freedreno/batch: Move query_buf allocation

This lets us move fd_batch_update_queries() after resource tracking.
Which will become needed in the next patch which adds validation to
assert needed BOs are attached to the submit.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25465>

src/gallium/drivers/freedreno/freedreno_batch_cache.c

@@ -440,6 +440,34 @@ alloc_batch_locked(struct fd_batch_cache *cache, struct fd_context *ctx,
    return batch;
 }
 
+static void
+alloc_query_buf(struct fd_context *ctx, struct fd_batch *batch)
+{
+   if (batch->query_buf)
+      return;
+
+   if ((ctx->screen->gen < 3) || (ctx->screen->gen > 4))
+      return;
+
+   /* For gens that use fd_hw_query, pre-allocate an initially zero-sized
+    * (unbacked) query buffer.  This simplifies draw/grid/etc-time resource
+    * tracking.
+    */
+   struct pipe_screen *pscreen = &ctx->screen->base;
+   struct pipe_resource templ = {
+      .target = PIPE_BUFFER,
+      .format = PIPE_FORMAT_R8_UNORM,
+      .bind = PIPE_BIND_QUERY_BUFFER,
+      .width0 = 0, /* create initially zero size buffer */
+      .height0 = 1,
+      .depth0 = 1,
+      .array_size = 1,
+      .last_level = 0,
+      .nr_samples = 1,
+   };
+   batch->query_buf = pscreen->resource_create(pscreen, &templ);
+}
+
 struct fd_batch *
 fd_bc_alloc_batch(struct fd_context *ctx, bool nondraw)
 {
@@ -457,6 +485,8 @@ fd_bc_alloc_batch(struct fd_context *ctx, bool nondraw)
    batch = alloc_batch_locked(cache, ctx, nondraw);
    fd_screen_unlock(ctx->screen);
 
+   alloc_query_buf(ctx, batch);
+
    if (batch && nondraw)
       fd_context_switch_to(ctx, batch);
 
@@ -552,6 +582,8 @@ fd_batch_from_fb(struct fd_context *ctx,
    struct fd_batch *batch = batch_from_key(ctx, key);
    fd_screen_unlock(ctx->screen);
 
+   alloc_query_buf(ctx, batch);
+
    fd_batch_set_fb(batch, pfb);
 
    return batch;

src/gallium/drivers/freedreno/freedreno_draw.c

@@ -234,13 +234,8 @@ batch_draw_tracking(struct fd_batch *batch, const struct pipe_draw_info *info,
 {
    struct fd_context *ctx = batch->ctx;
 
-   /* NOTE: needs to be before resource_written(batch->query_buf), otherwise
-    * query_buf may not be created yet.
-    */
-   fd_batch_update_queries(batch);
-
    if (!needs_draw_tracking(batch, info, indirect))
-      return;
+      goto out;
 
    /*
     * Figure out the buffers/features we need:
@@ -268,6 +263,9 @@ batch_draw_tracking(struct fd_batch *batch, const struct pipe_draw_info *info,
    resource_written(batch, batch->query_buf);
 
    fd_screen_unlock(ctx->screen);
+
+out:
+   fd_batch_update_queries(batch);
 }
 
 static void
@@ -585,11 +583,6 @@ fd_launch_grid(struct pipe_context *pctx,
    fd_batch_reference(&save_batch, ctx->batch);
    fd_batch_reference(&ctx->batch, batch);
 
-   /* NOTE: needs to be before resource_written(batch->query_buf), otherwise
-    * query_buf may not be created yet.
-    */
-   fd_batch_update_queries(batch);
-
    fd_screen_lock(ctx->screen);
 
    /* Mark SSBOs */
@@ -636,6 +629,8 @@ fd_launch_grid(struct pipe_context *pctx,
 
    fd_screen_unlock(ctx->screen);
 
+   fd_batch_update_queries(batch);
+
    DBG("%p: work_dim=%u, block=%ux%ux%u, grid=%ux%ux%u",
        batch, info->work_dim,
        info->block[0], info->block[1], info->block[2],

src/gallium/drivers/freedreno/freedreno_query_hw.c

@@ -299,22 +299,6 @@ fd_hw_sample_init(struct fd_batch *batch, uint32_t size)
    samp->tile_stride = 0;
    batch->next_sample_offset += size;
 
-   if (!batch->query_buf) {
-      struct pipe_screen *pscreen = &batch->ctx->screen->base;
-      struct pipe_resource templ = {
-         .target = PIPE_BUFFER,
-         .format = PIPE_FORMAT_R8_UNORM,
-         .bind = PIPE_BIND_QUERY_BUFFER,
-         .width0 = 0, /* create initially zero size buffer */
-         .height0 = 1,
-         .depth0 = 1,
-         .array_size = 1,
-         .last_level = 0,
-         .nr_samples = 1,
-      };
-      batch->query_buf = pscreen->resource_create(pscreen, &templ);
-   }
-
    pipe_resource_reference(&samp->prsc, batch->query_buf);
 
    return samp;