#include "main/glheader.h"
#include "main/macros.h"
#include "main/enums.h"
-#include "shader/program.h"
+#include "program/program.h"
#include "intel_batchbuffer.h"
#include "brw_defines.h"
#include "brw_context.h"
#include "brw_eu.h"
-#include "brw_util.h"
#include "brw_clip.h"
/* 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 (i = 0; i < c->nr_attrs; i++) {
- GLuint delta = i*16 + 32;
+ GLuint delta = c->offset[c->idx_to_attr[i]];
- if (delta == c->offset[VERT_RESULT_EDGE]) {
+ if (c->idx_to_attr[i] == VERT_RESULT_EDGE) {
if (force_edgeflag)
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(1));
else
}
if (i & 1) {
- GLuint delta = i*16 + 32;
+ GLuint delta = c->offset[c->idx_to_attr[c->nr_attrs - 1]] + ATTR_SIZE;
+
brw_MOV(p, deref_4f(dest_ptr, delta), brw_imm_f(0));
}
brw_clip_project_vertex(c, dest_ptr );
}
-
-
-
-#define MAX_MRF 16
-
void brw_clip_emit_vue(struct brw_clip_compile *c,
struct brw_indirect vert,
GLboolean allocate,
GLuint header)
{
struct brw_compile *p = &c->func;
- GLuint start = c->last_mrf;
+
+ brw_clip_ff_sync(c);
assert(!(allocate && eot));
-
- /* Cycle through mrf regs - probably futile as we have to wait for
- * the allocation response anyway. Also, the order this function
- * is invoked doesn't correspond to the order the instructions will
- * be executed, so it won't have any effect in many cases.
- */
-#if 0
- if (start + c->nr_regs + 1 >= MAX_MRF)
- start = 0;
- c->last_mrf = start + c->nr_regs + 1;
-#endif
-
/* Copy the vertex from vertn into m1..mN+1:
*/
- brw_copy_from_indirect(p, brw_message_reg(start+1), vert, c->nr_regs);
+ brw_copy_from_indirect(p, brw_message_reg(1), vert, c->nr_regs);
/* Overwrite PrimType and PrimStart in the message header, for
* each vertex in turn:
*/
brw_urb_WRITE(p,
allocate ? c->reg.R0 : retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
- start,
+ 0,
c->reg.R0,
allocate,
1, /* used */
{
struct brw_compile *p = &c->func;
+ brw_clip_ff_sync(c);
/* Send an empty message to kill the thread and release any
* allocated urb entry:
*/
}
}
+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;
+ 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, /* allocate */
+ 1, /* response length */
+ 0 /* eot */);
+ }
+ brw_ENDIF(p, need_ff_sync);
+ brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+ }
+}
+
+void brw_clip_init_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_MOV(p, c->reg.ff_sync, brw_imm_ud(0));
+ }
+}