freedreno: move more batch related tracking to fd_batch

To flush batches out of order, the gmem code needs to not depend on
state from fd_context (since that may apply to a more recent batch).
So this all moves into batch.

The one exception is the gmem/pipe/tile state itself.  But this is
only used from gmem code (and batches are flushed serially).  The
alternative would be having to re-calculate GMEM layout on every
batch, even if the dimensions of the render targets are the same.

Note: This opens up the possibility of pushing gmem/submit into a
helper thread.

Signed-off-by: Rob Clark <robdclark@gmail.com>

23 files changed, 420 insertions, 398 deletions

diff --git a/src/gallium/drivers/freedreno/a2xx/fd2_draw.c b/src/gallium/drivers/freedreno/a2xx/fd2_draw.c
index 030e6f69b71..a824018174a 100644
--- a/src/gallium/drivers/freedreno/a2xx/fd2_draw.c
+++ b/src/gallium/drivers/freedreno/a2xx/fd2_draw.c
@@ -107,7 +107,7 @@ fd2_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info)
 	OUT_RING(ring, info->max_index);        /* VGT_MAX_VTX_INDX */
 	OUT_RING(ring, info->min_index);        /* VGT_MIN_VTX_INDX */
 
-	fd_draw_emit(ctx, ring, ctx->primtypes[info->mode],
+	fd_draw_emit(ctx->batch, ring, ctx->primtypes[info->mode],
 				 IGNORE_VISIBILITY, info);
 
 	OUT_PKT3(ring, CP_SET_CONSTANT, 2);
@@ -126,7 +126,7 @@ fd2_clear(struct fd_context *ctx, unsigned buffers,
 {
 	struct fd2_context *fd2_ctx = fd2_context(ctx);
 	struct fd_ringbuffer *ring = ctx->batch->draw;
-	struct pipe_framebuffer_state *fb = &ctx->framebuffer;
+	struct pipe_framebuffer_state *fb = &ctx->batch->framebuffer;
 	uint32_t reg, colr = 0;
 
 	if ((buffers & PIPE_CLEAR_COLOR) && fb->nr_cbufs)
@@ -266,7 +266,7 @@ fd2_clear(struct fd_context *ctx, unsigned buffers,
 	OUT_RING(ring, 3);                 /* VGT_MAX_VTX_INDX */
 	OUT_RING(ring, 0);                 /* VGT_MIN_VTX_INDX */
 
-	fd_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
+	fd_draw(ctx->batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 3, 0, INDEX_SIZE_IGN, 0, 0, NULL);
 
 	OUT_PKT3(ring, CP_SET_CONSTANT, 2);
diff --git a/src/gallium/drivers/freedreno/a2xx/fd2_emit.c b/src/gallium/drivers/freedreno/a2xx/fd2_emit.c
index 032780364df..b3a1b3d29ab 100644
--- a/src/gallium/drivers/freedreno/a2xx/fd2_emit.c
+++ b/src/gallium/drivers/freedreno/a2xx/fd2_emit.c
@@ -250,10 +250,10 @@ fd2_emit_state(struct fd_context *ctx, uint32_t dirty)
 		OUT_RING(ring, xy2d(scissor->maxx,       /* PA_SC_WINDOW_SCISSOR_BR */
 				scissor->maxy));
 
-		ctx->max_scissor.minx = MIN2(ctx->max_scissor.minx, scissor->minx);
-		ctx->max_scissor.miny = MIN2(ctx->max_scissor.miny, scissor->miny);
-		ctx->max_scissor.maxx = MAX2(ctx->max_scissor.maxx, scissor->maxx);
-		ctx->max_scissor.maxy = MAX2(ctx->max_scissor.maxy, scissor->maxy);
+		ctx->batch->max_scissor.minx = MIN2(ctx->batch->max_scissor.minx, scissor->minx);
+		ctx->batch->max_scissor.miny = MIN2(ctx->batch->max_scissor.miny, scissor->miny);
+		ctx->batch->max_scissor.maxx = MAX2(ctx->batch->max_scissor.maxx, scissor->maxx);
+		ctx->batch->max_scissor.maxy = MAX2(ctx->batch->max_scissor.maxy, scissor->maxy);
 	}
 
 	if (dirty & FD_DIRTY_VIEWPORT) {
diff --git a/src/gallium/drivers/freedreno/a2xx/fd2_gmem.c b/src/gallium/drivers/freedreno/a2xx/fd2_gmem.c
index eba2cec82f4..6dc639670a1 100644
--- a/src/gallium/drivers/freedreno/a2xx/fd2_gmem.c
+++ b/src/gallium/drivers/freedreno/a2xx/fd2_gmem.c
@@ -56,10 +56,10 @@ static uint32_t fmt2swap(enum pipe_format format)
 /* transfer from gmem to system memory (ie. normal RAM) */
 
 static void
-emit_gmem2mem_surf(struct fd_context *ctx, uint32_t base,
+emit_gmem2mem_surf(struct fd_batch *batch, uint32_t base,
 		struct pipe_surface *psurf)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	struct fd_resource *rsc = fd_resource(psurf->texture);
 	uint32_t swap = fmt2swap(psurf->format);
 
@@ -90,16 +90,17 @@ emit_gmem2mem_surf(struct fd_context *ctx, uint32_t base,
 	OUT_RING(ring, 3);                 /* VGT_MAX_VTX_INDX */
 	OUT_RING(ring, 0);                 /* VGT_MIN_VTX_INDX */
 
-	fd_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
+	fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 3, 0, INDEX_SIZE_IGN, 0, 0, NULL);
 }
 
 static void
-fd2_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
+fd2_emit_tile_gmem2mem(struct fd_batch *batch, struct fd_tile *tile)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd2_context *fd2_ctx = fd2_context(ctx);
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 
 	fd2_emit_vertex_bufs(ring, 0x9c, (struct fd2_vertex_buf[]) {
 			{ .prsc = fd2_ctx->solid_vertexbuf, .size = 48 },
@@ -159,11 +160,11 @@ fd2_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
 	OUT_RING(ring, A2XX_RB_COPY_DEST_OFFSET_X(tile->xoff) |
 			A2XX_RB_COPY_DEST_OFFSET_Y(tile->yoff));
 
-	if (ctx->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL))
-		emit_gmem2mem_surf(ctx, tile->bin_w * tile->bin_h, pfb->zsbuf);
+	if (batch->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL))
+		emit_gmem2mem_surf(batch, tile->bin_w * tile->bin_h, pfb->zsbuf);
 
-	if (ctx->resolve & FD_BUFFER_COLOR)
-		emit_gmem2mem_surf(ctx, 0, pfb->cbufs[0]);
+	if (batch->resolve & FD_BUFFER_COLOR)
+		emit_gmem2mem_surf(batch, 0, pfb->cbufs[0]);
 
 	OUT_PKT3(ring, CP_SET_CONSTANT, 2);
 	OUT_RING(ring, CP_REG(REG_A2XX_RB_MODECONTROL));
@@ -173,10 +174,10 @@ fd2_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
 /* transfer from system memory to gmem */
 
 static void
-emit_mem2gmem_surf(struct fd_context *ctx, uint32_t base,
+emit_mem2gmem_surf(struct fd_batch *batch, uint32_t base,
 		struct pipe_surface *psurf)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	struct fd_resource *rsc = fd_resource(psurf->texture);
 	uint32_t swiz;
 
@@ -212,16 +213,17 @@ emit_mem2gmem_surf(struct fd_context *ctx, uint32_t base,
 	OUT_RING(ring, 3);                 /* VGT_MAX_VTX_INDX */
 	OUT_RING(ring, 0);                 /* VGT_MIN_VTX_INDX */
 
-	fd_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
+	fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 3, 0, INDEX_SIZE_IGN, 0, 0, NULL);
 }
 
 static void
-fd2_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
+fd2_emit_tile_mem2gmem(struct fd_batch *batch, struct fd_tile *tile)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd2_context *fd2_ctx = fd2_context(ctx);
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	unsigned bin_w = tile->bin_w;
 	unsigned bin_h = tile->bin_h;
 	float x0, y0, x1, y1;
@@ -317,26 +319,27 @@ fd2_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
 	OUT_RING(ring, CP_REG(REG_A2XX_PA_CL_CLIP_CNTL));
 	OUT_RING(ring, 0x00000000);
 
-	if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL))
-		emit_mem2gmem_surf(ctx, bin_w * bin_h, pfb->zsbuf);
+	if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL))
+		emit_mem2gmem_surf(batch, bin_w * bin_h, pfb->zsbuf);
 
-	if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_COLOR))
-		emit_mem2gmem_surf(ctx, 0, pfb->cbufs[0]);
+	if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_COLOR))
+		emit_mem2gmem_surf(batch, 0, pfb->cbufs[0]);
 
 	/* TODO blob driver seems to toss in a CACHE_FLUSH after each DRAW_INDX.. */
 }
 
 /* before first tile */
 static void
-fd2_emit_tile_init(struct fd_context *ctx)
+fd2_emit_tile_init(struct fd_batch *batch)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_context *ctx = batch->ctx;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
 	enum pipe_format format = pipe_surface_format(pfb->cbufs[0]);
 	uint32_t reg;
 
-	fd2_emit_restore(ctx, ctx->ring);
+	fd2_emit_restore(ctx, ring);
 
 	OUT_PKT3(ring, CP_SET_CONSTANT, 4);
 	OUT_RING(ring, CP_REG(REG_A2XX_RB_SURFACE_INFO));
@@ -351,10 +354,10 @@ fd2_emit_tile_init(struct fd_context *ctx)
 
 /* before mem2gmem */
 static void
-fd2_emit_tile_prep(struct fd_context *ctx, struct fd_tile *tile)
+fd2_emit_tile_prep(struct fd_batch *batch, struct fd_tile *tile)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	enum pipe_format format = pipe_surface_format(pfb->cbufs[0]);
 
 	OUT_PKT3(ring, CP_SET_CONSTANT, 2);
@@ -373,10 +376,10 @@ fd2_emit_tile_prep(struct fd_context *ctx, struct fd_tile *tile)
 
 /* before IB to rendering cmds: */
 static void
-fd2_emit_tile_renderprep(struct fd_context *ctx, struct fd_tile *tile)
+fd2_emit_tile_renderprep(struct fd_batch *batch, struct fd_tile *tile)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	enum pipe_format format = pipe_surface_format(pfb->cbufs[0]);
 
 	OUT_PKT3(ring, CP_SET_CONSTANT, 2);
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_context.c b/src/gallium/drivers/freedreno/a3xx/fd3_context.c
index e47bbff5643..2ffc2e0e53a 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_context.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_context.c
@@ -43,8 +43,6 @@ fd3_context_destroy(struct pipe_context *pctx)
 {
 	struct fd3_context *fd3_ctx = fd3_context(fd_context(pctx));
 
-	util_dynarray_fini(&fd3_ctx->rbrc_patches);
-
 	fd_bo_del(fd3_ctx->vs_pvt_mem);
 	fd_bo_del(fd3_ctx->fs_pvt_mem);
 	fd_bo_del(fd3_ctx->vsc_size_mem);
@@ -127,8 +125,6 @@ fd3_context_create(struct pipe_screen *pscreen, void *priv, unsigned flags)
 	if (!pctx)
 		return NULL;
 
-	util_dynarray_init(&fd3_ctx->rbrc_patches);
-
 	fd3_ctx->vs_pvt_mem = fd_bo_new(screen->dev, 0x2000,
 			DRM_FREEDRENO_GEM_TYPE_KMEM);
 
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_context.h b/src/gallium/drivers/freedreno/a3xx/fd3_context.h
index b4c2ebe570c..07354110361 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_context.h
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_context.h
@@ -41,11 +41,6 @@
 struct fd3_context {
 	struct fd_context base;
 
-	/* Keep track of writes to RB_RENDER_CONTROL which need to be patched
-	 * once we know whether or not to use GMEM, and GMEM tile pitch.
-	 */
-	struct util_dynarray rbrc_patches;
-
 	struct fd_bo *vs_pvt_mem, *fs_pvt_mem;
 
 	/* This only needs to be 4 * num_of_pipes bytes (ie. 32 bytes).  We
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_draw.c b/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
index 0593b253f1a..34d782b95ce 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_draw.c
@@ -84,8 +84,7 @@ draw_impl(struct fd_context *ctx, struct fd_ringbuffer *ring,
 			(info->mode == PIPE_PRIM_POINTS))
 		primtype = DI_PT_POINTLIST_PSIZE;
 
-	fd_draw_emit(ctx, ring,
-			primtype,
+	fd_draw_emit(ctx->batch, ring, primtype,
 			emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
 			info);
 }
@@ -223,7 +222,7 @@ fd3_clear_binning(struct fd_context *ctx, unsigned dirty)
 
 	fd3_emit_state(ctx, ring, &emit);
 	fd3_emit_vertex_bufs(ring, &emit);
-	reset_viewport(ring, &ctx->framebuffer);
+	reset_viewport(ring, &ctx->batch->framebuffer);
 
 	OUT_PKT0(ring, REG_A3XX_PC_PRIM_VTX_CNTL, 1);
 	OUT_RING(ring, A3XX_PC_PRIM_VTX_CNTL_STRIDE_IN_VPC(0) |
@@ -240,7 +239,7 @@ fd3_clear_binning(struct fd_context *ctx, unsigned dirty)
 
 	fd_event_write(ctx, ring, PERFCOUNTER_STOP);
 
-	fd_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
+	fd_draw(ctx->batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 2, 0, INDEX_SIZE_IGN, 0, 0, NULL);
 }
 
@@ -249,7 +248,7 @@ fd3_clear(struct fd_context *ctx, unsigned buffers,
 		const union pipe_color_union *color, double depth, unsigned stencil)
 {
 	struct fd3_context *fd3_ctx = fd3_context(ctx);
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
 	struct fd_ringbuffer *ring = ctx->batch->draw;
 	unsigned dirty = ctx->dirty;
 	unsigned i;
@@ -270,7 +269,7 @@ fd3_clear(struct fd_context *ctx, unsigned buffers,
 
 	/* emit generic state now: */
 	fd3_emit_state(ctx, ring, &emit);
-	reset_viewport(ring, &ctx->framebuffer);
+	reset_viewport(ring, &ctx->batch->framebuffer);
 
 	OUT_PKT0(ring, REG_A3XX_RB_BLEND_ALPHA, 1);
 	OUT_RING(ring, A3XX_RB_BLEND_ALPHA_UINT(0xff) |
@@ -278,7 +277,7 @@ fd3_clear(struct fd_context *ctx, unsigned buffers,
 
 	OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
 	OUT_RINGP(ring, A3XX_RB_RENDER_CONTROL_ALPHA_TEST_FUNC(FUNC_NEVER),
-			&fd3_ctx->rbrc_patches);
+			&ctx->batch->rbrc_patches);
 
 	if (buffers & PIPE_CLEAR_DEPTH) {
 		OUT_PKT0(ring, REG_A3XX_RB_DEPTH_CONTROL, 1);
@@ -374,7 +373,7 @@ fd3_clear(struct fd_context *ctx, unsigned buffers,
 
 	fd_event_write(ctx, ring, PERFCOUNTER_STOP);
 
-	fd_draw(ctx, ring, DI_PT_RECTLIST, USE_VISIBILITY,
+	fd_draw(ctx->batch, ring, DI_PT_RECTLIST, USE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 2, 0, INDEX_SIZE_IGN, 0, 0, NULL);
 }
 
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
index 45185edb505..eef5b52f12c 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
@@ -516,7 +516,7 @@ fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		 */
 
 		OUT_PKT0(ring, REG_A3XX_RB_RENDER_CONTROL, 1);
-		OUT_RINGP(ring, val, &fd3_context(ctx)->rbrc_patches);
+		OUT_RINGP(ring, val, &ctx->batch->rbrc_patches);
 	}
 
 	if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_STENCIL_REF)) {
@@ -631,10 +631,10 @@ fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		OUT_RING(ring, A3XX_GRAS_SC_WINDOW_SCISSOR_BR_X(scissor->maxx - 1) |
 				A3XX_GRAS_SC_WINDOW_SCISSOR_BR_Y(scissor->maxy - 1));
 
-		ctx->max_scissor.minx = MIN2(ctx->max_scissor.minx, scissor->minx);
-		ctx->max_scissor.miny = MIN2(ctx->max_scissor.miny, scissor->miny);
-		ctx->max_scissor.maxx = MAX2(ctx->max_scissor.maxx, scissor->maxx);
-		ctx->max_scissor.maxy = MAX2(ctx->max_scissor.maxy, scissor->maxy);
+		ctx->batch->max_scissor.minx = MIN2(ctx->batch->max_scissor.minx, scissor->minx);
+		ctx->batch->max_scissor.miny = MIN2(ctx->batch->max_scissor.miny, scissor->miny);
+		ctx->batch->max_scissor.maxx = MAX2(ctx->batch->max_scissor.maxx, scissor->maxx);
+		ctx->batch->max_scissor.maxy = MAX2(ctx->batch->max_scissor.maxy, scissor->maxy);
 	}
 
 	if (dirty & FD_DIRTY_VIEWPORT) {
@@ -649,7 +649,7 @@ fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	}
 
 	if (dirty & (FD_DIRTY_PROG | FD_DIRTY_FRAMEBUFFER | FD_DIRTY_BLEND_DUAL)) {
-		struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+		struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
 		int nr_cbufs = pfb->nr_cbufs;
 		if (fd3_blend_stateobj(ctx->blend)->rb_render_control &
 			A3XX_RB_RENDER_CONTROL_DUAL_COLOR_IN_ENABLE)
@@ -673,7 +673,8 @@ fd3_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		uint32_t i;
 
 		for (i = 0; i < ARRAY_SIZE(blend->rb_mrt); i++) {
-			enum pipe_format format = pipe_surface_format(ctx->framebuffer.cbufs[i]);
+			enum pipe_format format =
+				pipe_surface_format(ctx->batch->framebuffer.cbufs[i]);
 			const struct util_format_description *desc =
 				util_format_description(format);
 			bool is_float = util_format_is_float(format);
diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_gmem.c b/src/gallium/drivers/freedreno/a3xx/fd3_gmem.c
index 2449a84c721..b9af45683f9 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_gmem.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_gmem.c
@@ -128,9 +128,9 @@ emit_mrt(struct fd_ringbuffer *ring, unsigned nr_bufs,
 }
 
 static bool
-use_hw_binning(struct fd_context *ctx)
+use_hw_binning(struct fd_batch *batch)
 {
-	struct fd_gmem_stateobj *gmem = &ctx->gmem;
+	struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
 
 	/* workaround: combining scissor optimization and hw binning
 	 * seems problematic.  Seems like we end up with a mismatch
@@ -153,13 +153,14 @@ use_hw_binning(struct fd_context *ctx)
 }
 
 /* workaround for (hlsq?) lockup with hw binning on a3xx patchlevel 0 */
-static void update_vsc_pipe(struct fd_context *ctx);
+static void update_vsc_pipe(struct fd_batch *batch);
 static void
-emit_binning_workaround(struct fd_context *ctx)
+emit_binning_workaround(struct fd_batch *batch)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd3_context *fd3_ctx = fd3_context(ctx);
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	struct fd3_emit emit = {
 			.debug = &ctx->debug,
 			.vtx = &fd3_ctx->solid_vbuf_state,
@@ -308,12 +309,12 @@ emit_binning_workaround(struct fd_context *ctx)
 /* transfer from gmem to system memory (ie. normal RAM) */
 
 static void
-emit_gmem2mem_surf(struct fd_context *ctx,
+emit_gmem2mem_surf(struct fd_batch *batch,
 				   enum adreno_rb_copy_control_mode mode,
 				   bool stencil,
 				   uint32_t base, struct pipe_surface *psurf)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	struct fd_resource *rsc = fd_resource(psurf->texture);
 	enum pipe_format format = psurf->format;
 	if (stencil) {
@@ -342,16 +343,17 @@ emit_gmem2mem_surf(struct fd_context *ctx,
 			A3XX_RB_COPY_DEST_INFO_ENDIAN(ENDIAN_NONE) |
 			A3XX_RB_COPY_DEST_INFO_SWAP(fd3_pipe2swap(format)));
 
-	fd_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
+	fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 2, 0, INDEX_SIZE_IGN, 0, 0, NULL);
 }
 
 static void
-fd3_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
+fd3_emit_tile_gmem2mem(struct fd_batch *batch, struct fd_tile *tile)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd3_context *fd3_ctx = fd3_context(ctx);
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	struct fd3_emit emit = {
 			.debug = &ctx->debug,
 			.vtx = &fd3_ctx->solid_vbuf_state,
@@ -437,23 +439,23 @@ fd3_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
 	fd3_program_emit(ring, &emit, 0, NULL);
 	fd3_emit_vertex_bufs(ring, &emit);
 
-	if (ctx->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
+	if (batch->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
 		struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
-		if (!rsc->stencil || ctx->resolve & FD_BUFFER_DEPTH)
-			emit_gmem2mem_surf(ctx, RB_COPY_DEPTH_STENCIL, false,
+		if (!rsc->stencil || batch->resolve & FD_BUFFER_DEPTH)
+			emit_gmem2mem_surf(batch, RB_COPY_DEPTH_STENCIL, false,
 							   ctx->gmem.zsbuf_base[0], pfb->zsbuf);
-		if (rsc->stencil && ctx->resolve & FD_BUFFER_STENCIL)
-			emit_gmem2mem_surf(ctx, RB_COPY_DEPTH_STENCIL, true,
+		if (rsc->stencil && batch->resolve & FD_BUFFER_STENCIL)
+			emit_gmem2mem_surf(batch, RB_COPY_DEPTH_STENCIL, true,
 							   ctx->gmem.zsbuf_base[1], pfb->zsbuf);
 	}
 
-	if (ctx->resolve & FD_BUFFER_COLOR) {
+	if (batch->resolve & FD_BUFFER_COLOR) {
 		for (i = 0; i < pfb->nr_cbufs; i++) {
 			if (!pfb->cbufs[i])
 				continue;
-			if (!(ctx->resolve & (PIPE_CLEAR_COLOR0 << i)))
+			if (!(batch->resolve & (PIPE_CLEAR_COLOR0 << i)))
 				continue;
-			emit_gmem2mem_surf(ctx, RB_COPY_RESOLVE, false,
+			emit_gmem2mem_surf(batch, RB_COPY_RESOLVE, false,
 							   ctx->gmem.cbuf_base[i], pfb->cbufs[i]);
 		}
 	}
@@ -472,10 +474,10 @@ fd3_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
 /* transfer from system memory to gmem */
 
 static void
-emit_mem2gmem_surf(struct fd_context *ctx, uint32_t bases[],
+emit_mem2gmem_surf(struct fd_batch *batch, uint32_t bases[],
 		struct pipe_surface **psurf, uint32_t bufs, uint32_t bin_w)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	struct pipe_surface *zsbufs[2];
 
 	assert(bufs > 0);
@@ -502,7 +504,7 @@ emit_mem2gmem_surf(struct fd_context *ctx, uint32_t bases[],
 		OUT_PKT0(ring, REG_A3XX_RB_DEPTH_INFO, 2);
 		OUT_RING(ring, A3XX_RB_DEPTH_INFO_DEPTH_BASE(bases[0]) |
 				 A3XX_RB_DEPTH_INFO_DEPTH_FORMAT(DEPTHX_32));
-		OUT_RING(ring, A3XX_RB_DEPTH_PITCH(4 * ctx->gmem.bin_w));
+		OUT_RING(ring, A3XX_RB_DEPTH_PITCH(4 * batch->ctx->gmem.bin_w));
 
 		if (psurf[0]->format == PIPE_FORMAT_Z32_FLOAT) {
 			OUT_PKT0(ring, REG_A3XX_RB_MRT_CONTROL(0), 1);
@@ -523,17 +525,18 @@ emit_mem2gmem_surf(struct fd_context *ctx, uint32_t bases[],
 
 	fd3_emit_gmem_restore_tex(ring, psurf, bufs);
 
-	fd_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
+	fd_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 2, 0, INDEX_SIZE_IGN, 0, 0, NULL);
 }
 
 static void
-fd3_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
+fd3_emit_tile_mem2gmem(struct fd_batch *batch, struct fd_tile *tile)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd3_context *fd3_ctx = fd3_context(ctx);
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	struct fd3_emit emit = {
 			.debug = &ctx->debug,
 			.vtx = &fd3_ctx->blit_vbuf_state,
@@ -655,14 +658,14 @@ fd3_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
 	bin_w = gmem->bin_w;
 	bin_h = gmem->bin_h;
 
-	if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_COLOR)) {
+	if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_COLOR)) {
 		emit.prog = &ctx->blit_prog[pfb->nr_cbufs - 1];
 		emit.fp = NULL;      /* frag shader changed so clear cache */
 		fd3_program_emit(ring, &emit, pfb->nr_cbufs, pfb->cbufs);
-		emit_mem2gmem_surf(ctx, gmem->cbuf_base, pfb->cbufs, pfb->nr_cbufs, bin_w);
+		emit_mem2gmem_surf(batch, gmem->cbuf_base, pfb->cbufs, pfb->nr_cbufs, bin_w);
 	}
 
-	if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
+	if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
 		if (pfb->zsbuf->format != PIPE_FORMAT_Z32_FLOAT_S8X24_UINT &&
 			pfb->zsbuf->format != PIPE_FORMAT_Z32_FLOAT) {
 			/* Non-float can use a regular color write. It's split over 8-bit
@@ -680,7 +683,7 @@ fd3_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
 		}
 		emit.fp = NULL;      /* frag shader changed so clear cache */
 		fd3_program_emit(ring, &emit, 1, &pfb->zsbuf);
-		emit_mem2gmem_surf(ctx, gmem->zsbuf_base, &pfb->zsbuf, 1, bin_w);
+		emit_mem2gmem_surf(batch, gmem->zsbuf_base, &pfb->zsbuf, 1, bin_w);
 	}
 
 	OUT_PKT0(ring, REG_A3XX_GRAS_SC_CONTROL, 1);
@@ -695,34 +698,33 @@ fd3_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
 }
 
 static void
-patch_draws(struct fd_context *ctx, enum pc_di_vis_cull_mode vismode)
+patch_draws(struct fd_batch *batch, enum pc_di_vis_cull_mode vismode)
 {
 	unsigned i;
-	for (i = 0; i < fd_patch_num_elements(&ctx->draw_patches); i++) {
-		struct fd_cs_patch *patch = fd_patch_element(&ctx->draw_patches, i);
+	for (i = 0; i < fd_patch_num_elements(&batch->draw_patches); i++) {
+		struct fd_cs_patch *patch = fd_patch_element(&batch->draw_patches, i);
 		*patch->cs = patch->val | DRAW(0, 0, 0, vismode, 0);
 	}
-	util_dynarray_resize(&ctx->draw_patches, 0);
+	util_dynarray_resize(&batch->draw_patches, 0);
 }
 
 static void
-patch_rbrc(struct fd_context *ctx, uint32_t val)
+patch_rbrc(struct fd_batch *batch, uint32_t val)
 {
-	struct fd3_context *fd3_ctx = fd3_context(ctx);
 	unsigned i;
-	for (i = 0; i < fd_patch_num_elements(&fd3_ctx->rbrc_patches); i++) {
-		struct fd_cs_patch *patch = fd_patch_element(&fd3_ctx->rbrc_patches, i);
+	for (i = 0; i < fd_patch_num_elements(&batch->rbrc_patches); i++) {
+		struct fd_cs_patch *patch = fd_patch_element(&batch->rbrc_patches, i);
 		*patch->cs = patch->val | val;
 	}
-	util_dynarray_resize(&fd3_ctx->rbrc_patches, 0);
+	util_dynarray_resize(&batch->rbrc_patches, 0);
 }
 
 /* for rendering directly to system memory: */
 static void
-fd3_emit_sysmem_prep(struct fd_context *ctx)
+fd3_emit_sysmem_prep(struct fd_batch *batch)
 {
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
 	uint32_t i, pitch = 0;
 
 	for (i = 0; i < pfb->nr_cbufs; i++) {
@@ -732,7 +734,7 @@ fd3_emit_sysmem_prep(struct fd_context *ctx)
 		pitch = fd_resource(psurf->texture)->slices[psurf->u.tex.level].pitch;
 	}
 
-	fd3_emit_restore(ctx, ring);
+	fd3_emit_restore(batch->ctx, ring);
 
 	OUT_PKT0(ring, REG_A3XX_RB_FRAME_BUFFER_DIMENSION, 1);
 	OUT_RING(ring, A3XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
@@ -757,15 +759,16 @@ fd3_emit_sysmem_prep(struct fd_context *ctx)
 			A3XX_RB_MODE_CONTROL_MARB_CACHE_SPLIT_MODE |
 			A3XX_RB_MODE_CONTROL_MRT(MAX2(1, pfb->nr_cbufs) - 1));
 
-	patch_draws(ctx, IGNORE_VISIBILITY);
-	patch_rbrc(ctx, A3XX_RB_RENDER_CONTROL_BIN_WIDTH(pitch));
+	patch_draws(batch, IGNORE_VISIBILITY);
+	patch_rbrc(batch, A3XX_RB_RENDER_CONTROL_BIN_WIDTH(pitch));
 }
 
 static void
-update_vsc_pipe(struct fd_context *ctx)
+update_vsc_pipe(struct fd_batch *batch)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd3_context *fd3_ctx = fd3_context(ctx);
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	int i;
 
 	OUT_PKT0(ring, REG_A3XX_VSC_SIZE_ADDRESS, 1);
@@ -790,12 +793,12 @@ update_vsc_pipe(struct fd_context *ctx)
 }
 
 static void
-emit_binning_pass(struct fd_context *ctx)
+emit_binning_pass(struct fd_batch *batch)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
-	struct fd_batch *batch = ctx->batch;
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
 	int i;
 
 	uint32_t x1 = gmem->minx;
@@ -804,7 +807,7 @@ emit_binning_pass(struct fd_context *ctx)
 	uint32_t y2 = gmem->miny + gmem->height - 1;
 
 	if (ctx->screen->gpu_id == 320) {
-		emit_binning_workaround(ctx);
+		emit_binning_workaround(batch);
 		fd_wfi(ctx, ring);
 		OUT_PKT3(ring, CP_INVALIDATE_STATE, 1);
 		OUT_RING(ring, 0x00007fff);
@@ -912,19 +915,19 @@ emit_binning_pass(struct fd_context *ctx)
 	fd_wfi(ctx, ring);
 
 	if (ctx->screen->gpu_id == 320) {
-		emit_binning_workaround(ctx);
+		emit_binning_workaround(batch);
 	}
 }
 
 /* before first tile */
 static void
-fd3_emit_tile_init(struct fd_context *ctx)
+fd3_emit_tile_init(struct fd_batch *batch)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct fd_gmem_stateobj *gmem = &ctx->gmem;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
 	uint32_t rb_render_control;
 
-	fd3_emit_restore(ctx, ring);
+	fd3_emit_restore(batch->ctx, ring);
 
 	/* note: use gmem->bin_w/h, the bin_w/h parameters may be truncated
 	 * at the right and bottom edge tiles
@@ -933,29 +936,30 @@ fd3_emit_tile_init(struct fd_context *ctx)
 	OUT_RING(ring, A3XX_VSC_BIN_SIZE_WIDTH(gmem->bin_w) |
 			A3XX_VSC_BIN_SIZE_HEIGHT(gmem->bin_h));
 
-	update_vsc_pipe(ctx);
+	update_vsc_pipe(batch);
 
-	if (use_hw_binning(ctx)) {
+	if (use_hw_binning(batch)) {
 		/* emit hw binning pass: */
-		emit_binning_pass(ctx);
+		emit_binning_pass(batch);
 
-		patch_draws(ctx, USE_VISIBILITY);
+		patch_draws(batch, USE_VISIBILITY);
 	} else {
-		patch_draws(ctx, IGNORE_VISIBILITY);
+		patch_draws(batch, IGNORE_VISIBILITY);
 	}
 
 	rb_render_control = A3XX_RB_RENDER_CONTROL_ENABLE_GMEM |
 			A3XX_RB_RENDER_CONTROL_BIN_WIDTH(gmem->bin_w);
 
-	patch_rbrc(ctx, rb_render_control);
+	patch_rbrc(batch, rb_render_control);
 }
 
 /* before mem2gmem */
 static void
-fd3_emit_tile_prep(struct fd_context *ctx, struct fd_tile *tile)
+fd3_emit_tile_prep(struct fd_batch *batch, struct fd_tile *tile)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_context *ctx = batch->ctx;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 
 	if (ctx->needs_rb_fbd) {
 		fd_wfi(ctx, ring);
@@ -973,12 +977,13 @@ fd3_emit_tile_prep(struct fd_context *ctx, struct fd_tile *tile)
 
 /* before IB to rendering cmds: */
 static void
-fd3_emit_tile_renderprep(struct fd_context *ctx, struct fd_tile *tile)
+fd3_emit_tile_renderprep(struct fd_batch *batch, struct fd_tile *tile)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd3_context *fd3_ctx = fd3_context(ctx);
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 
 	uint32_t x1 = tile->xoff;
 	uint32_t y1 = tile->yoff;
@@ -1005,7 +1010,7 @@ fd3_emit_tile_renderprep(struct fd_context *ctx, struct fd_tile *tile)
 		OUT_RING(ring, 0x00000000);
 	}
 
-	if (use_hw_binning(ctx)) {
+	if (use_hw_binning(batch)) {
 		struct fd_vsc_pipe *pipe = &ctx->pipe[tile->p];
 
 		assert(pipe->w * pipe->h);
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_draw.c b/src/gallium/drivers/freedreno/a4xx/fd4_draw.c
index e0513860a71..8270c4f0cff 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_draw.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_draw.c
@@ -68,8 +68,7 @@ draw_impl(struct fd_context *ctx, struct fd_ringbuffer *ring,
 			(info->mode == PIPE_PRIM_POINTS))
 		primtype = DI_PT_POINTLIST_PSIZE;
 
-	fd4_draw_emit(ctx, ring,
-			primtype,
+	fd4_draw_emit(ctx->batch, ring, primtype,
 			emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
 			info);
 }
@@ -233,7 +232,7 @@ fd4_clear_binning(struct fd_context *ctx, unsigned dirty)
 
 	fd4_emit_state(ctx, ring, &emit);
 	fd4_emit_vertex_bufs(ring, &emit);
-	reset_viewport(ring, &ctx->framebuffer);
+	reset_viewport(ring, &ctx->batch->framebuffer);
 
 	OUT_PKT0(ring, REG_A4XX_PC_PRIM_VTX_CNTL, 2);
 	OUT_RING(ring, A4XX_PC_PRIM_VTX_CNTL_VAROUT(0) |
@@ -244,7 +243,7 @@ fd4_clear_binning(struct fd_context *ctx, unsigned dirty)
 	OUT_PKT0(ring, REG_A4XX_GRAS_ALPHA_CONTROL, 1);
 	OUT_RING(ring, 0x00000002);
 
-	fd4_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
+	fd4_draw(ctx->batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 2, 1, INDEX_SIZE_IGN, 0, 0, NULL);
 }
 
@@ -254,7 +253,7 @@ fd4_clear(struct fd_context *ctx, unsigned buffers,
 {
 	struct fd4_context *fd4_ctx = fd4_context(ctx);
 	struct fd_ringbuffer *ring = ctx->batch->draw;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
 	unsigned char mrt_comp[A4XX_MAX_RENDER_TARGETS] = {0};
 	unsigned dirty = ctx->dirty;
 	unsigned i;
@@ -390,7 +389,7 @@ fd4_clear(struct fd_context *ctx, unsigned buffers,
 	OUT_PKT3(ring, CP_UNKNOWN_1A, 1);
 	OUT_RING(ring, 0x00000001);
 
-	fd4_draw(ctx, ring, DI_PT_RECTLIST, USE_VISIBILITY,
+	fd4_draw(ctx->batch, ring, DI_PT_RECTLIST, USE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 2, 1, INDEX_SIZE_IGN, 0, 0, NULL);
 
 	OUT_PKT3(ring, CP_UNKNOWN_1A, 1);
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_draw.h b/src/gallium/drivers/freedreno/a4xx/fd4_draw.h
index 5f5f6cd1e5f..57b16870fab 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_draw.h
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_draw.h
@@ -48,7 +48,7 @@ static inline uint32_t DRAW4(enum pc_di_primtype prim_type,
 }
 
 static inline void
-fd4_draw(struct fd_context *ctx, struct fd_ringbuffer *ring,
+fd4_draw(struct fd_batch *batch, struct fd_ringbuffer *ring,
 		enum pc_di_primtype primtype,
 		enum pc_di_vis_cull_mode vismode,
 		enum pc_di_src_sel src_sel, uint32_t count,
@@ -70,7 +70,7 @@ fd4_draw(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		 * we know if we are binning or not
 		 */
 		OUT_RINGP(ring, DRAW4(primtype, src_sel, idx_type, 0),
-				&ctx->draw_patches);
+				&batch->draw_patches);
 	} else {
 		OUT_RING(ring, DRAW4(primtype, src_sel, idx_type, vismode));
 	}
@@ -84,7 +84,7 @@ fd4_draw(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	emit_marker(ring, 7);
 
-	fd_reset_wfi(ctx);
+	fd_reset_wfi(batch->ctx);
 }
 
 
@@ -101,18 +101,19 @@ fd4_size2indextype(unsigned index_size)
 	return INDEX4_SIZE_32_BIT;
 }
 static inline void
-fd4_draw_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
+fd4_draw_emit(struct fd_batch *batch, struct fd_ringbuffer *ring,
 		enum pc_di_primtype primtype,
 		enum pc_di_vis_cull_mode vismode,
 		const struct pipe_draw_info *info)
 {
-	struct pipe_index_buffer *idx = &ctx->indexbuf;
 	struct pipe_resource *idx_buffer = NULL;
 	enum a4xx_index_size idx_type;
 	enum pc_di_src_sel src_sel;
 	uint32_t idx_size, idx_offset;
 
 	if (info->indexed) {
+		struct pipe_index_buffer *idx = &batch->ctx->indexbuf;
+
 		assert(!idx->user_buffer);
 
 		idx_buffer = idx->buffer;
@@ -128,7 +129,7 @@ fd4_draw_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		src_sel = DI_SRC_SEL_AUTO_INDEX;
 	}
 
-	fd4_draw(ctx, ring, primtype, vismode, src_sel,
+	fd4_draw(batch, ring, primtype, vismode, src_sel,
 			info->count, info->instance_count,
 			idx_type, idx_size, idx_offset, idx_buffer);
 }
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
index 5bb712c755c..88e1a40ec90 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_emit.c
@@ -505,7 +505,7 @@ fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	emit_marker(ring, 5);
 
 	if ((dirty & FD_DIRTY_FRAMEBUFFER) && !emit->key.binning_pass) {
-		struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+		struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
 		unsigned char mrt_comp[A4XX_MAX_RENDER_TARGETS] = {0};
 
 		for (unsigned i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
@@ -525,7 +525,7 @@ fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_FRAMEBUFFER)) {
 		struct fd4_zsa_stateobj *zsa = fd4_zsa_stateobj(ctx->zsa);
-		struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+		struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
 		uint32_t rb_alpha_control = zsa->rb_alpha_control;
 
 		if (util_format_is_pure_integer(pipe_surface_format(pfb->cbufs[0])))
@@ -625,10 +625,10 @@ fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		OUT_RING(ring, A4XX_GRAS_SC_WINDOW_SCISSOR_TL_X(scissor->minx) |
 				A4XX_GRAS_SC_WINDOW_SCISSOR_TL_Y(scissor->miny));
 
-		ctx->max_scissor.minx = MIN2(ctx->max_scissor.minx, scissor->minx);
-		ctx->max_scissor.miny = MIN2(ctx->max_scissor.miny, scissor->miny);
-		ctx->max_scissor.maxx = MAX2(ctx->max_scissor.maxx, scissor->maxx);
-		ctx->max_scissor.maxy = MAX2(ctx->max_scissor.maxy, scissor->maxy);
+		ctx->batch->max_scissor.minx = MIN2(ctx->batch->max_scissor.minx, scissor->minx);
+		ctx->batch->max_scissor.miny = MIN2(ctx->batch->max_scissor.miny, scissor->miny);
+		ctx->batch->max_scissor.maxx = MAX2(ctx->batch->max_scissor.maxx, scissor->maxx);
+		ctx->batch->max_scissor.maxy = MAX2(ctx->batch->max_scissor.maxy, scissor->maxy);
 	}
 
 	if (dirty & FD_DIRTY_VIEWPORT) {
@@ -643,7 +643,7 @@ fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	}
 
 	if (dirty & (FD_DIRTY_PROG | FD_DIRTY_FRAMEBUFFER)) {
-		struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+		struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
 		unsigned n = pfb->nr_cbufs;
 		/* if we have depth/stencil, we need at least on MRT: */
 		if (pfb->zsbuf)
@@ -663,7 +663,7 @@ fd4_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 		for (i = 0; i < A4XX_MAX_RENDER_TARGETS; i++) {
 			enum pipe_format format = pipe_surface_format(
-					ctx->framebuffer.cbufs[i]);
+					ctx->batch->framebuffer.cbufs[i]);
 			bool is_int = util_format_is_pure_integer(format);
 			bool has_alpha = util_format_has_alpha(format);
 			uint32_t control = blend->rb_mrt[i].control;
diff --git a/src/gallium/drivers/freedreno/a4xx/fd4_gmem.c b/src/gallium/drivers/freedreno/a4xx/fd4_gmem.c
index 524c35a18bd..afd37a88f43 100644
--- a/src/gallium/drivers/freedreno/a4xx/fd4_gmem.c
+++ b/src/gallium/drivers/freedreno/a4xx/fd4_gmem.c
@@ -132,10 +132,10 @@ emit_mrt(struct fd_ringbuffer *ring, unsigned nr_bufs,
 }
 
 static bool
-use_hw_binning(struct fd_context *ctx)
+use_hw_binning(struct fd_batch *batch)
 {
-	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 
 	/* this seems to be a hw bug.. but this hack fixes piglit fbo-maxsize: */
 	if ((pfb->width > 4096) && (pfb->height > 4096))
@@ -147,10 +147,10 @@ use_hw_binning(struct fd_context *ctx)
 /* transfer from gmem to system memory (ie. normal RAM) */
 
 static void
-emit_gmem2mem_surf(struct fd_context *ctx, bool stencil,
+emit_gmem2mem_surf(struct fd_batch *batch, bool stencil,
 		uint32_t base, struct pipe_surface *psurf)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	struct fd_resource *rsc = fd_resource(psurf->texture);
 	enum pipe_format pformat = psurf->format;
 	struct fd_resource_slice *slice;
@@ -180,17 +180,18 @@ emit_gmem2mem_surf(struct fd_context *ctx, bool stencil,
 			A4XX_RB_COPY_DEST_INFO_ENDIAN(ENDIAN_NONE) |
 			A4XX_RB_COPY_DEST_INFO_SWAP(fd4_pipe2swap(pformat)));
 
-	fd4_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
+	fd4_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 2, 1, INDEX_SIZE_IGN, 0, 0, NULL);
 }
 
 static void
-fd4_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
+fd4_emit_tile_gmem2mem(struct fd_batch *batch, struct fd_tile *tile)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd4_context *fd4_ctx = fd4_context(ctx);
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	struct fd4_emit emit = {
 			.debug = &ctx->debug,
 			.vtx = &fd4_ctx->solid_vbuf_state,
@@ -269,22 +270,22 @@ fd4_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
 	fd4_program_emit(ring, &emit, 0, NULL);
 	fd4_emit_vertex_bufs(ring, &emit);
 
-	if (ctx->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
+	if (batch->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
 		struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
-		if (!rsc->stencil || (ctx->resolve & FD_BUFFER_DEPTH))
-			emit_gmem2mem_surf(ctx, false, ctx->gmem.zsbuf_base[0], pfb->zsbuf);
-		if (rsc->stencil && (ctx->resolve & FD_BUFFER_STENCIL))
-			emit_gmem2mem_surf(ctx, true, ctx->gmem.zsbuf_base[1], pfb->zsbuf);
+		if (!rsc->stencil || (batch->resolve & FD_BUFFER_DEPTH))
+			emit_gmem2mem_surf(batch, false, ctx->gmem.zsbuf_base[0], pfb->zsbuf);
+		if (rsc->stencil && (batch->resolve & FD_BUFFER_STENCIL))
+			emit_gmem2mem_surf(batch, true, ctx->gmem.zsbuf_base[1], pfb->zsbuf);
 	}
 
-	if (ctx->resolve & FD_BUFFER_COLOR) {
+	if (batch->resolve & FD_BUFFER_COLOR) {
 		unsigned i;
 		for (i = 0; i < pfb->nr_cbufs; i++) {
 			if (!pfb->cbufs[i])
 				continue;
-			if (!(ctx->resolve & (PIPE_CLEAR_COLOR0 << i)))
+			if (!(batch->resolve & (PIPE_CLEAR_COLOR0 << i)))
 				continue;
-			emit_gmem2mem_surf(ctx, false, gmem->cbuf_base[i], pfb->cbufs[i]);
+			emit_gmem2mem_surf(batch, false, gmem->cbuf_base[i], pfb->cbufs[i]);
 		}
 	}
 
@@ -298,10 +299,10 @@ fd4_emit_tile_gmem2mem(struct fd_context *ctx, struct fd_tile *tile)
 /* transfer from system memory to gmem */
 
 static void
-emit_mem2gmem_surf(struct fd_context *ctx, uint32_t *bases,
+emit_mem2gmem_surf(struct fd_batch *batch, uint32_t *bases,
 		struct pipe_surface **bufs, uint32_t nr_bufs, uint32_t bin_w)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	struct pipe_surface *zsbufs[2];
 
 	emit_mrt(ring, nr_bufs, bufs, bases, bin_w, false);
@@ -318,17 +319,18 @@ emit_mem2gmem_surf(struct fd_context *ctx, uint32_t *bases,
 
 	fd4_emit_gmem_restore_tex(ring, nr_bufs, bufs);
 
-	fd4_draw(ctx, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
+	fd4_draw(batch, ring, DI_PT_RECTLIST, IGNORE_VISIBILITY,
 			DI_SRC_SEL_AUTO_INDEX, 2, 1, INDEX_SIZE_IGN, 0, 0, NULL);
 }
 
 static void
-fd4_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
+fd4_emit_tile_mem2gmem(struct fd_batch *batch, struct fd_tile *tile)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd4_context *fd4_ctx = fd4_context(ctx);
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	struct fd4_emit emit = {
 			.debug = &ctx->debug,
 			.vtx = &fd4_ctx->blit_vbuf_state,
@@ -455,14 +457,14 @@ fd4_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
 	bin_w = gmem->bin_w;
 	bin_h = gmem->bin_h;
 
-	if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_COLOR)) {
+	if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_COLOR)) {
 		emit.prog = &ctx->blit_prog[pfb->nr_cbufs - 1];
 		emit.fp = NULL;      /* frag shader changed so clear cache */
 		fd4_program_emit(ring, &emit, pfb->nr_cbufs, pfb->cbufs);
-		emit_mem2gmem_surf(ctx, gmem->cbuf_base, pfb->cbufs, pfb->nr_cbufs, bin_w);
+		emit_mem2gmem_surf(batch, gmem->cbuf_base, pfb->cbufs, pfb->nr_cbufs, bin_w);
 	}
 
-	if (fd_gmem_needs_restore(ctx, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
+	if (fd_gmem_needs_restore(batch, tile, FD_BUFFER_DEPTH | FD_BUFFER_STENCIL)) {
 		switch (pfb->zsbuf->format) {
 		case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
 		case PIPE_FORMAT_Z32_FLOAT:
@@ -493,7 +495,7 @@ fd4_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
 		}
 		emit.fp = NULL;      /* frag shader changed so clear cache */
 		fd4_program_emit(ring, &emit, 1, &pfb->zsbuf);
-		emit_mem2gmem_surf(ctx, gmem->zsbuf_base, &pfb->zsbuf, 1, bin_w);
+		emit_mem2gmem_surf(batch, gmem->zsbuf_base, &pfb->zsbuf, 1, bin_w);
 	}
 
 	OUT_PKT0(ring, REG_A4XX_GRAS_SC_CONTROL, 1);
@@ -508,24 +510,24 @@ fd4_emit_tile_mem2gmem(struct fd_context *ctx, struct fd_tile *tile)
 }
 
 static void
-patch_draws(struct fd_context *ctx, enum pc_di_vis_cull_mode vismode)
+patch_draws(struct fd_batch *batch, enum pc_di_vis_cull_mode vismode)
 {
 	unsigned i;
-	for (i = 0; i < fd_patch_num_elements(&ctx->draw_patches); i++) {
-		struct fd_cs_patch *patch = fd_patch_element(&ctx->draw_patches, i);
+	for (i = 0; i < fd_patch_num_elements(&batch->draw_patches); i++) {
+		struct fd_cs_patch *patch = fd_patch_element(&batch->draw_patches, i);
 		*patch->cs = patch->val | DRAW4(0, 0, 0, vismode);
 	}
-	util_dynarray_resize(&ctx->draw_patches, 0);
+	util_dynarray_resize(&batch->draw_patches, 0);
 }
 
 /* for rendering directly to system memory: */
 static void
-fd4_emit_sysmem_prep(struct fd_context *ctx)
+fd4_emit_sysmem_prep(struct fd_batch *batch)
 {
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
 
-	fd4_emit_restore(ctx, ring);
+	fd4_emit_restore(batch->ctx, ring);
 
 	OUT_PKT0(ring, REG_A4XX_RB_FRAME_BUFFER_DIMENSION, 1);
 	OUT_RING(ring, A4XX_RB_FRAME_BUFFER_DIMENSION_WIDTH(pfb->width) |
@@ -552,14 +554,15 @@ fd4_emit_sysmem_prep(struct fd_context *ctx)
 	OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL, 1);
 	OUT_RING(ring, 0x8);
 
-	patch_draws(ctx, IGNORE_VISIBILITY);
+	patch_draws(batch, IGNORE_VISIBILITY);
 }
 
 static void
-update_vsc_pipe(struct fd_context *ctx)
+update_vsc_pipe(struct fd_batch *batch)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd4_context *fd4_ctx = fd4_context(ctx);
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	int i;
 
 	OUT_PKT0(ring, REG_A4XX_VSC_SIZE_ADDRESS, 1);
@@ -592,12 +595,12 @@ update_vsc_pipe(struct fd_context *ctx)
 }
 
 static void
-emit_binning_pass(struct fd_context *ctx)
+emit_binning_pass(struct fd_batch *batch)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
-	struct fd_batch *batch = ctx->batch;
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
+	struct fd_ringbuffer *ring = batch->gmem;
 	int i;
 
 	uint32_t x1 = gmem->minx;
@@ -658,20 +661,20 @@ emit_binning_pass(struct fd_context *ctx)
 
 /* before first tile */
 static void
-fd4_emit_tile_init(struct fd_context *ctx)
+fd4_emit_tile_init(struct fd_batch *batch)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct fd_gmem_stateobj *gmem = &ctx->gmem;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct fd_gmem_stateobj *gmem = &batch->ctx->gmem;
 
-	fd4_emit_restore(ctx, ring);
+	fd4_emit_restore(batch->ctx, ring);
 
 	OUT_PKT0(ring, REG_A4XX_VSC_BIN_SIZE, 1);
 	OUT_RING(ring, A4XX_VSC_BIN_SIZE_WIDTH(gmem->bin_w) |
 			A4XX_VSC_BIN_SIZE_HEIGHT(gmem->bin_h));
 
-	update_vsc_pipe(ctx);
+	update_vsc_pipe(batch);
 
-	if (use_hw_binning(ctx)) {
+	if (use_hw_binning(batch)) {
 		OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
 		OUT_RING(ring, A4XX_RB_MODE_CONTROL_WIDTH(gmem->bin_w) |
 				A4XX_RB_MODE_CONTROL_HEIGHT(gmem->bin_h));
@@ -682,11 +685,11 @@ fd4_emit_tile_init(struct fd_context *ctx)
 				0x8);
 
 		/* emit hw binning pass: */
-		emit_binning_pass(ctx);
+		emit_binning_pass(batch);
 
-		patch_draws(ctx, USE_VISIBILITY);
+		patch_draws(batch, USE_VISIBILITY);
 	} else {
-		patch_draws(ctx, IGNORE_VISIBILITY);
+		patch_draws(batch, IGNORE_VISIBILITY);
 	}
 
 	OUT_PKT0(ring, REG_A4XX_RB_MODE_CONTROL, 1);
@@ -697,10 +700,11 @@ fd4_emit_tile_init(struct fd_context *ctx)
 
 /* before mem2gmem */
 static void
-fd4_emit_tile_prep(struct fd_context *ctx, struct fd_tile *tile)
+fd4_emit_tile_prep(struct fd_batch *batch, struct fd_tile *tile)
 {
-	struct fd_ringbuffer *ring = ctx->ring;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_context *ctx = batch->ctx;
+	struct fd_ringbuffer *ring = batch->gmem;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
 
 	if (pfb->zsbuf) {
@@ -752,19 +756,20 @@ fd4_emit_tile_prep(struct fd_context *ctx, struct fd_tile *tile)
 
 /* before IB to rendering cmds: */
 static void
-fd4_emit_tile_renderprep(struct fd_context *ctx, struct fd_tile *tile)
+fd4_emit_tile_renderprep(struct fd_batch *batch, struct fd_tile *tile)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd4_context *fd4_ctx = fd4_context(ctx);
-	struct fd_ringbuffer *ring = ctx->ring;
+	struct fd_ringbuffer *ring = batch->gmem;
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 
 	uint32_t x1 = tile->xoff;
 	uint32_t y1 = tile->yoff;
 	uint32_t x2 = tile->xoff + tile->bin_w - 1;
 	uint32_t y2 = tile->yoff + tile->bin_h - 1;
 
-	if (use_hw_binning(ctx)) {
+	if (use_hw_binning(batch)) {
 		struct fd_vsc_pipe *pipe = &ctx->pipe[tile->p];
 
 		assert(pipe->w * pipe->h);
diff --git a/src/gallium/drivers/freedreno/freedreno_batch.c b/src/gallium/drivers/freedreno/freedreno_batch.c
index 6d17a422fc4..1fbce43f62c 100644
--- a/src/gallium/drivers/freedreno/freedreno_batch.c
+++ b/src/gallium/drivers/freedreno/freedreno_batch.c
@@ -64,16 +64,31 @@ fd_batch_create(struct fd_context *ctx)
 
 	list_inithead(&batch->used_resources);
 
+	/* reset maximal bounds: */
+	batch->max_scissor.minx = batch->max_scissor.miny = ~0;
+	batch->max_scissor.maxx = batch->max_scissor.maxy = 0;
+
+	util_dynarray_init(&batch->draw_patches);
+
+	if (is_a3xx(ctx->screen))
+		util_dynarray_init(&batch->rbrc_patches);
+
 	return batch;
 }
 
 void
 __fd_batch_destroy(struct fd_batch *batch)
 {
+	util_copy_framebuffer_state(&batch->framebuffer, NULL);
 	fd_ringbuffer_del(batch->draw);
 	fd_ringbuffer_del(batch->binning);
 	fd_ringbuffer_del(batch->gmem);
 
+	util_dynarray_fini(&batch->draw_patches);
+
+	if (is_a3xx(batch->ctx->screen))
+		util_dynarray_fini(&batch->rbrc_patches);
+
 	free(batch);
 }
 
@@ -88,7 +103,12 @@ fd_batch_flush(struct fd_batch *batch)
 {
 	struct fd_resource *rsc, *rsc_tmp;
 
-	fd_gmem_render_tiles(batch->ctx);
+	DBG("%p: needs_flush=%d", batch, batch->needs_flush);
+
+	if (!batch->needs_flush)
+		return;
+
+	fd_gmem_render_tiles(batch);
 
 	/* go through all the used resources and clear their reading flag */
 	LIST_FOR_EACH_ENTRY_SAFE(rsc, rsc_tmp, &batch->used_resources, list) {
diff --git a/src/gallium/drivers/freedreno/freedreno_batch.h b/src/gallium/drivers/freedreno/freedreno_batch.h
index 69779d8c8f4..4607250d3af 100644
--- a/src/gallium/drivers/freedreno/freedreno_batch.h
+++ b/src/gallium/drivers/freedreno/freedreno_batch.h
@@ -42,8 +42,74 @@ enum fd_resource_status;
 struct fd_batch {
 	struct pipe_reference reference;
 	unsigned seqno;
+
 	struct fd_context *ctx;
 
+	/* do we need to mem2gmem before rendering.  We don't, if for example,
+	 * there was a glClear() that invalidated the entire previous buffer
+	 * contents.  Keep track of which buffer(s) are cleared, or needs
+	 * restore.  Masks of PIPE_CLEAR_*
+	 *
+	 * The 'cleared' bits will be set for buffers which are *entirely*
+	 * cleared, and 'partial_cleared' bits will be set if you must
+	 * check cleared_scissor.
+	 */
+	enum {
+		/* align bitmask values w/ PIPE_CLEAR_*.. since that is convenient.. */
+		FD_BUFFER_COLOR   = PIPE_CLEAR_COLOR,
+		FD_BUFFER_DEPTH   = PIPE_CLEAR_DEPTH,
+		FD_BUFFER_STENCIL = PIPE_CLEAR_STENCIL,
+		FD_BUFFER_ALL     = FD_BUFFER_COLOR | FD_BUFFER_DEPTH | FD_BUFFER_STENCIL,
+	} cleared, partial_cleared, restore, resolve;
+
+	bool needs_flush;
+
+	/* To decide whether to render to system memory, keep track of the
+	 * number of draws, and whether any of them require multisample,
+	 * depth_test (or depth write), stencil_test, blending, and
+	 * color_logic_Op (since those functions are disabled when by-
+	 * passing GMEM.
+	 */
+	enum {
+		FD_GMEM_CLEARS_DEPTH_STENCIL = 0x01,
+		FD_GMEM_DEPTH_ENABLED        = 0x02,
+		FD_GMEM_STENCIL_ENABLED      = 0x04,
+
+		FD_GMEM_MSAA_ENABLED         = 0x08,
+		FD_GMEM_BLEND_ENABLED        = 0x10,
+		FD_GMEM_LOGICOP_ENABLED      = 0x20,
+	} gmem_reason;
+	unsigned num_draws;   /* number of draws in current batch */
+
+	/* Track the maximal bounds of the scissor of all the draws within a
+	 * batch.  Used at the tile rendering step (fd_gmem_render_tiles(),
+	 * mem2gmem/gmem2mem) to avoid needlessly moving data in/out of gmem.
+	 */
+	struct pipe_scissor_state max_scissor;
+
+	/* Track the cleared scissor for color/depth/stencil, so we know
+	 * which, if any, tiles need to be restored (mem2gmem).  Only valid
+	 * if the corresponding bit in ctx->cleared is set.
+	 */
+	struct {
+		struct pipe_scissor_state color, depth, stencil;
+	} cleared_scissor;
+
+	/* Keep track of DRAW initiators that need to be patched up depending
+	 * on whether we using binning or not:
+	 */
+	struct util_dynarray draw_patches;
+
+	/* Keep track of writes to RB_RENDER_CONTROL which need to be patched
+	 * once we know whether or not to use GMEM, and GMEM tile pitch.
+	 *
+	 * (only for a3xx.. but having gen specific subclasses of fd_batch
+	 * seemed overkill for now)
+	 */
+	struct util_dynarray rbrc_patches;
+
+	struct pipe_framebuffer_state framebuffer;
+
 	/** draw pass cmdstream: */
 	struct fd_ringbuffer *draw;
 	/** binning pass cmdstream: */
diff --git a/src/gallium/drivers/freedreno/freedreno_context.c b/src/gallium/drivers/freedreno/freedreno_context.c
index 3614370bf33..b9a1fe97745 100644
--- a/src/gallium/drivers/freedreno/freedreno_context.c
+++ b/src/gallium/drivers/freedreno/freedreno_context.c
@@ -45,21 +45,14 @@ void
 fd_context_render(struct pipe_context *pctx)
 {
 	struct fd_context *ctx = fd_context(pctx);
-
-	DBG("needs_flush: %d", ctx->needs_flush);
-
-	if (!ctx->needs_flush)
-		return;
+	struct fd_batch *new_batch;
 
 	fd_batch_flush(ctx->batch);
 
+	new_batch = fd_batch_create(ctx);
+	util_copy_framebuffer_state(&new_batch->framebuffer, &ctx->batch->framebuffer);
 	fd_batch_reference(&ctx->batch, NULL);
-	ctx->batch = fd_batch_create(ctx);
-
-	ctx->needs_flush = false;
-	ctx->cleared = ctx->partial_cleared = ctx->restore = ctx->resolve = 0;
-	ctx->gmem_reason = 0;
-	ctx->num_draws = 0;
+	ctx->batch = new_batch;
 }
 
 static void
@@ -120,8 +113,6 @@ fd_context_destroy(struct pipe_context *pctx)
 	fd_prog_fini(pctx);
 	fd_hw_query_fini(pctx);
 
-	util_dynarray_fini(&ctx->draw_patches);
-
 	if (ctx->blitter)
 		util_blitter_destroy(ctx->blitter);
 
@@ -190,8 +181,6 @@ fd_context_init(struct fd_context *ctx, struct pipe_screen *pscreen,
 
 	fd_reset_wfi(ctx);
 
-	util_dynarray_init(&ctx->draw_patches);
-
 	util_slab_create(&ctx->transfer_pool, sizeof(struct fd_transfer),
 			16, UTIL_SLAB_SINGLETHREADED);
 
diff --git a/src/gallium/drivers/freedreno/freedreno_context.h b/src/gallium/drivers/freedreno/freedreno_context.h
index 88e103ea6c3..9401367dca5 100644
--- a/src/gallium/drivers/freedreno/freedreno_context.h
+++ b/src/gallium/drivers/freedreno/freedreno_context.h
@@ -199,42 +199,6 @@ struct fd_context {
 	struct fd_program_stateobj blit_prog[MAX_RENDER_TARGETS]; // TODO move to screen?
 	struct fd_program_stateobj blit_z, blit_zs;
 
-	/* do we need to mem2gmem before rendering.  We don't, if for example,
-	 * there was a glClear() that invalidated the entire previous buffer
-	 * contents.  Keep track of which buffer(s) are cleared, or needs
-	 * restore.  Masks of PIPE_CLEAR_*
-	 *
-	 * The 'cleared' bits will be set for buffers which are *entirely*
-	 * cleared, and 'partial_cleared' bits will be set if you must
-	 * check cleared_scissor.
-	 */
-	enum {
-		/* align bitmask values w/ PIPE_CLEAR_*.. since that is convenient.. */
-		FD_BUFFER_COLOR   = PIPE_CLEAR_COLOR,
-		FD_BUFFER_DEPTH   = PIPE_CLEAR_DEPTH,
-		FD_BUFFER_STENCIL = PIPE_CLEAR_STENCIL,
-		FD_BUFFER_ALL     = FD_BUFFER_COLOR | FD_BUFFER_DEPTH | FD_BUFFER_STENCIL,
-	} cleared, partial_cleared, restore, resolve;
-
-	bool needs_flush;
-
-	/* To decide whether to render to system memory, keep track of the
-	 * number of draws, and whether any of them require multisample,
-	 * depth_test (or depth write), stencil_test, blending, and
-	 * color_logic_Op (since those functions are disabled when by-
-	 * passing GMEM.
-	 */
-	enum {
-		FD_GMEM_CLEARS_DEPTH_STENCIL = 0x01,
-		FD_GMEM_DEPTH_ENABLED        = 0x02,
-		FD_GMEM_STENCIL_ENABLED      = 0x04,
-
-		FD_GMEM_MSAA_ENABLED         = 0x08,
-		FD_GMEM_BLEND_ENABLED        = 0x10,
-		FD_GMEM_LOGICOP_ENABLED      = 0x20,
-	} gmem_reason;
-	unsigned num_draws;   /* number of draws in current batch */
-
 	/* Stats/counters:
 	 */
 	struct {
@@ -244,12 +208,6 @@ struct fd_context {
 		uint64_t batch_total, batch_sysmem, batch_gmem, batch_restore;
 	} stats;
 
-	/* TODO get rid of this.. only used in gmem/tiling code paths (and
-	 * NULL the rest of the time).  Just leaving for now to reduce some
-	 * churn..
-	 */
-	struct fd_ringbuffer *ring;
-
 	/* Current batch.. the rule here is that you can deref ctx->batch
 	 * in codepaths from pipe_context entrypoints.  But not in code-
 	 * paths from fd_batch_flush() (basically, the stuff that gets
@@ -269,11 +227,6 @@ struct fd_context {
 	 * */
 	bool needs_rb_fbd;
 
-	/* Keep track of DRAW initiators that need to be patched up depending
-	 * on whether we using binning or not:
-	 */
-	struct util_dynarray draw_patches;
-
 	struct pipe_scissor_state scissor;
 
 	/* we don't have a disable/enable bit for scissor, so instead we keep
@@ -282,22 +235,11 @@ struct fd_context {
 	 */
 	struct pipe_scissor_state disabled_scissor;
 
-	/* Track the maximal bounds of the scissor of all the draws within a
-	 * batch.  Used at the tile rendering step (fd_gmem_render_tiles(),
-	 * mem2gmem/gmem2mem) to avoid needlessly moving data in/out of gmem.
-	 */
-	struct pipe_scissor_state max_scissor;
-
-	/* Track the cleared scissor for color/depth/stencil, so we know
-	 * which, if any, tiles need to be restored (mem2gmem).  Only valid
-	 * if the corresponding bit in ctx->cleared is set.
-	 */
-	struct {
-		struct pipe_scissor_state color, depth, stencil;
-	} cleared_scissor;
-
 	/* Current gmem/tiling configuration.. gets updated on render_tiles()
 	 * if out of date with current maximal-scissor/cpp:
+	 *
+	 * (NOTE: this is kind of related to the batch, but moving it there
+	 * means we'd always have to recalc tiles ever batch)
 	 */
 	struct fd_gmem_stateobj gmem;
 	struct fd_vsc_pipe      pipe[8];
@@ -346,7 +288,6 @@ struct fd_context {
 	struct pipe_blend_color blend_color;
 	struct pipe_stencil_ref stencil_ref;
 	unsigned sample_mask;
-	struct pipe_framebuffer_state framebuffer;
 	struct pipe_poly_stipple stipple;
 	struct pipe_viewport_state viewport;
 	struct fd_constbuf_stateobj constbuf[PIPE_SHADER_TYPES];
@@ -361,14 +302,14 @@ struct fd_context {
 	struct pipe_debug_callback debug;
 
 	/* GMEM/tile handling fxns: */
-	void (*emit_tile_init)(struct fd_context *ctx);
-	void (*emit_tile_prep)(struct fd_context *ctx, struct fd_tile *tile);
-	void (*emit_tile_mem2gmem)(struct fd_context *ctx, struct fd_tile *tile);
-	void (*emit_tile_renderprep)(struct fd_context *ctx, struct fd_tile *tile);
-	void (*emit_tile_gmem2mem)(struct fd_context *ctx, struct fd_tile *tile);
+	void (*emit_tile_init)(struct fd_batch *batch);
+	void (*emit_tile_prep)(struct fd_batch *batch, struct fd_tile *tile);
+	void (*emit_tile_mem2gmem)(struct fd_batch *batch, struct fd_tile *tile);
+	void (*emit_tile_renderprep)(struct fd_batch *batch, struct fd_tile *tile);
+	void (*emit_tile_gmem2mem)(struct fd_batch *batch, struct fd_tile *tile);
 
 	/* optional, for GMEM bypass: */
-	void (*emit_sysmem_prep)(struct fd_context *ctx);
+	void (*emit_sysmem_prep)(struct fd_batch *batch);
 
 	/* draw: */
 	bool (*draw_vbo)(struct fd_context *ctx, const struct pipe_draw_info *info);
diff --git a/src/gallium/drivers/freedreno/freedreno_draw.c b/src/gallium/drivers/freedreno/freedreno_draw.c
index 2c76333cf12..b9477620176 100644
--- a/src/gallium/drivers/freedreno/freedreno_draw.c
+++ b/src/gallium/drivers/freedreno/freedreno_draw.c
@@ -40,26 +40,27 @@
 #include "freedreno_util.h"
 
 static void
-resource_read(struct fd_context *ctx, struct pipe_resource *prsc)
+resource_read(struct fd_batch *batch, struct pipe_resource *prsc)
 {
 	if (!prsc)
 		return;
-	fd_batch_resource_used(ctx->batch, fd_resource(prsc), FD_PENDING_READ);
+	fd_batch_resource_used(batch, fd_resource(prsc), FD_PENDING_READ);
 }
 
 static void
-resource_written(struct fd_context *ctx, struct pipe_resource *prsc)
+resource_written(struct fd_batch *batch, struct pipe_resource *prsc)
 {
 	if (!prsc)
 		return;
-	fd_batch_resource_used(ctx->batch, fd_resource(prsc), FD_PENDING_WRITE);
+	fd_batch_resource_used(batch, fd_resource(prsc), FD_PENDING_WRITE);
 }
 
 static void
 fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 {
 	struct fd_context *ctx = fd_context(pctx);
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_batch *batch = ctx->batch;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
 	unsigned i, prims, buffers = 0;
 
@@ -89,18 +90,18 @@ fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 
 	if (fd_depth_enabled(ctx)) {
 		buffers |= FD_BUFFER_DEPTH;
-		resource_written(ctx, pfb->zsbuf->texture);
-		ctx->gmem_reason |= FD_GMEM_DEPTH_ENABLED;
+		resource_written(batch, pfb->zsbuf->texture);
+		batch->gmem_reason |= FD_GMEM_DEPTH_ENABLED;
 	}
 
 	if (fd_stencil_enabled(ctx)) {
 		buffers |= FD_BUFFER_STENCIL;
-		resource_written(ctx, pfb->zsbuf->texture);
-		ctx->gmem_reason |= FD_GMEM_STENCIL_ENABLED;
+		resource_written(batch, pfb->zsbuf->texture);
+		batch->gmem_reason |= FD_GMEM_STENCIL_ENABLED;
 	}
 
 	if (fd_logicop_enabled(ctx))
-		ctx->gmem_reason |= FD_GMEM_LOGICOP_ENABLED;
+		batch->gmem_reason |= FD_GMEM_LOGICOP_ENABLED;
 
 	for (i = 0; i < pfb->nr_cbufs; i++) {
 		struct pipe_resource *surf;
@@ -110,45 +111,45 @@ fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 
 		surf = pfb->cbufs[i]->texture;
 
-		resource_written(ctx, surf);
+		resource_written(batch, surf);
 		buffers |= PIPE_CLEAR_COLOR0 << i;
 
 		if (surf->nr_samples > 1)
-			ctx->gmem_reason |= FD_GMEM_MSAA_ENABLED;
+			batch->gmem_reason |= FD_GMEM_MSAA_ENABLED;
 
 		if (fd_blend_enabled(ctx, i))
-			ctx->gmem_reason |= FD_GMEM_BLEND_ENABLED;
+			batch->gmem_reason |= FD_GMEM_BLEND_ENABLED;
 	}
 
 	/* Skip over buffer 0, that is sent along with the command stream */
 	for (i = 1; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
-		resource_read(ctx, ctx->constbuf[PIPE_SHADER_VERTEX].cb[i].buffer);
-		resource_read(ctx, ctx->constbuf[PIPE_SHADER_FRAGMENT].cb[i].buffer);
+		resource_read(batch, ctx->constbuf[PIPE_SHADER_VERTEX].cb[i].buffer);
+		resource_read(batch, ctx->constbuf[PIPE_SHADER_FRAGMENT].cb[i].buffer);
 	}
 
 	/* Mark VBOs as being read */
 	for (i = 0; i < ctx->vtx.vertexbuf.count; i++) {
 		assert(!ctx->vtx.vertexbuf.vb[i].user_buffer);
-		resource_read(ctx, ctx->vtx.vertexbuf.vb[i].buffer);
+		resource_read(batch, ctx->vtx.vertexbuf.vb[i].buffer);
 	}
 
 	/* Mark index buffer as being read */
-	resource_read(ctx, ctx->indexbuf.buffer);
+	resource_read(batch, ctx->indexbuf.buffer);
 
 	/* Mark textures as being read */
 	for (i = 0; i < ctx->verttex.num_textures; i++)
 		if (ctx->verttex.textures[i])
-			resource_read(ctx, ctx->verttex.textures[i]->texture);
+			resource_read(batch, ctx->verttex.textures[i]->texture);
 	for (i = 0; i < ctx->fragtex.num_textures; i++)
 		if (ctx->fragtex.textures[i])
-			resource_read(ctx, ctx->fragtex.textures[i]->texture);
+			resource_read(batch, ctx->fragtex.textures[i]->texture);
 
 	/* Mark streamout buffers as being written.. */
 	for (i = 0; i < ctx->streamout.num_targets; i++)
 		if (ctx->streamout.targets[i])
-			resource_written(ctx, ctx->streamout.targets[i]->buffer);
+			resource_written(batch, ctx->streamout.targets[i]->buffer);
 
-	ctx->num_draws++;
+	batch->num_draws++;
 
 	prims = u_reduced_prims_for_vertices(info->mode, info->count);
 
@@ -165,17 +166,17 @@ fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 	ctx->stats.prims_generated += prims;
 
 	/* any buffers that haven't been cleared yet, we need to restore: */
-	ctx->restore |= buffers & (FD_BUFFER_ALL & ~ctx->cleared);
+	batch->restore |= buffers & (FD_BUFFER_ALL & ~batch->cleared);
 	/* and any buffers used, need to be resolved: */
-	ctx->resolve |= buffers;
+	batch->resolve |= buffers;
 
-	DBG("%x num_draws=%u (%s/%s)", buffers, ctx->num_draws,
+	DBG("%x num_draws=%u (%s/%s)", buffers, batch->num_draws,
 		util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 		util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 
-	fd_hw_query_set_stage(ctx, ctx->batch->draw, FD_STAGE_DRAW);
+	fd_hw_query_set_stage(ctx, batch->draw, FD_STAGE_DRAW);
 	if (ctx->draw_vbo(ctx, info))
-		ctx->needs_flush = true;
+		batch->needs_flush = true;
 
 	for (i = 0; i < ctx->streamout.num_targets; i++)
 		ctx->streamout.offsets[i] += info->count;
@@ -183,7 +184,7 @@ fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 	if (fd_mesa_debug & FD_DBG_DDRAW)
 		ctx->dirty = 0xffffffff;
 
-	fd_batch_check_size(ctx->batch);
+	fd_batch_check_size(batch);
 }
 
 /* TODO figure out how to make better use of existing state mechanism
@@ -197,7 +198,8 @@ fd_clear(struct pipe_context *pctx, unsigned buffers,
 		const union pipe_color_union *color, double depth, unsigned stencil)
 {
 	struct fd_context *ctx = fd_context(pctx);
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct fd_batch *batch = ctx->batch;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
 	unsigned cleared_buffers;
 	int i;
@@ -213,38 +215,38 @@ fd_clear(struct pipe_context *pctx, unsigned buffers,
 	 * something like alpha-test causes side effects from the draw in
 	 * the depth buffer, etc)
 	 */
-	cleared_buffers = buffers & (FD_BUFFER_ALL & ~ctx->restore);
+	cleared_buffers = buffers & (FD_BUFFER_ALL & ~batch->restore);
 
 	/* do we have full-screen scissor? */
 	if (!memcmp(scissor, &ctx->disabled_scissor, sizeof(*scissor))) {
-		ctx->cleared |= cleared_buffers;
+		batch->cleared |= cleared_buffers;
 	} else {
-		ctx->partial_cleared |= cleared_buffers;
+		batch->partial_cleared |= cleared_buffers;
 		if (cleared_buffers & PIPE_CLEAR_COLOR)
-			ctx->cleared_scissor.color = *scissor;
+			batch->cleared_scissor.color = *scissor;
 		if (cleared_buffers & PIPE_CLEAR_DEPTH)
-			ctx->cleared_scissor.depth = *scissor;
+			batch->cleared_scissor.depth = *scissor;
 		if (cleared_buffers & PIPE_CLEAR_STENCIL)
-			ctx->cleared_scissor.stencil = *scissor;
+			batch->cleared_scissor.stencil = *scissor;
 	}
-	ctx->resolve |= buffers;
-	ctx->needs_flush = true;
+	batch->resolve |= buffers;
+	batch->needs_flush = true;
 
 	if (buffers & PIPE_CLEAR_COLOR)
 		for (i = 0; i < pfb->nr_cbufs; i++)
 			if (buffers & (PIPE_CLEAR_COLOR0 << i))
-				resource_written(ctx, pfb->cbufs[i]->texture);
+				resource_written(batch, pfb->cbufs[i]->texture);
 
 	if (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
-		resource_written(ctx, pfb->zsbuf->texture);
-		ctx->gmem_reason |= FD_GMEM_CLEARS_DEPTH_STENCIL;
+		resource_written(batch, pfb->zsbuf->texture);
+		batch->gmem_reason |= FD_GMEM_CLEARS_DEPTH_STENCIL;
 	}
 
 	DBG("%x depth=%f, stencil=%u (%s/%s)", buffers, depth, stencil,
 		util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 		util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 
-	fd_hw_query_set_stage(ctx, ctx->batch->draw, FD_STAGE_CLEAR);
+	fd_hw_query_set_stage(ctx, batch->draw, FD_STAGE_CLEAR);
 
 	ctx->clear(ctx, buffers, color, depth, stencil);
 
diff --git a/src/gallium/drivers/freedreno/freedreno_draw.h b/src/gallium/drivers/freedreno/freedreno_draw.h
index 7a970a28b31..60cd9c09881 100644
--- a/src/gallium/drivers/freedreno/freedreno_draw.h
+++ b/src/gallium/drivers/freedreno/freedreno_draw.h
@@ -42,7 +42,7 @@ struct fd_ringbuffer;
 void fd_draw_init(struct pipe_context *pctx);
 
 static inline void
-fd_draw(struct fd_context *ctx, struct fd_ringbuffer *ring,
+fd_draw(struct fd_batch *batch, struct fd_ringbuffer *ring,
 		enum pc_di_primtype primtype,
 		enum pc_di_vis_cull_mode vismode,
 		enum pc_di_src_sel src_sel, uint32_t count,
@@ -59,7 +59,7 @@ fd_draw(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	 */
 	emit_marker(ring, 7);
 
-	if (is_a3xx_p0(ctx->screen)) {
+	if (is_a3xx_p0(batch->ctx->screen)) {
 		/* dummy-draw workaround: */
 		OUT_PKT3(ring, CP_DRAW_INDX, 3);
 		OUT_RING(ring, 0x00000000);
@@ -81,7 +81,7 @@ fd_draw(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		 * we know if we are binning or not
 		 */
 		OUT_RINGP(ring, DRAW(primtype, src_sel, idx_type, 0, instances),
-				&ctx->draw_patches);
+				&batch->draw_patches);
 	} else {
 		OUT_RING(ring, DRAW(primtype, src_sel, idx_type, vismode, instances));
 	}
@@ -93,7 +93,7 @@ fd_draw(struct fd_context *ctx, struct fd_ringbuffer *ring,
 
 	emit_marker(ring, 7);
 
-	fd_reset_wfi(ctx);
+	fd_reset_wfi(batch->ctx);
 }
 
 
@@ -112,18 +112,19 @@ size2indextype(unsigned index_size)
 
 /* this is same for a2xx/a3xx, so split into helper: */
 static inline void
-fd_draw_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
+fd_draw_emit(struct fd_batch *batch, struct fd_ringbuffer *ring,
 		enum pc_di_primtype primtype,
 		enum pc_di_vis_cull_mode vismode,
 		const struct pipe_draw_info *info)
 {
-	struct pipe_index_buffer *idx = &ctx->indexbuf;
 	struct pipe_resource *idx_buffer = NULL;
 	enum pc_di_index_size idx_type = INDEX_SIZE_IGN;
 	enum pc_di_src_sel src_sel;
 	uint32_t idx_size, idx_offset;
 
 	if (info->indexed) {
+		struct pipe_index_buffer *idx = &batch->ctx->indexbuf;
+
 		assert(!idx->user_buffer);
 
 		idx_buffer = idx->buffer;
@@ -139,7 +140,7 @@ fd_draw_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
 		src_sel = DI_SRC_SEL_AUTO_INDEX;
 	}
 
-	fd_draw(ctx, ring, primtype, vismode, src_sel,
+	fd_draw(batch, ring, primtype, vismode, src_sel,
 			info->count, info->instance_count - 1,
 			idx_type, idx_size, idx_offset, idx_buffer);
 }
diff --git a/src/gallium/drivers/freedreno/freedreno_gmem.c b/src/gallium/drivers/freedreno/freedreno_gmem.c
index 54a3247b933..9ca7f5f93ff 100644
--- a/src/gallium/drivers/freedreno/freedreno_gmem.c
+++ b/src/gallium/drivers/freedreno/freedreno_gmem.c
@@ -67,11 +67,11 @@
  * resolve.
  */
 
-static uint32_t bin_width(struct fd_context *ctx)
+static uint32_t bin_width(struct fd_screen *screen)
 {
-	if (is_a4xx(ctx->screen))
+	if (is_a4xx(screen))
 		return 1024;
-	if (is_a3xx(ctx->screen))
+	if (is_a3xx(screen))
 		return 992;
 	return 512;
 }
@@ -103,20 +103,21 @@ total_size(uint8_t cbuf_cpp[], uint8_t zsbuf_cpp[2],
 }
 
 static void
-calculate_tiles(struct fd_context *ctx)
+calculate_tiles(struct fd_batch *batch)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
-	struct pipe_scissor_state *scissor = &ctx->max_scissor;
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
+	struct pipe_scissor_state *scissor = &batch->max_scissor;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	uint32_t gmem_size = ctx->screen->gmemsize_bytes;
 	uint32_t minx, miny, width, height;
 	uint32_t nbins_x = 1, nbins_y = 1;
 	uint32_t bin_w, bin_h;
-	uint32_t max_width = bin_width(ctx);
+	uint32_t max_width = bin_width(ctx->screen);
 	uint8_t cbuf_cpp[MAX_RENDER_TARGETS] = {0}, zsbuf_cpp[2] = {0};
 	uint32_t i, j, t, xoff, yoff;
 	uint32_t tpp_x, tpp_y;
-	bool has_zs = !!(ctx->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL));
+	bool has_zs = !!(batch->resolve & (FD_BUFFER_DEPTH | FD_BUFFER_STENCIL));
 	int tile_n[ARRAY_SIZE(ctx->pipe)];
 
 	if (has_zs) {
@@ -302,14 +303,15 @@ calculate_tiles(struct fd_context *ctx)
 }
 
 static void
-render_tiles(struct fd_context *ctx)
+render_tiles(struct fd_batch *batch)
 {
+	struct fd_context *ctx = batch->ctx;
 	struct fd_gmem_stateobj *gmem = &ctx->gmem;
 	int i;
 
-	ctx->emit_tile_init(ctx);
+	ctx->emit_tile_init(batch);
 
-	if (ctx->restore)
+	if (batch->restore)
 		ctx->stats.batch_restore++;
 
 	for (i = 0; i < (gmem->nbins_x * gmem->nbins_y); i++) {
@@ -318,52 +320,54 @@ render_tiles(struct fd_context *ctx)
 		DBG("bin_h=%d, yoff=%d, bin_w=%d, xoff=%d",
 			tile->bin_h, tile->yoff, tile->bin_w, tile->xoff);
 
-		ctx->emit_tile_prep(ctx, tile);
+		ctx->emit_tile_prep(batch, tile);
 
-		if (ctx->restore) {
-			fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_MEM2GMEM);
-			ctx->emit_tile_mem2gmem(ctx, tile);
-			fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_NULL);
+		if (batch->restore) {
+			fd_hw_query_set_stage(ctx, batch->gmem, FD_STAGE_MEM2GMEM);
+			ctx->emit_tile_mem2gmem(batch, tile);
+			fd_hw_query_set_stage(ctx, batch->gmem, FD_STAGE_NULL);
 		}
 
-		ctx->emit_tile_renderprep(ctx, tile);
+		ctx->emit_tile_renderprep(batch, tile);
 
-		fd_hw_query_prepare_tile(ctx, i, ctx->ring);
+		fd_hw_query_prepare_tile(ctx, i, batch->gmem);
 
 		/* emit IB to drawcmds: */
-		ctx->emit_ib(ctx->ring, ctx->batch->draw);
+		ctx->emit_ib(batch->gmem, batch->draw);
 		fd_reset_wfi(ctx);
 
 		/* emit gmem2mem to transfer tile back to system memory: */
-		fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_GMEM2MEM);
-		ctx->emit_tile_gmem2mem(ctx, tile);
-		fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_NULL);
+		fd_hw_query_set_stage(ctx, batch->gmem, FD_STAGE_GMEM2MEM);
+		ctx->emit_tile_gmem2mem(batch, tile);
+		fd_hw_query_set_stage(ctx, batch->gmem, FD_STAGE_NULL);
 	}
 }
 
 static void
-render_sysmem(struct fd_context *ctx)
+render_sysmem(struct fd_batch *batch)
 {
-	ctx->emit_sysmem_prep(ctx);
+	struct fd_context *ctx = batch->ctx;
 
-	fd_hw_query_prepare_tile(ctx, 0, ctx->ring);
+	ctx->emit_sysmem_prep(batch);
+
+	fd_hw_query_prepare_tile(ctx, 0, batch->gmem);
 
 	/* emit IB to drawcmds: */
-	ctx->emit_ib(ctx->ring, ctx->batch->draw);
+	ctx->emit_ib(batch->gmem, batch->draw);
 	fd_reset_wfi(ctx);
 }
 
 void
-fd_gmem_render_tiles(struct fd_context *ctx)
+fd_gmem_render_tiles(struct fd_batch *batch)
 {
-	struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
-	struct fd_batch *batch = ctx->batch;
+	struct fd_context *ctx = batch->ctx;
+	struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 	bool sysmem = false;
 
 	if (ctx->emit_sysmem_prep) {
-		if (ctx->cleared || ctx->gmem_reason || (ctx->num_draws > 5)) {
+		if (batch->cleared || batch->gmem_reason || (batch->num_draws > 5)) {
 			DBG("GMEM: cleared=%x, gmem_reason=%x, num_draws=%u",
-				ctx->cleared, ctx->gmem_reason, ctx->num_draws);
+				batch->cleared, batch->gmem_reason, batch->num_draws);
 		} else if (!(fd_mesa_debug & FD_DBG_NOBYPASS)) {
 			sysmem = true;
 		}
@@ -378,36 +382,28 @@ fd_gmem_render_tiles(struct fd_context *ctx)
 
 	ctx->stats.batch_total++;
 
-	ctx->ring = batch->gmem;
-
 	if (sysmem) {
 		DBG("rendering sysmem (%s/%s)",
 			util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 			util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 		fd_hw_query_prepare(ctx, 1);
-		render_sysmem(ctx);
+		render_sysmem(batch);
 		ctx->stats.batch_sysmem++;
 	} else {
 		struct fd_gmem_stateobj *gmem = &ctx->gmem;
-		calculate_tiles(ctx);
+		calculate_tiles(batch);
 		DBG("rendering %dx%d tiles (%s/%s)", gmem->nbins_x, gmem->nbins_y,
 			util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 			util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 		fd_hw_query_prepare(ctx, gmem->nbins_x * gmem->nbins_y);
-		render_tiles(ctx);
+		render_tiles(batch);
 		ctx->stats.batch_gmem++;
 	}
 
 	fd_ringbuffer_flush(batch->gmem);
 
-	ctx->ring = NULL;
-
 	fd_reset_wfi(ctx);
 
-	/* reset maximal bounds: */
-	ctx->max_scissor.minx = ctx->max_scissor.miny = ~0;
-	ctx->max_scissor.maxx = ctx->max_scissor.maxy = 0;
-
 	ctx->dirty = ~0;
 }
 
@@ -431,26 +427,26 @@ skip_restore(struct pipe_scissor_state *scissor, struct fd_tile *tile)
  * case would be a single clear.
  */
 bool
-fd_gmem_needs_restore(struct fd_context *ctx, struct fd_tile *tile,
+fd_gmem_needs_restore(struct fd_batch *batch, struct fd_tile *tile,
 		uint32_t buffers)
 {
-	if (!(ctx->restore & buffers))
+	if (!(batch->restore & buffers))
 		return false;
 
 	/* if buffers partially cleared, then slow-path to figure out
 	 * if this particular tile needs restoring:
 	 */
 	if ((buffers & FD_BUFFER_COLOR) &&
-			(ctx->partial_cleared & FD_BUFFER_COLOR) &&
-			skip_restore(&ctx->cleared_scissor.color, tile))
+			(batch->partial_cleared & FD_BUFFER_COLOR) &&
+			skip_restore(&batch->cleared_scissor.color, tile))
 		return false;
 	if ((buffers & FD_BUFFER_DEPTH) &&
-			(ctx->partial_cleared & FD_BUFFER_DEPTH) &&
-			skip_restore(&ctx->cleared_scissor.depth, tile))
+			(batch->partial_cleared & FD_BUFFER_DEPTH) &&
+			skip_restore(&batch->cleared_scissor.depth, tile))
 		return false;
 	if ((buffers & FD_BUFFER_STENCIL) &&
-			(ctx->partial_cleared & FD_BUFFER_STENCIL) &&
-			skip_restore(&ctx->cleared_scissor.stencil, tile))
+			(batch->partial_cleared & FD_BUFFER_STENCIL) &&
+			skip_restore(&batch->cleared_scissor.stencil, tile))
 		return false;
 
 	return true;
diff --git a/src/gallium/drivers/freedreno/freedreno_gmem.h b/src/gallium/drivers/freedreno/freedreno_gmem.h
index 38b557eb077..116423a40e2 100644
--- a/src/gallium/drivers/freedreno/freedreno_gmem.h
+++ b/src/gallium/drivers/freedreno/freedreno_gmem.h
@@ -59,11 +59,11 @@ struct fd_gmem_stateobj {
 	uint16_t width, height;
 };
 
-struct fd_context;
+struct fd_batch;
 
-void fd_gmem_render_tiles(struct fd_context *ctx);
+void fd_gmem_render_tiles(struct fd_batch *batch);
 
-bool fd_gmem_needs_restore(struct fd_context *ctx, struct fd_tile *tile,
+bool fd_gmem_needs_restore(struct fd_batch *batch, struct fd_tile *tile,
 		uint32_t buffers);
 
 #endif /* FREEDRENO_GMEM_H_ */
diff --git a/src/gallium/drivers/freedreno/freedreno_query_hw.c b/src/gallium/drivers/freedreno/freedreno_query_hw.c
index 817f129a404..76d90d60410 100644
--- a/src/gallium/drivers/freedreno/freedreno_query_hw.c
+++ b/src/gallium/drivers/freedreno/freedreno_query_hw.c
@@ -71,7 +71,7 @@ get_sample(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	if (!ctx->sample_cache[idx]) {
 		ctx->sample_cache[idx] =
 			ctx->sample_providers[idx]->get_sample(ctx, ring);
-		ctx->needs_flush = true;
+		ctx->batch->needs_flush = true;
 	}
 
 	fd_hw_sample_reference(ctx, &samp, ctx->sample_cache[idx]);
@@ -213,7 +213,7 @@ fd_hw_get_query_result(struct fd_context *ctx, struct fd_query *q,
 		/* if app didn't actually trigger any cmdstream, then
 		 * we have nothing to do:
 		 */
-		if (!ctx->needs_flush)
+		if (!ctx->batch->needs_flush)
 			return true;
 		DBG("reading query result forces flush!");
 		fd_context_render(&ctx->base);
diff --git a/src/gallium/drivers/freedreno/freedreno_resource.c b/src/gallium/drivers/freedreno/freedreno_resource.c
index eea53c60696..20d68fed9d5 100644
--- a/src/gallium/drivers/freedreno/freedreno_resource.c
+++ b/src/gallium/drivers/freedreno/freedreno_resource.c
@@ -821,7 +821,8 @@ fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond)
 	util_blitter_save_depth_stencil_alpha(ctx->blitter, ctx->zsa);
 	util_blitter_save_stencil_ref(ctx->blitter, &ctx->stencil_ref);
 	util_blitter_save_sample_mask(ctx->blitter, ctx->sample_mask);
-	util_blitter_save_framebuffer(ctx->blitter, &ctx->framebuffer);
+	util_blitter_save_framebuffer(ctx->blitter,
+			ctx->batch ? &ctx->batch->framebuffer : NULL);
 	util_blitter_save_fragment_sampler_states(ctx->blitter,
 			ctx->fragtex.num_samplers,
 			(void **)ctx->fragtex.samplers);
diff --git a/src/gallium/drivers/freedreno/freedreno_state.c b/src/gallium/drivers/freedreno/freedreno_state.c
index 252d153e341..98b56c7d512 100644
--- a/src/gallium/drivers/freedreno/freedreno_state.c
+++ b/src/gallium/drivers/freedreno/freedreno_state.c
@@ -115,13 +115,15 @@ fd_set_framebuffer_state(struct pipe_context *pctx,
 		const struct pipe_framebuffer_state *framebuffer)
 {
 	struct fd_context *ctx = fd_context(pctx);
-	struct pipe_framebuffer_state *cso = &ctx->framebuffer;
+	struct pipe_framebuffer_state *cso;
 
-	DBG("%d: cbufs[0]=%p, zsbuf=%p", ctx->needs_flush,
+	DBG("%d: cbufs[0]=%p, zsbuf=%p", ctx->batch->needs_flush,
 			framebuffer->cbufs[0], framebuffer->zsbuf);
 
 	fd_context_render(pctx);
 
+	cso = &ctx->batch->framebuffer;
+
 	if ((cso->width != framebuffer->width) ||
 			(cso->height != framebuffer->height))
 		ctx->needs_rb_fbd = true;