5 files changed, 36 insertions, 0 deletions

diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
index 811f58bbba2..8c37992e17d 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
@@ -33,6 +33,7 @@
 #include "util/u_format.h"
 
 #include "freedreno_resource.h"
+#include "freedreno_query_hw.h"
 
 #include "fd3_emit.h"
 #include "fd3_blend.h"
@@ -888,6 +889,8 @@ fd3_emit_restore(struct fd_context *ctx)
 
 	fd_wfi(ctx, ring);
 
+	fd_hw_query_enable(ctx, ring);
+
 	ctx->needs_rb_fbd = true;
 }
 
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
index 4a3f1da30ed..72154bf286a 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
@@ -33,6 +33,7 @@
 #include "util/u_format.h"
 
 #include "freedreno_resource.h"
+#include "freedreno_query_hw.h"
 
 #include "fd4_emit.h"
 #include "fd4_blend.h"
@@ -882,6 +883,8 @@ fd4_emit_restore(struct fd_context *ctx)
 	OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
 	OUT_RING(ring, 0x0);
 
+	fd_hw_query_enable(ctx, ring);
+
 	ctx->needs_rb_fbd = true;
 }
 
diff --git a/src/gallium/drivers/freedreno/freedreno_context.h b/src/gallium/drivers/freedreno/freedreno_context.h
index 9e7130ab915..85ce97c16b7 100644
--- a/src/gallium/drivers/freedreno/freedreno_context.h
+++ b/src/gallium/drivers/freedreno/freedreno_context.h
@@ -164,6 +164,9 @@ struct fd_context {
 	 */
 	struct fd_hw_sample *sample_cache[MAX_HW_SAMPLE_PROVIDERS];
 
+	/* which sample providers were active in the current batch: */
+	uint32_t active_providers;
+
 	/* tracking for current stage, to know when to start/stop
 	 * any active queries:
 	 */
diff --git a/src/gallium/drivers/freedreno/freedreno_query_hw.c b/src/gallium/drivers/freedreno/freedreno_query_hw.c
index 027fdc9de23..7e99c35d132 100644
--- a/src/gallium/drivers/freedreno/freedreno_query_hw.c
+++ b/src/gallium/drivers/freedreno/freedreno_query_hw.c
@@ -89,7 +89,9 @@ static void
 resume_query(struct fd_context *ctx, struct fd_hw_query *hq,
 		struct fd_ringbuffer *ring)
 {
+	int idx = pidx(hq->provider->query_type);
 	assert(!hq->period);
+	ctx->active_providers |= (1 << idx);
 	hq->period = util_slab_alloc(&ctx->sample_period_pool);
 	list_inithead(&hq->period->list);
 	hq->period->start = get_sample(ctx, ring, hq->base.type);
@@ -101,7 +103,9 @@ static void
 pause_query(struct fd_context *ctx, struct fd_hw_query *hq,
 		struct fd_ringbuffer *ring)
 {
+	int idx = pidx(hq->provider->query_type);
 	assert(hq->period && !hq->period->end);
+	assert(ctx->active_providers & (1 << idx));
 	hq->period->end = get_sample(ctx, ring, hq->base.type);
 	list_addtail(&hq->period->list, &hq->current_periods);
 	hq->period = NULL;
@@ -431,6 +435,23 @@ fd_hw_query_set_stage(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	ctx->stage = stage;
 }
 
+/* call the provider->enable() for all the hw queries that were active
+ * in the current batch.  This sets up perfctr selector regs statically
+ * for the duration of the batch.
+ */
+void
+fd_hw_query_enable(struct fd_context *ctx, struct fd_ringbuffer *ring)
+{
+	for (int idx = 0; idx < MAX_HW_SAMPLE_PROVIDERS; idx++) {
+		if (ctx->active_providers & (1 << idx)) {
+			assert(ctx->sample_providers[idx]);
+			if (ctx->sample_providers[idx]->enable)
+				ctx->sample_providers[idx]->enable(ctx, ring);
+		}
+	}
+	ctx->active_providers = 0;  /* clear it for next frame */
+}
+
 void
 fd_hw_query_register_provider(struct pipe_context *pctx,
 		const struct fd_hw_sample_provider *provider)
diff --git a/src/gallium/drivers/freedreno/freedreno_query_hw.h b/src/gallium/drivers/freedreno/freedreno_query_hw.h
index 8f4b1f58ee5..8a5d114d806 100644
--- a/src/gallium/drivers/freedreno/freedreno_query_hw.h
+++ b/src/gallium/drivers/freedreno/freedreno_query_hw.h
@@ -76,6 +76,11 @@ struct fd_hw_sample_provider {
 	/* stages applicable to the query type: */
 	enum fd_render_stage active;
 
+	/* Optional hook for enabling a counter.  Guaranteed to happen
+	 * at least once before the first ->get_sample() in a batch.
+	 */
+	void (*enable)(struct fd_context *ctx, struct fd_ringbuffer *ring);
+
 	/* when a new sample is required, emit appropriate cmdstream
 	 * and return a sample object:
 	 */
@@ -144,6 +149,7 @@ void fd_hw_query_prepare_tile(struct fd_context *ctx, uint32_t n,
 		struct fd_ringbuffer *ring);
 void fd_hw_query_set_stage(struct fd_context *ctx,
 		struct fd_ringbuffer *ring, enum fd_render_stage stage);
+void fd_hw_query_enable(struct fd_context *ctx, struct fd_ringbuffer *ring);
 void fd_hw_query_register_provider(struct pipe_context *pctx,
 		const struct fd_hw_sample_provider *provider);
 void fd_hw_query_init(struct pipe_context *pctx);