/* Points setup - several simplifications as all attributes are
* constant across the face of the point (point sprites excluded!)
*/
void brw_emit_point_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);
brw_MOV(p, c->m1Cx, brw_imm_ud(0)); /* zero - move out of loop */
brw_MOV(p, c->m2Cy, brw_imm_ud(0)); /* zero - move out of loop */
for (i = 0; i < c->nr_setup_regs; 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)
{
/* This seems odd as the values are all constant, but the
* fragment shader will be expecting it:
*/
brw_set_predicate_control_flag_value(p, pc_persp);
brw_MUL(p, a0, a0, c->inv_w[0]);
}
/* The delta values are always zero, just send the starting
* coordinate. Again, this is to fit in with the interpolation
* code in the fragment shader.
*/
{
brw_set_predicate_control_flag_value(p, pc);
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);
}
}
}
/* Sets the destination channels to 1.0 or 0.0 according to glFrontFacing. */
void emit_frontfacing(struct brw_compile *p,
const struct brw_reg *dst,
GLuint mask)
{
struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD);
GLuint i;
if (!(mask & WRITEMASK_XYZW))
return;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
brw_MOV(p, dst[i], brw_imm_f(0.0));
}
}
/* bit 31 is "primitive is back face", so checking < (1 << 31) gives
* us front face
*/
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, r1_6ud, brw_imm_ud(1 << 31));
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
brw_MOV(p, dst[i], brw_imm_f(1.0));
}
}
brw_set_predicate_control_flag_value(p, 0xff);
}
/* 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 intel_context *intel = &p->brw->intel;
struct brw_reg pixelmask;
GLuint i, j;
if (intel->gen >= 6)
pixelmask = retype(brw_vec1_grf(1, 7), BRW_REGISTER_TYPE_UW);
else
pixelmask = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
for (i = 0; i < 4; i++) {
/* Check if we've already done the comparison for this reg
* -- common when someone does KIL TEMP.wwww.
*/
for (j = 0; j < i; j++) {
if (memcmp(&arg0[j], &arg0[i], sizeof(arg0[0])) == 0)
break;
}
if (j != i)
continue;
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, pixelmask, brw_flag_reg(), pixelmask);
brw_pop_insn_state(p);
}
}
void
brw_init_compile(struct brw_context *brw, struct brw_compile *p, void *mem_ctx)
{
p->brw = brw;
/*
* Set the initial instruction store array size to 1024, if found that
* isn't enough, then it will double the store size at brw_next_insn()
* until out of memory.
*/
p->store_size = 1024;
p->store = rzalloc_array(mem_ctx, struct brw_instruction, p->store_size);
p->nr_insn = 0;
p->current = p->stack;
p->compressed = false;
memset(p->current, 0, sizeof(p->current[0]));
p->mem_ctx = mem_ctx;
/* Some defaults?
*/
brw_set_mask_control(p, BRW_MASK_ENABLE); /* what does this do? */
brw_set_saturate(p, 0);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_predicate_control_flag_value(p, 0xff);
/* Set up control flow stack */
p->if_stack_depth = 0;
p->if_stack_array_size = 16;
p->if_stack = rzalloc_array(mem_ctx, int, p->if_stack_array_size);
p->loop_stack_depth = 0;
p->loop_stack_array_size = 16;
p->loop_stack = rzalloc_array(mem_ctx, int, p->loop_stack_array_size);
p->if_depth_in_loop = rzalloc_array(mem_ctx, int, p->loop_stack_array_size);
}
static void emit_min(struct brw_wm_compile *c,
struct prog_instruction *inst)
{
struct brw_compile *p = &c->func;
GLuint mask = inst->DstReg.WriteMask;
struct brw_reg src0, src1, dst;
int i;
brw_push_insn_state(p);
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_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
brw_MOV(p, dst, src0);
brw_set_saturate(p, 0);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, src1, src0);
brw_set_saturate(p, (inst->SaturateMode != SATURATE_OFF) ? 1 : 0);
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
brw_MOV(p, dst, src1);
brw_set_saturate(p, 0);
brw_set_predicate_control_flag_value(p, 0xff);
}
}
brw_pop_insn_state(p);
}
void brw_init_compile( struct brw_compile *p )
{
p->nr_insn = 0;
p->current = p->stack;
memset(p->current, 0, sizeof(p->current[0]));
/* Some defaults?
*/
brw_set_mask_control(p, BRW_MASK_ENABLE); /* what does this do? */
brw_set_saturate(p, 0);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_predicate_control_flag_value(p, 0xff);
}
/**
* Extended math function, float[16].
* Use 2 send instructions.
*/
void brw_math_16( struct brw_compile *p,
struct brw_reg dest,
GLuint function,
GLuint saturate,
GLuint msg_reg_nr,
struct brw_reg src,
GLuint precision )
{
struct brw_instruction *insn;
GLuint msg_length = (function == BRW_MATH_FUNCTION_POW) ? 2 : 1;
GLuint response_length = (function == BRW_MATH_FUNCTION_SINCOS) ? 2 : 1;
/* First instruction:
*/
brw_push_insn_state(p);
brw_set_predicate_control_flag_value(p, 0xff);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
insn = next_insn(p, BRW_OPCODE_SEND);
insn->header.destreg__conditionalmod = msg_reg_nr;
brw_set_dest(insn, dest);
brw_set_src0(insn, src);
brw_set_math_message(p->brw,
insn,
msg_length, response_length,
function,
BRW_MATH_INTEGER_UNSIGNED,
precision,
saturate,
BRW_MATH_DATA_VECTOR);
/* Second instruction:
*/
insn = next_insn(p, BRW_OPCODE_SEND);
insn->header.compression_control = BRW_COMPRESSION_2NDHALF;
insn->header.destreg__conditionalmod = msg_reg_nr+1;
brw_set_dest(insn, offset(dest,1));
brw_set_src0(insn, src);
brw_set_math_message(p->brw,
insn,
msg_length, response_length,
function,
BRW_MATH_INTEGER_UNSIGNED,
precision,
saturate,
BRW_MATH_DATA_VECTOR);
brw_pop_insn_state(p);
}
static void emit_sop( struct brw_compile *p,
const struct brw_reg *dst,
GLuint mask,
GLuint cond,
const struct brw_reg *arg0,
const struct brw_reg *arg1 )
{
GLuint i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
brw_MOV(p, dst[i], brw_imm_f(0));
brw_CMP(p, brw_null_reg(), cond, arg0[i], arg1[i]);
brw_MOV(p, dst[i], brw_imm_f(1.0));
brw_set_predicate_control_flag_value(p, 0xff);
}
}
}
void emit_min(struct brw_compile *p,
const struct brw_reg *dst,
GLuint mask,
const struct brw_reg *arg0,
const struct brw_reg *arg1)
{
GLuint i;
for (i = 0; i < 4; i++) {
if (mask & (1<<i)) {
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0[i], arg1[i]);
brw_set_saturate(p, (mask & SATURATE) ? 1 : 0);
brw_SEL(p, dst[i], arg0[i], arg1[i]);
brw_set_saturate(p, 0);
brw_set_predicate_control_flag_value(p, 0xff);
}
}
}
/* 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);
}
}
void brw_compile_init(struct brw_compile *p, int gen, void *store)
{
assert(gen);
p->gen = gen;
p->store = store;
p->nr_insn = 0;
p->current = p->stack;
p->compressed = false;
memset(p->current, 0, sizeof(p->current[0]));
/* Some defaults?
*/
brw_set_mask_control(p, BRW_MASK_ENABLE); /* what does this do? */
brw_set_saturate(p, 0);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_predicate_control_flag_value(p, 0xff);
p->if_stack_depth = 0;
p->if_stack_array_size = 0;
p->if_stack = NULL;
}
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_reg a0 = offset(c->vert[0], i);
GLushort pc, pc_persp, pc_linear, pc_coord_replace;
GLboolean 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) {
brw_set_predicate_control_flag_value(p, 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) {
brw_set_predicate_control_flag_value(p, pc_coord_replace);
/* 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);
brw_set_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_access_mode(p, BRW_ALIGN_1);
}
if (pc & ~pc_coord_replace) {
brw_set_predicate_control_flag_value(p, 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 */
}
brw_set_predicate_control_flag_value(p, pc);
/* 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);
}
}
void brw_emit_line_setup( struct brw_sf_compile *c, GLboolean allocate)
{
struct brw_compile *p = &c->func;
GLuint i;
c->nr_verts = 2;
if (allocate)
alloc_regs(c);
invert_det(c);
copy_z_inv_w(c);
if (c->key.do_flat_shading)
do_flatshade_line(c);
for (i = 0; i < c->nr_setup_regs; i++)
{
/* Pair of incoming attributes:
*/
struct brw_reg a0 = offset(c->vert[0], i);
struct brw_reg a1 = offset(c->vert[1], i);
GLushort pc, pc_persp, pc_linear;
GLboolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
if (pc_persp)
{
brw_set_predicate_control_flag_value(p, pc_persp);
brw_MUL(p, a0, a0, c->inv_w[0]);
brw_MUL(p, a1, a1, c->inv_w[1]);
}
/* Calculate coefficients for position, color:
*/
if (pc_linear) {
brw_set_predicate_control_flag_value(p, pc_linear);
brw_ADD(p, c->a1_sub_a0, a1, negate(a0));
brw_MUL(p, c->tmp, c->a1_sub_a0, c->dx0);
brw_MUL(p, c->m1Cx, c->tmp, c->inv_det);
brw_MUL(p, c->tmp, c->a1_sub_a0, c->dy0);
brw_MUL(p, c->m2Cy, c->tmp, c->inv_det);
}
{
brw_set_predicate_control_flag_value(p, pc);
/* start point for interpolation
*/
brw_MOV(p, c->m3C0, a0);
/* 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);
}
}
}
void brw_emit_tri_setup( struct brw_sf_compile *c, GLboolean allocate)
{
struct brw_compile *p = &c->func;
GLuint i;
c->nr_verts = 3;
if (allocate)
alloc_regs(c);
invert_det(c);
copy_z_inv_w(c);
if (c->key.do_twoside_color)
do_twoside_color(c);
if (c->key.do_flat_shading)
do_flatshade_triangle(c);
for (i = 0; i < c->nr_setup_regs; i++)
{
/* Pair of incoming attributes:
*/
struct brw_reg a0 = offset(c->vert[0], i);
struct brw_reg a1 = offset(c->vert[1], i);
struct brw_reg a2 = offset(c->vert[2], i);
GLushort pc, pc_persp, pc_linear;
GLboolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
if (pc_persp)
{
brw_set_predicate_control_flag_value(p, pc_persp);
brw_MUL(p, a0, a0, c->inv_w[0]);
brw_MUL(p, a1, a1, c->inv_w[1]);
brw_MUL(p, a2, a2, c->inv_w[2]);
}
/* Calculate coefficients for interpolated values:
*/
if (pc_linear)
{
brw_set_predicate_control_flag_value(p, pc_linear);
brw_ADD(p, c->a1_sub_a0, a1, negate(a0));
brw_ADD(p, c->a2_sub_a0, a2, negate(a0));
/* calculate dA/dx
*/
brw_MUL(p, brw_null_reg(), c->a1_sub_a0, c->dy2);
brw_MAC(p, c->tmp, c->a2_sub_a0, negate(c->dy0));
brw_MUL(p, c->m1Cx, c->tmp, c->inv_det);
/* calculate dA/dy
*/
brw_MUL(p, brw_null_reg(), c->a2_sub_a0, c->dx0);
brw_MAC(p, c->tmp, c->a1_sub_a0, negate(c->dx2));
brw_MUL(p, c->m2Cy, c->tmp, c->inv_det);
}
{
brw_set_predicate_control_flag_value(p, pc);
/* start point for interpolation
*/
brw_MOV(p, c->m3C0, a0);
/* Copy m0..m3 to URB. m0 is implicitly copied from r0 in
* the send instruction:
*/
brw_urb_WRITE(p,
brw_null_reg(),
0,
brw_vec8_grf(0, 0), /* r0, will be copied to m0 */
0, /* allocate */
1, /* used */
4, /* msg len */
0, /* response len */
last, /* eot */
last, /* writes complete */
i*4, /* offset */
BRW_URB_SWIZZLE_TRANSPOSE); /* XXX: Swizzle control "SF to windower" */
}
}
}
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);
}
}
}
