diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index 3911ee73d302eb28324091580dd947bd3bfe72c1..2fdea6cc35f405ddeb6e1fa33ec7e409ef83f02a 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -1727,10 +1727,14 @@ cso_draw_vbo(struct cso_context *cso,
    struct u_vbuf *vbuf = cso->vbuf_current;
 
    /* We can't have both indirect drawing and SO-vertex-count drawing */
-   assert(info->indirect == NULL || info->count_from_stream_output == NULL);
+   assert(!info->indirect ||
+          info->indirect->buffer == NULL ||
+          info->indirect->count_from_stream_output == NULL);
 
    /* We can't have SO-vertex-count drawing with an index buffer */
-   assert(info->count_from_stream_output == NULL || info->index_size == 0);
+   assert(info->index_size == 0 ||
+          !info->indirect ||
+          info->indirect->count_from_stream_output == NULL);
 
    if (vbuf) {
       u_vbuf_draw_vbo(vbuf, info);
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 0ea1f14c1e0eef159f0ba694209c91c22cb0ffd4..a1626c7171f995291349ea29b6580655462ec6ee 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -445,9 +445,9 @@ resolve_draw_info(const struct pipe_draw_info *raw_info,
 {
    memcpy(info, raw_info, sizeof(struct pipe_draw_info));
 
-   if (raw_info->count_from_stream_output) {
+   if (raw_info->indirect && raw_info->indirect->count_from_stream_output) {
       struct draw_so_target *target =
-         (struct draw_so_target *)info->count_from_stream_output;
+         (struct draw_so_target *)info->indirect->count_from_stream_output;
       assert(vertex_buffer != NULL);
       info->count = vertex_buffer->stride == 0 ? 0 :
                        target->internal_offset / vertex_buffer->stride;
@@ -455,6 +455,7 @@ resolve_draw_info(const struct pipe_draw_info *raw_info,
       /* Stream output draw can not be indexed */
       debug_assert(!info->index_size);
       info->max_index = info->count - 1;
+      info->indirect = NULL;
    }
 }
 
diff --git a/src/gallium/auxiliary/driver_ddebug/dd_draw.c b/src/gallium/auxiliary/driver_ddebug/dd_draw.c
index b93890f4f99457660e0f898453b199f60bd6f239..e9e3b493d89178becb62ed4c9615f6f9fd2975ed 100644
--- a/src/gallium/auxiliary/driver_ddebug/dd_draw.c
+++ b/src/gallium/auxiliary/driver_ddebug/dd_draw.c
@@ -356,13 +356,14 @@ dd_dump_draw_vbo(struct dd_draw_state *dstate, struct pipe_draw_info *info, FILE
    int sh, i;
 
    DUMP(draw_info, info);
-   if (info->count_from_stream_output)
-      DUMP_M(stream_output_target, info,
-             count_from_stream_output);
    if (info->indirect) {
-      DUMP_M(resource, info, indirect->buffer);
+      if (info->indirect->buffer)
+         DUMP_M(resource, info, indirect->buffer);
       if (info->indirect->indirect_draw_count)
          DUMP_M(resource, info, indirect->indirect_draw_count);
+      if (info->indirect->count_from_stream_output)
+         DUMP_M(stream_output_target, info,
+                indirect->count_from_stream_output);
    }
 
    fprintf(f, "\n");
@@ -705,7 +706,7 @@ dd_unreference_copy_of_call(struct dd_call *dst)
    case CALL_FLUSH:
       break;
    case CALL_DRAW_VBO:
-      pipe_so_target_reference(&dst->info.draw_vbo.draw.count_from_stream_output, NULL);
+      pipe_so_target_reference(&dst->info.draw_vbo.indirect.count_from_stream_output, NULL);
       pipe_resource_reference(&dst->info.draw_vbo.indirect.buffer, NULL);
       pipe_resource_reference(&dst->info.draw_vbo.indirect.indirect_draw_count, NULL);
       if (dst->info.draw_vbo.draw.index_size &&
@@ -1305,9 +1306,6 @@ dd_context_draw_vbo(struct pipe_context *_pipe,
 
    record->call.type = CALL_DRAW_VBO;
    record->call.info.draw_vbo.draw = *info;
-   record->call.info.draw_vbo.draw.count_from_stream_output = NULL;
-   pipe_so_target_reference(&record->call.info.draw_vbo.draw.count_from_stream_output,
-                            info->count_from_stream_output);
    if (info->index_size && !info->has_user_indices) {
       record->call.info.draw_vbo.draw.index.resource = NULL;
       pipe_resource_reference(&record->call.info.draw_vbo.draw.index.resource,
@@ -1324,6 +1322,9 @@ dd_context_draw_vbo(struct pipe_context *_pipe,
       record->call.info.draw_vbo.indirect.indirect_draw_count = NULL;
       pipe_resource_reference(&record->call.info.draw_vbo.indirect.indirect_draw_count,
                               info->indirect->indirect_draw_count);
+      record->call.info.draw_vbo.indirect.count_from_stream_output = NULL;
+      pipe_so_target_reference(&record->call.info.draw_vbo.indirect.count_from_stream_output,
+                               info->indirect->count_from_stream_output);
    } else {
       memset(&record->call.info.draw_vbo.indirect, 0, sizeof(*info->indirect));
    }
diff --git a/src/gallium/auxiliary/driver_trace/tr_dump_state.c b/src/gallium/auxiliary/driver_trace/tr_dump_state.c
index 49d2109101a11274670601d81093042c283397b4..c63c373f9235552ef33bcd487e5c6a52bfee53fe 100644
--- a/src/gallium/auxiliary/driver_trace/tr_dump_state.c
+++ b/src/gallium/auxiliary/driver_trace/tr_dump_state.c
@@ -793,7 +793,6 @@ void trace_dump_draw_info(const struct pipe_draw_info *state)
    trace_dump_member(uint, state, restart_index);
 
    trace_dump_member(ptr, state, index.resource);
-   trace_dump_member(ptr, state, count_from_stream_output);
 
    if (!state->indirect) {
       trace_dump_member(ptr, state, indirect);
@@ -804,6 +803,7 @@ void trace_dump_draw_info(const struct pipe_draw_info *state)
       trace_dump_member(uint, state, indirect->indirect_draw_count_offset);
       trace_dump_member(ptr, state, indirect->buffer);
       trace_dump_member(ptr, state, indirect->indirect_draw_count);
+      trace_dump_member(ptr, state, indirect->count_from_stream_output);
    }
 
    trace_dump_struct_end();
diff --git a/src/gallium/auxiliary/util/u_draw.c b/src/gallium/auxiliary/util/u_draw.c
index 90d01297c490cc58123283c7b36177e5a3babb21..a6835204c72aee819d9d185ee4f7ef747c997db0 100644
--- a/src/gallium/auxiliary/util/u_draw.c
+++ b/src/gallium/auxiliary/util/u_draw.c
@@ -139,7 +139,7 @@ util_draw_indirect(struct pipe_context *pipe,
    unsigned num_params = info_in->index_size ? 5 : 4;
 
    assert(info_in->indirect);
-   assert(!info_in->count_from_stream_output);
+   assert(!info_in->indirect->count_from_stream_output);
 
    memcpy(&info, info_in, sizeof(info));
 
diff --git a/src/gallium/auxiliary/util/u_dump_state.c b/src/gallium/auxiliary/util/u_dump_state.c
index f57a5923ad0a1084267378ff515f79aef24d42c0..3cbd779bfd14734a86b4d28d73b9aeb9a6120da2 100644
--- a/src/gallium/auxiliary/util/u_dump_state.c
+++ b/src/gallium/auxiliary/util/u_dump_state.c
@@ -941,7 +941,6 @@ util_dump_draw_info(FILE *stream, const struct pipe_draw_info *state)
       else
          util_dump_member(stream, ptr, state, index.resource);
    }
-   util_dump_member(stream, ptr, state, count_from_stream_output);
 
    if (!state->indirect) {
       util_dump_member(stream, ptr, state, indirect);
@@ -952,6 +951,7 @@ util_dump_draw_info(FILE *stream, const struct pipe_draw_info *state)
       util_dump_member(stream, uint, state, indirect->indirect_draw_count_offset);
       util_dump_member(stream, ptr, state, indirect->buffer);
       util_dump_member(stream, ptr, state, indirect->indirect_draw_count);
+      util_dump_member(stream, ptr, state, indirect->count_from_stream_output);
    }
 
    util_dump_struct_end(stream);
diff --git a/src/gallium/auxiliary/util/u_prim_restart.c b/src/gallium/auxiliary/util/u_prim_restart.c
index eef2b4c622cb00d1c82d3878c0e1fe03bb998e8d..f7b34f9ebc8ddcc60152b7434a18e9334d6422c7 100644
--- a/src/gallium/auxiliary/util/u_prim_restart.c
+++ b/src/gallium/auxiliary/util/u_prim_restart.c
@@ -114,7 +114,7 @@ util_translate_prim_restart_ib(struct pipe_context *context,
    dst_index_size = MAX2(2, info->index_size);
    assert(dst_index_size == 2 || dst_index_size == 4);
 
-   if (info->indirect) {
+   if (info->indirect && info->indirect->buffer) {
       indirect = read_indirect_elements(context, info->indirect);
       count = indirect.count;
       start = indirect.firstIndex;
@@ -235,7 +235,7 @@ util_draw_vbo_without_prim_restart(struct pipe_context *context,
    assert(info->index_size);
    assert(info->primitive_restart);
 
-   if (info->indirect) {
+   if (info->indirect && info->indirect->buffer) {
       indirect = read_indirect_elements(context, info->indirect);
       info_count = indirect.count;
       info_start = indirect.firstIndex;
diff --git a/src/gallium/auxiliary/util/u_threaded_context.c b/src/gallium/auxiliary/util/u_threaded_context.c
index 5da9a177fa8c1c9a44d14c3325b13c5f39d485a0..f66280da6f62f429aa0e17669aceade35e8ae49f 100644
--- a/src/gallium/auxiliary/util/u_threaded_context.c
+++ b/src/gallium/auxiliary/util/u_threaded_context.c
@@ -126,7 +126,6 @@ tc_batch_execute(void *job, UNUSED int thread_index)
          if (next != last && next->call_id == TC_CALL_draw_vbo &&
              first_info->draw.drawid == 0 &&
              !first_info->draw.indirect &&
-             !first_info->draw.count_from_stream_output &&
              is_next_call_a_mergeable_draw(first_info, next, &next_info)) {
             /* Merge up to 256 draw calls. */
             struct pipe_draw_start_count multi[256];
@@ -2199,12 +2198,12 @@ tc_call_draw_vbo(struct pipe_context *pipe, union tc_payload *payload)
    struct tc_full_draw_info *info = (struct tc_full_draw_info*)payload;
 
    pipe->draw_vbo(pipe, &info->draw);
-   pipe_so_target_reference(&info->draw.count_from_stream_output, NULL);
    if (info->draw.index_size)
       pipe_resource_reference(&info->draw.index.resource, NULL);
    if (info->draw.indirect) {
       pipe_resource_reference(&info->indirect.buffer, NULL);
       pipe_resource_reference(&info->indirect.indirect_draw_count, NULL);
+      pipe_so_target_reference(&info->indirect.count_from_stream_output, NULL);
    }
 }
 
@@ -2243,9 +2242,6 @@ tc_draw_vbo(struct pipe_context *_pipe, const struct pipe_draw_info *info)
          return;
 
       struct tc_full_draw_info *p = tc_add_draw_vbo(_pipe, false);
-      p->draw.count_from_stream_output = NULL;
-      pipe_so_target_reference(&p->draw.count_from_stream_output,
-                               info->count_from_stream_output);
       memcpy(&p->draw, info, sizeof(*info));
       p->draw.has_user_indices = false;
       p->draw.index.resource = buffer;
@@ -2253,9 +2249,6 @@ tc_draw_vbo(struct pipe_context *_pipe, const struct pipe_draw_info *info)
    } else {
       /* Non-indexed call or indexed with a real index buffer. */
       struct tc_full_draw_info *p = tc_add_draw_vbo(_pipe, indirect != NULL);
-      p->draw.count_from_stream_output = NULL;
-      pipe_so_target_reference(&p->draw.count_from_stream_output,
-                               info->count_from_stream_output);
       if (index_size) {
          tc_set_resource_reference(&p->draw.index.resource,
                                    info->index.resource);
@@ -2266,6 +2259,9 @@ tc_draw_vbo(struct pipe_context *_pipe, const struct pipe_draw_info *info)
          tc_set_resource_reference(&p->draw.indirect->buffer, indirect->buffer);
          tc_set_resource_reference(&p->indirect.indirect_draw_count,
                                    indirect->indirect_draw_count);
+         p->indirect.count_from_stream_output = NULL;
+         pipe_so_target_reference(&p->indirect.count_from_stream_output,
+                                  indirect->count_from_stream_output);
          memcpy(&p->indirect, indirect, sizeof(*indirect));
          p->draw.indirect = &p->indirect;
       }
diff --git a/src/gallium/auxiliary/util/u_vbuf.c b/src/gallium/auxiliary/util/u_vbuf.c
index 511a97dd806a9f7bbc529520105e960416c324f0..19898dc7dc5b18d874c68e391e347b7f2f7da27c 100644
--- a/src/gallium/auxiliary/util/u_vbuf.c
+++ b/src/gallium/auxiliary/util/u_vbuf.c
@@ -1306,7 +1306,7 @@ void u_vbuf_draw_vbo(struct u_vbuf *mgr, const struct pipe_draw_info *info)
    new_info = *info;
 
    /* Handle indirect (multi)draws. */
-   if (new_info.indirect) {
+   if (new_info.indirect && new_info.indirect->buffer) {
       const struct pipe_draw_indirect_info *indirect = new_info.indirect;
       unsigned draw_count = 0;
 
diff --git a/src/gallium/drivers/etnaviv/etnaviv_context.c b/src/gallium/drivers/etnaviv/etnaviv_context.c
index 9c334a450c67fa4c63eb26fb9f2066917386a6ba..40d9656affdb4b78c6191f0f625e831eaa09a01b 100644
--- a/src/gallium/drivers/etnaviv/etnaviv_context.c
+++ b/src/gallium/drivers/etnaviv/etnaviv_context.c
@@ -231,7 +231,7 @@ etna_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
    uint32_t draw_mode;
    unsigned i;
 
-   if (!info->count_from_stream_output && !info->indirect &&
+   if (!info->indirect &&
        !info->primitive_restart &&
        !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
       return;
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_draw.h b/src/gallium/drivers/freedreno/a4xx/fd4_draw.h
index 57aabf529d124f195c98c83e7527369dae30e1f4..bf3a6346aa6d218e4978bd5a1f080e69460ee152 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_draw.h
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_draw.h
@@ -97,7 +97,7 @@ fd4_draw_emit(struct fd_batch *batch, struct fd_ringbuffer *ring,
 	enum pc_di_src_sel src_sel;
 	uint32_t idx_size, idx_offset;
 
-	if (info->indirect) {
+	if (info->indirect && info->indirect->buffer) {
 		struct fd_resource *ind = fd_resource(info->indirect->buffer);
 
 		emit_marker(ring, 7);
diff --git a/src/gallium/drivers/freedreno/a5xx/fd5_draw.h b/src/gallium/drivers/freedreno/a5xx/fd5_draw.h
index 2dfd95304af50c345fab6a19a77ce1f1ec22a220..71ede21a96fcb8dcdf35ea625720a95540f86365 100644
--- a/src/gallium/drivers/freedreno/a5xx/fd5_draw.h
+++ b/src/gallium/drivers/freedreno/a5xx/fd5_draw.h
@@ -91,7 +91,7 @@ fd5_draw_emit(struct fd_batch *batch, struct fd_ringbuffer *ring,
 	enum pc_di_src_sel src_sel;
 	uint32_t max_indices, idx_offset;
 
-	if (info->indirect) {
+	if (info->indirect && info->indirect->buffer) {
 		struct fd_resource *ind = fd_resource(info->indirect->buffer);
 
 		emit_marker5(ring, 7);
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_draw.c b/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
index c68ae9445cc76b683e7a66b791564e072a4ae5d3..3f23289baa346f1de1c6defa2224a1a5b6b906ce 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
@@ -195,7 +195,7 @@ fd6_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
 	if (emit.key.gs)
 		emit.key.key.has_gs = true;
 
-	if (!(emit.key.hs || emit.key.ds || emit.key.gs || info->indirect))
+	if (!(emit.key.hs || emit.key.ds || emit.key.gs || (info->indirect && info->indirect->buffer)))
 		fd6_vsc_update_sizes(ctx->batch, info);
 
 	fixup_shader_state(ctx, &emit.key.key);
@@ -315,7 +315,7 @@ fd6_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
 	 */
 	emit_marker6(ring, 7);
 
-	if (info->indirect) {
+	if (info->indirect && info->indirect->buffer) {
 		draw_emit_indirect(ring, &draw0, info, index_offset);
 	} else {
 		draw_emit(ring, &draw0, info, index_offset);
diff --git a/src/gallium/drivers/freedreno/freedreno_draw.c b/src/gallium/drivers/freedreno/freedreno_draw.c
index 9437c7d1eae58f7e62148651d27af18a6e095281..9e432b0e173e8968618d700ecec85ec49111d88b 100644
--- a/src/gallium/drivers/freedreno/freedreno_draw.c
+++ b/src/gallium/drivers/freedreno/freedreno_draw.c
@@ -175,7 +175,7 @@ batch_draw_tracking(struct fd_batch *batch, const struct pipe_draw_info *info)
 		resource_read(batch, info->index.resource);
 
 	/* Mark indirect draw buffer as being read */
-	if (info->indirect)
+	if (info->indirect && info->indirect->buffer)
 		resource_read(batch, info->indirect->buffer);
 
 	/* Mark textures as being read */
@@ -218,13 +218,13 @@ fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 	 * to be able to emulate it, to determine if game is feeding us
 	 * bogus data:
 	 */
-	if (info->indirect && (fd_mesa_debug & FD_DBG_NOINDR)) {
+	if (info->indirect && info->indirect->buffer && (fd_mesa_debug & FD_DBG_NOINDR)) {
 		util_draw_indirect(pctx, info);
 		return;
 	}
 
 	if (info->mode != PIPE_PRIM_MAX &&
-	    !info->count_from_stream_output && !info->indirect &&
+	    !info->indirect &&
 	    !info->primitive_restart &&
 	    !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
 		return;
diff --git a/src/gallium/drivers/iris/iris_draw.c b/src/gallium/drivers/iris/iris_draw.c
index fc545eb7d36d79237b78f04b4d6bdc253bfa335c..171ce4f5a6e98ecaf18677a0cee43f9c452b197a 100644
--- a/src/gallium/drivers/iris/iris_draw.c
+++ b/src/gallium/drivers/iris/iris_draw.c
@@ -118,7 +118,7 @@ iris_update_draw_parameters(struct iris_context *ice,
    if (ice->state.vs_uses_draw_params) {
       struct iris_state_ref *draw_params = &ice->draw.draw_params;
 
-      if (info->indirect) {
+      if (info->indirect && info->indirect->buffer) {
          pipe_resource_reference(&draw_params->res, info->indirect->buffer);
          draw_params->offset =
             info->indirect->offset + (info->index_size ? 12 : 8);
@@ -192,7 +192,7 @@ iris_indirect_draw_vbo(struct iris_context *ice,
 
       iris_update_draw_parameters(ice, &info);
 
-      batch->screen->vtbl.upload_render_state(ice, batch, &info);
+      batch->screen->vtbl.upload_render_state(ice, batch, &info, info.indirect);
 
       ice->state.dirty &= ~IRIS_ALL_DIRTY_FOR_RENDER;
       ice->state.stage_dirty &= ~IRIS_ALL_STAGE_DIRTY_FOR_RENDER;
@@ -221,7 +221,7 @@ iris_simple_draw_vbo(struct iris_context *ice,
 
    iris_update_draw_parameters(ice, draw);
 
-   batch->screen->vtbl.upload_render_state(ice, batch, draw);
+   batch->screen->vtbl.upload_render_state(ice, batch, draw, draw->indirect);
 }
 
 /**
@@ -234,6 +234,7 @@ iris_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
    struct iris_screen *screen = (struct iris_screen*)ice->ctx.screen;
    const struct gen_device_info *devinfo = &screen->devinfo;
    struct iris_batch *batch = &ice->batches[IRIS_BATCH_RENDER];
+   struct pipe_draw_indirect_info *indirect = info->indirect;
 
    if (ice->state.predicate == IRIS_PREDICATE_STATE_DONT_RENDER)
       return;
@@ -269,7 +270,7 @@ iris_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
 
    iris_handle_always_flush_cache(batch);
 
-   if (info->indirect)
+   if (indirect && indirect->buffer)
       iris_indirect_draw_vbo(ice, info);
    else
       iris_simple_draw_vbo(ice, info);
diff --git a/src/gallium/drivers/iris/iris_screen.h b/src/gallium/drivers/iris/iris_screen.h
index 1a03a3f46df5ca615260f4b744a78df1f37a2dd6..420c347a7ce23465e5fc40abc1f94c24a497a4b8 100644
--- a/src/gallium/drivers/iris/iris_screen.h
+++ b/src/gallium/drivers/iris/iris_screen.h
@@ -60,7 +60,8 @@ struct iris_vtable {
    void (*init_compute_context)(struct iris_batch *batch);
    void (*upload_render_state)(struct iris_context *ice,
                                struct iris_batch *batch,
-                               const struct pipe_draw_info *draw);
+                               const struct pipe_draw_info *draw,
+                               const struct pipe_draw_indirect_info *indirect);
    void (*update_surface_base_address)(struct iris_batch *batch,
                                        struct iris_binder *binder);
    void (*upload_compute_state)(struct iris_context *ice,
diff --git a/src/gallium/drivers/iris/iris_state.c b/src/gallium/drivers/iris/iris_state.c
index 59a63f7bbabe72b97170248468449203cf3bb822..cc41a07f6d477946f50d9af7320848d82e2f3c22 100644
--- a/src/gallium/drivers/iris/iris_state.c
+++ b/src/gallium/drivers/iris/iris_state.c
@@ -6417,7 +6417,8 @@ iris_upload_dirty_render_state(struct iris_context *ice,
 static void
 iris_upload_render_state(struct iris_context *ice,
                          struct iris_batch *batch,
-                         const struct pipe_draw_info *draw)
+                         const struct pipe_draw_info *draw,
+                         const struct pipe_draw_indirect_info *indirect)
 {
    bool use_predicate = ice->state.predicate == IRIS_PREDICATE_STATE_USE_BIT;
 
@@ -6492,14 +6493,14 @@ iris_upload_render_state(struct iris_context *ice,
 #define _3DPRIM_START_INSTANCE      0x243C
 #define _3DPRIM_BASE_VERTEX         0x2440
 
-   if (draw->indirect) {
-      if (draw->indirect->indirect_draw_count) {
+   if (indirect && !indirect->count_from_stream_output) {
+      if (indirect->indirect_draw_count) {
          use_predicate = true;
 
          struct iris_bo *draw_count_bo =
-            iris_resource_bo(draw->indirect->indirect_draw_count);
+            iris_resource_bo(indirect->indirect_draw_count);
          unsigned draw_count_offset =
-            draw->indirect->indirect_draw_count_offset;
+            indirect->indirect_draw_count_offset;
 
          iris_emit_pipe_control_flush(batch,
                                       "ensure indirect draw buffer is flushed",
@@ -6551,43 +6552,43 @@ iris_upload_render_state(struct iris_context *ice,
             iris_batch_emit(batch, &mi_predicate, sizeof(uint32_t));
          }
       }
-      struct iris_bo *bo = iris_resource_bo(draw->indirect->buffer);
+      struct iris_bo *bo = iris_resource_bo(indirect->buffer);
       assert(bo);
 
       iris_emit_cmd(batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
          lrm.RegisterAddress = _3DPRIM_VERTEX_COUNT;
-         lrm.MemoryAddress = ro_bo(bo, draw->indirect->offset + 0);
+         lrm.MemoryAddress = ro_bo(bo, indirect->offset + 0);
       }
       iris_emit_cmd(batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
          lrm.RegisterAddress = _3DPRIM_INSTANCE_COUNT;
-         lrm.MemoryAddress = ro_bo(bo, draw->indirect->offset + 4);
+         lrm.MemoryAddress = ro_bo(bo, indirect->offset + 4);
       }
       iris_emit_cmd(batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
          lrm.RegisterAddress = _3DPRIM_START_VERTEX;
-         lrm.MemoryAddress = ro_bo(bo, draw->indirect->offset + 8);
+         lrm.MemoryAddress = ro_bo(bo, indirect->offset + 8);
       }
       if (draw->index_size) {
          iris_emit_cmd(batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
             lrm.RegisterAddress = _3DPRIM_BASE_VERTEX;
-            lrm.MemoryAddress = ro_bo(bo, draw->indirect->offset + 12);
+            lrm.MemoryAddress = ro_bo(bo, indirect->offset + 12);
          }
          iris_emit_cmd(batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
             lrm.RegisterAddress = _3DPRIM_START_INSTANCE;
-            lrm.MemoryAddress = ro_bo(bo, draw->indirect->offset + 16);
+            lrm.MemoryAddress = ro_bo(bo, indirect->offset + 16);
          }
       } else {
          iris_emit_cmd(batch, GENX(MI_LOAD_REGISTER_MEM), lrm) {
             lrm.RegisterAddress = _3DPRIM_START_INSTANCE;
-            lrm.MemoryAddress = ro_bo(bo, draw->indirect->offset + 12);
+            lrm.MemoryAddress = ro_bo(bo, indirect->offset + 12);
          }
          iris_emit_cmd(batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
             lri.RegisterOffset = _3DPRIM_BASE_VERTEX;
             lri.DataDWord = 0;
          }
       }
-   } else if (draw->count_from_stream_output) {
+   } else if (indirect && indirect->count_from_stream_output) {
       struct iris_stream_output_target *so =
-         (void *) draw->count_from_stream_output;
+         (void *) indirect->count_from_stream_output;
 
       /* XXX: Replace with actual cache tracking */
       iris_emit_pipe_control_flush(batch,
@@ -6615,7 +6616,7 @@ iris_upload_render_state(struct iris_context *ice,
       prim.VertexAccessType = draw->index_size > 0 ? RANDOM : SEQUENTIAL;
       prim.PredicateEnable = use_predicate;
 
-      if (draw->indirect || draw->count_from_stream_output) {
+      if (indirect) {
          prim.IndirectParameterEnable = true;
       } else {
          prim.StartInstanceLocation = draw->start_instance;
diff --git a/src/gallium/drivers/llvmpipe/lp_draw_arrays.c b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
index d9700cfd6d78a580e381cd3c65f38412d0306fc3..4ee88c888e8deee802547dad25a7d25d424ba17f 100644
--- a/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
+++ b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
@@ -61,7 +61,7 @@ llvmpipe_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
    if (!llvmpipe_check_render_cond(lp))
       return;
 
-   if (info->indirect) {
+   if (info->indirect && info->indirect->buffer) {
       util_draw_indirect(pipe, info);
       return;
    }
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_push.c b/src/gallium/drivers/nouveau/nv50/nv50_push.c
index d835e9d696a1a8db3afa4f2c920b027f6e396dd1..d10399530c9afa87bea1e04847689431552e51c3 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_push.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_push.c
@@ -292,10 +292,10 @@ nv50_push_vbo(struct nv50_context *nv50, const struct pipe_draw_info *info)
       ctx.primitive_restart = info->primitive_restart;
       ctx.restart_index = info->restart_index;
    } else {
-      if (unlikely(info->count_from_stream_output)) {
+      if (unlikely(info->indirect && info->indirect->count_from_stream_output)) {
          struct pipe_context *pipe = &nv50->base.pipe;
          struct nv50_so_target *targ;
-         targ = nv50_so_target(info->count_from_stream_output);
+         targ = nv50_so_target(info->indirect->count_from_stream_output);
          if (!targ->pq) {
             NOUVEAU_ERR("draw_stream_output not supported on pre-NVA0 cards\n");
             return;
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_vbo.c b/src/gallium/drivers/nouveau/nv50/nv50_vbo.c
index a8e2a1b3b27b108e9299b5d6e8f82bf8afadd889..7d9c47df037450e7c68dfc62e2c167347c4cbdd7 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_vbo.c
@@ -702,7 +702,7 @@ nva0_draw_stream_output(struct nv50_context *nv50,
                         const struct pipe_draw_info *info)
 {
    struct nouveau_pushbuf *push = nv50->base.pushbuf;
-   struct nv50_so_target *so = nv50_so_target(info->count_from_stream_output);
+   struct nv50_so_target *so = nv50_so_target(info->indirect->count_from_stream_output);
    struct nv04_resource *res = nv04_resource(so->pipe.buffer);
    unsigned num_instances = info->instance_count;
    unsigned mode = nv50_prim_gl(info->mode);
@@ -874,7 +874,7 @@ nv50_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
                          info->mode, info->start, info->count,
                          info->instance_count, info->index_bias, info->index_size);
    } else
-   if (unlikely(info->count_from_stream_output)) {
+   if (unlikely(info->indirect && info->indirect->count_from_stream_output)) {
       nva0_draw_stream_output(nv50, info);
    } else {
       nv50_draw_arrays(nv50,
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c
index 8287d8431b109f49d0a04257d9d0441c9734a8bf..3b44d6605c988796f4122ca94977b0a0310519be 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c
@@ -769,7 +769,7 @@ nvc0_draw_stream_output(struct nvc0_context *nvc0,
                         const struct pipe_draw_info *info)
 {
    struct nouveau_pushbuf *push = nvc0->base.pushbuf;
-   struct nvc0_so_target *so = nvc0_so_target(info->count_from_stream_output);
+   struct nvc0_so_target *so = nvc0_so_target(info->indirect->count_from_stream_output);
    struct nv04_resource *res = nv04_resource(so->pipe.buffer);
    unsigned mode = nvc0_prim_gl(info->mode);
    unsigned num_instances = info->instance_count;
@@ -938,7 +938,7 @@ nvc0_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
     * if index count is larger and we expect repeated vertices, suggest upload.
     */
    nvc0->vbo_push_hint =
-      !info->indirect && info->index_size &&
+      (!info->indirect || info->indirect->count_from_stream_output) && info->index_size &&
       (nvc0->vb_elt_limit >= (info->count * 2));
 
    /* Check whether we want to switch vertex-submission mode. */
@@ -1005,7 +1005,7 @@ nvc0_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 
    nvc0_state_validate_3d(nvc0, ~0);
 
-   if (nvc0->vertprog->vp.need_draw_parameters && !info->indirect) {
+   if (nvc0->vertprog->vp.need_draw_parameters && (!info->indirect || info->indirect->count_from_stream_output)) {
       PUSH_SPACE(push, 9);
       BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
       PUSH_DATA (push, NVC0_CB_AUX_SIZE);
@@ -1057,7 +1057,7 @@ nvc0_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
    }
 
    if (nvc0->state.vbo_mode) {
-      if (info->indirect)
+      if (info->indirect && info->indirect->buffer)
          nvc0_push_vbo_indirect(nvc0, info);
       else
          nvc0_push_vbo(nvc0, info);
@@ -1088,10 +1088,10 @@ nvc0_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
       nvc0->base.vbo_dirty = false;
    }
 
-   if (unlikely(info->indirect)) {
+   if (unlikely(info->indirect && info->indirect->buffer)) {
       nvc0_draw_indirect(nvc0, info);
    } else
-   if (unlikely(info->count_from_stream_output)) {
+   if (unlikely(info->indirect && info->indirect->count_from_stream_output)) {
       nvc0_draw_stream_output(nvc0, info);
    } else
    if (info->index_size) {
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo_translate.c b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo_translate.c
index d49a5dfd2cfd0284e5c4b190e554192d1aa57821..de08b9e6a19fb190643de0f7dcbb224959528f0b 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo_translate.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo_translate.c
@@ -596,10 +596,10 @@ nvc0_push_vbo(struct nvc0_context *nvc0, const struct pipe_draw_info *info)
       nvc0_push_map_idxbuf(&ctx, nvc0, info);
       index_size = info->index_size;
    } else {
-      if (unlikely(info->count_from_stream_output)) {
+      if (unlikely(info->indirect && info->indirect->count_from_stream_output)) {
          struct pipe_context *pipe = &nvc0->base.pipe;
          struct nvc0_so_target *targ;
-         targ = nvc0_so_target(info->count_from_stream_output);
+         targ = nvc0_so_target(info->indirect->count_from_stream_output);
          pipe->get_query_result(pipe, targ->pq, true, (void *)&vert_count);
          vert_count /= targ->stride;
       }
diff --git a/src/gallium/drivers/panfrost/pan_context.c b/src/gallium/drivers/panfrost/pan_context.c
index c451bd78f083ef69953fbcc494df63a2c17e56a8..56f10361064adaab40de88a97e174efc486a5db3 100644
--- a/src/gallium/drivers/panfrost/pan_context.c
+++ b/src/gallium/drivers/panfrost/pan_context.c
@@ -363,8 +363,8 @@ panfrost_draw_emit_tiler(struct panfrost_batch *batch,
                         cfg.base_vertex_offset = info->index_bias - ctx->offset_start;
                         cfg.index_count = info->count;
                 } else {
-                        cfg.index_count = info->count_from_stream_output ?
-                                          pan_so_target(info->count_from_stream_output)->offset :
+                        cfg.index_count = info->indirect && info->indirect->count_from_stream_output ?
+                                          pan_so_target(info->indirect->count_from_stream_output)->offset :
                                           ctx->vertex_count;
                 }
         }
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 32cc7441ff172c52de867872696ed00ff2f2b6fb..a192d457b6d43ed1781ebe6e45059b3dea89b0d0 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -2065,8 +2065,15 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 	int index_bias;
 	struct r600_shader_atomic combined_atomics[8];
 	uint8_t atomic_used_mask = 0;
+	struct pipe_draw_indirect_info *indirect = info->indirect;
+	struct pipe_stream_output_target *count_from_so = NULL;
 
-	if (!info->indirect && !info->count && (index_size || !info->count_from_stream_output)) {
+	if (indirect && indirect->count_from_stream_output) {
+		count_from_so = indirect->count_from_stream_output;
+		indirect = NULL;
+	}
+
+	if (!indirect && !info->count && (index_size || !count_from_so)) {
 		return;
 	}
 
@@ -2137,17 +2144,17 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 			void *ptr;
 			unsigned start, count;
 
-			if (likely(!info->indirect)) {
+			if (likely(!indirect)) {
 				start = 0;
 				count = info->count;
 			}
 			else {
 				/* Have to get start/count from indirect buffer, slow path ahead... */
-				struct r600_resource *indirect_resource = (struct r600_resource *)info->indirect->buffer;
+				struct r600_resource *indirect_resource = (struct r600_resource *)indirect->buffer;
 				unsigned *data = r600_buffer_map_sync_with_rings(&rctx->b, indirect_resource,
 					PIPE_MAP_READ);
 				if (data) {
-					data += info->indirect->offset / sizeof(unsigned);
+					data += indirect->offset / sizeof(unsigned);
 					start = data[2] * index_size;
 					count = data[0];
 				}
@@ -2176,7 +2183,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 		 * and the indices are emitted via PKT3_DRAW_INDEX_IMMD.
 		 * Indirect draws never use immediate indices.
 		 * Note: Instanced rendering in combination with immediate indices hangs. */
-		if (has_user_indices && (R600_BIG_ENDIAN || info->indirect ||
+		if (has_user_indices && (R600_BIG_ENDIAN || indirect ||
 						 info->instance_count > 1 ||
 						 info->count*index_size > 20)) {
 			indexbuf = NULL;
@@ -2194,7 +2201,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 	if (rctx->vgt_state.vgt_multi_prim_ib_reset_en != info->primitive_restart ||
 	    rctx->vgt_state.vgt_multi_prim_ib_reset_indx != info->restart_index ||
 	    rctx->vgt_state.vgt_indx_offset != index_bias ||
-	    (rctx->vgt_state.last_draw_was_indirect && !info->indirect)) {
+	    (rctx->vgt_state.last_draw_was_indirect && !indirect)) {
 		rctx->vgt_state.vgt_multi_prim_ib_reset_en = info->primitive_restart;
 		rctx->vgt_state.vgt_multi_prim_ib_reset_indx = info->restart_index;
 		rctx->vgt_state.vgt_indx_offset = index_bias;
@@ -2274,7 +2281,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 	}
 
 	/* Update start instance. */
-	if (!info->indirect && rctx->last_start_instance != info->start_instance) {
+	if (!indirect && rctx->last_start_instance != info->start_instance) {
 		radeon_set_ctl_const(cs, R_03CFF4_SQ_VTX_START_INST_LOC, info->start_instance);
 		rctx->last_start_instance = info->start_instance;
 	}
@@ -2289,11 +2296,11 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 	}
 
 	/* Draw packets. */
-	if (likely(!info->indirect)) {
+	if (likely(!indirect)) {
 		radeon_emit(cs, PKT3(PKT3_NUM_INSTANCES, 0, 0));
 		radeon_emit(cs, info->instance_count);
 	} else {
-		uint64_t va = r600_resource(info->indirect->buffer)->gpu_address;
+		uint64_t va = r600_resource(indirect->buffer)->gpu_address;
 		assert(rctx->b.chip_class >= EVERGREEN);
 
 		// Invalidate so non-indirect draw calls reset this state
@@ -2307,7 +2314,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 
 		radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
 		radeon_emit(cs, radeon_add_to_buffer_list(&rctx->b, &rctx->b.gfx,
-							  (struct r600_resource*)info->indirect->buffer,
+							  (struct r600_resource*)indirect->buffer,
 							  RADEON_USAGE_READ,
                                                           RADEON_PRIO_DRAW_INDIRECT));
 	}
@@ -2328,7 +2335,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 		} else {
 			uint64_t va = r600_resource(indexbuf)->gpu_address + index_offset;
 
-			if (likely(!info->indirect)) {
+			if (likely(!indirect)) {
 				radeon_emit(cs, PKT3(PKT3_DRAW_INDEX, 3, render_cond_bit));
 				radeon_emit(cs, va);
 				radeon_emit(cs, (va >> 32UL) & 0xFF);
@@ -2357,13 +2364,13 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 				radeon_emit(cs, max_size);
 
 				radeon_emit(cs, PKT3(EG_PKT3_DRAW_INDEX_INDIRECT, 1, render_cond_bit));
-				radeon_emit(cs, info->indirect->offset);
+				radeon_emit(cs, indirect->offset);
 				radeon_emit(cs, V_0287F0_DI_SRC_SEL_DMA);
 			}
 		}
 	} else {
-		if (unlikely(info->count_from_stream_output)) {
-			struct r600_so_target *t = (struct r600_so_target*)info->count_from_stream_output;
+		if (unlikely(count_from_so)) {
+			struct r600_so_target *t = (struct r600_so_target*)count_from_so;
 			uint64_t va = t->buf_filled_size->gpu_address + t->buf_filled_size_offset;
 
 			radeon_set_context_reg(cs, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE, t->stride_in_dw);
@@ -2381,16 +2388,16 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 								  RADEON_PRIO_SO_FILLED_SIZE));
 		}
 
-		if (likely(!info->indirect)) {
+		if (likely(!indirect)) {
 			radeon_emit(cs, PKT3(PKT3_DRAW_INDEX_AUTO, 1, render_cond_bit));
 			radeon_emit(cs, info->count);
 		}
 		else {
 			radeon_emit(cs, PKT3(EG_PKT3_DRAW_INDIRECT, 1, render_cond_bit));
-			radeon_emit(cs, info->indirect->offset);
+			radeon_emit(cs, indirect->offset);
 		}
 		radeon_emit(cs, V_0287F0_DI_SRC_SEL_AUTO_INDEX |
-				(info->count_from_stream_output ? S_0287F0_USE_OPAQUE(1) : 0));
+				(count_from_so ? S_0287F0_USE_OPAQUE(1) : 0));
 	}
 
 	/* SMX returns CONTEXT_DONE too early workaround */
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 18f5c2632d3945df9a6e4c0a51748533a8a86a2d..695995a3baf6500733d3c30dfcdc1fa97aff4797 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -477,6 +477,22 @@ static bool si_is_line_stipple_enabled(struct si_context *sctx)
           (rs->polygon_mode_is_lines || util_prim_is_lines(sctx->current_rast_prim));
 }
 
+static bool num_instanced_prims_less_than(const struct pipe_draw_info *info,
+                                          const struct pipe_draw_indirect_info *indirect,
+                                          enum pipe_prim_type prim,
+                                          unsigned min_vertex_count,
+                                          unsigned instance_count,
+                                          unsigned num_prims)
+{
+   if (indirect) {
+      return indirect->buffer ||
+             (instance_count > 1 && indirect->count_from_stream_output);
+   } else {
+      return instance_count > 1 &&
+             si_num_prims_for_vertices(prim, min_vertex_count, info->vertices_per_patch) < num_prims;
+   }
+}
+
 ALWAYS_INLINE
 static unsigned si_get_ia_multi_vgt_param(struct si_context *sctx,
                                           const struct pipe_draw_info *info,
@@ -484,6 +500,7 @@ static unsigned si_get_ia_multi_vgt_param(struct si_context *sctx,
                                           unsigned instance_count, bool primitive_restart,
                                           unsigned min_vertex_count)
 {
+   const struct pipe_draw_indirect_info *indirect = info->indirect;
    union si_vgt_param_key key = sctx->ia_multi_vgt_param_key;
    unsigned primgroup_size;
    unsigned ia_multi_vgt_param;
@@ -497,14 +514,11 @@ static unsigned si_get_ia_multi_vgt_param(struct si_context *sctx,
    }
 
    key.u.prim = prim;
-   key.u.uses_instancing = info->indirect || instance_count > 1;
+   key.u.uses_instancing = (indirect && indirect->buffer) || instance_count > 1;
    key.u.multi_instances_smaller_than_primgroup =
-      info->indirect ||
-      (instance_count > 1 &&
-       (info->count_from_stream_output ||
-        si_num_prims_for_vertices(prim, min_vertex_count, info->vertices_per_patch) < primgroup_size));
+      num_instanced_prims_less_than(info, indirect, prim, min_vertex_count, instance_count, primgroup_size);
    key.u.primitive_restart = primitive_restart;
-   key.u.count_from_stream_output = info->count_from_stream_output != NULL;
+   key.u.count_from_stream_output = indirect && indirect->count_from_stream_output;
    key.u.line_stipple_enabled = si_is_line_stipple_enabled(sctx);
 
    ia_multi_vgt_param =
@@ -521,11 +535,7 @@ static unsigned si_get_ia_multi_vgt_param(struct si_context *sctx,
        * only applies it to Hawaii. Do what Vulkan does.
        */
       if (sctx->family == CHIP_HAWAII && G_028AA8_SWITCH_ON_EOI(ia_multi_vgt_param) &&
-          (info->indirect ||
-           (instance_count > 1 &&
-            (info->count_from_stream_output ||
-             si_num_prims_for_vertices(prim, min_vertex_count, info->vertices_per_patch) <= 1))))
-
+          num_instanced_prims_less_than(info, indirect, prim, min_vertex_count, instance_count, 2))
          sctx->flags |= SI_CONTEXT_VGT_FLUSH;
    }
 
@@ -771,15 +781,18 @@ static void si_emit_draw_packets(struct si_context *sctx, const struct pipe_draw
    unsigned sh_base_reg = sctx->shader_pointers.sh_base[PIPE_SHADER_VERTEX];
    bool render_cond_bit = sctx->render_cond && !sctx->render_cond_force_off;
    uint32_t index_max_size = 0;
+   uint32_t use_opaque = 0;
    uint64_t index_va = 0;
 
-   if (info->count_from_stream_output) {
-      struct si_streamout_target *t = (struct si_streamout_target *)info->count_from_stream_output;
+   if (indirect && indirect->count_from_stream_output) {
+      struct si_streamout_target *t = (struct si_streamout_target *)indirect->count_from_stream_output;
 
       radeon_set_context_reg(cs, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE, t->stride_in_dw);
       si_cp_copy_data(sctx, sctx->gfx_cs, COPY_DATA_REG, NULL,
                       R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE >> 2, COPY_DATA_SRC_MEM,
                       t->buf_filled_size, t->buf_filled_size_offset);
+      use_opaque = S_0287F0_USE_OPAQUE(1);
+      indirect = NULL;
    }
 
    /* draw packet */
@@ -975,8 +988,7 @@ static void si_emit_draw_packets(struct si_context *sctx, const struct pipe_draw
 
             radeon_emit(cs, PKT3(PKT3_DRAW_INDEX_AUTO, 1, render_cond_bit));
             radeon_emit(cs, draws[i].count);
-            radeon_emit(cs, V_0287F0_DI_SRC_SEL_AUTO_INDEX |
-                            S_0287F0_USE_OPAQUE(!!info->count_from_stream_output));
+            radeon_emit(cs, V_0287F0_DI_SRC_SEL_AUTO_INDEX | use_opaque);
          }
          if (num_draws > 1 && !sctx->num_vs_blit_sgprs)
             sctx->last_base_vertex = draws[num_draws - 1].start;
@@ -1563,7 +1575,7 @@ static void si_get_draw_start_count(struct si_context *sctx, const struct pipe_d
 {
    struct pipe_draw_indirect_info *indirect = info->indirect;
 
-   if (indirect) {
+   if (indirect && !indirect->count_from_stream_output) {
       unsigned indirect_count;
       struct pipe_transfer *transfer;
       unsigned begin, end;
@@ -1761,11 +1773,12 @@ static void si_multi_draw_vbo(struct pipe_context *ctx,
 {
    struct si_context *sctx = (struct si_context *)ctx;
    struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
+   struct pipe_draw_indirect_info *indirect = info->indirect;
    struct pipe_resource *indexbuf = info->index.resource;
    unsigned dirty_tex_counter, dirty_buf_counter;
    enum pipe_prim_type rast_prim, prim = info->mode;
    unsigned index_size = info->index_size;
-   unsigned index_offset = info->indirect ? draws[0].start * index_size : 0;
+   unsigned index_offset = indirect && indirect->buffer ? draws[0].start * index_size : 0;
    unsigned instance_count = info->instance_count;
    bool primitive_restart =
       info->primitive_restart &&
@@ -1776,7 +1789,7 @@ static void si_multi_draw_vbo(struct pipe_context *ctx,
     * no workaround for indirect draws, but we can at least skip
     * direct draws.
     */
-   if (unlikely(!info->indirect && !instance_count))
+   if (unlikely(!indirect && !instance_count))
       return;
 
    struct si_shader_selector *vs = sctx->vs_shader.cso;
@@ -1892,7 +1905,7 @@ static void si_multi_draw_vbo(struct pipe_context *ctx,
       } else if (info->has_user_indices) {
          unsigned start_offset;
 
-         assert(!info->indirect);
+         assert(!indirect);
          assert(num_draws == 1);
          start_offset = draws[0].start * index_size;
 
@@ -1920,15 +1933,14 @@ static void si_multi_draw_vbo(struct pipe_context *ctx,
    unsigned min_direct_count = 0;
    unsigned total_direct_count = 0;
 
-   if (info->indirect) {
-      struct pipe_draw_indirect_info *indirect = info->indirect;
-
+   if (indirect) {
       /* Add the buffer size for memory checking in need_cs_space. */
-      si_context_add_resource_size(sctx, indirect->buffer);
+      if (indirect->buffer)
+         si_context_add_resource_size(sctx, indirect->buffer);
 
       /* Indirect buffers use TC L2 on GFX9, but not older hw. */
       if (sctx->chip_class <= GFX8) {
-         if (si_resource(indirect->buffer)->TC_L2_dirty) {
+         if (indirect->buffer && si_resource(indirect->buffer)->TC_L2_dirty) {
             sctx->flags |= SI_CONTEXT_WB_L2;
             si_resource(indirect->buffer)->TC_L2_dirty = false;
          }
@@ -1956,7 +1968,6 @@ static void si_multi_draw_vbo(struct pipe_context *ctx,
            : /* Add, then return true. */
            (sctx->compute_num_verts_ineligible += total_direct_count,
             false)) && /* Add, then return false. */
-       (!info->count_from_stream_output || pd_msg("draw_opaque")) &&
        (primitive_restart ?
                           /* Supported prim types with primitive restart: */
            (prim == PIPE_PRIM_TRIANGLE_STRIP || pd_msg("bad prim type with primitive restart")) &&
@@ -2126,7 +2137,7 @@ static void si_multi_draw_vbo(struct pipe_context *ctx,
       masked_atoms |= si_get_atom_bit(sctx, &sctx->atoms.s.scissors);
       gfx9_scissor_bug = true;
 
-      if (info->count_from_stream_output ||
+      if ((indirect && indirect->count_from_stream_output) ||
           sctx->dirty_atoms & si_atoms_that_always_roll_context() ||
           sctx->dirty_states & si_states_that_always_roll_context())
          sctx->context_roll = true;
diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c
index 91427348a25732c26a0460d92ea8473b36c0e64a..1dd42594da5429c590bdad81df3206f714b6dbeb 100644
--- a/src/gallium/drivers/softpipe/sp_draw_arrays.c
+++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c
@@ -69,7 +69,7 @@ softpipe_draw_vbo(struct pipe_context *pipe,
    if (!softpipe_check_render_cond(sp))
       return;
 
-   if (info->indirect) {
+   if (info->indirect && info->indirect->buffer) {
       util_draw_indirect(pipe, info);
       return;
    }
diff --git a/src/gallium/drivers/svga/svga_pipe_draw.c b/src/gallium/drivers/svga/svga_pipe_draw.c
index 06540e46dcfd401f3f7bb6857bb302ea033ae905..0986a1800d6c06c691fe297d28b3b96cc912a5c8 100644
--- a/src/gallium/drivers/svga/svga_pipe_draw.c
+++ b/src/gallium/drivers/svga/svga_pipe_draw.c
@@ -83,7 +83,7 @@ retry_draw_auto(struct svga_context *svga,
                 const struct pipe_draw_info *info)
 {
    assert(svga_have_sm5(svga));
-   assert(info->count_from_stream_output);
+   assert(info->indirect->count_from_stream_output);
    assert(info->instance_count == 1);
    /* SO drawing implies core profile and none of these prim types */
    assert(info->mode != PIPE_PRIM_QUADS &&
@@ -117,7 +117,7 @@ retry_draw_auto(struct svga_context *svga,
                   0,    /* start instance */
                   1,    /* only 1 instance supported */
                   NULL, /* indirect drawing info */
-                  info->count_from_stream_output));
+                  info->indirect->count_from_stream_output));
 
       return PIPE_OK;
    }
@@ -132,7 +132,7 @@ retry_draw_indirect(struct svga_context *svga,
                     const struct pipe_draw_info *info)
 {
    assert(svga_have_sm5(svga));
-   assert(info->indirect);
+   assert(info->indirect && info->indirect->buffer);
    /* indirect drawing implies core profile and none of these prim types */
    assert(info->mode != PIPE_PRIM_QUADS &&
           info->mode != PIPE_PRIM_QUAD_STRIP &&
@@ -273,8 +273,7 @@ svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
       goto done;
    }
 
-   if (!info->indirect && !info->count_from_stream_output &&
-       !u_trim_pipe_prim(info->mode, &count))
+   if (!info->indirect && !u_trim_pipe_prim(info->mode, &count))
       goto done;
 
    needed_swtnl = svga->state.sw.need_swtnl;
@@ -318,7 +317,7 @@ svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
                                svga_is_using_flat_shading(svga),
                                svga->curr.rast->templ.flatshade_first);
 
-      if (info->count_from_stream_output) {
+      if (info->indirect && info->indirect->count_from_stream_output) {
          unsigned stream = 0;
          assert(count == 0);
 
@@ -332,7 +331,7 @@ svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 
          /* Check the stream index of the specified stream output target */
          for (unsigned i = 0; i < ARRAY_SIZE(svga->so_targets); i++) {
-            if (svga->vcount_so_targets[i] == info->count_from_stream_output) {
+            if (svga->vcount_so_targets[i] == info->indirect->count_from_stream_output) {
                stream = (svga->vcount_buffer_stream >> (i * 4)) & 0xf;
                break;
             }
@@ -342,10 +341,10 @@ svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
          }
       }
 
-      if (info->count_from_stream_output && count == 0) {
+      if (info->indirect && info->indirect->count_from_stream_output && count == 0) {
          ret = retry_draw_auto(svga, info);
       }
-      else if (info->indirect) {
+      else if (info->indirect && info->indirect->buffer) {
          ret = retry_draw_indirect(svga, info);
       }
       else if (info->index_size) {
diff --git a/src/gallium/drivers/swr/swr_draw.cpp b/src/gallium/drivers/swr/swr_draw.cpp
index 399821c32eaba9acb66262adb8535242437b2c43..0bf29ef77a3cb18dad6f1394a662b334edf5f2c7 100644
--- a/src/gallium/drivers/swr/swr_draw.cpp
+++ b/src/gallium/drivers/swr/swr_draw.cpp
@@ -41,7 +41,7 @@ swr_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 {
    struct swr_context *ctx = swr_context(pipe);
 
-   if (!info->count_from_stream_output && !info->indirect &&
+   if (!info->indirect &&
        !info->primitive_restart &&
        !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
       return;
@@ -49,7 +49,7 @@ swr_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
    if (!swr_check_render_cond(pipe))
       return;
 
-   if (info->indirect) {
+   if (info->indirect && info->indirect->buffer) {
       util_draw_indirect(pipe, info);
       return;
    }
@@ -66,11 +66,12 @@ swr_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 
    struct pipe_draw_info resolved_info;
    /* DrawTransformFeedback */
-   if (info->count_from_stream_output) {
+   if (info->indirect && info->indirect->count_from_stream_output) {
       // trick copied from softpipe to modify const struct *info
       memcpy(&resolved_info, (void*)info, sizeof(struct pipe_draw_info));
       resolved_info.count = ctx->so_primCounter * resolved_info.vertices_per_patch;
       resolved_info.max_index = resolved_info.count - 1;
+      resolved_info.indirect = NULL;
       info = &resolved_info;
    }
 
diff --git a/src/gallium/drivers/tegra/tegra_context.c b/src/gallium/drivers/tegra/tegra_context.c
index c3ae90e4d6db95ad2e9acb59af0e2394da08a7c7..1b15ba2fce10baca29ca1fa595f82b4b8f6f5f9d 100644
--- a/src/gallium/drivers/tegra/tegra_context.c
+++ b/src/gallium/drivers/tegra/tegra_context.c
@@ -52,10 +52,10 @@ tegra_draw_vbo(struct pipe_context *pcontext,
    struct pipe_draw_indirect_info indirect;
    struct pipe_draw_info info;
 
-   if (pinfo && (pinfo->indirect || pinfo->index_size)) {
+   if (pinfo && ((pinfo->indirect && pinfo->indirect->buffer) || pinfo->index_size)) {
       memcpy(&info, pinfo, sizeof(info));
 
-      if (pinfo->indirect) {
+      if (pinfo->indirect && pinfo->indirect->buffer) {
          memcpy(&indirect, pinfo->indirect, sizeof(indirect));
          indirect.buffer = tegra_resource_unwrap(info.indirect->buffer);
          info.indirect = &indirect;
diff --git a/src/gallium/drivers/v3d/v3dx_draw.c b/src/gallium/drivers/v3d/v3dx_draw.c
index 968918ab70f298afe1dbfccae891aebeaa274046..1957f88d127fe13b09b0e29aadcd87235d11b9a0 100644
--- a/src/gallium/drivers/v3d/v3dx_draw.c
+++ b/src/gallium/drivers/v3d/v3dx_draw.c
@@ -1088,7 +1088,7 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 {
         struct v3d_context *v3d = v3d_context(pctx);
 
-        if (!info->count_from_stream_output && !info->indirect &&
+        if (!info->indirect &&
             !info->primitive_restart &&
             !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
                 return;
@@ -1127,7 +1127,7 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
         for (int s = 0; s < PIPE_SHADER_COMPUTE; s++)
                 v3d_predraw_check_stage_inputs(pctx, s);
 
-        if (info->indirect) {
+        if (info->indirect && info->indirect->buffer) {
                 v3d_flush_jobs_writing_resource(v3d, info->indirect->buffer,
                                                 V3D_FLUSH_DEFAULT, false);
         }
@@ -1155,7 +1155,7 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
          * on the last submitted render, rather than tracking the last
          * rendering to each texture's BO.
          */
-        if (v3d->tex[PIPE_SHADER_VERTEX].num_textures || info->indirect) {
+        if (v3d->tex[PIPE_SHADER_VERTEX].num_textures || (info->indirect && info->indirect->buffer)) {
                 perf_debug("Blocking binner on last render "
                            "due to vertex texturing or indirect drawing.\n");
                 job->submit.in_sync_bcl = v3d->out_sync;
@@ -1284,7 +1284,7 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
                 }
 #endif
 
-                if (info->indirect) {
+                if (info->indirect && info->indirect->buffer) {
                         cl_emit(&job->bcl, INDIRECT_INDEXED_INSTANCED_PRIM_LIST, prim) {
                                 prim.index_type = ffs(info->index_size) - 1;
 #if V3D_VERSION < 40
@@ -1335,7 +1335,7 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
                 if (info->has_user_indices)
                         pipe_resource_reference(&prsc, NULL);
         } else {
-                if (info->indirect) {
+                if (info->indirect && info->indirect->buffer) {
                         cl_emit(&job->bcl, INDIRECT_VERTEX_ARRAY_INSTANCED_PRIMS, prim) {
                                 prim.mode = hw_prim_type | prim_tf_enable;
                                 prim.number_of_draw_indirect_array_records = info->indirect->draw_count;
@@ -1346,7 +1346,8 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
                         }
                 } else if (info->instance_count > 1) {
                         struct pipe_stream_output_target *so =
-                                info->count_from_stream_output;
+                                info->indirect && info->indirect->count_from_stream_output ?
+                                        info->indirect->count_from_stream_output : NULL;
                         uint32_t vert_count = so ?
                                 v3d_stream_output_target_get_vertex_count(so) :
                                 info->count;
@@ -1358,7 +1359,8 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
                         }
                 } else {
                         struct pipe_stream_output_target *so =
-                                info->count_from_stream_output;
+                                info->indirect && info->indirect->count_from_stream_output ?
+                                        info->indirect->count_from_stream_output : NULL;
                         uint32_t vert_count = so ?
                                 v3d_stream_output_target_get_vertex_count(so) :
                                 info->count;
diff --git a/src/gallium/drivers/vc4/vc4_draw.c b/src/gallium/drivers/vc4/vc4_draw.c
index b4523bb12500b39b93811c598f8cbb4470d9bd08..bd54df0cdda541643017d130392c818602e13c03 100644
--- a/src/gallium/drivers/vc4/vc4_draw.c
+++ b/src/gallium/drivers/vc4/vc4_draw.c
@@ -291,7 +291,7 @@ vc4_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
         struct vc4_context *vc4 = vc4_context(pctx);
         struct pipe_draw_info local_info;
 
-	if (!info->count_from_stream_output && !info->indirect &&
+	if (!info->indirect &&
 	    !info->primitive_restart &&
 	    !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
 		return;
diff --git a/src/gallium/drivers/virgl/virgl_context.c b/src/gallium/drivers/virgl/virgl_context.c
index 4fabc89b1bd510067dc542b3697a49467525b22c..8aaceda2fb4e62bb5940ce6706c97131db188631 100644
--- a/src/gallium/drivers/virgl/virgl_context.c
+++ b/src/gallium/drivers/virgl/virgl_context.c
@@ -855,7 +855,7 @@ static void virgl_draw_vbo(struct pipe_context *ctx,
    struct virgl_indexbuf ib = {};
    struct pipe_draw_info info = *dinfo;
 
-   if (!dinfo->count_from_stream_output && !dinfo->indirect &&
+   if (!dinfo->indirect &&
        !dinfo->primitive_restart &&
        !u_trim_pipe_prim(dinfo->mode, (unsigned*)&dinfo->count))
       return;
diff --git a/src/gallium/drivers/virgl/virgl_encode.c b/src/gallium/drivers/virgl/virgl_encode.c
index ee75debddf49964fe42f8a2f4cadd99c82e58b4d..dc51178f802b94221b347e2f71e3268ea51321cc 100644
--- a/src/gallium/drivers/virgl/virgl_encode.c
+++ b/src/gallium/drivers/virgl/virgl_encode.c
@@ -712,7 +712,7 @@ int virgl_encoder_draw_vbo(struct virgl_context *ctx,
    uint32_t length = VIRGL_DRAW_VBO_SIZE;
    if (info->mode == PIPE_PRIM_PATCHES)
       length = VIRGL_DRAW_VBO_SIZE_TESS;
-   if (info->indirect)
+   if (info->indirect && info->indirect->buffer)
       length = VIRGL_DRAW_VBO_SIZE_INDIRECT;
    virgl_encoder_write_cmd_dword(ctx, VIRGL_CMD0(VIRGL_CCMD_DRAW_VBO, 0, length));
    virgl_encoder_write_dword(ctx->cbuf, info->start);
@@ -726,8 +726,8 @@ int virgl_encoder_draw_vbo(struct virgl_context *ctx,
    virgl_encoder_write_dword(ctx->cbuf, info->restart_index);
    virgl_encoder_write_dword(ctx->cbuf, info->min_index);
    virgl_encoder_write_dword(ctx->cbuf, info->max_index);
-   if (info->count_from_stream_output)
-      virgl_encoder_write_dword(ctx->cbuf, info->count_from_stream_output->buffer_size);
+   if (info->indirect && info->indirect->count_from_stream_output)
+      virgl_encoder_write_dword(ctx->cbuf, info->indirect->count_from_stream_output->buffer_size);
    else
       virgl_encoder_write_dword(ctx->cbuf, 0);
    if (length >= VIRGL_DRAW_VBO_SIZE_TESS) {
diff --git a/src/gallium/drivers/zink/zink_draw.c b/src/gallium/drivers/zink/zink_draw.c
index 9ff6e6420a9e9c566e2c58754cfcba262f0af225..ae57550fed0af06c8103327d34d3dda10d3f4aa0 100644
--- a/src/gallium/drivers/zink/zink_draw.c
+++ b/src/gallium/drivers/zink/zink_draw.c
@@ -214,7 +214,9 @@ zink_draw_vbo(struct pipe_context *pctx,
    struct zink_screen *screen = zink_screen(pctx->screen);
    struct zink_rasterizer_state *rast_state = ctx->rast_state;
    struct zink_depth_stencil_alpha_state *dsa_state = ctx->dsa_state;
-   struct zink_so_target *so_target = zink_so_target(dinfo->count_from_stream_output);
+   struct zink_so_target *so_target =
+      dinfo->indirect && dinfo->indirect->count_from_stream_output ?
+         zink_so_target(dinfo->indirect->count_from_stream_output) : NULL;
    VkBuffer counter_buffers[PIPE_MAX_SO_OUTPUTS];
    VkDeviceSize counter_buffer_offsets[PIPE_MAX_SO_OUTPUTS] = {};
    bool need_index_buffer_unref = false;
@@ -514,7 +516,7 @@ zink_draw_vbo(struct pipe_context *pctx,
       struct zink_resource *res = zink_resource(index_buffer);
       vkCmdBindIndexBuffer(batch->cmdbuf, res->buffer, index_offset, index_type);
       zink_batch_reference_resource_rw(batch, res, false);
-      if (dinfo->indirect) {
+      if (dinfo->indirect && dinfo->indirect->buffer) {
          struct zink_resource *indirect = zink_resource(dinfo->indirect->buffer);
          zink_batch_reference_resource_rw(batch, indirect, false);
          vkCmdDrawIndexedIndirect(batch->cmdbuf, indirect->buffer, dinfo->indirect->offset, dinfo->indirect->draw_count, dinfo->indirect->stride);
@@ -528,7 +530,7 @@ zink_draw_vbo(struct pipe_context *pctx,
          screen->vk_CmdDrawIndirectByteCountEXT(batch->cmdbuf, dinfo->instance_count, dinfo->start_instance,
                                        zink_resource(so_target->counter_buffer)->buffer, so_target->counter_buffer_offset, 0,
                                        MIN2(so_target->stride, screen->info.tf_props.maxTransformFeedbackBufferDataStride));
-      } else if (dinfo->indirect) {
+      } else if (dinfo->indirect && dinfo->indirect->buffer) {
          struct zink_resource *indirect = zink_resource(dinfo->indirect->buffer);
          zink_batch_reference_resource_rw(batch, indirect, false);
          vkCmdDrawIndirect(batch->cmdbuf, indirect->buffer, dinfo->indirect->offset, dinfo->indirect->draw_count, dinfo->indirect->stride);
diff --git a/src/gallium/frontends/nine/device9.c b/src/gallium/frontends/nine/device9.c
index 36f0c8f1266c5b2169a5f253d204262aef62d36c..a7587cc95170773195349ea50d43bee9ecd746e7 100644
--- a/src/gallium/frontends/nine/device9.c
+++ b/src/gallium/frontends/nine/device9.c
@@ -3174,7 +3174,6 @@ NineDevice9_ProcessVertices( struct NineDevice9 *This,
     draw.start_instance = 0;
     draw.primitive_restart = FALSE;
     draw.restart_index = 0;
-    draw.count_from_stream_output = NULL;
     draw.indirect = NULL;
     draw.instance_count = 1;
     draw.index_size = 0;
diff --git a/src/gallium/frontends/nine/nine_state.c b/src/gallium/frontends/nine/nine_state.c
index 9dae199d5f80f98dffb1880649638f38734ab6c9..e4d72c62ae8774230d2d70e1aa8bec29000d0c94 100644
--- a/src/gallium/frontends/nine/nine_state.c
+++ b/src/gallium/frontends/nine/nine_state.c
@@ -2318,7 +2318,6 @@ init_draw_info(struct pipe_draw_info *info,
     info->primitive_restart = FALSE;
     info->has_user_indices = FALSE;
     info->restart_index = 0;
-    info->count_from_stream_output = NULL;
     info->indirect = NULL;
 }
 
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index fbf463618da2f75be263ad0de714142b580e0827..ad5701ba9c54c76d68ce8235ee3defe90dcb76cd 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -712,6 +712,8 @@ struct pipe_draw_indirect_info
     *     uint32_t start;
     *     uint32_t start_instance;
     *  };
+    *
+    * If NULL, count_from_stream_output != NULL.
     */
    struct pipe_resource *buffer;
 
@@ -719,6 +721,22 @@ struct pipe_draw_indirect_info
     * is to be used as the real draw_count.
     */
    struct pipe_resource *indirect_draw_count;
+
+   /**
+    * Stream output target. If not NULL, it's used to provide the 'count'
+    * parameter based on the number vertices captured by the stream output
+    * stage. (or generally, based on the number of bytes captured)
+    *
+    * Only 'mode', 'start_instance', and 'instance_count' are taken into
+    * account, all the other variables from pipe_draw_info are ignored.
+    *
+    * 'start' is implicitly 0 and 'count' is set as discussed above.
+    * The draw command is non-indexed.
+    *
+    * Note that this only provides the count. The vertex buffers must
+    * be set via set_vertex_buffers manually.
+    */
+   struct pipe_stream_output_target *count_from_stream_output;
 };
 
 struct pipe_draw_start_count {
@@ -777,22 +795,6 @@ struct pipe_draw_info
    } index;
 
    struct pipe_draw_indirect_info *indirect; /**< Indirect draw. */
-
-   /**
-    * Stream output target. If not NULL, it's used to provide the 'count'
-    * parameter based on the number vertices captured by the stream output
-    * stage. (or generally, based on the number of bytes captured)
-    *
-    * Only 'mode', 'start_instance', and 'instance_count' are taken into
-    * account, all the other variables from pipe_draw_info are ignored.
-    *
-    * 'start' is implicitly 0 and 'count' is set as discussed above.
-    * The draw command is non-indexed.
-    *
-    * Note that this only provides the count. The vertex buffers must
-    * be set via set_vertex_buffers manually.
-    */
-   struct pipe_stream_output_target *count_from_stream_output;
 };
 
 
diff --git a/src/mesa/state_tracker/st_cb_xformfb.c b/src/mesa/state_tracker/st_cb_xformfb.c
index 4126e64345b35d0c5a624181d50501c60c2b4b1d..8ecf91c5e2b40c3d09587acb324b55b99c1ac71a 100644
--- a/src/mesa/state_tracker/st_cb_xformfb.c
+++ b/src/mesa/state_tracker/st_cb_xformfb.c
@@ -214,7 +214,8 @@ st_end_transform_feedback(struct gl_context *ctx,
 
 bool
 st_transform_feedback_draw_init(struct gl_transform_feedback_object *obj,
-                                unsigned stream, struct pipe_draw_info *out)
+                                unsigned stream,
+                                struct pipe_draw_indirect_info *out)
 {
    struct st_transform_feedback_object *sobj =
          st_transform_feedback_object(obj);
diff --git a/src/mesa/state_tracker/st_cb_xformfb.h b/src/mesa/state_tracker/st_cb_xformfb.h
index 5db4094755201cc1d825df7556c37831ea5a7bca..635ebd8eed695c4736d77b46ff13ceec87da3eec 100644
--- a/src/mesa/state_tracker/st_cb_xformfb.h
+++ b/src/mesa/state_tracker/st_cb_xformfb.h
@@ -31,14 +31,15 @@
 
 struct dd_function_table;
 struct gl_transform_feedback_object;
-struct pipe_draw_info;
+struct pipe_draw_indirect_info;
 
 extern void
 st_init_xformfb_functions(struct dd_function_table *functions);
 
 extern bool
 st_transform_feedback_draw_init(struct gl_transform_feedback_object *obj,
-                                unsigned stream, struct pipe_draw_info *out);
+                                unsigned stream,
+                                struct pipe_draw_indirect_info *out);
 
 
 #endif /* ST_CB_XFORMFB_H */
diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c
index 1745a8d194805a01b773da413a191ad0da4ac226..ef8d5b0b1493693730cc9d416bcd921f35c6abfa 100644
--- a/src/mesa/state_tracker/st_draw.c
+++ b/src/mesa/state_tracker/st_draw.c
@@ -177,7 +177,6 @@ st_draw_vbo(struct gl_context *ctx,
    info.primitive_restart = false;
    info.vertices_per_patch = ctx->TessCtrlProgram.patch_vertices;
    info.indirect = NULL;
-   info.count_from_stream_output = NULL;
    info.restart_index = 0;
    info.start_instance = base_instance;
    info.instance_count = num_instances;
@@ -330,15 +329,18 @@ st_draw_transform_feedback(struct gl_context *ctx, GLenum mode,
 {
    struct st_context *st = st_context(ctx);
    struct pipe_draw_info info;
+   struct pipe_draw_indirect_info indirect;
 
    prepare_draw(st, ctx);
 
+   memset(&indirect, 0, sizeof(indirect));
    util_draw_init_info(&info);
    info.start = 0; /* index offset / index size */
    info.max_index = ~0u; /* so that u_vbuf can tell that it's unknown */
    info.mode = translate_prim(ctx, mode);
    info.vertices_per_patch = ctx->TessCtrlProgram.patch_vertices;
    info.instance_count = num_instances;
+   info.indirect = &indirect;
 
    if (ST_DEBUG & DEBUG_DRAW) {
       debug_printf("st/draw transform feedback: mode %s\n",
@@ -347,7 +349,7 @@ st_draw_transform_feedback(struct gl_context *ctx, GLenum mode,
 
    /* Transform feedback drawing is always non-indexed. */
    /* Set info.count_from_stream_output. */
-   if (!st_transform_feedback_draw_init(tfb_vertcount, stream, &info))
+   if (!st_transform_feedback_draw_init(tfb_vertcount, stream, &indirect))
       return;
 
    cso_draw_vbo(st->cso_context, &info);