freedreno: re-order support for hw queries

Push query state down to batch, and use the resource tracking to figure
out which batch(es) need to be flushed to get the query result.

This means we actually need to allocate the prsc up front, before we
know the size.  So we have to add a special way to allocate an un-
backed resource, and then later allocate the backing storage.

Signed-off-by: Rob Clark <[email protected]>

4 files changed, 14 insertions, 13 deletions

diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
index 88e1a40ec90..9ce93f6e33f 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
@@ -736,8 +736,9 @@ fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
  * state, there could have been a context switch between ioctls):
  */
 void
-fd4_emit_restore(struct fd_context *ctx, struct fd_ringbuffer *ring)
+fd4_emit_restore(struct fd_batch *batch, struct fd_ringbuffer *ring)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd4_context *fd4_ctx = fd4_context(ctx);
 
 	OUT_PKT0(ring, REG_A4XX_RBBM_PERFCTR_CTL, 1);
@@ -885,7 +886,7 @@ fd4_emit_restore(struct fd_context *ctx, struct fd_ringbuffer *ring)
 	OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
 	OUT_RING(ring, 0x0);
 
-	fd_hw_query_enable(ctx, ring);
+	fd_hw_query_enable(batch, ring);
 
 	ctx->needs_rb_fbd = true;
 }
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.h b/src/gallium/drivers/freedreno/a4xx/fd4_emit.h
index 89dc51ad1ee..42e0e5e645a 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.h
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.h
@@ -102,7 +102,7 @@ void fd4_emit_vertex_bufs(struct fd_ringbuffer *ring, struct fd4_emit *emit);
 void fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		struct fd4_emit *emit);
 
-void fd4_emit_restore(struct fd_context *ctx, struct fd_ringbuffer *ring);
+void fd4_emit_restore(struct fd_batch *batch, struct fd_ringbuffer *ring);
 
 void fd4_emit_init(struct pipe_context *pctx);
 
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_gmem.c b/src/gallium/drivers/freedreno/a4xx/fd4_gmem.c
index afd37a88f43..3f3847c2a28 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_gmem.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_gmem.c
@@ -527,7 +527,7 @@ fd4_emit_sysmem_prep(struct fd_batch *batch)
 	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	struct fd_ringbuffer *ring = batch->gmem;
 
-	fd4_emit_restore(batch->ctx, ring);
+	fd4_emit_restore(batch, ring);
 
 	OUT_PKT0(ring, REG_A4XX_RB_FRAME_BUFFER_DIMENSION, 1);
 	OUT_RING(ring, A4XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
@@ -666,7 +666,7 @@ fd4_emit_tile_init(struct fd_batch *batch)
 	struct fd_ringbuffer *ring = batch->gmem;
 	struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
 
-	fd4_emit_restore(batch->ctx, ring);
+	fd4_emit_restore(batch, ring);
 
 	OUT_PKT0(ring, REG_A4XX_VSC_BIN_SIZE, 1);
 	OUT_RING(ring, A4XX_VSC_BIN_SIZE_WIDTH(gmem->bin_w) |
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_query.c b/src/gallium/drivers/freedreno/a4xx/fd4_query.c
index 41e3e6506bd..921384c1911 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_query.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_query.c
@@ -48,10 +48,10 @@ struct fd_rb_samp_ctrs {
  */
 
 static struct fd_hw_sample *
-occlusion_get_sample(struct fd_context *ctx, struct fd_ringbuffer *ring)
+occlusion_get_sample(struct fd_batch *batch, struct fd_ringbuffer *ring)
 {
 	struct fd_hw_sample *samp =
-			fd_hw_sample_init(ctx, sizeof(struct fd_rb_samp_ctrs));
+			fd_hw_sample_init(batch, sizeof(struct fd_rb_samp_ctrs));
 
 	/* low bits of sample addr should be zero (since they are control
 	 * flags in RB_SAMPLE_COUNT_CONTROL):
@@ -73,7 +73,7 @@ occlusion_get_sample(struct fd_context *ctx, struct fd_ringbuffer *ring)
 	OUT_RING(ring, 1);             /* NumInstances */
 	OUT_RING(ring, 0);             /* NumIndices */
 
-	fd_event_write(ctx, ring, ZPASS_DONE);
+	fd_event_write(batch->ctx, ring, ZPASS_DONE);
 
 	return samp;
 }
@@ -123,18 +123,18 @@ time_elapsed_enable(struct fd_context *ctx, struct fd_ringbuffer *ring)
 }
 
 static struct fd_hw_sample *
-time_elapsed_get_sample(struct fd_context *ctx, struct fd_ringbuffer *ring)
+time_elapsed_get_sample(struct fd_batch *batch, struct fd_ringbuffer *ring)
 {
-	struct fd_hw_sample *samp = fd_hw_sample_init(ctx, sizeof(uint64_t));
+	struct fd_hw_sample *samp = fd_hw_sample_init(batch, sizeof(uint64_t));
 
 	/* use unused part of vsc_size_mem as scratch space, to avoid
 	 * extra allocation:
 	 */
-	struct fd_bo *scratch_bo = fd4_context(ctx)->vsc_size_mem;
+	struct fd_bo *scratch_bo = fd4_context(batch->ctx)->vsc_size_mem;
 	const int sample_off = 128;
 	const int addr_off = sample_off + 8;
 
-	debug_assert(ctx->screen->max_freq > 0);
+	debug_assert(batch->ctx->screen->max_freq > 0);
 
 	/* Basic issue is that we need to read counter value to a relative
 	 * destination (with per-tile offset) rather than absolute dest
@@ -161,7 +161,7 @@ time_elapsed_get_sample(struct fd_context *ctx, struct fd_ringbuffer *ring)
 	 * shot, but that's really just polishing a turd..
 	 */
 
-	fd_wfi(ctx, ring);
+	fd_wfi(batch->ctx, ring);
 
 	/* copy sample counter _LO and _HI to scratch: */
 	OUT_PKT3(ring, CP_REG_TO_MEM, 2);