#include "intel_batchbuffer.h"
#include "intel_reg.h"
-/**
- * Emit PIPE_CONTROLs to write the current GPU timestamp into a buffer.
- */
-static void
-write_timestamp(struct brw_context *brw, drm_intel_bo *query_bo, int idx)
-{
- /* Emit workaround flushes: */
- if (brw->gen == 6) {
- /* The timestamp write below is a non-zero post-sync op, which on
- * Gen6 necessitates a CS stall. CS stalls need stall at scoreboard
- * set. See the comments for intel_emit_post_sync_nonzero_flush().
- */
- BEGIN_BATCH(4);
- OUT_BATCH(_3DSTATE_PIPE_CONTROL | (4 - 2));
- OUT_BATCH(PIPE_CONTROL_CS_STALL | PIPE_CONTROL_STALL_AT_SCOREBOARD);
- OUT_BATCH(0);
- OUT_BATCH(0);
- ADVANCE_BATCH();
- }
-
- BEGIN_BATCH(5);
- OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
- OUT_BATCH(PIPE_CONTROL_WRITE_TIMESTAMP);
- OUT_RELOC(query_bo,
- I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
- PIPE_CONTROL_GLOBAL_GTT_WRITE |
- idx * sizeof(uint64_t));
- OUT_BATCH(0);
- OUT_BATCH(0);
- ADVANCE_BATCH();
-}
-
-/**
- * Emit PIPE_CONTROLs to write the PS_DEPTH_COUNT register into a buffer.
- */
-static void
-write_depth_count(struct brw_context *brw, drm_intel_bo *query_bo, int idx)
-{
- /* Emit Sandybridge workaround flush: */
- if (brw->gen == 6)
- intel_emit_post_sync_nonzero_flush(brw);
-
- BEGIN_BATCH(5);
- OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
- OUT_BATCH(PIPE_CONTROL_DEPTH_STALL |
- PIPE_CONTROL_WRITE_DEPTH_COUNT);
- OUT_RELOC(query_bo,
- I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
- PIPE_CONTROL_GLOBAL_GTT_WRITE |
- (idx * sizeof(uint64_t)));
- OUT_BATCH(0);
- OUT_BATCH(0);
- ADVANCE_BATCH();
-}
-
/*
* Write an arbitrary 64-bit register to a buffer via MI_STORE_REGISTER_MEM.
*
/* MI_STORE_REGISTER_MEM only stores a single 32-bit value, so to
* read a full 64-bit register, we need to do two of them.
*/
- BEGIN_BATCH(6);
- OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
- OUT_BATCH(reg);
- OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
- idx * sizeof(uint64_t));
- OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
- OUT_BATCH(reg + sizeof(uint32_t));
- OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
- sizeof(uint32_t) + idx * sizeof(uint64_t));
- ADVANCE_BATCH();
+ if (brw->gen >= 8) {
+ BEGIN_BATCH(8);
+ OUT_BATCH(MI_STORE_REGISTER_MEM | (4 - 2));
+ OUT_BATCH(reg);
+ OUT_RELOC64(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ idx * sizeof(uint64_t));
+ OUT_BATCH(MI_STORE_REGISTER_MEM | (4 - 2));
+ OUT_BATCH(reg + sizeof(uint32_t));
+ OUT_RELOC64(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ sizeof(uint32_t) + idx * sizeof(uint64_t));
+ ADVANCE_BATCH();
+ } else {
+ BEGIN_BATCH(6);
+ OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
+ OUT_BATCH(reg);
+ OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ idx * sizeof(uint64_t));
+ OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
+ OUT_BATCH(reg + sizeof(uint32_t));
+ OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ sizeof(uint32_t) + idx * sizeof(uint64_t));
+ ADVANCE_BATCH();
+ }
}
static void
write_primitives_generated(struct brw_context *brw,
- drm_intel_bo *query_bo, int idx)
+ drm_intel_bo *query_bo, int stream, int idx)
{
intel_batchbuffer_emit_mi_flush(brw);
- brw_store_register_mem64(brw, query_bo, CL_INVOCATION_COUNT, idx);
+ if (brw->gen >= 7 && stream > 0) {
+ brw_store_register_mem64(brw, query_bo,
+ GEN7_SO_PRIM_STORAGE_NEEDED(stream), idx);
+ } else {
+ brw_store_register_mem64(brw, query_bo, CL_INVOCATION_COUNT, idx);
+ }
}
static void
write_xfb_primitives_written(struct brw_context *brw,
- drm_intel_bo *bo, int idx)
+ drm_intel_bo *bo, int stream, int idx)
{
intel_batchbuffer_emit_mi_flush(brw);
if (brw->gen >= 7) {
- brw_store_register_mem64(brw, bo, GEN7_SO_NUM_PRIMS_WRITTEN(0), idx);
+ brw_store_register_mem64(brw, bo, GEN7_SO_NUM_PRIMS_WRITTEN(stream), idx);
} else {
brw_store_register_mem64(brw, bo, GEN6_SO_NUM_PRIMS_WRITTEN, idx);
}
if (query->bo == NULL)
return;
- /* If the application has requested the query result, but this batch is
- * still contributing to it, flush it now so the results will be present
- * when mapped.
- */
- if (drm_intel_bo_references(brw->batch.bo, query->bo))
- intel_batchbuffer_flush(brw);
-
- if (unlikely(brw->perf_debug)) {
- if (drm_intel_bo_busy(query->bo)) {
- perf_debug("Stalling on the GPU waiting for a query object.\n");
- }
- }
-
- drm_intel_bo_map(query->bo, false);
+ brw_bo_map(brw, query->bo, false, "query object");
uint64_t *results = query->bo->virtual;
switch (query->Base.Target) {
case GL_TIME_ELAPSED:
break;
default:
- assert(!"Unrecognized query target in brw_queryobj_get_results()");
- break;
+ unreachable("Unrecognized query target in brw_queryobj_get_results()");
}
drm_intel_bo_unmap(query->bo);
*/
drm_intel_bo_unreference(query->bo);
query->bo = NULL;
+
+ query->Base.Ready = true;
}
/**
* obtain the time elapsed. Notably, this includes time elapsed while
* the system was doing other work, such as running other applications.
*/
- write_timestamp(brw, query->bo, 0);
+ brw_write_timestamp(brw, query->bo, 0);
break;
case GL_ANY_SAMPLES_PASSED:
case GL_ANY_SAMPLES_PASSED_CONSERVATIVE:
case GL_SAMPLES_PASSED_ARB:
- write_depth_count(brw, query->bo, 0);
+ brw_write_depth_count(brw, query->bo, 0);
break;
case GL_PRIMITIVES_GENERATED:
- write_primitives_generated(brw, query->bo, 0);
+ write_primitives_generated(brw, query->bo, query->Base.Stream, 0);
break;
case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
- write_xfb_primitives_written(brw, query->bo, 0);
+ write_xfb_primitives_written(brw, query->bo, query->Base.Stream, 0);
break;
default:
- assert(!"Unrecognized query target in brw_begin_query()");
- break;
+ unreachable("Unrecognized query target in brw_begin_query()");
}
}
switch (query->Base.Target) {
case GL_TIME_ELAPSED:
- write_timestamp(brw, query->bo, 1);
+ brw_write_timestamp(brw, query->bo, 1);
break;
case GL_ANY_SAMPLES_PASSED:
case GL_ANY_SAMPLES_PASSED_CONSERVATIVE:
case GL_SAMPLES_PASSED_ARB:
- write_depth_count(brw, query->bo, 1);
+ brw_write_depth_count(brw, query->bo, 1);
break;
case GL_PRIMITIVES_GENERATED:
- write_primitives_generated(brw, query->bo, 1);
+ write_primitives_generated(brw, query->bo, query->Base.Stream, 1);
break;
case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
- write_xfb_primitives_written(brw, query->bo, 1);
+ write_xfb_primitives_written(brw, query->bo, query->Base.Stream, 1);
break;
default:
- assert(!"Unrecognized query target in brw_end_query()");
- break;
+ unreachable("Unrecognized query target in brw_end_query()");
}
+
+ /* The current batch contains the commands to handle EndQuery(),
+ * but they won't actually execute until it is flushed.
+ */
+ query->flushed = false;
+}
+
+/**
+ * Flush the batch if it still references the query object BO.
+ */
+static void
+flush_batch_if_needed(struct brw_context *brw, struct brw_query_object *query)
+{
+ /* If the batch doesn't reference the BO, it must have been flushed
+ * (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);
+
+ if (!query->flushed)
+ intel_batchbuffer_flush(brw);
}
/**
*/
static void gen6_wait_query(struct gl_context *ctx, struct gl_query_object *q)
{
+ struct brw_context *brw = brw_context(ctx);
struct brw_query_object *query = (struct brw_query_object *)q;
+ /* If the application has requested the query result, but this batch is
+ * still contributing to it, flush it now to finish that work so the
+ * result will become available (eventually).
+ */
+ flush_batch_if_needed(brw, query);
+
gen6_queryobj_get_results(ctx, query);
- query->Base.Ready = true;
}
/**
struct brw_context *brw = brw_context(ctx);
struct brw_query_object *query = (struct brw_query_object *)q;
+ /* If query->bo is NULL, we've already gathered the results - this is a
+ * redundant CheckQuery call. Ignore it.
+ */
+ if (query->bo == NULL)
+ return;
+
/* From the GL_ARB_occlusion_query spec:
*
* "Instead of allowing for an infinite loop, performing a
* not ready yet on the first time it is queried. This ensures that
* the async query will return true in finite time.
*/
- if (query->bo && drm_intel_bo_references(brw->batch.bo, query->bo))
- intel_batchbuffer_flush(brw);
+ flush_batch_if_needed(brw, query);
- if (query->bo == NULL || !drm_intel_bo_busy(query->bo)) {
+ if (!drm_intel_bo_busy(query->bo)) {
gen6_queryobj_get_results(ctx, query);
- query->Base.Ready = true;
}
}