static void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim)
{
struct brw_compile *p = &c->func;
struct intel_context *intel = &c->func.brw->intel;
if (intel->gen < 6) {
brw_MOV(p, get_element_ud(c->reg.R0, 1), brw_imm_ud(num_prim));
brw_ff_sync(p,
c->reg.R0,
0,
c->reg.R0,
1, /* allocate */
1, /* response length */
0 /* eot */);
} else {
brw_MOV(p, retype(c->reg.temp, BRW_REGISTER_TYPE_UD),
retype(c->reg.R0, BRW_REGISTER_TYPE_UD));
brw_MOV(p, get_element_ud(c->reg.temp, 1), brw_imm_ud(num_prim));
brw_ff_sync(p,
c->reg.temp,
0,
c->reg.temp,
1, /* allocate */
1, /* response length */
0 /* eot */);
brw_MOV(p, get_element_ud(c->reg.R0, 0),
get_element_ud(c->reg.temp, 0));
}
}
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);
}
}
/**
* Send an FF_SYNC message to ensure that all previously spawned GS threads
* have finished sending primitives down the pipeline, and to allocate a URB
* entry for the first output vertex. Only needed when intel->needs_ff_sync
* is true.
*
* This function modifies c->reg.header: in DWORD 1, it stores num_prim (which
* is needed by the FF_SYNC message), and in DWORD 0, it stores the handle to
* the allocated URB entry (which will be needed by the URB_WRITE meesage that
* follows).
*/
static void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim)
{
struct brw_compile *p = &c->func;
brw_MOV(p, get_element_ud(c->reg.header, 1), brw_imm_ud(num_prim));
brw_ff_sync(p,
c->reg.temp,
0,
c->reg.header,
1, /* allocate */
1, /* response length */
0 /* eot */);
brw_MOV(p, get_element_ud(c->reg.header, 0),
get_element_ud(c->reg.temp, 0));
}
static void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim)
{
struct brw_compile *p = &c->func;
brw_MOV(p, get_element_ud(c->reg.R0, 1), brw_imm_ud(num_prim));
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);
}
void brw_clip_ff_sync(struct brw_clip_compile *c)
{
struct brw_codegen *p = &c->func;
if (p->devinfo->gen == 5) {
brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1));
brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_Z);
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_default_predicate_control(p, BRW_PREDICATE_NONE);
}
}
void brw_clip_ff_sync(struct brw_clip_compile *c)
{
struct brw_compile *p = &c->func;
struct brw_context *brw = p->brw;
if (brw->gen == 5) {
brw_AND(p, brw_null_reg(), c->reg.ff_sync, brw_imm_ud(0x1));
brw_last_inst->header.destreg__conditionalmod = BRW_CONDITIONAL_Z;
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_default_predicate_control(p, BRW_PREDICATE_NONE);
}
}
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);
}
}
