_imm_f(0));
brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_emit_vue(c, v0, BRW_URB_WRITE_ALLOCATE_COMPLETE,
(_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_START);
brw_clip_emit_vue(c, v1, BRW_URB_WRITE_ALLOCATE_COMPLETE,
(_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_END);
}
brw_ENDIF(p);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
brw_inst_set_cond_modifier(brw, brw_last_inst, BRW_CONDITIONAL_NZ);
}
brw_WHILE(p);
brw_inst_set_pred_control(brw, brw_last_inst, BRW_PREDICATE_NORMAL);
}
/* Interpolate between two vertices and put the result into a0.0.
* Increment a0.0 accordingly.
*/
void brw_clip_interp_vertex( struct brw_clip_compile *c,
struct brw_indirect dest_ptr,
struct brw_indirect v0_ptr, /* from */
struct brw_indirect v1_ptr, /* to */
struct brw_reg t0,
GLboolean force_edgeflag)
{
struct brw_compile *p = &c->func;
struct brw_reg tmp = get_tmp(c);
GLuint i;
/* Just copy the vertex header:
*/
/*
* After CLIP stage, only first 256 bits of the VUE are read
* back on IGDNG, so needn't change it
*/
brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
/* Iterate over each attribute (could be done in pairs?)
*/
for (i = 0; i < c->nr_attrs; i++) {
GLuint delta = i*16 + 32;
if (BRW_IS_IGDNG(p->brw))
delta = i * 16 + 32 * 3;
if (delta == c->offset[VERT_RESULT_EDGE]) {
if (force_edgeflag)
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
else
brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
}
else {
/* Interpolate:
*
* New = attr0 + t*attr1 - t*attr0
*/
brw_MUL(p,
vec4(brw_null_reg()),
deref_4f(v1_ptr, delta),
t0);
brw_MAC(p,
tmp,
negate(deref_4f(v0_ptr, delta)),
t0);
brw_ADD(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta),
tmp);
}
}
if (i & 1) {
GLuint delta = i*16 + 32;
if (BRW_IS_IGDNG(p->brw))
delta = i * 16 + 32 * 3;
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
}
release_tmp(c, tmp);
/* Recreate the projected (NDC) coordinate in the new vertex
* header:
*/
brw_clip_project_vertex(c, dest_ptr );
}
/***********************************************************************
* Output clipped polygon as an unfilled primitive:
*/
static void emit_lines(struct brw_clip_compile *c,
GLboolean do_offset)
{
struct brw_compile *p = &c->func;
struct brw_instruction *loop;
struct brw_instruction *draw_edge;
struct brw_indirect v0 = brw_indirect(0, 0);
struct brw_indirect v1 = brw_indirect(1, 0);
struct brw_indirect v0ptr = brw_indirect(2, 0);
struct brw_indirect v1ptr = brw_indirect(3, 0);
/* Need a seperate loop for offset:
*/
if (do_offset) {
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
loop = brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
apply_one_offset(c, v0);
brw_set_conditionalmod(p, BRW_CONDITIONAL_G);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p, loop);
}
/* v1ptr = &inlist[nr_verts]
* *v1ptr = v0
*/
brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v0ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v1ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
brw_MOV(p, deref_1uw(v1ptr, 0), deref_1uw(v0ptr, 0));
loop = brw_DO(p, BRW_EXECUTE_1);
{
brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
brw_MOV(p, get_addr_reg(v1), deref_1uw(v0ptr, 2));
brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
/* draw edge if edgeflag != 0 */
brw_CMP(p,
vec1(brw_null_reg()), BRW_CONDITIONAL_NZ,
deref_1f(v0, c->offset[VERT_RESULT_EDGE]),
brw_imm_f(0));
draw_edge = brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_START);
brw_clip_emit_vue(c, v1, 1, 0, (_3DPRIM_LINESTRIP << 2) | R02_PRIM_END);
}
brw_ENDIF(p, draw_edge);
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p, loop);
}
void brw_emit_point_sprite_setup( struct brw_sf_compile *c, GLboolean allocate)
{
struct brw_compile *p = &c->func;
GLuint i;
c->nr_verts = 1;
if (allocate)
alloc_regs(c);
copy_z_inv_w(c);
for (i = 0; i < c->nr_setup_regs; i++)
{
struct brw_sf_point_tex *tex = &c->point_attrs[c->idx_to_attr[2*i]];
struct brw_reg a0 = offset(c->vert[0], i);
GLushort pc, pc_persp, pc_linear;
GLboolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
if (pc_persp)
{
if (!tex->CoordReplace) {
brw_set_predicate_control_flag_value(p, pc_persp);
brw_MUL(p, a0, a0, c->inv_w[0]);
}
}
if (tex->CoordReplace) {
/* Caculate 1.0/PointWidth */
brw_math(&c->func,
c->tmp,
BRW_MATH_FUNCTION_INV,
BRW_MATH_SATURATE_NONE,
0,
c->dx0,
BRW_MATH_DATA_SCALAR,
BRW_MATH_PRECISION_FULL);
if (c->key.SpriteOrigin == GL_LOWER_LEFT) {
brw_MUL(p, c->m1Cx, c->tmp, c->inv_w[0]);
brw_MOV(p, vec1(suboffset(c->m1Cx, 1)), brw_imm_f(0.0));
brw_MUL(p, c->m2Cy, c->tmp, negate(c->inv_w[0]));
brw_MOV(p, vec1(suboffset(c->m2Cy, 0)), brw_imm_f(0.0));
} else {
brw_MUL(p, c->m1Cx, c->tmp, c->inv_w[0]);
brw_MOV(p, vec1(suboffset(c->m1Cx, 1)), brw_imm_f(0.0));
brw_MUL(p, c->m2Cy, c->tmp, c->inv_w[0]);
brw_MOV(p, vec1(suboffset(c->m2Cy, 0)), brw_imm_f(0.0));
}
} else {
brw_MOV(p, c->m1Cx, brw_imm_ud(0));
brw_MOV(p, c->m2Cy, brw_imm_ud(0));
}
{
brw_set_predicate_control_flag_value(p, pc);
if (tex->CoordReplace) {
if (c->key.SpriteOrigin == GL_LOWER_LEFT) {
brw_MUL(p, c->m3C0, c->inv_w[0], brw_imm_f(1.0));
brw_MOV(p, vec1(suboffset(c->m3C0, 0)), brw_imm_f(0.0));
}
else
brw_MOV(p, c->m3C0, brw_imm_f(0.0));
} else {
brw_MOV(p, c->m3C0, a0); /* constant value */
}
/* Copy m0..m3 to URB.
*/
brw_urb_WRITE(p,
brw_null_reg(),
0,
brw_vec8_grf(0, 0),
0, /* allocate */
1, /* used */
4, /* msg len */
0, /* response len */
last, /* eot */
last, /* writes complete */
i*4, /* urb destination offset */
BRW_URB_SWIZZLE_TRANSPOSE);
}
}
}
/* Interpolate between two vertices and put the result into a0.0.
* Increment a0.0 accordingly.
*/
void brw_clip_interp_vertex( struct brw_clip_compile *c,
struct brw_indirect dest_ptr,
struct brw_indirect v0_ptr, /* from */
struct brw_indirect v1_ptr, /* to */
struct brw_reg t0,
bool force_edgeflag)
{
struct brw_compile *p = &c->func;
struct brw_reg tmp = get_tmp(c);
GLuint slot;
/* Just copy the vertex header:
*/
/*
* After CLIP stage, only first 256 bits of the VUE are read
* back on Ironlake, so needn't change it
*/
brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
/* Iterate over each attribute (could be done in pairs?)
*/
for (slot = 0; slot < c->vue_map.num_slots; slot++) {
int varying = c->vue_map.slot_to_varying[slot];
GLuint delta = brw_vue_slot_to_offset(slot);
if (varying == VARYING_SLOT_EDGE) {
if (force_edgeflag)
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
else
brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
} else if (varying == VARYING_SLOT_PSIZ ||
varying == VARYING_SLOT_CLIP_DIST0 ||
varying == VARYING_SLOT_CLIP_DIST1) {
/* PSIZ doesn't need interpolation because it isn't used by the
* fragment shader. CLIP_DIST0 and CLIP_DIST1 don't need
* intepolation because on pre-GEN6, these are just placeholder VUE
* slots that don't perform any action.
*/
} else if (varying < VARYING_SLOT_MAX) {
/* This is a true vertex result (and not a special value for the VUE
* header), so interpolate:
*
* New = attr0 + t*attr1 - t*attr0
*/
brw_MUL(p,
vec4(brw_null_reg()),
deref_4f(v1_ptr, delta),
t0);
brw_MAC(p,
tmp,
negate(deref_4f(v0_ptr, delta)),
t0);
brw_ADD(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta),
tmp);
}
}
if (c->vue_map.num_slots % 2) {
GLuint delta = brw_vue_slot_to_offset(c->vue_map.num_slots);
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
}
release_tmp(c, tmp);
/* Recreate the projected (NDC) coordinate in the new vertex
* header:
*/
brw_clip_project_vertex(c, dest_ptr );
}
void emit_txb(struct brw_wm_compile *c,
struct brw_reg *dst,
GLuint dst_flags,
struct brw_reg *arg,
struct brw_reg depth_payload,
GLuint tex_idx,
GLuint sampler)
{
struct brw_compile *p = &c->func;
struct intel_context *intel = &p->brw->intel;
GLuint msgLength;
GLuint msg_type;
GLuint mrf_per_channel;
GLuint response_length;
struct brw_reg dst_retyped;
/* The G45 and older chipsets don't support 8-wide dispatch for LOD biased
* samples, so we'll use the 16-wide instruction, leave the second halves
* undefined, and trust the execution mask to keep the undefined pixels
* from mattering.
*/
if (c->dispatch_width == 16 || intel->gen < 5) {
if (intel->gen >= 5)
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_BIAS;
else
msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS;
mrf_per_channel = 2;
dst_retyped = retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW);
response_length = 8;
} else {
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_BIAS;
mrf_per_channel = 1;
dst_retyped = retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW);
response_length = 4;
}
/* Shadow ignored for txb. */
switch (tex_idx) {
case TEXTURE_1D_INDEX:
brw_MOV(p, brw_message_reg(2 + 0 * mrf_per_channel), arg[0]);
brw_MOV(p, brw_message_reg(2 + 1 * mrf_per_channel), brw_imm_f(0));
brw_MOV(p, brw_message_reg(2 + 2 * mrf_per_channel), brw_imm_f(0));
break;
case TEXTURE_2D_INDEX:
case TEXTURE_RECT_INDEX:
brw_MOV(p, brw_message_reg(2 + 0 * mrf_per_channel), arg[0]);
brw_MOV(p, brw_message_reg(2 + 1 * mrf_per_channel), arg[1]);
brw_MOV(p, brw_message_reg(2 + 2 * mrf_per_channel), brw_imm_f(0));
break;
case TEXTURE_3D_INDEX:
case TEXTURE_CUBE_INDEX:
brw_MOV(p, brw_message_reg(2 + 0 * mrf_per_channel), arg[0]);
brw_MOV(p, brw_message_reg(2 + 1 * mrf_per_channel), arg[1]);
brw_MOV(p, brw_message_reg(2 + 2 * mrf_per_channel), arg[2]);
break;
default:
/* unexpected target */
abort();
}
brw_MOV(p, brw_message_reg(2 + 3 * mrf_per_channel), arg[3]);
msgLength = 2 + 4 * mrf_per_channel - 1;
brw_SAMPLE(p,
dst_retyped,
1,
retype(depth_payload, BRW_REGISTER_TYPE_UW),
SURF_INDEX_TEXTURE(sampler),
sampler,
dst_flags & WRITEMASK_XYZW,
msg_type,
response_length,
msgLength,
1,
BRW_SAMPLER_SIMD_MODE_SIMD16,
BRW_SAMPLER_RETURN_FORMAT_FLOAT32);
}
void emit_tex(struct brw_wm_compile *c,
struct brw_reg *dst,
GLuint dst_flags,
struct brw_reg *arg,
struct brw_reg depth_payload,
GLuint tex_idx,
GLuint sampler,
bool shadow)
{
struct brw_compile *p = &c->func;
struct intel_context *intel = &p->brw->intel;
struct brw_reg dst_retyped;
GLuint cur_mrf = 2, response_length;
GLuint i, nr_texcoords;
GLuint emit;
GLuint msg_type;
GLuint mrf_per_channel;
GLuint simd_mode;
if (c->dispatch_width == 16) {
mrf_per_channel = 2;
response_length = 8;
dst_retyped = retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW);
simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD16;
} else {
mrf_per_channel = 1;
response_length = 4;
dst_retyped = retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW);
simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD8;
}
/* How many input regs are there?
*/
switch (tex_idx) {
case TEXTURE_1D_INDEX:
emit = WRITEMASK_X;
nr_texcoords = 1;
break;
case TEXTURE_2D_INDEX:
case TEXTURE_1D_ARRAY_INDEX:
case TEXTURE_RECT_INDEX:
emit = WRITEMASK_XY;
nr_texcoords = 2;
break;
case TEXTURE_3D_INDEX:
case TEXTURE_2D_ARRAY_INDEX:
case TEXTURE_CUBE_INDEX:
emit = WRITEMASK_XYZ;
nr_texcoords = 3;
break;
default:
/* unexpected target */
abort();
}
/* Pre-Ironlake, the 8-wide sampler always took u,v,r. */
if (intel->gen < 5 && c->dispatch_width == 8)
nr_texcoords = 3;
if (shadow) {
if (intel->gen < 7) {
/* For shadow comparisons, we have to supply u,v,r. */
nr_texcoords = 3;
} else {
/* On Ivybridge, the shadow comparitor comes first. Just load it. */
brw_MOV(p, brw_message_reg(cur_mrf), arg[2]);
cur_mrf += mrf_per_channel;
}
}
/* Emit the texcoords. */
for (i = 0; i < nr_texcoords; i++) {
if (c->key.tex.gl_clamp_mask[i] & (1 << sampler))
brw_set_saturate(p, true);
if (emit & (1<<i))
brw_MOV(p, brw_message_reg(cur_mrf), arg[i]);
else
brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0));
cur_mrf += mrf_per_channel;
brw_set_saturate(p, false);
}
/* Fill in the shadow comparison reference value. */
if (shadow && intel->gen < 7) {
if (intel->gen >= 5) {
/* Fill in the cube map array index value. */
brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0));
cur_mrf += mrf_per_channel;
} else if (c->dispatch_width == 8) {
/* Fill in the LOD bias value. */
brw_MOV(p, brw_message_reg(cur_mrf), brw_imm_f(0));
cur_mrf += mrf_per_channel;
}
brw_MOV(p, brw_message_reg(cur_mrf), arg[2]);
cur_mrf += mrf_per_channel;
}
if (intel->gen >= 5) {
if (shadow)
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_COMPARE;
else
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE;
} else {
/* Note that G45 and older determines shadow compare and dispatch width
* from message length for most messages.
*/
if (c->dispatch_width == 16 && shadow)
msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE;
else
msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE;
}
brw_SAMPLE(p,
dst_retyped,
1,
retype(depth_payload, BRW_REGISTER_TYPE_UW),
SURF_INDEX_TEXTURE(sampler),
sampler,
dst_flags & WRITEMASK_XYZW,
msg_type,
response_length,
cur_mrf - 1,
1,
simd_mode,
BRW_SAMPLER_RETURN_FORMAT_FLOAT32);
}
/* Interpolate between two vertices and put the result into a0.0.
* Increment a0.0 accordingly.
*
* Beware that dest_ptr can be equal to v0_ptr!
*/
void brw_clip_interp_vertex( struct brw_clip_compile *c,
struct brw_indirect dest_ptr,
struct brw_indirect v0_ptr, /* from */
struct brw_indirect v1_ptr, /* to */
struct brw_reg t0,
bool force_edgeflag)
{
struct brw_codegen *p = &c->func;
struct brw_reg t_nopersp, v0_ndc_copy;
GLuint slot;
/* Just copy the vertex header:
*/
/*
* After CLIP stage, only first 256 bits of the VUE are read
* back on Ironlake, so needn't change it
*/
brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
/* First handle the 3D and NDC interpolation, in case we
* need noperspective interpolation. Doing it early has no
* performance impact in any case.
*/
/* Take a copy of the v0 NDC coordinates, in case dest == v0. */
if (c->has_noperspective_shading) {
GLuint offset = brw_varying_to_offset(&c->vue_map,
BRW_VARYING_SLOT_NDC);
v0_ndc_copy = get_tmp(c);
brw_MOV(p, v0_ndc_copy, deref_4f(v0_ptr, offset));
}
/* Compute the new 3D position
*
* dest_hpos = v0_hpos * (1 - t0) + v1_hpos * t0
*/
{
GLuint delta = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS);
struct brw_reg tmp = get_tmp(c);
brw_MUL(p, vec4(brw_null_reg()), deref_4f(v1_ptr, delta), t0);
brw_MAC(p, tmp, negate(deref_4f(v0_ptr, delta)), t0);
brw_ADD(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta), tmp);
release_tmp(c, tmp);
}
/* Recreate the projected (NDC) coordinate in the new vertex header */
brw_clip_project_vertex(c, dest_ptr);
/* If we have noperspective attributes,
* we need to compute the screen-space t
*/
if (c->has_noperspective_shading) {
GLuint delta = brw_varying_to_offset(&c->vue_map,
BRW_VARYING_SLOT_NDC);
struct brw_reg tmp = get_tmp(c);
t_nopersp = get_tmp(c);
/* t_nopersp = vec4(v1.xy, dest.xy) */
brw_MOV(p, t_nopersp, deref_4f(v1_ptr, delta));
brw_MOV(p, tmp, deref_4f(dest_ptr, delta));
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_MOV(p,
brw_writemask(t_nopersp, WRITEMASK_ZW),
brw_swizzle(tmp, 0, 1, 0, 1));
/* t_nopersp = vec4(v1.xy, dest.xy) - v0.xyxy */
brw_ADD(p, t_nopersp, t_nopersp,
negate(brw_swizzle(v0_ndc_copy, 0, 1, 0, 1)));
/* Add the absolute values of the X and Y deltas so that if
* the points aren't in the same place on the screen we get
* nonzero values to divide.
*
* After that, we have vert1 - vert0 in t_nopersp.x and
* vertnew - vert0 in t_nopersp.y
*
* t_nopersp = vec2(|v1.x -v0.x| + |v1.y -v0.y|,
* |dest.x-v0.x| + |dest.y-v0.y|)
*/
brw_ADD(p,
brw_writemask(t_nopersp, WRITEMASK_XY),
brw_abs(brw_swizzle(t_nopersp, 0, 2, 0, 0)),
brw_abs(brw_swizzle(t_nopersp, 1, 3, 0, 0)));
brw_set_default_access_mode(p, BRW_ALIGN_1);
/* If the points are in the same place, just substitute a
* value to avoid divide-by-zero
*/
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ,
vec1(t_nopersp),
brw_imm_f(0));
brw_IF(p, BRW_EXECUTE_1);
brw_MOV(p, t_nopersp, brw_imm_vf4(brw_float_to_vf(1.0),
brw_float_to_vf(0.0),
brw_float_to_vf(0.0),
brw_float_to_vf(0.0)));
brw_ENDIF(p);
/* Now compute t_nopersp = t_nopersp.y/t_nopersp.x and broadcast it. */
brw_math_invert(p, get_element(t_nopersp, 0), get_element(t_nopersp, 0));
brw_MUL(p, vec1(t_nopersp), vec1(t_nopersp),
vec1(suboffset(t_nopersp, 1)));
brw_set_default_access_mode(p, BRW_ALIGN_16);
brw_MOV(p, t_nopersp, brw_swizzle(t_nopersp, 0, 0, 0, 0));
brw_set_default_access_mode(p, BRW_ALIGN_1);
release_tmp(c, tmp);
release_tmp(c, v0_ndc_copy);
}
/* Now we can iterate over each attribute
* (could be done in pairs?)
*/
for (slot = 0; slot < c->vue_map.num_slots; slot++) {
int varying = c->vue_map.slot_to_varying[slot];
GLuint delta = brw_vue_slot_to_offset(slot);
/* HPOS, NDC already handled above */
if (varying == VARYING_SLOT_POS || varying == BRW_VARYING_SLOT_NDC)
continue;
if (varying == VARYING_SLOT_EDGE) {
if (force_edgeflag)
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
else
brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
} else if (varying == VARYING_SLOT_PSIZ) {
/* PSIZ doesn't need interpolation because it isn't used by the
* fragment shader.
*/
} else if (varying < VARYING_SLOT_MAX) {
/* This is a true vertex result (and not a special value for the VUE
* header), so interpolate:
*
* New = attr0 + t*attr1 - t*attr0
*
* Unless the attribute is flat shaded -- in which case just copy
* from one of the sources (doesn't matter which; already copied from pv)
*/
GLuint interp = c->key.interpolation_mode.mode[slot];
if (interp != INTERP_QUALIFIER_FLAT) {
struct brw_reg tmp = get_tmp(c);
struct brw_reg t =
interp == INTERP_QUALIFIER_NOPERSPECTIVE ? t_nopersp : t0;
brw_MUL(p,
vec4(brw_null_reg()),
deref_4f(v1_ptr, delta),
t);
brw_MAC(p,
tmp,
negate(deref_4f(v0_ptr, delta)),
t);
brw_ADD(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta),
tmp);
release_tmp(c, tmp);
}
else {
brw_MOV(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta));
}
}
}
if (c->vue_map.num_slots % 2) {
GLuint delta = brw_vue_slot_to_offset(c->vue_map.num_slots);
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
}
if (c->has_noperspective_shading)
release_tmp(c, t_nopersp);
}
void brw_clip_tri_alloc_regs( struct brw_clip_compile *c,
GLuint nr_verts )
{
struct intel_context *intel = &c->func.brw->intel;
GLuint i = 0,j;
/* Register usage is static, precompute here:
*/
c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++;
if (c->key.nr_userclip) {
c->reg.fixed_planes = brw_vec4_grf(i, 0);
i += (6 + c->key.nr_userclip + 1) / 2;
c->prog_data.curb_read_length = (6 + c->key.nr_userclip + 1) / 2;
}
else
c->prog_data.curb_read_length = 0;
/* Payload vertices plus space for more generated vertices:
*/
for (j = 0; j < nr_verts; j++) {
c->reg.vertex[j] = brw_vec4_grf(i, 0);
i += c->nr_regs;
}
if (c->vue_map.num_slots % 2) {
/* The VUE has an odd number of slots so the last register is only half
* used. Fill the second half with zero.
*/
for (j = 0; j < 3; j++) {
GLuint delta = brw_vue_slot_to_offset(c->vue_map.num_slots);
brw_MOV(&c->func, byte_offset(c->reg.vertex[j], delta), brw_imm_f(0));
}
}
c->reg.t = brw_vec1_grf(i, 0);
c->reg.loopcount = retype(brw_vec1_grf(i, 1), BRW_REGISTER_TYPE_D);
c->reg.nr_verts = retype(brw_vec1_grf(i, 2), BRW_REGISTER_TYPE_UD);
c->reg.planemask = retype(brw_vec1_grf(i, 3), BRW_REGISTER_TYPE_UD);
c->reg.plane_equation = brw_vec4_grf(i, 4);
i++;
c->reg.dpPrev = brw_vec1_grf(i, 0); /* fixme - dp4 will clobber r.1,2,3 */
c->reg.dp = brw_vec1_grf(i, 4);
i++;
c->reg.inlist = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
i++;
c->reg.outlist = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
i++;
c->reg.freelist = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
i++;
if (!c->key.nr_userclip) {
c->reg.fixed_planes = brw_vec8_grf(i, 0);
i++;
}
if (c->key.do_unfilled) {
c->reg.dir = brw_vec4_grf(i, 0);
c->reg.offset = brw_vec4_grf(i, 4);
i++;
c->reg.tmp0 = brw_vec4_grf(i, 0);
c->reg.tmp1 = brw_vec4_grf(i, 4);
i++;
}
if (intel->needs_ff_sync) {
c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD);
i++;
}
c->first_tmp = i;
c->last_tmp = i;
c->prog_data.urb_read_length = c->nr_regs; /* ? */
c->prog_data.total_grf = i;
}
void brw_clip_tri_alloc_regs( struct brw_clip_compile *c,
GLuint nr_verts )
{
GLuint i = 0,j;
/* Register usage is static, precompute here:
*/
c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++;
if (c->key.nr_userclip) {
c->reg.fixed_planes = brw_vec4_grf(i, 0);
i += (6 + c->key.nr_userclip + 1) / 2;
c->prog_data.curb_read_length = (6 + c->key.nr_userclip + 1) / 2;
}
else
c->prog_data.curb_read_length = 0;
/* Payload vertices plus space for more generated vertices:
*/
for (j = 0; j < nr_verts; j++) {
c->reg.vertex[j] = brw_vec4_grf(i, 0);
i += c->nr_regs;
}
if (c->key.nr_attrs & 1) {
for (j = 0; j < 3; j++) {
GLuint delta = c->key.nr_attrs*16 + 32;
if (c->chipset.is_igdng)
delta = c->key.nr_attrs * 16 + 32 * 3;
brw_MOV(&c->func, byte_offset(c->reg.vertex[j], delta), brw_imm_f(0));
}
}
c->reg.t = brw_vec1_grf(i, 0);
c->reg.loopcount = retype(brw_vec1_grf(i, 1), BRW_REGISTER_TYPE_D);
c->reg.nr_verts = retype(brw_vec1_grf(i, 2), BRW_REGISTER_TYPE_UD);
c->reg.planemask = retype(brw_vec1_grf(i, 3), BRW_REGISTER_TYPE_UD);
c->reg.plane_equation = brw_vec4_grf(i, 4);
i++;
c->reg.dpPrev = brw_vec1_grf(i, 0); /* fixme - dp4 will clobber r.1,2,3 */
c->reg.dp = brw_vec1_grf(i, 4);
i++;
c->reg.inlist = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
i++;
c->reg.outlist = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
i++;
c->reg.freelist = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
i++;
if (!c->key.nr_userclip) {
c->reg.fixed_planes = brw_vec8_grf(i, 0);
i++;
}
if (c->key.do_unfilled) {
c->reg.dir = brw_vec4_grf(i, 0);
c->reg.offset = brw_vec4_grf(i, 4);
i++;
c->reg.tmp0 = brw_vec4_grf(i, 0);
c->reg.tmp1 = brw_vec4_grf(i, 4);
i++;
}
if (c->need_ff_sync) {
c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD);
i++;
}
c->first_tmp = i;
c->last_tmp = i;
c->prog_data.urb_read_length = c->nr_regs; /* ? */
c->prog_data.total_grf = i;
}
void brw_emit_point_sprite_setup(struct brw_sf_compile *c, bool allocate)
{
struct brw_compile *p = &c->func;
GLuint i;
c->flag_value = 0xff;
c->nr_verts = 1;
if (allocate)
alloc_regs(c);
copy_z_inv_w(c);
for (i = 0; i < c->nr_setup_regs; i++)
{
struct brw_reg a0 = offset(c->vert[0], i);
GLushort pc, pc_persp, pc_linear, pc_coord_replace;
bool last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
pc_coord_replace = calculate_point_sprite_mask(c, i);
pc_persp &= ~pc_coord_replace;
if (pc_persp) {
set_predicate_control_flag_value(p, c, pc_persp);
brw_MUL(p, a0, a0, c->inv_w[0]);
}
/* Point sprite coordinate replacement: A texcoord with this
* enabled gets replaced with the value (x, y, 0, 1) where x and
* y vary from 0 to 1 across the horizontal and vertical of the
* point.
*/
if (pc_coord_replace) {
set_predicate_control_flag_value(p, c, pc_coord_replace);
/* Caculate 1.0/PointWidth */
gen4_math(&c->func,
c->tmp,
BRW_MATH_FUNCTION_INV,
0,
c->dx0,
BRW_MATH_DATA_SCALAR,
BRW_MATH_PRECISION_FULL);
brw_set_default_access_mode(p, BRW_ALIGN_16);
/* dA/dx, dA/dy */
brw_MOV(p, c->m1Cx, brw_imm_f(0.0));
brw_MOV(p, c->m2Cy, brw_imm_f(0.0));
brw_MOV(p, brw_writemask(c->m1Cx, WRITEMASK_X), c->tmp);
if (c->key.sprite_origin_lower_left) {
brw_MOV(p, brw_writemask(c->m2Cy, WRITEMASK_Y), negate(c->tmp));
} else {
brw_MOV(p, brw_writemask(c->m2Cy, WRITEMASK_Y), c->tmp);
}
/* attribute constant offset */
brw_MOV(p, c->m3C0, brw_imm_f(0.0));
if (c->key.sprite_origin_lower_left) {
brw_MOV(p, brw_writemask(c->m3C0, WRITEMASK_YW), brw_imm_f(1.0));
} else {
brw_MOV(p, brw_writemask(c->m3C0, WRITEMASK_W), brw_imm_f(1.0));
}
brw_set_default_access_mode(p, BRW_ALIGN_1);
}
if (pc & ~pc_coord_replace) {
set_predicate_control_flag_value(p, c, pc & ~pc_coord_replace);
brw_MOV(p, c->m1Cx, brw_imm_ud(0));
brw_MOV(p, c->m2Cy, brw_imm_ud(0));
brw_MOV(p, c->m3C0, a0); /* constant value */
}
set_predicate_control_flag_value(p, c, pc);
/* Copy m0..m3 to URB. */
brw_urb_WRITE(p,
brw_null_reg(),
0,
brw_vec8_grf(0, 0),
last ? BRW_URB_WRITE_EOT_COMPLETE
: BRW_URB_WRITE_NO_FLAGS,
4, /* msg len */
0, /* response len */
i*4, /* urb destination offset */
BRW_URB_SWIZZLE_TRANSPOSE);
}
brw_set_default_predicate_control(p, BRW_PREDICATE_NONE);
}
/* Interpolate between two vertices and put the result into a0.0.
* Increment a0.0 accordingly.
*/
void brw_clip_interp_vertex( struct brw_clip_compile *c,
struct brw_indirect dest_ptr,
struct brw_indirect v0_ptr, /* from */
struct brw_indirect v1_ptr, /* to */
struct brw_reg t0,
boolean force_edgeflag)
{
#if 0
struct brw_compile *p = &c->func;
struct brw_reg tmp = get_tmp(c);
unsigned i;
/* Just copy the vertex header:
*/
brw_copy_indirect_to_indirect(p, dest_ptr, v0_ptr, 1);
/* Iterate over each attribute (could be done in pairs?)
*/
for (i = 0; i < c->nr_attrs; i++) {
unsigned delta = i*16 + 32;
if (delta == c->offset[VERT_RESULT_EDGE]) {
if (force_edgeflag)
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
else
brw_MOV(p, deref_4f(dest_ptr, delta), deref_4f(v0_ptr, delta));
}
else {
/* Interpolate:
*
* New = attr0 + t*attr1 - t*attr0
*/
brw_MUL(p,
vec4(brw_null_reg()),
deref_4f(v1_ptr, delta),
t0);
brw_MAC(p,
tmp,
negate(deref_4f(v0_ptr, delta)),
t0);
brw_ADD(p,
deref_4f(dest_ptr, delta),
deref_4f(v0_ptr, delta),
tmp);
}
}
if (i & 1) {
unsigned delta = i*16 + 32;
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
}
release_tmp(c, tmp);
/* Recreate the projected (NDC) coordinate in the new vertex
* header:
*/
brw_clip_project_vertex(c, dest_ptr );
#else
#warning "disabled"
#endif
}
