r600g: add multiple stream support for geom shaders

This patch is taken from work by Glenn and myself,
and I've spent some time making it all work here.

This adds support for the multiple streams part of
ARB_gpu_shader5 to r600g.

It doesn't enable ARB_gpu_shader5 yet.

Signed-off-by: Dave Airlie <[email protected]>

5 files changed, 171 insertions, 74 deletions

diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
index 9ef92741879..0d4b5987246 100644
--- a/src/gallium/drivers/r600/evergreen_state.c
+++ b/src/gallium/drivers/r600/evergreen_state.c
@@ -3012,8 +3012,12 @@ void evergreen_update_gs_state(struct pipe_context *ctx, struct r600_pipe_shader
 	struct r600_command_buffer *cb = &shader->command_buffer;
 	struct r600_shader *rshader = &shader->shader;
 	struct r600_shader *cp_shader = &shader->gs_copy_shader->shader;
-	unsigned gsvs_itemsize =
-			(cp_shader->ring_item_size * shader->selector->gs_max_out_vertices) >> 2;
+	unsigned gsvs_itemsizes[4] = {
+			(cp_shader->ring_item_sizes[0] * shader->selector->gs_max_out_vertices) >> 2,
+			(cp_shader->ring_item_sizes[1] * shader->selector->gs_max_out_vertices) >> 2,
+			(cp_shader->ring_item_sizes[2] * shader->selector->gs_max_out_vertices) >> 2,
+			(cp_shader->ring_item_sizes[3] * shader->selector->gs_max_out_vertices) >> 2
+	};
 
 	r600_init_command_buffer(cb, 64);
 
@@ -3032,21 +3036,24 @@ void evergreen_update_gs_state(struct pipe_context *ctx, struct r600_pipe_shader
 				S_028B90_ENABLE(shader->selector->gs_num_invocations > 0));
 	}
 	r600_store_context_reg_seq(cb, R_02891C_SQ_GS_VERT_ITEMSIZE, 4);
-	r600_store_value(cb, cp_shader->ring_item_size >> 2);
-	r600_store_value(cb, 0);
-	r600_store_value(cb, 0);
-	r600_store_value(cb, 0);
+	r600_store_value(cb, cp_shader->ring_item_sizes[0] >> 2);
+	r600_store_value(cb, cp_shader->ring_item_sizes[1] >> 2);
+	r600_store_value(cb, cp_shader->ring_item_sizes[2] >> 2);
+	r600_store_value(cb, cp_shader->ring_item_sizes[3] >> 2);
 
 	r600_store_context_reg(cb, R_028900_SQ_ESGS_RING_ITEMSIZE,
-			       (rshader->ring_item_size) >> 2);
+			       (rshader->ring_item_sizes[0]) >> 2);
 
 	r600_store_context_reg(cb, R_028904_SQ_GSVS_RING_ITEMSIZE,
-			       gsvs_itemsize);
+			       gsvs_itemsizes[0] +
+			       gsvs_itemsizes[1] +
+			       gsvs_itemsizes[2] +
+			       gsvs_itemsizes[3]);
 
 	r600_store_context_reg_seq(cb, R_02892C_SQ_GSVS_RING_OFFSET_1, 3);
-	r600_store_value(cb, gsvs_itemsize);
-	r600_store_value(cb, gsvs_itemsize);
-	r600_store_value(cb, gsvs_itemsize);
+	r600_store_value(cb, gsvs_itemsizes[0]);
+	r600_store_value(cb, gsvs_itemsizes[0] + gsvs_itemsizes[1]);
+	r600_store_value(cb, gsvs_itemsizes[0] + gsvs_itemsizes[1] + gsvs_itemsizes[2]);
 
 	/* FIXME calculate these values somehow ??? */
 	r600_store_context_reg_seq(cb, R_028A54_GS_PER_ES, 3);
diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
index f6efaa312f2..fd9c16c6a96 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -358,7 +358,7 @@ static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
 	case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
 		return 16384;
 	case PIPE_CAP_MAX_VERTEX_STREAMS:
-		return 1;
+		return family >= CHIP_CEDAR ? 4 : 1;
 
 	case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
 		return 2047;
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 153f9a22cea..25a251e8187 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -311,7 +311,8 @@ struct r600_shader_ctx {
 	int					gs_out_ring_offset;
 	int					gs_next_vertex;
 	struct r600_shader	*gs_for_vs;
-	int					gs_export_gpr_treg;
+	int					gs_export_gpr_tregs[4];
+	const struct pipe_stream_output_info	*gs_stream_output_info;
 	unsigned				enabled_stream_buffers_mask;
 };
 
@@ -320,7 +321,7 @@ struct r600_shader_tgsi_instruction {
 	int (*process)(struct r600_shader_ctx *ctx);
 };
 
-static int emit_gs_ring_writes(struct r600_shader_ctx *ctx, bool ind);
+static int emit_gs_ring_writes(struct r600_shader_ctx *ctx, const struct pipe_stream_output_info *so, int stream, bool ind);
 static const struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[], eg_shader_tgsi_instruction[], cm_shader_tgsi_instruction[];
 static int tgsi_helper_tempx_replicate(struct r600_shader_ctx *ctx);
 static inline void callstack_push(struct r600_shader_ctx *ctx, unsigned reason);
@@ -1337,9 +1338,11 @@ static int process_twoside_color_inputs(struct r600_shader_ctx *ctx)
 	return 0;
 }
 
-static int emit_streamout(struct r600_shader_ctx *ctx, struct pipe_stream_output_info *so)
+static int emit_streamout(struct r600_shader_ctx *ctx, struct pipe_stream_output_info *so,
+						  int stream, unsigned *stream_item_size)
 {
 	unsigned so_gpr[PIPE_MAX_SHADER_OUTPUTS];
+	unsigned start_comp[PIPE_MAX_SHADER_OUTPUTS];
 	int i, j, r;
 
 	/* Sanity checking. */
@@ -1359,8 +1362,9 @@ static int emit_streamout(struct r600_shader_ctx *ctx, struct pipe_stream_output
 
 	/* Initialize locations where the outputs are stored. */
 	for (i = 0; i < so->num_outputs; i++) {
-		so_gpr[i] = ctx->shader->output[so->output[i].register_index].gpr;
 
+		so_gpr[i] = ctx->shader->output[so->output[i].register_index].gpr;
+		start_comp[i] = so->output[i].start_component;
 		/* Lower outputs with dst_offset < start_component.
 		 *
 		 * We can only output 4D vectors with a write mask, e.g. we can
@@ -1386,7 +1390,7 @@ static int emit_streamout(struct r600_shader_ctx *ctx, struct pipe_stream_output
 				if (r)
 					return r;
 			}
-			so->output[i].start_component = 0;
+			start_comp[i] = 0;
 			so_gpr[i] = tmp;
 		}
 	}
@@ -1395,18 +1399,21 @@ static int emit_streamout(struct r600_shader_ctx *ctx, struct pipe_stream_output
 	for (i = 0; i < so->num_outputs; i++) {
 		struct r600_bytecode_output output;
 
+		if (stream != -1 && stream != so->output[i].output_buffer)
+			continue;
+
 		memset(&output, 0, sizeof(struct r600_bytecode_output));
 		output.gpr = so_gpr[i];
-		output.elem_size = so->output[i].num_components;
-		output.array_base = so->output[i].dst_offset - so->output[i].start_component;
+		output.elem_size = so->output[i].num_components - 1;
+		if (output.elem_size == 2)
+			output.elem_size = 3; // 3 not supported, write 4 with junk at end
+		output.array_base = so->output[i].dst_offset - start_comp[i];
 		output.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE;
 		output.burst_count = 1;
 		/* array_size is an upper limit for the burst_count
 		 * with MEM_STREAM instructions */
 		output.array_size = 0xFFF;
-		output.comp_mask = ((1 << so->output[i].num_components) - 1) << so->output[i].start_component;
-
-		ctx->enabled_stream_buffers_mask |= (1 << so->output[i].output_buffer);
+		output.comp_mask = ((1 << so->output[i].num_components) - 1) << start_comp[i];
 
 		if (ctx->bc->chip_class >= EVERGREEN) {
 			switch (so->output[i].output_buffer) {
@@ -1423,6 +1430,9 @@ static int emit_streamout(struct r600_shader_ctx *ctx, struct pipe_stream_output
 				output.op = CF_OP_MEM_STREAM0_BUF3;
 				break;
 			}
+			output.op += so->output[i].stream * 4;
+			assert(output.op >= CF_OP_MEM_STREAM0_BUF0 && output.op <= CF_OP_MEM_STREAM3_BUF3);
+			ctx->enabled_stream_buffers_mask |= (1 << so->output[i].output_buffer) << so->output[i].stream * 4;
 		} else {
 			switch (so->output[i].output_buffer) {
 			case 0:
@@ -1438,6 +1448,7 @@ static int emit_streamout(struct r600_shader_ctx *ctx, struct pipe_stream_output
 				output.op = CF_OP_MEM_STREAM3;
 					break;
 			}
+			ctx->enabled_stream_buffers_mask |= 1 << so->output[i].output_buffer;
 		}
 		r = r600_bytecode_add_output(ctx->bc, &output);
 		if (r)
@@ -1490,7 +1501,8 @@ static int generate_gs_copy_shader(struct r600_context *rctx,
 	struct r600_bytecode_output output;
 	struct r600_bytecode_cf *cf_jump, *cf_pop,
 		*last_exp_pos = NULL, *last_exp_param = NULL;
-	int i, next_clip_pos = 61, next_param = 0;
+	int i, j, next_clip_pos = 61, next_param = 0;
+	int ring;
 
 	cshader = calloc(1, sizeof(struct r600_pipe_shader));
 	if (!cshader)
@@ -1510,6 +1522,9 @@ static int generate_gs_copy_shader(struct r600_context *rctx,
 
 	ctx.bc->isa = rctx->isa;
 
+	cf_jump = NULL;
+	memset(cshader->shader.ring_item_sizes, 0, sizeof(cshader->shader.ring_item_sizes));
+
 	/* R0.x = R0.x & 0x3fffffff */
 	memset(&alu, 0, sizeof(alu));
 	alu.op = ALU_OP2_AND_INT;
@@ -1528,22 +1543,10 @@ static int generate_gs_copy_shader(struct r600_context *rctx,
 	alu.last = 1;
 	r600_bytecode_add_alu(ctx.bc, &alu);
 
-	/* PRED_SETE_INT __, R0.y, 0 */
-	memset(&alu, 0, sizeof(alu));
-	alu.op = ALU_OP2_PRED_SETE_INT;
-	alu.src[0].chan = 1;
-	alu.src[1].sel = V_SQ_ALU_SRC_0;
-	alu.execute_mask = 1;
-	alu.update_pred = 1;
-	alu.last = 1;
-	r600_bytecode_add_alu_type(ctx.bc, &alu, CF_OP_ALU_PUSH_BEFORE);
-
-	r600_bytecode_add_cfinst(ctx.bc, CF_OP_JUMP);
-	cf_jump = ctx.bc->cf_last;
-
 	/* fetch vertex data from GSVS ring */
 	for (i = 0; i < ocnt; ++i) {
 		struct r600_shader_io *out = &ctx.shader->output[i];
+
 		out->gpr = i + 1;
 		out->ring_offset = i * 16;
 
@@ -1553,6 +1556,7 @@ static int generate_gs_copy_shader(struct r600_context *rctx,
 		vtx.fetch_type = SQ_VTX_FETCH_NO_INDEX_OFFSET;
 		vtx.offset = out->ring_offset;
 		vtx.dst_gpr = out->gpr;
+		vtx.src_gpr = 0;
 		vtx.dst_sel_x = 0;
 		vtx.dst_sel_y = 1;
 		vtx.dst_sel_z = 2;
@@ -1565,18 +1569,68 @@ static int generate_gs_copy_shader(struct r600_context *rctx,
 
 		r600_bytecode_add_vtx(ctx.bc, &vtx);
 	}
+	ctx.temp_reg = i + 1;
+	for (ring = 3; ring >= 0; --ring) {
+		bool enabled = false;
+		for (i = 0; i < so->num_outputs; i++) {
+			if (so->output[i].stream == ring) {
+				enabled = true;
+				break;
+			}
+		}
+		if (ring != 0 && !enabled) {
+			cshader->shader.ring_item_sizes[ring] = 0;
+			continue;
+		}
+
+		if (cf_jump) {
+			// Patch up jump label
+			r600_bytecode_add_cfinst(ctx.bc, CF_OP_POP);
+			cf_pop = ctx.bc->cf_last;
+
+			cf_jump->cf_addr = cf_pop->id + 2;
+			cf_jump->pop_count = 1;
+			cf_pop->cf_addr = cf_pop->id + 2;
+			cf_pop->pop_count = 1;
+		}
+
+		/* PRED_SETE_INT __, R0.y, ring */
+		memset(&alu, 0, sizeof(alu));
+		alu.op = ALU_OP2_PRED_SETE_INT;
+		alu.src[0].chan = 1;
+		alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
+		alu.src[1].value = ring;
+		alu.execute_mask = 1;
+		alu.update_pred = 1;
+		alu.last = 1;
+		r600_bytecode_add_alu_type(ctx.bc, &alu, CF_OP_ALU_PUSH_BEFORE);
 
-	/* XXX handle clipvertex, streamout? */
-	emit_streamout(&ctx, so);
+		r600_bytecode_add_cfinst(ctx.bc, CF_OP_JUMP);
+		cf_jump = ctx.bc->cf_last;
+
+		if (enabled)
+			emit_streamout(&ctx, so, ring, &cshader->shader.ring_item_sizes[ring]);
+		cshader->shader.ring_item_sizes[ring] = ocnt * 16;
+	}
 
 	/* export vertex data */
 	/* XXX factor out common code with r600_shader_from_tgsi ? */
 	for (i = 0; i < ocnt; ++i) {
 		struct r600_shader_io *out = &ctx.shader->output[i];
-
+		bool instream0 = true;
 		if (out->name == TGSI_SEMANTIC_CLIPVERTEX)
 			continue;
 
+		for (j = 0; j < so->num_outputs; j++) {
+			if (so->output[j].register_index == i) {
+				if (so->output[j].stream == 0)
+					break;
+				if (so->output[j].stream > 0)
+					instream0 = false;
+			}
+		}
+		if (!instream0)
+			continue;
 		memset(&output, 0, sizeof(output));
 		output.gpr = out->gpr;
 		output.elem_size = 3;
@@ -1722,19 +1776,19 @@ static int generate_gs_copy_shader(struct r600_context *rctx,
 	}
 
 	gs->gs_copy_shader = cshader;
+	cshader->enabled_stream_buffers_mask = ctx.enabled_stream_buffers_mask;
 
 	ctx.bc->nstack = 1;
 
-	cshader->enabled_stream_buffers_mask = ctx.enabled_stream_buffers_mask;
-	cshader->shader.ring_item_size = ocnt * 16;
-
 	return r600_bytecode_build(ctx.bc);
 }
 
-static int emit_gs_ring_writes(struct r600_shader_ctx *ctx, bool ind)
+static int emit_gs_ring_writes(struct r600_shader_ctx *ctx, const struct pipe_stream_output_info *so, int stream, bool ind)
 {
 	struct r600_bytecode_output output;
 	int i, k, ring_offset;
+	int effective_stream = stream == -1 ? 0 : stream;
+	int idx = 0;
 
 	for (i = 0; i < ctx->shader->noutput; i++) {
 		if (ctx->gs_for_vs) {
@@ -1751,15 +1805,18 @@ static int emit_gs_ring_writes(struct r600_shader_ctx *ctx, bool ind)
 
 			if (ring_offset == -1)
 				continue;
-		} else
-			ring_offset = i * 16;
+		} else {
+			ring_offset = idx * 16;
+			idx++;
+		}
 
+		if (stream > 0 && ctx->shader->output[i].name == TGSI_SEMANTIC_POSITION)
+			continue;
 		/* next_ring_offset after parsing input decls contains total size of
 		 * single vertex data, gs_next_vertex - current vertex index */
 		if (!ind)
 			ring_offset += ctx->gs_out_ring_offset * ctx->gs_next_vertex;
 
-		/* get a temp and add the ring offset to the next vertex base in the shader */
 		memset(&output, 0, sizeof(struct r600_bytecode_output));
 		output.gpr = ctx->shader->output[i].gpr;
 		output.elem_size = 3;
@@ -1770,28 +1827,39 @@ static int emit_gs_ring_writes(struct r600_shader_ctx *ctx, bool ind)
 			output.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE_IND;
 		else
 			output.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE;
-		output.op = CF_OP_MEM_RING;
 
+		switch (stream) {
+		default:
+		case 0:
+			output.op = CF_OP_MEM_RING; break;
+		case 1:
+			output.op = CF_OP_MEM_RING1; break;
+		case 2:
+			output.op = CF_OP_MEM_RING2; break;
+		case 3:
+			output.op = CF_OP_MEM_RING3; break;
+		}
 
 		if (ind) {
 			output.array_base = ring_offset >> 2; /* in dwords */
 			output.array_size = 0xfff;
-			output.index_gpr = ctx->gs_export_gpr_treg;
+			output.index_gpr = ctx->gs_export_gpr_tregs[effective_stream];
 		} else
 			output.array_base = ring_offset >> 2; /* in dwords */
 		r600_bytecode_add_output(ctx->bc, &output);
 	}
 
 	if (ind) {
+		/* get a temp and add the ring offset to the next vertex base in the shader */
 		struct r600_bytecode_alu alu;
 		int r;
 
 		memset(&alu, 0, sizeof(struct r600_bytecode_alu));
 		alu.op = ALU_OP2_ADD_INT;
-		alu.src[0].sel = ctx->gs_export_gpr_treg;
+		alu.src[0].sel = ctx->gs_export_gpr_tregs[effective_stream];
 		alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
 		alu.src[1].value = ctx->gs_out_ring_offset >> 4;
-		alu.dst.sel = ctx->gs_export_gpr_treg;
+		alu.dst.sel = ctx->gs_export_gpr_tregs[effective_stream];
 		alu.dst.write = 1;
 		alu.last = 1;
 		r = r600_bytecode_add_alu(ctx->bc, &alu);
@@ -1856,6 +1924,7 @@ static int r600_shader_from_tgsi(struct r600_context *rctx,
 	ctx.next_ring_offset = 0;
 	ctx.gs_out_ring_offset = 0;
 	ctx.gs_next_vertex = 0;
+	ctx.gs_stream_output_info = &so;
 
 	shader->uses_index_registers = false;
 	ctx.face_gpr = -1;
@@ -1942,8 +2011,11 @@ static int r600_shader_from_tgsi(struct r600_context *rctx,
 	ctx.bc->index_reg[1] = ctx.bc->ar_reg + 2;
 
 	if (ctx.type == TGSI_PROCESSOR_GEOMETRY) {
-		ctx.gs_export_gpr_treg = ctx.bc->ar_reg + 3;
-		ctx.temp_reg = ctx.bc->ar_reg + 4;
+		ctx.gs_export_gpr_tregs[0] = ctx.bc->ar_reg + 3;
+		ctx.gs_export_gpr_tregs[1] = ctx.bc->ar_reg + 4;
+		ctx.gs_export_gpr_tregs[2] = ctx.bc->ar_reg + 5;
+		ctx.gs_export_gpr_tregs[3] = ctx.bc->ar_reg + 6;
+		ctx.temp_reg = ctx.bc->ar_reg + 7;
 	} else {
 		ctx.temp_reg = ctx.bc->ar_reg + 3;
 	}
@@ -2006,7 +2078,10 @@ static int r600_shader_from_tgsi(struct r600_context *rctx,
 		}
 	}
 	
-	shader->ring_item_size = ctx.next_ring_offset;
+	shader->ring_item_sizes[0] = ctx.next_ring_offset;
+	shader->ring_item_sizes[1] = 0;
+	shader->ring_item_sizes[2] = 0;
+	shader->ring_item_sizes[3] = 0;
 
 	/* Process two side if needed */
 	if (shader->two_side && ctx.colors_used) {
@@ -2129,17 +2204,18 @@ static int r600_shader_from_tgsi(struct r600_context *rctx,
 		if (ctx.type == TGSI_PROCESSOR_GEOMETRY) {
 			struct r600_bytecode_alu alu;
 			int r;
-
-			memset(&alu, 0, sizeof(struct r600_bytecode_alu));
-			alu.op = ALU_OP1_MOV;
-			alu.src[0].sel = V_SQ_ALU_SRC_LITERAL;
-			alu.src[0].value = 0;
-			alu.dst.sel = ctx.gs_export_gpr_treg;
-			alu.dst.write = 1;
-			alu.last = 1;
-			r = r600_bytecode_add_alu(ctx.bc, &alu);
-			if (r)
-				return r;
+			for (j = 0; j < 4; j++) {
+				memset(&alu, 0, sizeof(struct r600_bytecode_alu));
+				alu.op = ALU_OP1_MOV;
+				alu.src[0].sel = V_SQ_ALU_SRC_LITERAL;
+				alu.src[0].value = 0;
+				alu.dst.sel = ctx.gs_export_gpr_tregs[j];
+				alu.dst.write = 1;
+				alu.last = 1;
+				r = r600_bytecode_add_alu(ctx.bc, &alu);
+				if (r)
+					return r;
+			}
 		}
 		if (shader->two_side && ctx.colors_used) {
 			if ((r = process_twoside_color_inputs(&ctx)))
@@ -2240,14 +2316,20 @@ static int r600_shader_from_tgsi(struct r600_context *rctx,
 	/* Add stream outputs. */
 	if (!ring_outputs && ctx.type == TGSI_PROCESSOR_VERTEX &&
 	    so.num_outputs && !use_llvm)
-		emit_streamout(&ctx, &so);
+		emit_streamout(&ctx, &so, -1, NULL);
 
 	pipeshader->enabled_stream_buffers_mask = ctx.enabled_stream_buffers_mask;
 	convert_edgeflag_to_int(&ctx);
 
 	if (ring_outputs) {
-		if (key.vs.as_es)
-			emit_gs_ring_writes(&ctx, FALSE);
+		if (key.vs.as_es) {
+			ctx.gs_export_gpr_tregs[0] = r600_get_temp(&ctx);
+			ctx.gs_export_gpr_tregs[1] = -1;
+			ctx.gs_export_gpr_tregs[2] = -1;
+			ctx.gs_export_gpr_tregs[3] = -1;
+
+			emit_gs_ring_writes(&ctx, &so, -1, FALSE);
+		}
 	} else {
 		/* Export output */
 		next_clip_base = shader->vs_out_misc_write ? 62 : 61;
@@ -7198,10 +7280,17 @@ static int tgsi_loop_brk_cont(struct r600_shader_ctx *ctx)
 
 static int tgsi_gs_emit(struct r600_shader_ctx *ctx)
 {
+	struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
+	int stream = ctx->literals[inst->Src[0].Register.Index * 4 + inst->Src[0].Register.SwizzleX];
+	int r;
+
 	if (ctx->inst_info->op == CF_OP_EMIT_VERTEX)
-		emit_gs_ring_writes(ctx, TRUE);
+		emit_gs_ring_writes(ctx, ctx->gs_stream_output_info, stream, TRUE);
 
-	return r600_bytecode_add_cfinst(ctx->bc, ctx->inst_info->op);
+	r = r600_bytecode_add_cfinst(ctx->bc, ctx->inst_info->op);
+	if (!r)
+		ctx->bc->cf_last->count = stream; // Count field for CUT/EMIT_VERTEX indicates which stream
+	return r;
 }
 
 static int tgsi_umad(struct r600_shader_ctx *ctx)
diff --git a/src/gallium/drivers/r600/r600_shader.h b/src/gallium/drivers/r600/r600_shader.h
index f5ca9d67f1e..f49e1c1d64e 100644
--- a/src/gallium/drivers/r600/r600_shader.h
+++ b/src/gallium/drivers/r600/r600_shader.h
@@ -78,8 +78,9 @@ struct r600_shader {
 	/* Temporarily workaround SB not handling CF_INDEX_[01] index registers */
 	boolean			uses_index_registers;
 
-	/* size in bytes of a data item in the ring (single vertex data) */
-	unsigned		ring_item_size;
+	/* Size in bytes of a data item in the ring(s) (single vertex data).
+	   Stages with only one ring items 123 will be set to 0. */
+	unsigned		ring_item_sizes[4];
 
 	unsigned		indirect_files;
 	unsigned		max_arrays;
diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index aff8f03f9b1..f2d24a3e37a 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -2650,7 +2650,7 @@ void r600_update_gs_state(struct pipe_context *ctx, struct r600_pipe_shader *sha
 	struct r600_shader *rshader = &shader->shader;
 	struct r600_shader *cp_shader = &shader->gs_copy_shader->shader;
 	unsigned gsvs_itemsize =
-			(cp_shader->ring_item_size * shader->selector->gs_max_out_vertices) >> 2;
+			(cp_shader->ring_item_sizes[0] * shader->selector->gs_max_out_vertices) >> 2;
 
 	r600_init_command_buffer(cb, 64);
 
@@ -2665,10 +2665,10 @@ void r600_update_gs_state(struct pipe_context *ctx, struct r600_pipe_shader *sha
 			       r600_conv_prim_to_gs_out(shader->selector->gs_output_prim));
 
 	r600_store_context_reg(cb, R_0288C8_SQ_GS_VERT_ITEMSIZE,
-	                       cp_shader->ring_item_size >> 2);
+	                       cp_shader->ring_item_sizes[0] >> 2);
 
 	r600_store_context_reg(cb, R_0288A8_SQ_ESGS_RING_ITEMSIZE,
-			       (rshader->ring_item_size) >> 2);
+			       (rshader->ring_item_sizes[0]) >> 2);
 
 	r600_store_context_reg(cb, R_0288AC_SQ_GSVS_RING_ITEMSIZE,
 			       gsvs_itemsize);