freedreno/a6xx: add LRZ support

As with a5xx, hidden behind FD_MESA_DEBUG=lrz due to being paranoid
about z-fighting issues with some games (in particular, this was
observed with 0ad on a5xx.. but I think the proper solution to enable
this by default is to figure out how to do driver specific driconf
options).

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

8 files changed, 104 insertions, 132 deletions

diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_blend.c b/src/gallium/drivers/freedreno/a6xx/fd6_blend.c
index e62163e797c..185b061cd1e 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_blend.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_blend.c
@@ -91,8 +91,6 @@ fd6_blend_state_create(struct pipe_context *pctx,
 
 	so->base = *cso;
 
-	so->lrz_write = true;  /* unless blend enabled for any MRT */
-
 	for (i = 0; i < ARRAY_SIZE(so->rb_mrt); i++) {
 		const struct pipe_rt_blend_state *rt;
 
@@ -128,7 +126,6 @@ fd6_blend_state_create(struct pipe_context *pctx,
 					A6XX_RB_MRT_CONTROL_BLEND |
 					A6XX_RB_MRT_CONTROL_BLEND2;
 			mrt_blend |= (1 << i);
-			so->lrz_write = false;
 		}
 
 		if (reads_dest) {
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_blend.h b/src/gallium/drivers/freedreno/a6xx/fd6_blend.h
index 45ad4ea071f..d59b6628c5b 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_blend.h
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_blend.h
@@ -48,7 +48,6 @@ struct fd6_blend_stateobj {
 	} rb_mrt[A6XX_MAX_RENDER_TARGETS];
 	uint32_t rb_blend_cntl;
 	uint32_t sp_blend_cntl;
-	bool lrz_write;
 };
 
 static inline struct fd6_blend_stateobj *
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_draw.c b/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
index 16985cb6e33..252fdc354b4 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_draw.c
@@ -248,8 +248,7 @@ fd6_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
 	/* figure out whether we need to disable LRZ write for binning
 	 * pass using draw pass's fp:
 	 */
-	// TODO disable until lrz is wired up:
-	emit.no_lrz_write = true; // fp->writes_pos || fp->has_kill;
+	emit.no_lrz_write = fp->writes_pos || fp->has_kill;
 
 	emit.key.binning_pass = false;
 	emit.dirty = dirty;
@@ -290,12 +289,10 @@ static bool is_z32(enum pipe_format format)
 	}
 }
 
-#if 0
 static void
 fd6_clear_lrz(struct fd_batch *batch, struct fd_resource *zsbuf, double depth)
 {
 	struct fd_ringbuffer *ring;
-	uint32_t clear = util_pack_z(PIPE_FORMAT_Z16_UNORM, depth);
 
 	// TODO mid-frame clears (ie. app doing crazy stuff)??  Maybe worth
 	// splitting both clear and lrz clear out into their own rb's.  And
@@ -310,137 +307,105 @@ fd6_clear_lrz(struct fd_batch *batch, struct fd_resource *zsbuf, double depth)
 
 	ring = batch->lrz_clear;
 
-	OUT_WFI5(ring);
+	emit_marker6(ring, 7);
+	OUT_PKT7(ring, CP_SET_MARKER, 1);
+	OUT_RING(ring, A2XX_CP_SET_MARKER_0_MODE(RM6_BYPASS));
+	emit_marker6(ring, 7);
 
 	OUT_PKT4(ring, REG_A6XX_RB_CCU_CNTL, 1);
 	OUT_RING(ring, 0x10000000);
 
 	OUT_PKT4(ring, REG_A6XX_HLSQ_UPDATE_CNTL, 1);
-	OUT_RING(ring, 0x20fffff);
+	OUT_RING(ring, 0x7ffff);
 
-	OUT_PKT4(ring, REG_A6XX_GRAS_SU_CNTL, 1);
-	OUT_RING(ring, A6XX_GRAS_SU_CNTL_LINEHALFWIDTH(0.0));
-
-	OUT_PKT4(ring, REG_A6XX_GRAS_CNTL, 1);
-	OUT_RING(ring, 0x00000000);
+	emit_marker6(ring, 7);
+	OUT_PKT7(ring, CP_SET_MARKER, 1);
+	OUT_RING(ring, A2XX_CP_SET_MARKER_0_MODE(0xc));
+	emit_marker6(ring, 7);
 
-	OUT_PKT4(ring, REG_A6XX_GRAS_CL_CNTL, 1);
-	OUT_RING(ring, 0x00000181);
+	OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8C01, 1);
+	OUT_RING(ring, 0x0);
 
-	OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_CNTL, 1);
+	OUT_PKT4(ring, REG_A6XX_SP_PS_2D_SRC_INFO, 13);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
 	OUT_RING(ring, 0x00000000);
-
-	OUT_PKT4(ring, REG_A6XX_RB_MRT_BUF_INFO(0), 5);
-	OUT_RING(ring, A6XX_RB_MRT_BUF_INFO_COLOR_FORMAT(RB5_R16_UNORM) |
-			A6XX_RB_MRT_BUF_INFO_COLOR_TILE_MODE(TILE6_LINEAR) |
-			A6XX_RB_MRT_BUF_INFO_COLOR_SWAP(WZYX));
-	OUT_RING(ring, A6XX_RB_MRT_PITCH(zsbuf->lrz_pitch * 2));
-	OUT_RING(ring, A6XX_RB_MRT_ARRAY_PITCH(fd_bo_size(zsbuf->lrz)));
-	OUT_RELOCW(ring, zsbuf->lrz, 0x1000, 0, 0);
-
-	OUT_PKT4(ring, REG_A6XX_RB_RENDER_CNTL, 1);
 	OUT_RING(ring, 0x00000000);
 
-	OUT_PKT4(ring, REG_A6XX_RB_DEST_MSAA_CNTL, 1);
-	OUT_RING(ring, A6XX_RB_DEST_MSAA_CNTL_SAMPLES(MSAA_ONE));
-
-	OUT_PKT4(ring, REG_A6XX_RB_BLIT_CNTL, 1);
-	OUT_RING(ring, A6XX_RB_BLIT_CNTL_BUF(BLIT_MRT0));
-
-	OUT_PKT4(ring, REG_A6XX_RB_CLEAR_CNTL, 1);
-	OUT_RING(ring, A6XX_RB_CLEAR_CNTL_FAST_CLEAR |
-			A6XX_RB_CLEAR_CNTL_MASK(0xf));
-
-	OUT_PKT4(ring, REG_A6XX_RB_CLEAR_COLOR_DW0, 1);
-	OUT_RING(ring, clear);  /* RB_CLEAR_COLOR_DW0 */
-
-	OUT_PKT4(ring, REG_A6XX_VSC_RESOLVE_CNTL, 2);
-	OUT_RING(ring, A6XX_VSC_RESOLVE_CNTL_X(zsbuf->lrz_width) |
-			 A6XX_VSC_RESOLVE_CNTL_Y(zsbuf->lrz_height));
-	OUT_RING(ring, 0x00000000);   // XXX UNKNOWN_0CDE
+	OUT_PKT4(ring, REG_A6XX_SP_UNKNOWN_ACC0, 1);
+	OUT_RING(ring, 0x0000f410);
 
-	OUT_PKT4(ring, REG_A6XX_RB_CNTL, 1);
-	OUT_RING(ring, A6XX_RB_CNTL_BYPASS);
+	OUT_PKT4(ring, REG_A6XX_GRAS_2D_BLIT_CNTL, 1);
+	OUT_RING(ring, A6XX_GRAS_2D_BLIT_CNTL_COLOR_FORMAT(RB6_R16_UNORM) |
+			0x4f00080);
 
-	OUT_PKT4(ring, REG_A6XX_RB_RESOLVE_CNTL_1, 2);
-	OUT_RING(ring, A6XX_RB_RESOLVE_CNTL_1_X(0) |
-			A6XX_RB_RESOLVE_CNTL_1_Y(0));
-	OUT_RING(ring, A6XX_RB_RESOLVE_CNTL_2_X(zsbuf->lrz_width - 1) |
-			A6XX_RB_RESOLVE_CNTL_2_Y(zsbuf->lrz_height - 1));
+	OUT_PKT4(ring, REG_A6XX_RB_2D_BLIT_CNTL, 1);
+	OUT_RING(ring, A6XX_RB_2D_BLIT_CNTL_COLOR_FORMAT(RB6_R16_UNORM) |
+			0x4f00080);
 
-	fd6_emit_blit(batch->ctx, ring);
-}
-#endif
+	fd6_event_write(batch, ring, UNK_1D, true);
+	fd6_event_write(batch, ring, PC_CCU_INVALIDATE_COLOR, false);
 
-#if 0
-clear_with_cp_blit()
-{
-	/* Clear with CP_BLIT */
-	WRITE(REG_A6XX_GRAS_2D_BLIT_CNTL, 0x10f43180);
+	OUT_PKT4(ring, REG_A6XX_RB_2D_SRC_SOLID_C0, 4);
+	OUT_RING(ring, fui(depth));
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
 
-	OUT_PKT4(ring, REG_A6XX_SP_PS_2D_SRC_INFO, 7);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
+	OUT_PKT4(ring, REG_A6XX_RB_2D_DST_INFO, 9);
+	OUT_RING(ring, A6XX_RB_2D_DST_INFO_COLOR_FORMAT(RB6_R16_UNORM) |
+			A6XX_RB_2D_DST_INFO_TILE_MODE(TILE6_LINEAR) |
+			A6XX_RB_2D_DST_INFO_COLOR_SWAP(WZYX));
+	OUT_RELOCW(ring, zsbuf->lrz, 0, 0, 0);
+	OUT_RING(ring, A6XX_RB_2D_DST_SIZE_PITCH(zsbuf->lrz_pitch * 2));
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
+	OUT_RING(ring, 0x00000000);
 
-	WRITE(0xacc0, 0xf181);
-	WRITE(0xacc0, 0xf181);
+	OUT_PKT4(ring, REG_A6XX_GRAS_2D_SRC_TL_X, 4);
+	OUT_RING(ring, A6XX_GRAS_2D_SRC_TL_X_X(0));
+	OUT_RING(ring, A6XX_GRAS_2D_SRC_BR_X_X(0));
+	OUT_RING(ring, A6XX_GRAS_2D_SRC_TL_Y_Y(0));
+	OUT_RING(ring, A6XX_GRAS_2D_SRC_BR_Y_Y(0));
 
-	WRITE(REG_A6XX_GRAS_2D_BLIT_CNTL, 0x10f43180);
-	WRITE(REG_A6XX_RB_2D_BLIT_CNTL, 0x10f43180);
+	OUT_PKT4(ring, REG_A6XX_GRAS_2D_DST_TL, 2);
+	OUT_RING(ring, A6XX_GRAS_2D_DST_TL_X(0) |
+			A6XX_GRAS_2D_DST_TL_Y(0));
+	OUT_RING(ring, A6XX_GRAS_2D_DST_BR_X(zsbuf->lrz_width - 1) |
+			A6XX_GRAS_2D_DST_BR_Y(zsbuf->lrz_height - 1));
 
-	OUT_PKT4(ring, REG_A6XX_RB_2D_SRC_SOLID_C0, 4);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0xff);
-	OUT_RING(ring, 0);
+	fd6_event_write(batch, ring, 0x3f, false);
 
-	DBG("%x %x %x %x\n", color->ui[0], color->ui[1], color->ui[2], color->ui[3]);
+	OUT_WFI5(ring);
 
-	struct pipe_surface *psurf = pfb->cbufs[0];
-	struct fd_resource *rsc = fd_resource(psurf->texture);
-	struct fd_resource_slice *slice = fd_resource_slice(rsc, psurf->u.tex.level);
+	OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
+	OUT_RING(ring, 0x1000000);
 
-	uint32_t offset = fd_resource_offset(rsc, psurf->u.tex.level,
-										 psurf->u.tex.first_layer);
-	uint32_t stride = slice->pitch * rsc->cpp;
+	OUT_PKT7(ring, CP_BLIT, 1);
+	OUT_RING(ring, CP_BLIT_0_OP(BLIT_OP_SCALE));
 
-	enum a6xx_color_fmt format = fd6_pipe2color(pfmt);
-	OUT_PKT4(ring, REG_A6XX_RB_2D_DST_INFO, 9);
-	OUT_RING(ring,
-			 A6XX_RB_2D_DST_INFO_COLOR_FORMAT(format) |
-			 A6XX_RB_2D_DST_INFO_TILE_MODE(TILE6_LINEAR) |
-			 A6XX_RB_2D_DST_INFO_COLOR_SWAP(WXYZ));
-	OUT_RELOCW(ring, rsc->bo, offset, 0, 0);  /* RB_2D_DST_LO/HI */
-	OUT_RING(ring, A6XX_RB_2D_DST_SIZE_PITCH(stride));
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
+	OUT_WFI5(ring);
 
-	OUT_PKT4(ring, REG_A6XX_GRAS_2D_SRC_TL_X, 4);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
-	OUT_RING(ring, 0);
+	OUT_PKT4(ring, REG_A6XX_RB_UNKNOWN_8E04, 1);
+	OUT_RING(ring, 0x0);
 
-	OUT_PKT4(ring, REG_A6XX_GRAS_2D_DST_TL, 2);
-	OUT_RING(ring,
-			 A6XX_GRAS_2D_DST_TL_X(ctx->batch->max_scissor.minx) |
-			 A6XX_GRAS_2D_DST_TL_Y(ctx->batch->max_scissor.miny));
-	OUT_RING(ring,
-			 A6XX_GRAS_2D_DST_BR_X(ctx->batch->max_scissor.maxx) |
-			 A6XX_GRAS_2D_DST_BR_Y(ctx->batch->max_scissor.maxy));
+	fd6_event_write(batch, ring, UNK_1D, true);
+	fd6_event_write(batch, ring, FACENESS_FLUSH, true);
+	fd6_event_write(batch, ring, CACHE_FLUSH_TS, true);
 
-	OUT_PKT7(ring, CP_BLIT, 1);
-	OUT_RING(ring, CP_BLIT_0_OP(BLIT_OP_SCALE));
+	fd6_cache_flush(batch, ring);
 }
-#endif
 
 static bool
 fd6_clear(struct fd_context *ctx, unsigned buffers,
@@ -567,7 +532,6 @@ fd6_clear(struct fd_context *ctx, unsigned buffers,
 
 		fd6_emit_blit(ctx->batch, ring);
 
-#if 0
 		if (pfb->zsbuf && (buffers & PIPE_CLEAR_DEPTH)) {
 			struct fd_resource *zsbuf = fd_resource(pfb->zsbuf->texture);
 			if (zsbuf->lrz) {
@@ -575,7 +539,6 @@ fd6_clear(struct fd_context *ctx, unsigned buffers,
 				fd6_clear_lrz(ctx->batch, zsbuf, depth);
 			}
 		}
-#endif
 	}
 
 	return true;
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
index 7433e39f168..75a3e64267a 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
@@ -632,20 +632,25 @@ fd6_emit_state(struct fd_context *ctx, struct fd_ringbuffer *ring,
 	}
 
 	if (dirty & (FD_DIRTY_ZSA | FD_DIRTY_BLEND | FD_DIRTY_PROG)) {
-		struct fd6_blend_stateobj *blend = fd6_blend_stateobj(ctx->blend);
 		struct fd6_zsa_stateobj *zsa = fd6_zsa_stateobj(ctx->zsa);
 
 		if (pfb->zsbuf) {
 			struct fd_resource *rsc = fd_resource(pfb->zsbuf->texture);
 			uint32_t gras_lrz_cntl = zsa->gras_lrz_cntl;
+			uint32_t rb_lrz_cntl = zsa->rb_lrz_cntl;
 
-			if (emit->no_lrz_write || !rsc->lrz || !rsc->lrz_valid)
+			if (emit->no_lrz_write || !rsc->lrz || !rsc->lrz_valid) {
 				gras_lrz_cntl = 0;
-			else if (emit->key.binning_pass && blend->lrz_write && zsa->lrz_write)
+				rb_lrz_cntl = 0;
+			} else if (emit->key.binning_pass && zsa->lrz_write) {
 				gras_lrz_cntl |= A6XX_GRAS_LRZ_CNTL_LRZ_WRITE;
+			}
 
 			OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_CNTL, 1);
 			OUT_RING(ring, gras_lrz_cntl);
+
+			OUT_PKT4(ring, REG_A6XX_RB_LRZ_CNTL, 1);
+			OUT_RING(ring, rb_lrz_cntl);
 		}
 	}
 
@@ -1158,6 +1163,12 @@ t7              opcode: CP_WAIT_FOR_IDLE (26) (1 dwords)
 
 	OUT_PKT4(ring, REG_A6XX_SP_GS_CTRL_REG0, 1);
 	OUT_RING(ring, 0x00000000);
+
+	OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_CNTL, 1);
+	OUT_RING(ring, 0x00000000);
+
+	OUT_PKT4(ring, REG_A6XX_RB_LRZ_CNTL, 1);
+	OUT_RING(ring, 0x00000000);
 }
 
 static void
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.h b/src/gallium/drivers/freedreno/a6xx/fd6_emit.h
index bfff7b30266..743e5df196b 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.h
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.h
@@ -153,17 +153,8 @@ fd6_emit_render_cntl(struct fd_context *ctx, bool blit, bool binning)
 static inline void
 fd6_emit_lrz_flush(struct fd_ringbuffer *ring)
 {
-	/* TODO I think the extra writes to GRAS_LRZ_CNTL are probably
-	 * a workaround and not needed on all a5xx.
-	 */
-	OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_CNTL, 1);
-	OUT_RING(ring, A6XX_GRAS_LRZ_CNTL_ENABLE);
-
 	OUT_PKT7(ring, CP_EVENT_WRITE, 1);
 	OUT_RING(ring, LRZ_FLUSH);
-
-	OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_CNTL, 1);
-	OUT_RING(ring, 0x0);
 }
 
 static inline enum a6xx_state_block
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_gmem.c b/src/gallium/drivers/freedreno/a6xx/fd6_gmem.c
index d82b71954b4..0c96250f974 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_gmem.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_gmem.c
@@ -159,9 +159,12 @@ emit_zs(struct fd_ringbuffer *ring, struct pipe_surface *zsbuf,
 
 		if (rsc->lrz) {
 			OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_BUFFER_BASE_LO, 5);
-			OUT_RELOCW(ring, rsc->lrz, 0x1000, 0, 0);
+			OUT_RELOCW(ring, rsc->lrz, 0, 0, 0);
 			OUT_RING(ring, A6XX_GRAS_LRZ_BUFFER_PITCH_PITCH(rsc->lrz_pitch));
-			OUT_RELOCW(ring, rsc->lrz, 0, 0, 0); /* GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_LO/HI */
+			//OUT_RELOCW(ring, rsc->lrz, 0, 0, 0); /* GRAS_LRZ_FAST_CLEAR_BUFFER_BASE_LO/HI */
+			// XXX a6xx seems to use a different buffer here.. not sure what for..
+			OUT_RING(ring, 0x00000000);
+			OUT_RING(ring, 0x00000000);
 		} else {
 			OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_BUFFER_BASE_LO, 5);
 			OUT_RING(ring, 0x00000000);
@@ -441,11 +444,11 @@ fd6_emit_tile_init(struct fd_batch *batch)
 
 	fd6_emit_restore(batch, ring);
 
+	fd6_emit_lrz_flush(ring);
+
 	if (batch->lrz_clear)
 		ctx->emit_ib(ring, batch->lrz_clear);
 
-	fd6_emit_lrz_flush(ring);
-
 	fd6_cache_flush(batch, ring);
 
 	OUT_PKT7(ring, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
@@ -468,7 +471,6 @@ fd6_emit_tile_init(struct fd_batch *batch)
 				A6XX_RB_BIN_CONTROL_BINNING_PASS | 0x6000000);
 		update_render_cntl(batch, true);
 		emit_binning_pass(batch);
-		fd6_emit_lrz_flush(ring);
 		patch_draws(batch, USE_VISIBILITY);
 
 		set_bin_size(ring, gmem->bin_w, gmem->bin_h,
@@ -791,6 +793,9 @@ fd6_emit_tile_fini(struct fd_batch *batch)
 {
 	struct fd_ringbuffer *ring = batch->gmem;
 
+	OUT_PKT4(ring, REG_A6XX_GRAS_LRZ_CNTL, 1);
+	OUT_RING(ring, A6XX_GRAS_LRZ_CNTL_ENABLE | A6XX_GRAS_LRZ_CNTL_UNK3);
+
 	fd6_emit_lrz_flush(ring);
 
 	fd6_event_write(batch, ring, CACHE_FLUSH_TS, true);
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_zsa.c b/src/gallium/drivers/freedreno/a6xx/fd6_zsa.c
index b7ce799eb40..01599bb1b6e 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_zsa.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_zsa.c
@@ -50,11 +50,13 @@ fd6_zsa_state_create(struct pipe_context *pctx,
 	case PIPE_FUNC_LESS:
 	case PIPE_FUNC_LEQUAL:
 		so->gras_lrz_cntl = A6XX_GRAS_LRZ_CNTL_ENABLE;
+		so->rb_lrz_cntl = A6XX_RB_LRZ_CNTL_ENABLE;
 		break;
 
 	case PIPE_FUNC_GREATER:
 	case PIPE_FUNC_GEQUAL:
 		so->gras_lrz_cntl = A6XX_GRAS_LRZ_CNTL_ENABLE | A6XX_GRAS_LRZ_CNTL_GREATER;
+		so->rb_lrz_cntl = A6XX_RB_LRZ_CNTL_ENABLE;
 		break;
 
 	default:
@@ -63,8 +65,11 @@ fd6_zsa_state_create(struct pipe_context *pctx,
 		break;
 	}
 
-	if (!(cso->stencil->enabled || cso->alpha.enabled || !cso->depth.writemask))
+	if (cso->depth.writemask) {
+		if (cso->depth.enabled)
+			so->gras_lrz_cntl |= A6XX_GRAS_LRZ_CNTL_UNK4;
 		so->lrz_write = true;
+	}
 
 	so->rb_depth_cntl |=
 		A6XX_RB_DEPTH_CNTL_ZFUNC(cso->depth.func); /* maps 1:1 */
diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_zsa.h b/src/gallium/drivers/freedreno/a6xx/fd6_zsa.h
index 1c626890981..ca777e0f74c 100644
--- a/src/gallium/drivers/freedreno/a6xx/fd6_zsa.h
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_zsa.h
@@ -43,6 +43,7 @@ struct fd6_zsa_stateobj {
 	uint32_t rb_stencilmask;
 	uint32_t rb_stencilwrmask;
 	uint32_t gras_lrz_cntl;
+	uint32_t rb_lrz_cntl;
 	bool lrz_write;
 };