*/
if (brw->ctx.Extensions.ARB_query_buffer_object &&
brw_is_query_pipelined(query)) {
- brw_emit_pipe_control_write(brw,
- PIPE_CONTROL_WRITE_IMMEDIATE,
+ unsigned flags = PIPE_CONTROL_WRITE_IMMEDIATE;
+
+ if (available) {
+ /* Order available *after* the query results. */
+ flags |= PIPE_CONTROL_FLUSH_ENABLE;
+ } else {
+ /* Make it unavailable *before* any pipelined reads. */
+ flags |= PIPE_CONTROL_CS_STALL;
+ }
+
+ brw_emit_pipe_control_write(brw, flags,
query->bo, 2 * sizeof(uint64_t),
- available, 0);
+ available);
}
}
static void
write_primitives_generated(struct brw_context *brw,
- drm_intel_bo *query_bo, int stream, int idx)
+ struct brw_bo *query_bo, int stream, int idx)
{
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
+
brw_emit_mi_flush(brw);
- if (brw->gen >= 7 && stream > 0) {
+ if (devinfo->gen >= 7 && stream > 0) {
brw_store_register_mem64(brw, query_bo,
GEN7_SO_PRIM_STORAGE_NEEDED(stream),
idx * sizeof(uint64_t));
static void
write_xfb_primitives_written(struct brw_context *brw,
- drm_intel_bo *bo, int stream, int idx)
+ struct brw_bo *bo, int stream, int idx)
{
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
+
brw_emit_mi_flush(brw);
- if (brw->gen >= 7) {
+ if (devinfo->gen >= 7) {
brw_store_register_mem64(brw, bo, GEN7_SO_NUM_PRIMS_WRITTEN(stream),
idx * sizeof(uint64_t));
} else {
}
}
+static void
+write_xfb_overflow_streams(struct gl_context *ctx,
+ struct brw_bo *bo, int stream, int count,
+ int idx)
+{
+ struct brw_context *brw = brw_context(ctx);
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
+
+ brw_emit_mi_flush(brw);
+
+ for (int i = 0; i < count; i++) {
+ int w_idx = 4 * i + idx;
+ int g_idx = 4 * i + idx + 2;
+
+ if (devinfo->gen >= 7) {
+ brw_store_register_mem64(brw, bo,
+ GEN7_SO_NUM_PRIMS_WRITTEN(stream + i),
+ g_idx * sizeof(uint64_t));
+ brw_store_register_mem64(brw, bo,
+ GEN7_SO_PRIM_STORAGE_NEEDED(stream + i),
+ w_idx * sizeof(uint64_t));
+ } else {
+ brw_store_register_mem64(brw, bo,
+ GEN6_SO_NUM_PRIMS_WRITTEN,
+ g_idx * sizeof(uint64_t));
+ brw_store_register_mem64(brw, bo,
+ GEN6_SO_PRIM_STORAGE_NEEDED,
+ w_idx * sizeof(uint64_t));
+ }
+ }
+}
+
+static bool
+check_xfb_overflow_streams(uint64_t *results, int count)
+{
+ bool overflow = false;
+
+ for (int i = 0; i < count; i++) {
+ uint64_t *result_i = &results[4 * i];
+
+ if ((result_i[3] - result_i[2]) != (result_i[1] - result_i[0])) {
+ overflow = true;
+ break;
+ }
+ }
+
+ return overflow;
+}
+
static inline int
pipeline_target_to_index(int target)
{
}
static void
-emit_pipeline_stat(struct brw_context *brw, drm_intel_bo *bo,
+emit_pipeline_stat(struct brw_context *brw, struct brw_bo *bo,
int stream, int target, int idx)
{
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
+
/* One source of confusion is the tessellation shader statistics. The
* hardware has no statistics specific to the TE unit. Ideally we could have
* the HS primitives for TESS_CONTROL_SHADER_PATCHES_ARB, and the DS
/* Gen6 GS code counts full primitives, that is, it won't count individual
* triangles in a triangle strip. Use CL_INVOCATION_COUNT for that.
*/
- if (brw->gen == 6 && target == GL_GEOMETRY_SHADER_PRIMITIVES_EMITTED_ARB)
+ if (devinfo->gen == 6 && target == GL_GEOMETRY_SHADER_PRIMITIVES_EMITTED_ARB)
reg = CL_INVOCATION_COUNT;
assert(reg != 0);
struct brw_query_object *query)
{
struct brw_context *brw = brw_context(ctx);
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
if (query->bo == NULL)
return;
- brw_bo_map(brw, query->bo, false, "query object");
- uint64_t *results = query->bo->virtual;
+ uint64_t *results = brw_bo_map(brw, query->bo, MAP_READ);
switch (query->Base.Target) {
case GL_TIME_ELAPSED:
/* The query BO contains the starting and ending timestamps.
* Subtract the two and convert to nanoseconds.
*/
- query->Base.Result += 80 * (results[1] - results[0]);
+ query->Base.Result = brw_raw_timestamp_delta(brw, results[0], results[1]);
+ query->Base.Result = gen_device_info_timebase_scale(devinfo, query->Base.Result);
break;
case GL_TIMESTAMP:
- /* Our timer is a clock that increments every 80ns (regardless of
- * other clock scaling in the system). The timestamp register we can
- * read for glGetTimestamp() masks out the top 32 bits, so we do that
- * here too to let the two counters be compared against each other.
- *
- * If we just multiplied that 32 bits of data by 80, it would roll
- * over at a non-power-of-two, so an application couldn't use
- * GL_QUERY_COUNTER_BITS to handle rollover correctly. Instead, we
- * report 36 bits and truncate at that (rolling over 5 times as often
- * as the HW counter), and when the 32-bit counter rolls over, it
- * happens to also be at a rollover in the reported value from near
- * (1<<36) to 0.
- *
- * The low 32 bits rolls over in ~343 seconds. Our 36-bit result
- * rolls over every ~69 seconds.
- *
- * The query BO contains a single timestamp value in results[0].
+ /* The query BO contains a single timestamp value in results[0]. */
+ query->Base.Result = gen_device_info_timebase_scale(devinfo, results[0]);
+
+ /* Ensure the scaled timestamp overflows according to
+ * GL_QUERY_COUNTER_BITS
*/
- query->Base.Result = 80 * (results[0] & 0xffffffff);
- query->Base.Result &= (1ull << 36) - 1;
+ query->Base.Result &= (1ull << ctx->Const.QueryCounterBits.Timestamp) - 1;
break;
case GL_SAMPLES_PASSED_ARB:
query->Base.Result = results[1] - results[0];
break;
+ case GL_TRANSFORM_FEEDBACK_STREAM_OVERFLOW_ARB:
+ query->Base.Result = check_xfb_overflow_streams(results, 1);
+ break;
+
+ case GL_TRANSFORM_FEEDBACK_OVERFLOW_ARB:
+ query->Base.Result = check_xfb_overflow_streams(results, MAX_VERTEX_STREAMS);
+ break;
+
case GL_FRAGMENT_SHADER_INVOCATIONS_ARB:
query->Base.Result = (results[1] - results[0]);
/* Implement the "WaDividePSInvocationCountBy4:HSW,BDW" workaround:
* and correctly emitted the number of pixel shader invocations, but,
* whomever forgot to undo the multiply by 4.
*/
- if (brw->gen == 8 || brw->is_haswell)
+ if (devinfo->gen == 8 || devinfo->is_haswell)
query->Base.Result /= 4;
break;
default:
unreachable("Unrecognized query target in brw_queryobj_get_results()");
}
- drm_intel_bo_unmap(query->bo);
+ brw_bo_unmap(query->bo);
/* Now that we've processed the data stored in the query's buffer object,
* we can release it.
*/
- drm_intel_bo_unreference(query->bo);
+ brw_bo_unreference(query->bo);
query->bo = NULL;
query->Base.Ready = true;
struct brw_query_object *query = (struct brw_query_object *)q;
/* Since we're starting a new query, we need to throw away old results. */
- drm_intel_bo_unreference(query->bo);
- query->bo = drm_intel_bo_alloc(brw->bufmgr, "query results", 4096, 4096);
+ brw_bo_unreference(query->bo);
+ query->bo =
+ brw_bo_alloc(brw->bufmgr, "query results", 4096, BRW_MEMZONE_OTHER);
/* For ARB_query_buffer_object: The result is not available */
set_query_availability(brw, query, false);
write_xfb_primitives_written(brw, query->bo, query->Base.Stream, 0);
break;
+ case GL_TRANSFORM_FEEDBACK_STREAM_OVERFLOW_ARB:
+ write_xfb_overflow_streams(ctx, query->bo, query->Base.Stream, 1, 0);
+ break;
+
+ case GL_TRANSFORM_FEEDBACK_OVERFLOW_ARB:
+ write_xfb_overflow_streams(ctx, query->bo, 0, MAX_VERTEX_STREAMS, 0);
+ break;
+
case GL_VERTICES_SUBMITTED_ARB:
case GL_PRIMITIVES_SUBMITTED_ARB:
case GL_VERTEX_SHADER_INVOCATIONS_ARB:
write_xfb_primitives_written(brw, query->bo, query->Base.Stream, 1);
break;
+ case GL_TRANSFORM_FEEDBACK_STREAM_OVERFLOW_ARB:
+ write_xfb_overflow_streams(ctx, query->bo, query->Base.Stream, 1, 1);
+ break;
+
+ case GL_TRANSFORM_FEEDBACK_OVERFLOW_ARB:
+ write_xfb_overflow_streams(ctx, query->bo, 0, MAX_VERTEX_STREAMS, 1);
+ break;
+
+ /* calculate overflow here */
case GL_VERTICES_SUBMITTED_ARB:
case GL_PRIMITIVES_SUBMITTED_ARB:
case GL_VERTEX_SHADER_INVOCATIONS_ARB:
* (for example, due to being full). Record that it's been flushed.
*/
query->flushed = query->flushed ||
- !drm_intel_bo_references(brw->batch.bo, query->bo);
+ !brw_batch_references(&brw->batch, query->bo);
if (!query->flushed)
intel_batchbuffer_flush(brw);
*/
flush_batch_if_needed(brw, query);
- if (!drm_intel_bo_busy(query->bo)) {
+ if (!brw_bo_busy(query->bo)) {
gen6_queryobj_get_results(ctx, query);
}
}