1 files changed, 87 insertions, 24 deletions

diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index dc8d4cb315..956c7e7930 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -155,11 +155,14 @@ static int r600_pipe_shader_vs(struct pipe_context *ctx, struct r600_context_sta
 
 static int r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_context_state *rpshader)
 {
+	const struct pipe_rasterizer_state *rasterizer;
 	struct r600_screen *rscreen = r600_screen(ctx->screen);
 	struct r600_shader *rshader = &rpshader->shader;
+	struct r600_context *rctx = r600_context(ctx);
 	struct radeon_state *state;
 	unsigned i, tmp, exports_ps, num_cout;
 
+	rasterizer = &rctx->rasterizer->state.rasterizer;
 	rpshader->rstate = radeon_state_decref(rpshader->rstate);
 	state = radeon_state(rscreen->rw, R600_PS_SHADER_TYPE, R600_PS_SHADER);
 	if (state == NULL)
@@ -171,6 +174,9 @@ static int r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_context_sta
 			rshader->input[i].name == TGSI_SEMANTIC_BCOLOR) {
 			tmp |= S_028644_FLAT_SHADE(rshader->flat_shade);
 		}
+		if (rasterizer->sprite_coord_enable & (1 << i)) {
+			tmp |= S_028644_PT_SPRITE_TEX(1);
+		}
 		state->states[R600_PS_SHADER__SPI_PS_INPUT_CNTL_0 + i] = tmp;
 	}
 
@@ -339,7 +345,8 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
 {
 	struct tgsi_full_immediate *immediate;
 	struct r600_shader_ctx ctx;
-	struct r600_bc_output output;
+	struct r600_bc_output output[32];
+	unsigned output_done, noutput;
 	unsigned opcode;
 	int i, r = 0, pos0;
 
@@ -417,34 +424,41 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
 		}
 	}
 	/* export output */
-	for (i = 0, pos0 = 0; i < shader->noutput; i++) {
-		memset(&output, 0, sizeof(struct r600_bc_output));
-		output.gpr = shader->output[i].gpr;
-		output.elem_size = 3;
-		output.swizzle_x = 0;
-		output.swizzle_y = 1;
-		output.swizzle_z = 2;
-		output.swizzle_w = 3;
-		output.barrier = 1;
-		output.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM;
-		output.array_base = i - pos0;
-		output.inst = V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE;
-		switch (ctx.type == TGSI_PROCESSOR_VERTEX) {
+	noutput = shader->noutput;
+	for (i = 0, pos0 = 0; i < noutput; i++) {
+		memset(&output[i], 0, sizeof(struct r600_bc_output));
+		output[i].gpr = shader->output[i].gpr;
+		output[i].elem_size = 3;
+		output[i].swizzle_x = 0;
+		output[i].swizzle_y = 1;
+		output[i].swizzle_z = 2;
+		output[i].swizzle_w = 3;
+		output[i].barrier = 1;
+		output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM;
+		output[i].array_base = i - pos0;
+		output[i].inst = V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT;
+		switch (ctx.type) {
 		case TGSI_PROCESSOR_VERTEX:
 			if (shader->output[i].name == TGSI_SEMANTIC_POSITION) {
-				output.array_base = 60;
-				output.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS;
+				output[i].array_base = 60;
+				output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS;
 				/* position doesn't count in array_base */
-				pos0 = 1;
+				pos0++;
+			}
+			if (shader->output[i].name == TGSI_SEMANTIC_PSIZE) {
+				output[i].array_base = 61;
+				output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS;
+				/* position doesn't count in array_base */
+				pos0++;
 			}
 			break;
 		case TGSI_PROCESSOR_FRAGMENT:
 			if (shader->output[i].name == TGSI_SEMANTIC_COLOR) {
-				output.array_base = 0;
-				output.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
+				output[i].array_base = shader->output[i].sid;
+				output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
 			} else if (shader->output[i].name == TGSI_SEMANTIC_POSITION) {
-				output.array_base = 61;
-				output.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
+				output[i].array_base = 61;
+				output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
 			} else {
 				R600_ERR("unsupported fragment output name %d\n", shader->output[i].name);
 				r = -EINVAL;
@@ -456,10 +470,58 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
 			r = -EINVAL;
 			goto out_err;
 		}
-		if (i == (shader->noutput - 1)) {
-			output.end_of_program = 1;
+	}
+	/* add fake param output for vertex shader if no param is exported */
+	if (ctx.type == TGSI_PROCESSOR_VERTEX) {
+		for (i = 0, pos0 = 0; i < noutput; i++) {
+			if (output[i].type == V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM) {
+				pos0 = 1;
+				break;
+			}
+		}
+		if (!pos0) {
+			memset(&output[i], 0, sizeof(struct r600_bc_output));
+			output[i].gpr = 0;
+			output[i].elem_size = 3;
+			output[i].swizzle_x = 0;
+			output[i].swizzle_y = 1;
+			output[i].swizzle_z = 2;
+			output[i].swizzle_w = 3;
+			output[i].barrier = 1;
+			output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM;
+			output[i].array_base = 0;
+			output[i].inst = V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT;
+			noutput++;
 		}
-		r = r600_bc_add_output(ctx.bc, &output);
+	}
+	/* add fake pixel export */
+	if (ctx.type == TGSI_PROCESSOR_FRAGMENT && !noutput) {
+		memset(&output[0], 0, sizeof(struct r600_bc_output));
+		output[0].gpr = 0;
+		output[0].elem_size = 3;
+		output[0].swizzle_x = 7;
+		output[0].swizzle_y = 7;
+		output[0].swizzle_z = 7;
+		output[0].swizzle_w = 7;
+		output[0].barrier = 1;
+		output[0].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM;
+		output[0].array_base = 0;
+		output[0].inst = V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT;
+		noutput++;
+	}
+	/* set export done on last export of each type */
+	for (i = noutput - 1, output_done = 0; i >= 0; i--) {
+		if (i == (noutput - 1)) {
+			output[i].end_of_program = 1;
+		}
+		if (!(output_done & (1 << output[i].type))) {
+			output_done |= (1 << output[i].type);
+			output[i].inst = V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE;
+		}
+	}
+	/* add output to bytecode */
+	for (i = 0; i < noutput; i++) {
+		r = r600_bc_add_output(ctx.bc, &output[i]);
 		if (r)
 			goto out_err;
 	}
@@ -490,6 +552,7 @@ static int tgsi_src(struct r600_shader_ctx *ctx,
 	if (tgsi_src->Register.File == TGSI_FILE_IMMEDIATE) {
 		r600_src->sel = 0;
 	}
+	r600_src->neg = tgsi_src->Register.Negate;
 	r600_src->sel += ctx->file_offset[tgsi_src->Register.File];
 	return 0;
 }