5 files changed, 17 insertions, 9 deletions

diff --git a/src/gallium/drivers/radeonsi/radeonsi_pipe.h b/src/gallium/drivers/radeonsi/radeonsi_pipe.h
index 388f6df0518..c5b33f721a4 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_pipe.h
+++ b/src/gallium/drivers/radeonsi/radeonsi_pipe.h
@@ -140,7 +140,6 @@ struct r600_context {
 	unsigned			pa_sc_line_stipple;
 	unsigned			pa_su_sc_mode_cntl;
 	unsigned			pa_cl_clip_cntl;
-	unsigned			pa_cl_vs_out_cntl;
 	/* for saving when using blitter */
 	struct pipe_stencil_ref		stencil_ref;
 	struct si_pipe_shader_selector	*ps_shader;
diff --git a/src/gallium/drivers/radeonsi/radeonsi_shader.c b/src/gallium/drivers/radeonsi/radeonsi_shader.c
index bea2895bf0c..e6ed5459cea 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_shader.c
+++ b/src/gallium/drivers/radeonsi/radeonsi_shader.c
@@ -612,7 +612,9 @@ static void si_llvm_emit_epilogue(struct lp_build_tgsi_context * bld_base)
 			/* Select the correct target */
 			switch(d->Semantic.Name) {
 			case TGSI_SEMANTIC_PSIZE:
-				target = V_008DFC_SQ_EXP_POS;
+				shader->vs_out_misc_write = 1;
+				shader->vs_out_point_size = 1;
+				target = V_008DFC_SQ_EXP_POS + 1;
 				break;
 			case TGSI_SEMANTIC_POSITION:
 				if (si_shader_ctx->type == TGSI_PROCESSOR_VERTEX) {
diff --git a/src/gallium/drivers/radeonsi/radeonsi_shader.h b/src/gallium/drivers/radeonsi/radeonsi_shader.h
index 0fbd601ad34..667f2c33985 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_shader.h
+++ b/src/gallium/drivers/radeonsi/radeonsi_shader.h
@@ -110,6 +110,8 @@ struct si_shader {
 	bool			uses_kill;
 	bool			uses_instanceid;
 	bool			fs_write_all;
+	bool			vs_out_misc_write;
+	bool			vs_out_point_size;
 	unsigned		nr_cbufs;
 };
 
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index ed95b1d8d89..6d072ef725a 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -412,9 +412,6 @@ static void *si_create_rs_state(struct pipe_context *ctx,
 		S_028810_ZCLIP_NEAR_DISABLE(!state->depth_clip) |
 		S_028810_ZCLIP_FAR_DISABLE(!state->depth_clip) |
 		S_028810_DX_LINEAR_ATTR_CLIP_ENA(1);
-	rs->pa_cl_vs_out_cntl =
-		S_02881C_USE_VTX_POINT_SIZE(state->point_size_per_vertex) |
-		S_02881C_VS_OUT_MISC_VEC_ENA(state->point_size_per_vertex);
 
 	clip_rule = state->scissor ? 0xAAAA : 0xFFFF;
 
@@ -485,7 +482,6 @@ static void si_bind_rs_state(struct pipe_context *ctx, void *state)
 	rctx->pa_sc_line_stipple = rs->pa_sc_line_stipple;
 	rctx->pa_su_sc_mode_cntl = rs->pa_su_sc_mode_cntl;
 	rctx->pa_cl_clip_cntl = rs->pa_cl_clip_cntl;
-	rctx->pa_cl_vs_out_cntl = rs->pa_cl_vs_out_cntl;
 
 	si_pm4_bind_state(rctx, rasterizer, rs);
 	si_update_fb_rs_state(rctx);
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 8d169079f4b..a9ecc64afa4 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -55,8 +55,13 @@ static void si_pipe_shader_vs(struct pipe_context *ctx, struct si_pipe_shader *s
 	 * takes care of adding a dummy export.
 	 */
 	for (nparams = 0, i = 0 ; i < shader->shader.noutput; i++) {
-		if (shader->shader.output[i].name != TGSI_SEMANTIC_POSITION)
+		switch (shader->shader.output[i].name) {
+		case TGSI_SEMANTIC_POSITION:
+		case TGSI_SEMANTIC_PSIZE:
+			break;
+		default:
 			nparams++;
+		}
 	}
 	if (nparams < 1)
 		nparams = 1;
@@ -66,7 +71,9 @@ static void si_pipe_shader_vs(struct pipe_context *ctx, struct si_pipe_shader *s
 
 	si_pm4_set_reg(pm4, R_02870C_SPI_SHADER_POS_FORMAT,
 		       S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP) |
-		       S_02870C_POS1_EXPORT_FORMAT(V_02870C_SPI_SHADER_NONE) |
+		       S_02870C_POS1_EXPORT_FORMAT(shader->shader.vs_out_misc_write ?
+						   V_02870C_SPI_SHADER_4COMP :
+						   V_02870C_SPI_SHADER_NONE) |
 		       S_02870C_POS2_EXPORT_FORMAT(V_02870C_SPI_SHADER_NONE) |
 		       S_02870C_POS3_EXPORT_FORMAT(V_02870C_SPI_SHADER_NONE));
 
@@ -261,6 +268,7 @@ static bool si_update_draw_info_state(struct r600_context *rctx,
 			       const struct pipe_draw_info *info)
 {
 	struct si_pm4_state *pm4 = CALLOC_STRUCT(si_pm4_state);
+	struct si_shader *vs = &rctx->vs_shader->current->shader;
 	unsigned prim = si_conv_pipe_prim(info->mode);
 	unsigned ls_mask = 0;
 
@@ -297,7 +305,8 @@ static bool si_update_draw_info_state(struct r600_context *rctx,
 		si_pm4_set_reg(pm4, R_028814_PA_SU_SC_MODE_CNTL, rctx->pa_su_sc_mode_cntl);
         }
 	si_pm4_set_reg(pm4, R_02881C_PA_CL_VS_OUT_CNTL,
-		       prim == PIPE_PRIM_POINTS ? rctx->pa_cl_vs_out_cntl : 0
+		       S_02881C_USE_VTX_POINT_SIZE(vs->vs_out_point_size) |
+		       S_02881C_VS_OUT_MISC_VEC_ENA(vs->vs_out_misc_write)
 		       /*| (rctx->rasterizer->clip_plane_enable &
 		       rctx->vs_shader->shader.clip_dist_write)*/);
 	si_pm4_set_reg(pm4, R_028810_PA_CL_CLIP_CNTL, rctx->pa_cl_clip_cntl