[Mesa-dev] [PATCH 1/3] i965/gen6-7/sol: Restructure primitive counter into a separate type.
Francisco Jerez
currojerez at riseup.net
Fri Nov 17 22:28:47 UTC 2017
A primitive counter encapsulates a scalar aggregating counter for each
vertex stream along with a section within the primitive tally buffer
which hasn't been read out yet. Defining this as a separate type will
allow us to keep multiple counter objects around for the same
transform feedback object without any code duplication.
---
src/mesa/drivers/dri/i965/brw_context.h | 38 ++++++++++++++++++---
src/mesa/drivers/dri/i965/gen6_sol.c | 53 ++++++++++++++----------------
src/mesa/drivers/dri/i965/gen7_sol_state.c | 6 +---
3 files changed, 58 insertions(+), 39 deletions(-)
diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index b3d7c6baf8a..6a36cb12c84 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -490,6 +490,36 @@ struct intel_batchbuffer {
#define BRW_MAX_XFB_STREAMS 4
+struct brw_transform_feedback_counter {
+ /**
+ * Index of the first entry of this counter within the primitive count BO.
+ * An entry is considered to be an N-tuple of 64bit values, where N is the
+ * number of vertex streams supported by the platform.
+ */
+ unsigned bo_start;
+
+ /**
+ * Index one past the last entry of this counter within the primitive
+ * count BO.
+ */
+ unsigned bo_end;
+
+ /**
+ * Primitive count values accumulated while this counter was active,
+ * excluding any entries buffered between \c bo_start and \c bo_end, which
+ * haven't been accounted for yet.
+ */
+ uint64_t accum[BRW_MAX_XFB_STREAMS];
+};
+
+static inline void
+brw_reset_transform_feedback_counter(
+ struct brw_transform_feedback_counter *counter)
+{
+ counter->bo_start = counter->bo_end;
+ memset(&counter->accum, 0, sizeof(counter->accum));
+}
+
struct brw_transform_feedback_object {
struct gl_transform_feedback_object base;
@@ -508,14 +538,12 @@ struct brw_transform_feedback_object {
*/
unsigned max_index;
+ struct brw_bo *prim_count_bo;
+
/**
* Count of primitives generated during this transform feedback operation.
- * @{
*/
- uint64_t prims_generated[BRW_MAX_XFB_STREAMS];
- struct brw_bo *prim_count_bo;
- unsigned prim_count_buffer_index; /**< in number of uint64_t units */
- /** @} */
+ struct brw_transform_feedback_counter counter;
/**
* Number of vertices written between last Begin/EndTransformFeedback().
diff --git a/src/mesa/drivers/dri/i965/gen6_sol.c b/src/mesa/drivers/dri/i965/gen6_sol.c
index 7a510940c8e..a909339e166 100644
--- a/src/mesa/drivers/dri/i965/gen6_sol.c
+++ b/src/mesa/drivers/dri/i965/gen6_sol.c
@@ -233,37 +233,36 @@ brw_delete_transform_feedback(struct gl_context *ctx,
* Note that we expose one stream pre-Gen7, so the above is just (start, end).
*/
static void
-tally_prims_generated(struct brw_context *brw,
- struct brw_transform_feedback_object *obj)
+aggregate_transform_feedback_counter(
+ struct brw_context *brw,
+ struct brw_bo *bo,
+ struct brw_transform_feedback_counter *counter)
{
- const struct gl_context *ctx = &brw->ctx;
- const int streams = ctx->Const.MaxVertexStreams;
+ const unsigned streams = brw->ctx.Const.MaxVertexStreams;
/* If the current batch is still contributing to the number of primitives
* generated, flush it now so the results will be present when mapped.
*/
- if (brw_batch_references(&brw->batch, obj->prim_count_bo))
+ if (brw_batch_references(&brw->batch, bo))
intel_batchbuffer_flush(brw);
- if (unlikely(brw->perf_debug && brw_bo_busy(obj->prim_count_bo)))
+ if (unlikely(brw->perf_debug && brw_bo_busy(bo)))
perf_debug("Stalling for # of transform feedback primitives written.\n");
- uint64_t *prim_counts = brw_bo_map(brw, obj->prim_count_bo, MAP_READ);
+ uint64_t *prim_counts = brw_bo_map(brw, bo, MAP_READ);
+ prim_counts += counter->bo_start * streams;
- assert(obj->prim_count_buffer_index % (2 * streams) == 0);
- int pairs = obj->prim_count_buffer_index / (2 * streams);
+ for (unsigned i = counter->bo_start; i + 1 < counter->bo_end; i += 2) {
+ for (unsigned s = 0; s < streams; s++)
+ counter->accum[s] += prim_counts[streams + s] - prim_counts[s];
- for (int i = 0; i < pairs; i++) {
- for (int s = 0; s < streams; s++) {
- obj->prims_generated[s] += prim_counts[streams + s] - prim_counts[s];
- }
- prim_counts += 2 * streams; /* move to the next pair */
+ prim_counts += 2 * streams;
}
- brw_bo_unmap(obj->prim_count_bo);
+ brw_bo_unmap(bo);
/* We've already gathered up the old data; we can safely overwrite it now. */
- obj->prim_count_buffer_index = 0;
+ counter->bo_start = counter->bo_end = 0;
}
/**
@@ -288,9 +287,9 @@ brw_save_primitives_written_counters(struct brw_context *brw,
assert(obj->prim_count_bo != NULL);
/* Check if there's enough space for a new pair of four values. */
- if (obj->prim_count_buffer_index + 2 * streams >= 4096 / sizeof(uint64_t)) {
- /* Gather up the results so far and release the BO. */
- tally_prims_generated(brw, obj);
+ if ((obj->counter.bo_end + 2) * streams * sizeof(uint64_t) >= 4096) {
+ aggregate_transform_feedback_counter(brw, obj->prim_count_bo,
+ &obj->counter);
}
/* Flush any drawing so that the counters have the right values. */
@@ -299,7 +298,7 @@ brw_save_primitives_written_counters(struct brw_context *brw,
/* Emit MI_STORE_REGISTER_MEM commands to write the values. */
if (devinfo->gen >= 7) {
for (int i = 0; i < streams; i++) {
- int offset = (obj->prim_count_buffer_index + i) * sizeof(uint64_t);
+ int offset = (streams * obj->counter.bo_end + i) * sizeof(uint64_t);
brw_store_register_mem64(brw, obj->prim_count_bo,
GEN7_SO_NUM_PRIMS_WRITTEN(i),
offset);
@@ -307,11 +306,11 @@ brw_save_primitives_written_counters(struct brw_context *brw,
} else {
brw_store_register_mem64(brw, obj->prim_count_bo,
GEN6_SO_NUM_PRIMS_WRITTEN,
- obj->prim_count_buffer_index * sizeof(uint64_t));
+ obj->counter.bo_end * sizeof(uint64_t));
}
/* Update where to write data to. */
- obj->prim_count_buffer_index += streams;
+ obj->counter.bo_end++;
}
static void
@@ -337,10 +336,10 @@ compute_vertices_written_so_far(struct brw_context *brw,
}
/* Get the number of primitives generated. */
- tally_prims_generated(brw, obj);
+ aggregate_transform_feedback_counter(brw, obj->prim_count_bo, &obj->counter);
for (int i = 0; i < ctx->Const.MaxVertexStreams; i++) {
- vertices_written[i] = vertices_per_prim * obj->prims_generated[i];
+ vertices_written[i] = vertices_per_prim * obj->counter.accum[i];
}
}
@@ -437,11 +436,7 @@ brw_begin_transform_feedback(struct gl_context *ctx, GLenum mode,
* so we can't delay it any further.
*/
brw_compute_xfb_vertices_written(brw, brw_obj);
-
- /* No primitives have been generated yet. */
- for (int i = 0; i < BRW_MAX_XFB_STREAMS; i++) {
- brw_obj->prims_generated[i] = 0;
- }
+ brw_reset_transform_feedback_counter(&brw_obj->counter);
/* Store the starting value of the SO_NUM_PRIMS_WRITTEN counters. */
brw_save_primitives_written_counters(brw, brw_obj);
diff --git a/src/mesa/drivers/dri/i965/gen7_sol_state.c b/src/mesa/drivers/dri/i965/gen7_sol_state.c
index 2189ed1b74f..f097e2250be 100644
--- a/src/mesa/drivers/dri/i965/gen7_sol_state.c
+++ b/src/mesa/drivers/dri/i965/gen7_sol_state.c
@@ -50,11 +50,7 @@ gen7_begin_transform_feedback(struct gl_context *ctx, GLenum mode,
* so we can't delay it any further.
*/
brw_compute_xfb_vertices_written(brw, brw_obj);
-
- /* No primitives have been generated yet. */
- for (int i = 0; i < BRW_MAX_XFB_STREAMS; i++) {
- brw_obj->prims_generated[i] = 0;
- }
+ brw_reset_transform_feedback_counter(&brw_obj->counter);
/* Store the starting value of the SO_NUM_PRIMS_WRITTEN counters. */
brw_save_primitives_written_counters(brw, brw_obj);
--
2.14.2
More information about the mesa-dev
mailing list