static void merge_edgeflags( struct brw_clip_compile *c )
{
struct brw_compile *p = &c->func;
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_set_conditionalmod(p, BRW_CONDITIONAL_EQ);
brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<8));
brw_MOV(p, byte_offset(c->reg.vertex[0],
brw_varying_to_offset(&c->vue_map,
VARYING_SLOT_EDGE)),
brw_imm_f(0));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_set_conditionalmod(p, BRW_CONDITIONAL_EQ);
brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<9));
brw_MOV(p, byte_offset(c->reg.vertex[2],
brw_varying_to_offset(&c->vue_map,
VARYING_SLOT_EDGE)),
brw_imm_f(0));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
brw_ENDIF(p);
}
void brw_emit_anyprim_setup( struct brw_sf_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_reg ip = brw_ip_reg();
struct brw_reg payload_prim = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, 1, 0);
struct brw_reg payload_attr = get_element_ud(brw_vec1_reg(BRW_GENERAL_REGISTER_FILE, 1, 0), 0);
struct brw_reg primmask;
int jmp;
struct brw_reg v1_null_ud = vec1(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
c->nr_verts = 3;
alloc_regs(c);
primmask = retype(get_element(c->tmp, 0), BRW_REGISTER_TYPE_UD);
brw_MOV(p, primmask, brw_imm_ud(1));
brw_SHL(p, primmask, primmask, payload_prim);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, v1_null_ud, primmask, brw_imm_ud((1<<_3DPRIM_TRILIST) |
(1<<_3DPRIM_TRISTRIP) |
(1<<_3DPRIM_TRIFAN) |
(1<<_3DPRIM_TRISTRIP_REVERSE) |
(1<<_3DPRIM_POLYGON) |
(1<<_3DPRIM_RECTLIST) |
(1<<_3DPRIM_TRIFAN_NOSTIPPLE)));
jmp = brw_JMPI(p, ip, ip, brw_imm_d(0)) - p->store;
brw_emit_tri_setup(c, false);
brw_land_fwd_jump(p, jmp);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, v1_null_ud, primmask, brw_imm_ud((1<<_3DPRIM_LINELIST) |
(1<<_3DPRIM_LINESTRIP) |
(1<<_3DPRIM_LINELOOP) |
(1<<_3DPRIM_LINESTRIP_CONT) |
(1<<_3DPRIM_LINESTRIP_BF) |
(1<<_3DPRIM_LINESTRIP_CONT_BF)));
jmp = brw_JMPI(p, ip, ip, brw_imm_d(0)) - p->store;
brw_emit_line_setup(c, false);
brw_land_fwd_jump(p, jmp);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, v1_null_ud, payload_attr, brw_imm_ud(1<<BRW_SPRITE_POINT_ENABLE));
jmp = brw_JMPI(p, ip, ip, brw_imm_d(0)) - p->store;
brw_emit_point_sprite_setup(c, false);
brw_land_fwd_jump(p, jmp);
brw_emit_point_setup( c, false );
}
void brw_clip_ff_sync(struct brw_clip_compile *c)
{
if (c->need_ff_sync) {
struct brw_compile *p = &c->func;
struct brw_instruction *need_ff_sync;
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1));
need_ff_sync = brw_IF(p, BRW_EXECUTE_1);
{
brw_OR(p, c->reg.ff_sync, c->reg.ff_sync, brw_imm_ud(0x1));
brw_ff_sync(p,
c->reg.R0,
0,
c->reg.R0,
1,
1, /* used */
1, /* msg length */
1, /* response length */
0, /* eot */
1, /* write compelete */
0, /* urb offset */
BRW_URB_SWIZZLE_NONE);
}
brw_ENDIF(p, need_ff_sync);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
}
void brw_clip_tri_emit_polygon(struct brw_clip_compile *c)
{
struct brw_compile *p = &c->func;
/* for (loopcount = nr_verts-2; loopcount > 0; loopcount--)
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_G);
brw_ADD(p,
c->reg.loopcount,
c->reg.nr_verts,
brw_imm_d(-2));
brw_IF(p, BRW_EXECUTE_1);
{
struct brw_indirect v0 = brw_indirect(0, 0);
struct brw_indirect vptr = brw_indirect(1, 0);
brw_MOV(p, get_addr_reg(vptr), brw_address(c->reg.inlist));
brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0));
brw_clip_emit_vue(c, v0, 1, 0,
((_3DPRIM_TRIFAN << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_START));
brw_ADD(p, get_addr_reg(vptr), get_addr_reg(vptr), brw_imm_uw(2));
brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0));
brw_DO(p, BRW_EXECUTE_1);
{
brw_clip_emit_vue(c, v0, 1, 0,
(_3DPRIM_TRIFAN << URB_WRITE_PRIM_TYPE_SHIFT));
brw_ADD(p, get_addr_reg(vptr), get_addr_reg(vptr), brw_imm_uw(2));
brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0));
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p);
brw_clip_emit_vue(c, v0, 0, 1,
((_3DPRIM_TRIFAN << URB_WRITE_PRIM_TYPE_SHIFT)
| URB_WRITE_PRIM_END));
}
brw_ENDIF(p);
}
void brw_emit_tri_clip( struct brw_clip_compile *c )
{
struct brw_instruction *neg_rhw;
struct brw_compile *p = &c->func;
brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6);
brw_clip_tri_init_vertices(c);
brw_clip_init_clipmask(c);
brw_clip_init_ff_sync(c);
/* if -ve rhw workaround bit is set,
do cliptest */
if (c->chipset.is_965) {
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2),
brw_imm_ud(1<<20));
neg_rhw = brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_test(c);
}
brw_ENDIF(p, neg_rhw);
}
/* Can't push into do_clip_tri because with polygon (or quad)
* flatshading, need to apply the flatshade here because we don't
* respect the PV when converting to trifan for emit:
*/
if (c->key.do_flat_shading)
brw_clip_tri_flat_shade(c);
if ((c->key.clip_mode == BRW_CLIPMODE_NORMAL) ||
(c->key.clip_mode == BRW_CLIPMODE_KERNEL_CLIP))
do_clip_tri(c);
else
maybe_do_clip_tri(c);
brw_clip_tri_emit_polygon(c);
/* Send an empty message to kill the thread:
*/
brw_clip_kill_thread(c);
}
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_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_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);
}
void brw_clip_ff_sync(struct brw_clip_compile *c)
{
struct intel_context *intel = &c->func.brw->intel;
if (intel->needs_ff_sync) {
struct brw_compile *p = &c->func;
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1));
brw_IF(p, BRW_EXECUTE_1);
{
brw_OR(p, c->reg.ff_sync, c->reg.ff_sync, brw_imm_ud(0x1));
brw_ff_sync(p,
c->reg.R0,
0,
c->reg.R0,
1, /* allocate */
1, /* response length */
0 /* eot */);
}
brw_ENDIF(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
}
/* Post-fragment-program processing. Send the results to the
* framebuffer.
* \param arg0 the fragment color
* \param arg1 the pass-through depth value
* \param arg2 the shader-computed depth value
*/
void emit_fb_write(struct brw_wm_compile *c,
struct brw_reg *arg0,
struct brw_reg *arg1,
struct brw_reg *arg2,
GLuint target,
GLuint eot)
{
struct brw_compile *p = &c->func;
struct brw_context *brw = p->brw;
struct intel_context *intel = &brw->intel;
GLuint nr = 2;
GLuint channel;
/* Reserve a space for AA - may not be needed:
*/
if (c->aa_dest_stencil_reg)
nr += 1;
/* I don't really understand how this achieves the color interleave
* (ie RGBARGBA) in the result: [Do the saturation here]
*/
brw_push_insn_state(p);
if (c->key.clamp_fragment_color)
brw_set_saturate(p, 1);
for (channel = 0; channel < 4; channel++) {
if (intel->gen >= 6) {
/* gen6 SIMD16 single source DP write looks like:
* m + 0: r0
* m + 1: r1
* m + 2: g0
* m + 3: g1
* m + 4: b0
* m + 5: b1
* m + 6: a0
* m + 7: a1
*/
if (c->dispatch_width == 16) {
brw_MOV(p, brw_message_reg(nr + channel * 2), arg0[channel]);
} else {
brw_MOV(p, brw_message_reg(nr + channel), arg0[channel]);
}
} else if (c->dispatch_width == 16 && brw->has_compr4) {
/* pre-gen6 SIMD16 single source DP write looks like:
* m + 0: r0
* m + 1: g0
* m + 2: b0
* m + 3: a0
* m + 4: r1
* m + 5: g1
* m + 6: b1
* m + 7: a1
*
* By setting the high bit of the MRF register number, we indicate
* that we want COMPR4 mode - instead of doing the usual destination
* + 1 for the second half we get destination + 4.
*/
brw_MOV(p,
brw_message_reg(nr + channel + BRW_MRF_COMPR4),
arg0[channel]);
} else {
/* mov (8) m2.0<1>:ud r28.0<8;8,1>:ud { Align1 } */
/* mov (8) m6.0<1>:ud r29.0<8;8,1>:ud { Align1 SecHalf } */
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p,
brw_message_reg(nr + channel),
arg0[channel]);
if (c->dispatch_width == 16) {
brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
brw_MOV(p,
brw_message_reg(nr + channel + 4),
sechalf(arg0[channel]));
}
}
}
brw_set_saturate(p, 0);
/* skip over the regs populated above:
*/
if (c->dispatch_width == 16)
nr += 8;
else
nr += 4;
brw_pop_insn_state(p);
if (c->source_depth_to_render_target)
{
if (c->computes_depth)
brw_MOV(p, brw_message_reg(nr), arg2[2]);
else
brw_MOV(p, brw_message_reg(nr), arg1[1]); /* ? */
nr += 2;
}
if (c->dest_depth_reg)
{
GLuint comp = c->dest_depth_reg / 2;
GLuint off = c->dest_depth_reg % 2;
if (off != 0) {
brw_push_insn_state(p);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, brw_message_reg(nr), offset(arg1[comp],1));
/* 2nd half? */
brw_MOV(p, brw_message_reg(nr+1), arg1[comp+1]);
brw_pop_insn_state(p);
}
else {
brw_MOV(p, brw_message_reg(nr), arg1[comp]);
}
nr += 2;
}
if (intel->gen >= 6) {
/* Load the message header. There's no implied move from src0
* to the base mrf on gen6.
*/
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_MOV(p, retype(brw_message_reg(0), BRW_REGISTER_TYPE_UD),
retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
brw_pop_insn_state(p);
if (target != 0) {
brw_MOV(p, retype(brw_vec1_reg(BRW_MESSAGE_REGISTER_FILE,
0,
2), BRW_REGISTER_TYPE_UD),
brw_imm_ud(target));
}
}
if (!c->runtime_check_aads_emit) {
if (c->aa_dest_stencil_reg)
emit_aa(c, arg1, 2);
fire_fb_write(c, 0, nr, target, eot);
}
else {
struct brw_reg v1_null_ud = vec1(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
struct brw_reg ip = brw_ip_reg();
struct brw_instruction *jmp;
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p,
v1_null_ud,
get_element_ud(brw_vec8_grf(1,0), 6),
brw_imm_ud(1<<26));
jmp = brw_JMPI(p, ip, ip, brw_imm_w(0));
{
emit_aa(c, arg1, 2);
fire_fb_write(c, 0, nr, target, eot);
/* note - thread killed in subroutine */
}
brw_land_fwd_jump(p, jmp);
/* ELSE: Shuffle up one register to fill in the hole left for AA:
*/
fire_fb_write(c, 1, nr-1, target, eot);
}
}
void
vec4_generator::generate_code(exec_list *instructions)
{
int last_native_insn_offset = 0;
const char *last_annotation_string = NULL;
const void *last_annotation_ir = NULL;
if (unlikely(debug_flag)) {
if (shader_prog) {
fprintf(stderr, "Native code for %s vertex shader %d:\n",
shader_prog->Label ? shader_prog->Label : "unnamed",
shader_prog->Name);
} else {
fprintf(stderr, "Native code for vertex program %d:\n", prog->Id);
}
}
foreach_list(node, instructions) {
vec4_instruction *inst = (vec4_instruction *)node;
struct brw_reg src[3], dst;
if (unlikely(debug_flag)) {
if (last_annotation_ir != inst->ir) {
last_annotation_ir = inst->ir;
if (last_annotation_ir) {
fprintf(stderr, " ");
if (shader_prog) {
((ir_instruction *) last_annotation_ir)->fprint(stderr);
} else {
const prog_instruction *vpi;
vpi = (const prog_instruction *) inst->ir;
fprintf(stderr, "%d: ", (int)(vpi - prog->Instructions));
_mesa_fprint_instruction_opt(stderr, vpi, 0,
PROG_PRINT_DEBUG, NULL);
}
fprintf(stderr, "\n");
}
}
if (last_annotation_string != inst->annotation) {
last_annotation_string = inst->annotation;
if (last_annotation_string)
fprintf(stderr, " %s\n", last_annotation_string);
}
}
for (unsigned int i = 0; i < 3; i++) {
src[i] = inst->get_src(this->prog_data, i);
}
dst = inst->get_dst();
brw_set_conditionalmod(p, inst->conditional_mod);
brw_set_predicate_control(p, inst->predicate);
brw_set_predicate_inverse(p, inst->predicate_inverse);
brw_set_saturate(p, inst->saturate);
brw_set_mask_control(p, inst->force_writemask_all);
unsigned pre_emit_nr_insn = p->nr_insn;
generate_vec4_instruction(inst, dst, src);
if (inst->no_dd_clear || inst->no_dd_check) {
assert(p->nr_insn == pre_emit_nr_insn + 1 ||
!"no_dd_check or no_dd_clear set for IR emitting more "
"than 1 instruction");
struct brw_instruction *last = &p->store[pre_emit_nr_insn];
if (inst->no_dd_clear)
last->header.dependency_control |= BRW_DEPENDENCY_NOTCLEARED;
if (inst->no_dd_check)
last->header.dependency_control |= BRW_DEPENDENCY_NOTCHECKED;
}
if (unlikely(debug_flag)) {
brw_dump_compile(p, stderr,
last_native_insn_offset, p->next_insn_offset);
}
last_native_insn_offset = p->next_insn_offset;
}
/***********************************************************************
* 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;
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_set_conditionalmod(p, BRW_CONDITIONAL_G);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p);
}
/* 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_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p);
}
/* Post-fragment-program processing. Send the results to the
* framebuffer.
*/
static void emit_fb_write( struct brw_wm_compile *c,
struct brw_reg *arg0,
struct brw_reg *arg1,
struct brw_reg *arg2,
GLuint target,
GLuint eot)
{
struct brw_compile *p = &c->func;
GLuint nr = 2;
GLuint channel;
/* Reserve a space for AA - may not be needed:
*/
if (c->key.aa_dest_stencil_reg)
nr += 1;
/* I don't really understand how this achieves the color interleave
* (ie RGBARGBA) in the result: [Do the saturation here]
*/
{
brw_push_insn_state(p);
for (channel = 0; channel < 4; channel++) {
/* mov (8) m2.0<1>:ud r28.0<8;8,1>:ud { Align1 } */
/* mov (8) m6.0<1>:ud r29.0<8;8,1>:ud { Align1 SecHalf } */
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p,
brw_message_reg(nr + channel),
arg0[channel]);
brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
brw_MOV(p,
brw_message_reg(nr + channel + 4),
sechalf(arg0[channel]));
}
/* skip over the regs populated above:
*/
nr += 8;
brw_pop_insn_state(p);
}
if (c->key.source_depth_to_render_target)
{
if (c->key.computes_depth)
brw_MOV(p, brw_message_reg(nr), arg2[2]);
else
brw_MOV(p, brw_message_reg(nr), arg1[1]); /* ? */
nr += 2;
}
if (c->key.dest_depth_reg)
{
GLuint comp = c->key.dest_depth_reg / 2;
GLuint off = c->key.dest_depth_reg % 2;
if (off != 0) {
brw_push_insn_state(p);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, brw_message_reg(nr), offset(arg1[comp],1));
/* 2nd half? */
brw_MOV(p, brw_message_reg(nr+1), arg1[comp+1]);
brw_pop_insn_state(p);
}
else {
brw_MOV(p, brw_message_reg(nr), arg1[comp]);
}
nr += 2;
}
if (!c->key.runtime_check_aads_emit) {
if (c->key.aa_dest_stencil_reg)
emit_aa(c, arg1, 2);
fire_fb_write(c, 0, nr, target, eot);
}
else {
struct brw_reg v1_null_ud = vec1(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
struct brw_reg ip = brw_ip_reg();
struct brw_instruction *jmp;
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p,
v1_null_ud,
get_element_ud(brw_vec8_grf(1,0), 6),
brw_imm_ud(1<<26));
jmp = brw_JMPI(p, ip, ip, brw_imm_w(0));
{
emit_aa(c, arg1, 2);
fire_fb_write(c, 0, nr, target, eot);
/* note - thread killed in subroutine */
}
brw_land_fwd_jump(p, jmp);
/* ELSE: Shuffle up one register to fill in the hole left for AA:
*/
fire_fb_write(c, 1, nr-1, target, eot);
}
}
void
vec4_generator::generate_code(exec_list *instructions)
{
int last_native_insn_offset = 0;
const char *last_annotation_string = NULL;
const void *last_annotation_ir = NULL;
if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
if (shader) {
printf("Native code for vertex shader %d:\n", prog->Name);
} else {
printf("Native code for vertex program %d:\n", c->vp->program.Base.Id);
}
}
foreach_list(node, instructions) {
vec4_instruction *inst = (vec4_instruction *)node;
struct brw_reg src[3], dst;
if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
if (last_annotation_ir != inst->ir) {
last_annotation_ir = inst->ir;
if (last_annotation_ir) {
printf(" ");
if (shader) {
((ir_instruction *) last_annotation_ir)->print();
} else {
const prog_instruction *vpi;
vpi = (const prog_instruction *) inst->ir;
printf("%d: ", (int)(vpi - vp->Base.Instructions));
_mesa_fprint_instruction_opt(stdout, vpi, 0,
PROG_PRINT_DEBUG, NULL);
}
printf("\n");
}
}
if (last_annotation_string != inst->annotation) {
last_annotation_string = inst->annotation;
if (last_annotation_string)
printf(" %s\n", last_annotation_string);
}
}
for (unsigned int i = 0; i < 3; i++) {
src[i] = inst->get_src(i);
}
dst = inst->get_dst();
brw_set_conditionalmod(p, inst->conditional_mod);
brw_set_predicate_control(p, inst->predicate);
brw_set_predicate_inverse(p, inst->predicate_inverse);
brw_set_saturate(p, inst->saturate);
switch (inst->opcode) {
case BRW_OPCODE_MOV:
brw_MOV(p, dst, src[0]);
break;
case BRW_OPCODE_ADD:
brw_ADD(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MUL:
brw_MUL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MACH:
brw_set_acc_write_control(p, 1);
brw_MACH(p, dst, src[0], src[1]);
brw_set_acc_write_control(p, 0);
break;
case BRW_OPCODE_FRC:
brw_FRC(p, dst, src[0]);
break;
case BRW_OPCODE_RNDD:
brw_RNDD(p, dst, src[0]);
break;
case BRW_OPCODE_RNDE:
brw_RNDE(p, dst, src[0]);
break;
case BRW_OPCODE_RNDZ:
brw_RNDZ(p, dst, src[0]);
break;
case BRW_OPCODE_AND:
brw_AND(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_OR:
brw_OR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_XOR:
brw_XOR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_NOT:
brw_NOT(p, dst, src[0]);
break;
case BRW_OPCODE_ASR:
brw_ASR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_SHR:
brw_SHR(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_SHL:
brw_SHL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_CMP:
brw_CMP(p, dst, inst->conditional_mod, src[0], src[1]);
break;
case BRW_OPCODE_SEL:
brw_SEL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DPH:
brw_DPH(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP4:
brw_DP4(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP3:
brw_DP3(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_DP2:
brw_DP2(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_IF:
if (inst->src[0].file != BAD_FILE) {
/* The instruction has an embedded compare (only allowed on gen6) */
assert(intel->gen == 6);
gen6_IF(p, inst->conditional_mod, src[0], src[1]);
} else {
struct brw_instruction *brw_inst = brw_IF(p, BRW_EXECUTE_8);
brw_inst->header.predicate_control = inst->predicate;
}
break;
case BRW_OPCODE_ELSE:
brw_ELSE(p);
break;
case BRW_OPCODE_ENDIF:
brw_ENDIF(p);
break;
case BRW_OPCODE_DO:
brw_DO(p, BRW_EXECUTE_8);
break;
case BRW_OPCODE_BREAK:
brw_BREAK(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case BRW_OPCODE_CONTINUE:
/* FINISHME: We need to write the loop instruction support still. */
if (intel->gen >= 6)
gen6_CONT(p);
else
brw_CONT(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case BRW_OPCODE_WHILE:
brw_WHILE(p);
break;
default:
generate_vs_instruction(inst, dst, src);
break;
}
if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
brw_dump_compile(p, stdout,
last_native_insn_offset, p->next_insn_offset);
}
last_native_insn_offset = p->next_insn_offset;
}
/***********************************************************************
* 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);
}
/* Use mesa's clipping algorithms, translated to GEN4 assembly.
*/
void brw_clip_tri( struct brw_clip_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_indirect vtx = brw_indirect(0, 0);
struct brw_indirect vtxPrev = brw_indirect(1, 0);
struct brw_indirect vtxOut = brw_indirect(2, 0);
struct brw_indirect plane_ptr = brw_indirect(3, 0);
struct brw_indirect inlist_ptr = brw_indirect(4, 0);
struct brw_indirect outlist_ptr = brw_indirect(5, 0);
struct brw_indirect freelist_ptr = brw_indirect(6, 0);
struct brw_instruction *plane_loop;
struct brw_instruction *plane_active;
struct brw_instruction *vertex_loop;
struct brw_instruction *next_test;
struct brw_instruction *prev_test;
brw_MOV(p, get_addr_reg(vtxPrev), brw_address(c->reg.vertex[2]) );
brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c));
brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist));
brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
brw_MOV(p, get_addr_reg(freelist_ptr), brw_address(c->reg.vertex[3]) );
plane_loop = brw_DO(p, BRW_EXECUTE_1);
{
/* if (planemask & 1)
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, vec1(brw_null_reg()), c->reg.planemask, brw_imm_ud(1));
plane_active = brw_IF(p, BRW_EXECUTE_1);
{
/* vtxOut = freelist_ptr++
*/
brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(freelist_ptr) );
brw_ADD(p, get_addr_reg(freelist_ptr), get_addr_reg(freelist_ptr), brw_imm_uw(c->nr_regs * REG_SIZE));
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_MOV(p, c->reg.loopcount, c->reg.nr_verts);
brw_MOV(p, c->reg.nr_verts, brw_imm_ud(0));
vertex_loop = brw_DO(p, BRW_EXECUTE_1);
{
/* vtx = *input_ptr;
*/
brw_MOV(p, get_addr_reg(vtx), deref_1uw(inlist_ptr, 0));
/* IS_NEGATIVE(prev) */
brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
brw_DP4(p, vec4(c->reg.dpPrev), deref_4f(vtxPrev, c->offset_hpos), c->reg.plane_equation);
prev_test = brw_IF(p, BRW_EXECUTE_1);
{
/* IS_POSITIVE(next)
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_GE);
brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset_hpos), c->reg.plane_equation);
next_test = brw_IF(p, BRW_EXECUTE_1);
{
/* Coming back in.
*/
brw_ADD(p, c->reg.t, c->reg.dpPrev, negate(c->reg.dp));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dpPrev);
/* If (vtxOut == 0) vtxOut = vtxPrev
*/
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtxPrev) );
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_clip_interp_vertex(c, vtxOut, vtxPrev, vtx, c->reg.t, GL_FALSE);
/* *outlist_ptr++ = vtxOut;
* nr_verts++;
* vtxOut = 0;
*/
brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
}
brw_ENDIF(p, next_test);
}
prev_test = brw_ELSE(p, prev_test);
{
/* *outlist_ptr++ = vtxPrev;
* nr_verts++;
*/
brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxPrev));
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
/* IS_NEGATIVE(next)
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset_hpos), c->reg.plane_equation);
next_test = brw_IF(p, BRW_EXECUTE_1);
{
/* Going out of bounds. Avoid division by zero as we
* know dp != dpPrev from DIFFERENT_SIGNS, above.
*/
brw_ADD(p, c->reg.t, c->reg.dp, negate(c->reg.dpPrev));
brw_math_invert(p, c->reg.t, c->reg.t);
brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp);
/* If (vtxOut == 0) vtxOut = vtx
*/
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtx) );
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_clip_interp_vertex(c, vtxOut, vtx, vtxPrev, c->reg.t, GL_TRUE);
/* *outlist_ptr++ = vtxOut;
* nr_verts++;
* vtxOut = 0;
*/
brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
}
brw_ENDIF(p, next_test);
}
brw_ENDIF(p, prev_test);
/* vtxPrev = vtx;
* inlist_ptr++;
*/
brw_MOV(p, get_addr_reg(vtxPrev), get_addr_reg(vtx));
brw_ADD(p, get_addr_reg(inlist_ptr), get_addr_reg(inlist_ptr), brw_imm_uw(sizeof(short)));
/* while (--loopcount != 0)
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
}
brw_WHILE(p, vertex_loop);
/* vtxPrev = *(outlist_ptr-1) OR: outlist[nr_verts-1]
* inlist = outlist
* inlist_ptr = &inlist[0]
* outlist_ptr = &outlist[0]
*/
brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_w(-2));
brw_MOV(p, get_addr_reg(vtxPrev), deref_1uw(outlist_ptr, 0));
brw_MOV(p, brw_vec8_grf(c->reg.inlist.nr, 0), brw_vec8_grf(c->reg.outlist.nr, 0));
brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist));
brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
}
brw_ENDIF(p, plane_active);
/* plane_ptr++;
*/
brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c));
/* nr_verts >= 3
*/
brw_CMP(p,
vec1(brw_null_reg()),
BRW_CONDITIONAL_GE,
c->reg.nr_verts,
brw_imm_ud(3));
/* && (planemask>>=1) != 0
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
}
brw_WHILE(p, plane_loop);
}
static void brw_clip_test( struct brw_clip_compile *c )
{
struct brw_reg t = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
struct brw_reg t1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
struct brw_reg t2 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
struct brw_reg t3 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
struct brw_reg v0 = get_tmp(c);
struct brw_reg v1 = get_tmp(c);
struct brw_reg v2 = get_tmp(c);
struct brw_indirect vt0 = brw_indirect(0, 0);
struct brw_indirect vt1 = brw_indirect(1, 0);
struct brw_indirect vt2 = brw_indirect(2, 0);
struct brw_compile *p = &c->func;
struct brw_instruction *is_outside;
struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */
brw_MOV(p, get_addr_reg(vt0), brw_address(c->reg.vertex[0]));
brw_MOV(p, get_addr_reg(vt1), brw_address(c->reg.vertex[1]));
brw_MOV(p, get_addr_reg(vt2), brw_address(c->reg.vertex[2]));
brw_MOV(p, v0, deref_4f(vt0, c->offset_hpos));
brw_MOV(p, v1, deref_4f(vt1, c->offset_hpos));
brw_MOV(p, v2, deref_4f(vt2, c->offset_hpos));
brw_AND(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(~0x3f));
/* test nearz, xmin, ymin plane */
/* clip.xyz < -clip.w */
brw_CMP(p, t1, BRW_CONDITIONAL_L, v0, negate(get_element(v0, 3)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, t2, BRW_CONDITIONAL_L, v1, negate(get_element(v1, 3)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, t3, BRW_CONDITIONAL_L, v2, negate(get_element(v2, 3)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* All vertices are outside of a plane, rejected */
brw_AND(p, t, t1, t2);
brw_AND(p, t, t, t3);
brw_OR(p, tmp0, get_element(t, 0), get_element(t, 1));
brw_OR(p, tmp0, tmp0, get_element(t, 2));
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, brw_null_reg(), tmp0, brw_imm_ud(0x1));
is_outside = brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_kill_thread(c);
}
brw_ENDIF(p, is_outside);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* some vertices are inside a plane, some are outside,need to clip */
brw_XOR(p, t, t1, t2);
brw_XOR(p, t1, t2, t3);
brw_OR(p, t, t, t1);
brw_AND(p, t, t, brw_imm_ud(0x1));
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 0), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<5)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 1), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<3)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 2), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<1)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* test farz, xmax, ymax plane */
/* clip.xyz > clip.w */
brw_CMP(p, t1, BRW_CONDITIONAL_G, v0, get_element(v0, 3));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, t2, BRW_CONDITIONAL_G, v1, get_element(v1, 3));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, t3, BRW_CONDITIONAL_G, v2, get_element(v2, 3));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* All vertices are outside of a plane, rejected */
brw_AND(p, t, t1, t2);
brw_AND(p, t, t, t3);
brw_OR(p, tmp0, get_element(t, 0), get_element(t, 1));
brw_OR(p, tmp0, tmp0, get_element(t, 2));
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, brw_null_reg(), tmp0, brw_imm_ud(0x1));
is_outside = brw_IF(p, BRW_EXECUTE_1);
{
brw_clip_kill_thread(c);
}
brw_ENDIF(p, is_outside);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/* some vertices are inside a plane, some are outside,need to clip */
brw_XOR(p, t, t1, t2);
brw_XOR(p, t1, t2, t3);
brw_OR(p, t, t, t1);
brw_AND(p, t, t, brw_imm_ud(0x1));
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 0), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<4)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 1), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<2)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
get_element(t, 2), brw_imm_ud(0));
brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<0)));
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
release_tmps(c);
}
void brw_emit_anyprim_setup( struct brw_sf_compile *c )
{
struct brw_compile *p = &c->func;
struct brw_reg ip = brw_ip_reg();
struct brw_reg payload_prim = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, 1, 0);
struct brw_reg payload_attr = get_element_ud(brw_vec1_reg(BRW_GENERAL_REGISTER_FILE, 1, 0), 0);
struct brw_reg primmask;
struct brw_instruction *jmp;
struct brw_reg v1_null_ud = vec1(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
GLuint saveflag;
c->nr_verts = 3;
alloc_regs(c);
primmask = retype(get_element(c->tmp, 0), BRW_REGISTER_TYPE_UD);
brw_MOV(p, primmask, brw_imm_ud(1));
brw_SHL(p, primmask, primmask, payload_prim);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, v1_null_ud, primmask, brw_imm_ud((1<<_3DPRIM_TRILIST) |
(1<<_3DPRIM_TRISTRIP) |
(1<<_3DPRIM_TRIFAN) |
(1<<_3DPRIM_TRISTRIP_REVERSE) |
(1<<_3DPRIM_POLYGON) |
(1<<_3DPRIM_RECTLIST) |
(1<<_3DPRIM_TRIFAN_NOSTIPPLE)));
jmp = brw_JMPI(p, ip, ip, brw_imm_d(0));
{
saveflag = p->flag_value;
brw_push_insn_state(p);
brw_emit_tri_setup( c, GL_FALSE );
brw_pop_insn_state(p);
p->flag_value = saveflag;
/* note - thread killed in subroutine, so must
* restore the flag which is changed when building
* the subroutine. fix #13240
*/
}
brw_land_fwd_jump(p, jmp);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, v1_null_ud, primmask, brw_imm_ud((1<<_3DPRIM_LINELIST) |
(1<<_3DPRIM_LINESTRIP) |
(1<<_3DPRIM_LINELOOP) |
(1<<_3DPRIM_LINESTRIP_CONT) |
(1<<_3DPRIM_LINESTRIP_BF) |
(1<<_3DPRIM_LINESTRIP_CONT_BF)));
jmp = brw_JMPI(p, ip, ip, brw_imm_d(0));
{
saveflag = p->flag_value;
brw_push_insn_state(p);
brw_emit_line_setup( c, GL_FALSE );
brw_pop_insn_state(p);
p->flag_value = saveflag;
/* note - thread killed in subroutine */
}
brw_land_fwd_jump(p, jmp);
brw_set_conditionalmod(p, BRW_CONDITIONAL_Z);
brw_AND(p, v1_null_ud, payload_attr, brw_imm_ud(1<<BRW_SPRITE_POINT_ENABLE));
jmp = brw_JMPI(p, ip, ip, brw_imm_d(0));
{
saveflag = p->flag_value;
brw_push_insn_state(p);
brw_emit_point_sprite_setup( c, GL_FALSE );
brw_pop_insn_state(p);
p->flag_value = saveflag;
}
brw_land_fwd_jump(p, jmp);
brw_emit_point_setup( c, GL_FALSE );
}
/**
* Generate the geometry shader program used on Gen6 to perform stream output
* (transform feedback).
*/
void
gen6_sol_program(struct brw_gs_compile *c, struct brw_gs_prog_key *key,
unsigned num_verts, bool check_edge_flags)
{
struct brw_compile *p = &c->func;
c->prog_data.svbi_postincrement_value = num_verts;
brw_gs_alloc_regs(c, num_verts, true);
brw_gs_initialize_header(c);
if (key->num_transform_feedback_bindings > 0) {
unsigned vertex, binding;
struct brw_reg destination_indices_uw =
vec8(retype(c->reg.destination_indices, BRW_REGISTER_TYPE_UW));
/* Note: since we use the binding table to keep track of buffer offsets
* and stride, the GS doesn't need to keep track of a separate pointer
* into each buffer; it uses a single pointer which increments by 1 for
* each vertex. So we use SVBI0 for this pointer, regardless of whether
* transform feedback is in interleaved or separate attribs mode.
*
* Make sure that the buffers have enough room for all the vertices.
*/
brw_ADD(p, get_element_ud(c->reg.temp, 0),
get_element_ud(c->reg.SVBI, 0), brw_imm_ud(num_verts));
brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_LE,
get_element_ud(c->reg.temp, 0),
get_element_ud(c->reg.SVBI, 4));
brw_IF(p, BRW_EXECUTE_1);
/* Compute the destination indices to write to. Usually we use SVBI[0]
* + (0, 1, 2). However, for odd-numbered triangles in tristrips, the
* vertices come down the pipeline in reversed winding order, so we need
* to flip the order when writing to the transform feedback buffer. To
* ensure that flatshading accuracy is preserved, we need to write them
* in order SVBI[0] + (0, 2, 1) if we're using the first provoking
* vertex convention, and in order SVBI[0] + (1, 0, 2) if we're using
* the last provoking vertex convention.
*
* Note: since brw_imm_v can only be used in instructions in
* packed-word execution mode, and SVBI is a double-word, we need to
* first move the appropriate immediate constant ((0, 1, 2), (0, 2, 1),
* or (1, 0, 2)) to the destination_indices register, and then add SVBI
* using a separate instruction. Also, since the immediate constant is
* expressed as packed words, and we need to load double-words into
* destination_indices, we need to intersperse zeros to fill the upper
* halves of each double-word.
*/
brw_MOV(p, destination_indices_uw,
brw_imm_v(0x00020100)); /* (0, 1, 2) */
if (num_verts == 3) {
/* Get primitive type into temp register. */
brw_AND(p, get_element_ud(c->reg.temp, 0),
get_element_ud(c->reg.R0, 2), brw_imm_ud(0x1f));
/* Test if primitive type is TRISTRIP_REVERSE. We need to do this as
* an 8-wide comparison so that the conditional MOV that follows
* moves all 8 words correctly.
*/
brw_CMP(p, vec8(brw_null_reg()), BRW_CONDITIONAL_EQ,
get_element_ud(c->reg.temp, 0),
brw_imm_ud(_3DPRIM_TRISTRIP_REVERSE));
/* If so, then overwrite destination_indices_uw with the appropriate
* reordering.
*/
brw_MOV(p, destination_indices_uw,
brw_imm_v(key->pv_first ? 0x00010200 /* (0, 2, 1) */
: 0x00020001)); /* (1, 0, 2) */
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
brw_ADD(p, c->reg.destination_indices,
c->reg.destination_indices, get_element_ud(c->reg.SVBI, 0));
/* For each vertex, generate code to output each varying using the
* appropriate binding table entry.
*/
for (vertex = 0; vertex < num_verts; ++vertex) {
/* Set up the correct destination index for this vertex */
brw_MOV(p, get_element_ud(c->reg.header, 5),
get_element_ud(c->reg.destination_indices, vertex));
for (binding = 0; binding < key->num_transform_feedback_bindings;
++binding) {
unsigned char varying =
key->transform_feedback_bindings[binding];
unsigned char slot = c->vue_map.varying_to_slot[varying];
/* From the Sandybridge PRM, Volume 2, Part 1, Section 4.5.1:
*
* "Prior to End of Thread with a URB_WRITE, the kernel must
* ensure that all writes are complete by sending the final
* write as a committed write."
*/
bool final_write =
binding == key->num_transform_feedback_bindings - 1 &&
vertex == num_verts - 1;
struct brw_reg vertex_slot = c->reg.vertex[vertex];
vertex_slot.nr += slot / 2;
vertex_slot.subnr = (slot % 2) * 16;
/* gl_PointSize is stored in VARYING_SLOT_PSIZ.w. */
vertex_slot.dw1.bits.swizzle = varying == VARYING_SLOT_PSIZ
? BRW_SWIZZLE_WWWW : key->transform_feedback_swizzles[binding];
brw_set_access_mode(p, BRW_ALIGN_16);
brw_MOV(p, stride(c->reg.header, 4, 4, 1),
retype(vertex_slot, BRW_REGISTER_TYPE_UD));
brw_set_access_mode(p, BRW_ALIGN_1);
brw_svb_write(p,
final_write ? c->reg.temp : brw_null_reg(), /* dest */
1, /* msg_reg_nr */
c->reg.header, /* src0 */
SURF_INDEX_SOL_BINDING(binding), /* binding_table_index */
final_write); /* send_commit_msg */
}
}
brw_ENDIF(p);
/* Now, reinitialize the header register from R0 to restore the parts of
* the register that we overwrote while streaming out transform feedback
* data.
*/
brw_gs_initialize_header(c);
/* Finally, wait for the write commit to occur so that we can proceed to
* other things safely.
*
* From the Sandybridge PRM, Volume 4, Part 1, Section 3.3:
*
* The write commit does not modify the destination register, but
* merely clears the dependency associated with the destination
* register. Thus, a simple “mov” instruction using the register as a
* source is sufficient to wait for the write commit to occur.
*/
brw_MOV(p, c->reg.temp, c->reg.temp);
}
brw_gs_ff_sync(c, 1);
/* If RASTERIZER_DISCARD is enabled, we have nothing further to do, so
* release the URB that was just allocated, and terminate the thread.
*/
if (key->rasterizer_discard) {
brw_gs_terminate(c);
return;
}
brw_gs_overwrite_header_dw2_from_r0(c);
switch (num_verts) {
case 1:
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_START | URB_WRITE_PRIM_END);
brw_gs_emit_vue(c, c->reg.vertex[0], true);
break;
case 2:
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_START);
brw_gs_emit_vue(c, c->reg.vertex[0], false);
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_END - URB_WRITE_PRIM_START);
brw_gs_emit_vue(c, c->reg.vertex[1], true);
break;
case 3:
if (check_edge_flags) {
/* Only emit vertices 0 and 1 if this is the first triangle of the
* polygon. Otherwise they are redundant.
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
get_element_ud(c->reg.R0, 2),
brw_imm_ud(BRW_GS_EDGE_INDICATOR_0));
brw_IF(p, BRW_EXECUTE_1);
}
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_START);
brw_gs_emit_vue(c, c->reg.vertex[0], false);
brw_gs_offset_header_dw2(c, -URB_WRITE_PRIM_START);
brw_gs_emit_vue(c, c->reg.vertex[1], false);
if (check_edge_flags) {
brw_ENDIF(p);
/* Only emit vertex 2 in PRIM_END mode if this is the last triangle
* of the polygon. Otherwise leave the primitive incomplete because
* there are more polygon vertices coming.
*/
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
brw_AND(p, retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
get_element_ud(c->reg.R0, 2),
brw_imm_ud(BRW_GS_EDGE_INDICATOR_1));
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
}
brw_gs_offset_header_dw2(c, URB_WRITE_PRIM_END);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
brw_gs_emit_vue(c, c->reg.vertex[2], true);
break;
}
}
static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
{
#define MAX_IFSN 32
#define MAX_LOOP_DEPTH 32
struct brw_instruction *if_inst[MAX_IFSN], *loop_inst[MAX_LOOP_DEPTH];
struct brw_instruction *inst0, *inst1;
int i, if_insn = 0, loop_insn = 0;
struct brw_compile *p = &c->func;
struct brw_indirect stack_index = brw_indirect(0, 0);
c->reg_index = 0;
prealloc_reg(c);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack));
for (i = 0; i < c->nr_fp_insns; i++) {
struct prog_instruction *inst = &c->prog_instructions[i];
struct prog_instruction *orig_inst;
if ((orig_inst = inst->Data) != 0)
orig_inst->Data = current_insn(p);
if (inst->CondUpdate)
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
else
brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE);
switch (inst->Opcode) {
case WM_PIXELXY:
emit_pixel_xy(c, inst);
break;
case WM_DELTAXY:
emit_delta_xy(c, inst);
break;
case WM_PIXELW:
emit_pixel_w(c, inst);
break;
case WM_LINTERP:
emit_linterp(c, inst);
break;
case WM_PINTERP:
emit_pinterp(c, inst);
break;
case WM_CINTERP:
emit_cinterp(c, inst);
break;
case WM_WPOSXY:
emit_wpos_xy(c, inst);
break;
case WM_FB_WRITE:
emit_fb_write(c, inst);
break;
case OPCODE_ABS:
emit_abs(c, inst);
break;
case OPCODE_ADD:
emit_add(c, inst);
break;
case OPCODE_SUB:
emit_sub(c, inst);
break;
case OPCODE_FRC:
emit_frc(c, inst);
break;
case OPCODE_FLR:
emit_flr(c, inst);
break;
case OPCODE_LRP:
emit_lrp(c, inst);
break;
case OPCODE_INT:
emit_int(c, inst);
break;
case OPCODE_MOV:
emit_mov(c, inst);
break;
case OPCODE_DP3:
emit_dp3(c, inst);
break;
case OPCODE_DP4:
emit_dp4(c, inst);
break;
case OPCODE_XPD:
emit_xpd(c, inst);
break;
case OPCODE_DPH:
emit_dph(c, inst);
break;
case OPCODE_RCP:
emit_rcp(c, inst);
break;
case OPCODE_RSQ:
emit_rsq(c, inst);
break;
case OPCODE_SIN:
emit_sin(c, inst);
break;
case OPCODE_COS:
emit_cos(c, inst);
break;
case OPCODE_EX2:
emit_ex2(c, inst);
break;
case OPCODE_LG2:
emit_lg2(c, inst);
break;
case OPCODE_MAX:
emit_max(c, inst);
break;
case OPCODE_MIN:
emit_min(c, inst);
break;
case OPCODE_DDX:
emit_ddx(c, inst);
break;
case OPCODE_DDY:
emit_ddy(c, inst);
break;
case OPCODE_SLT:
emit_slt(c, inst);
break;
case OPCODE_SLE:
emit_sle(c, inst);
break;
case OPCODE_SGT:
emit_sgt(c, inst);
break;
case OPCODE_SGE:
emit_sge(c, inst);
break;
case OPCODE_SEQ:
emit_seq(c, inst);
break;
case OPCODE_SNE:
emit_sne(c, inst);
break;
case OPCODE_MUL:
emit_mul(c, inst);
break;
case OPCODE_POW:
emit_pow(c, inst);
break;
case OPCODE_MAD:
emit_mad(c, inst);
break;
case OPCODE_TEX:
emit_tex(c, inst);
break;
case OPCODE_TXB:
emit_txb(c, inst);
break;
case OPCODE_KIL_NV:
emit_kil(c);
break;
case OPCODE_IF:
assert(if_insn < MAX_IFSN);
if_inst[if_insn++] = brw_IF(p, BRW_EXECUTE_8);
break;
case OPCODE_ELSE:
if_inst[if_insn-1] = brw_ELSE(p, if_inst[if_insn-1]);
break;
case OPCODE_ENDIF:
assert(if_insn > 0);
brw_ENDIF(p, if_inst[--if_insn]);
break;
case OPCODE_BGNSUB:
case OPCODE_ENDSUB:
break;
case OPCODE_CAL:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_set_access_mode(p, BRW_ALIGN_1);
brw_ADD(p, deref_1ud(stack_index, 0), brw_ip_reg(), brw_imm_d(3*16));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(4));
orig_inst = inst->Data;
orig_inst->Data = &p->store[p->nr_insn];
brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
brw_pop_insn_state(p);
break;
case OPCODE_RET:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(-4));
brw_set_access_mode(p, BRW_ALIGN_1);
brw_MOV(p, brw_ip_reg(), deref_1ud(stack_index, 0));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_pop_insn_state(p);
break;
case OPCODE_BGNLOOP:
loop_inst[loop_insn++] = brw_DO(p, BRW_EXECUTE_8);
break;
case OPCODE_BRK:
brw_BREAK(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_CONT:
brw_CONT(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_ENDLOOP:
loop_insn--;
inst0 = inst1 = brw_WHILE(p, loop_inst[loop_insn]);
/* patch all the BREAK instructions from
last BEGINLOOP */
while (inst0 > loop_inst[loop_insn]) {
inst0--;
if (inst0->header.opcode == BRW_OPCODE_BREAK) {
inst0->bits3.if_else.jump_count = inst1 - inst0 + 1;
inst0->bits3.if_else.pop_count = 0;
} else if (inst0->header.opcode == BRW_OPCODE_CONTINUE) {
inst0->bits3.if_else.jump_count = inst1 - inst0;
inst0->bits3.if_else.pop_count = 0;
}
}
break;
default:
_mesa_printf("unsupported IR in fragment shader %d\n",
inst->Opcode);
}
if (inst->CondUpdate)
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
else
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
post_wm_emit(c);
for (i = 0; i < c->fp->program.Base.NumInstructions; i++)
c->fp->program.Base.Instructions[i].Data = NULL;
}
