/**
* Compute the number of primitives written during our most recent
* transform feedback activity (the current SO_NUM_PRIMS_WRITTEN value
* minus the stashed "start" value), and add it to our running tally.
*
* If \p finalize is true, also compute the number of vertices written
* (by multiplying by the number of vertices per primitive), and store
* that to the "final" location.
*
* Otherwise, just overwrite the old tally with the new one.
*/
static void
tally_prims_written(struct brw_context *brw,
struct brw_transform_feedback_object *obj,
bool finalize)
{
/* Flush any drawing so that the counters have the right values. */
brw_emit_mi_flush(brw);
for (int i = 0; i < BRW_MAX_XFB_STREAMS; i++) {
/* GPR0 = Tally */
brw_load_register_imm32(brw, HSW_CS_GPR(0) + 4, 0);
brw_load_register_mem(brw, HSW_CS_GPR(0), obj->prim_count_bo,
I915_GEM_DOMAIN_INSTRUCTION,
I915_GEM_DOMAIN_INSTRUCTION,
TALLY_OFFSET + i * sizeof(uint32_t));
if (!obj->base.Paused) {
/* GPR1 = Start Snapshot */
brw_load_register_mem64(brw, HSW_CS_GPR(1), obj->prim_count_bo,
I915_GEM_DOMAIN_INSTRUCTION,
I915_GEM_DOMAIN_INSTRUCTION,
START_OFFSET + i * sizeof(uint64_t));
/* GPR2 = Ending Snapshot */
brw_load_register_reg64(brw, GEN7_SO_NUM_PRIMS_WRITTEN(i), HSW_CS_GPR(2));
BEGIN_BATCH(9);
OUT_BATCH(HSW_MI_MATH | (9 - 2));
/* GPR1 = GPR2 (End) - GPR1 (Start) */
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCA, R2));
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCB, R1));
OUT_BATCH(MI_MATH_ALU0(SUB));
OUT_BATCH(MI_MATH_ALU2(STORE, R1, ACCU));
/* GPR0 = GPR0 (Tally) + GPR1 (Diff) */
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCA, R0));
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCB, R1));
OUT_BATCH(MI_MATH_ALU0(ADD));
OUT_BATCH(MI_MATH_ALU2(STORE, R0, ACCU));
ADVANCE_BATCH();
}
if (!finalize) {
/* Write back the new tally */
brw_store_register_mem32(brw, obj->prim_count_bo, HSW_CS_GPR(0),
TALLY_OFFSET + i * sizeof(uint32_t));
} else {
/* Convert the number of primitives to the number of vertices. */
if (obj->primitive_mode == GL_LINES) {
/* Double R0 (R0 = R0 + R0) */
BEGIN_BATCH(5);
OUT_BATCH(HSW_MI_MATH | (5 - 2));
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCA, R0));
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCB, R0));
OUT_BATCH(MI_MATH_ALU0(ADD));
OUT_BATCH(MI_MATH_ALU2(STORE, R0, ACCU));
ADVANCE_BATCH();
} else if (obj->primitive_mode == GL_TRIANGLES) {
/* Triple R0 (R1 = R0 + R0, R0 = R0 + R1) */
BEGIN_BATCH(9);
OUT_BATCH(HSW_MI_MATH | (9 - 2));
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCA, R0));
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCB, R0));
OUT_BATCH(MI_MATH_ALU0(ADD));
OUT_BATCH(MI_MATH_ALU2(STORE, R1, ACCU));
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCA, R0));
OUT_BATCH(MI_MATH_ALU2(LOAD, SRCB, R1));
OUT_BATCH(MI_MATH_ALU0(ADD));
OUT_BATCH(MI_MATH_ALU2(STORE, R0, ACCU));
ADVANCE_BATCH();
}
/* Store it to the final result */
brw_store_register_mem32(brw, obj->prim_count_bo, HSW_CS_GPR(0),
i * sizeof(uint32_t));
}
}
}
static void
brw_emit_prim(struct brw_context *brw,
const struct _mesa_prim *prim,
uint32_t hw_prim)
{
int verts_per_instance;
int vertex_access_type;
int indirect_flag;
DBG("PRIM: %s %d %d\n", _mesa_enum_to_string(prim->mode),
prim->start, prim->count);
int start_vertex_location = prim->start;
int base_vertex_location = prim->basevertex;
if (prim->indexed) {
vertex_access_type = brw->gen >= 7 ?
GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM :
GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
start_vertex_location += brw->ib.start_vertex_offset;
base_vertex_location += brw->vb.start_vertex_bias;
} else {
vertex_access_type = brw->gen >= 7 ?
GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL :
GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
start_vertex_location += brw->vb.start_vertex_bias;
}
/* We only need to trim the primitive count on pre-Gen6. */
if (brw->gen < 6)
verts_per_instance = trim(prim->mode, prim->count);
else
verts_per_instance = prim->count;
/* If nothing to emit, just return. */
if (verts_per_instance == 0 && !prim->is_indirect)
return;
/* If we're set to always flush, do it before and after the primitive emit.
* We want to catch both missed flushes that hurt instruction/state cache
* and missed flushes of the render cache as it heads to other parts of
* the besides the draw code.
*/
if (brw->always_flush_cache)
brw_emit_mi_flush(brw);
/* If indirect, emit a bunch of loads from the indirect BO. */
if (prim->is_indirect) {
struct gl_buffer_object *indirect_buffer = brw->ctx.DrawIndirectBuffer;
drm_intel_bo *bo = intel_bufferobj_buffer(brw,
intel_buffer_object(indirect_buffer),
prim->indirect_offset, 5 * sizeof(GLuint));
indirect_flag = GEN7_3DPRIM_INDIRECT_PARAMETER_ENABLE;
brw_load_register_mem(brw, GEN7_3DPRIM_VERTEX_COUNT, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 0);
brw_load_register_mem(brw, GEN7_3DPRIM_INSTANCE_COUNT, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 4);
brw_load_register_mem(brw, GEN7_3DPRIM_START_VERTEX, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 8);
if (prim->indexed) {
brw_load_register_mem(brw, GEN7_3DPRIM_BASE_VERTEX, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 12);
brw_load_register_mem(brw, GEN7_3DPRIM_START_INSTANCE, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 16);
} else {
brw_load_register_mem(brw, GEN7_3DPRIM_START_INSTANCE, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 12);
BEGIN_BATCH(3);
OUT_BATCH(MI_LOAD_REGISTER_IMM | (3 - 2));
OUT_BATCH(GEN7_3DPRIM_BASE_VERTEX);
OUT_BATCH(0);
ADVANCE_BATCH();
}
} else {
indirect_flag = 0;
}
BEGIN_BATCH(brw->gen >= 7 ? 7 : 6);
if (brw->gen >= 7) {
const int predicate_enable =
(brw->predicate.state == BRW_PREDICATE_STATE_USE_BIT)
? GEN7_3DPRIM_PREDICATE_ENABLE : 0;
OUT_BATCH(CMD_3D_PRIM << 16 | (7 - 2) | indirect_flag | predicate_enable);
OUT_BATCH(hw_prim | vertex_access_type);
} else {
OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
hw_prim << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
vertex_access_type);
}
OUT_BATCH(verts_per_instance);
OUT_BATCH(start_vertex_location);
OUT_BATCH(prim->num_instances);
OUT_BATCH(prim->base_instance);
OUT_BATCH(base_vertex_location);
ADVANCE_BATCH();
if (brw->always_flush_cache)
brw_emit_mi_flush(brw);
}
