r600g: evergreen interpolation support.

On evergreen, interpolation has moved into the fragment shader,
with the interpolation parmaters being passed via GPRs and LDS entries.

This works out the number of interps required and reserves GPR/LDS
storage for them, it also correctly routes face/position values which
aren't interpolated from the vertex shader.

Also if we noticed nothing is to be interpolated we always setup perspective
interpolation for one value otherwise the GPU appears to lockup.

This fixes about 15 piglit tests on evergreen.

3 files changed, 149 insertions, 22 deletions

diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
index 542df11db6..935496c04a 100644
--- a/src/gallium/drivers/r600/evergreen_state.c
+++ b/src/gallium/drivers/r600/evergreen_state.c
@@ -1529,23 +1529,39 @@ void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader
 	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
 	struct r600_pipe_state *rstate = &shader->rstate;
 	struct r600_shader *rshader = &shader->shader;
-	unsigned i, tmp, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z;
-	boolean have_pos = FALSE, have_face = FALSE;
+	unsigned i, tmp, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1;
+	int pos_index = -1, face_index = -1;
+	int ninterp = 0;
+	boolean have_linear = FALSE, have_centroid = FALSE, have_perspective = FALSE;
+	unsigned spi_baryc_cntl;
 
 	/* clear previous register */
 	rstate->nregs = 0;
 
 	for (i = 0; i < rshader->ninput; i++) {
 		tmp = S_028644_SEMANTIC(r600_find_vs_semantic_index(&rctx->vs_shader->shader, rshader, i));
+		/* evergreen NUM_INTERP only contains values interpolated into the LDS,
+		   POSITION goes via GPRs from the SC so isn't counted */
 		if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
-			have_pos = TRUE;
+			pos_index = i;
+		else if (rshader->input[i].name == TGSI_SEMANTIC_FACE)
+			face_index = i;
+		else {
+			if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR ||
+			    rshader->input[i].interpolate == TGSI_INTERPOLATE_PERSPECTIVE)
+				ninterp++;
+			if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR)
+				have_linear = TRUE;
+			if (rshader->input[i].interpolate == TGSI_INTERPOLATE_PERSPECTIVE)
+				have_perspective = TRUE;
+			if (rshader->input[i].centroid)
+				have_centroid = TRUE;
+		}
 		if (rshader->input[i].name == TGSI_SEMANTIC_COLOR ||
 		    rshader->input[i].name == TGSI_SEMANTIC_BCOLOR ||
 		    rshader->input[i].name == TGSI_SEMANTIC_POSITION) {
 			tmp |= S_028644_FLAT_SHADE(rshader->flat_shade);
 		}
-		if (rshader->input[i].name == TGSI_SEMANTIC_FACE)
-			have_face = TRUE;
 		if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC &&
 			rctx->sprite_coord_enable & (1 << rshader->input[i].sid)) {
 			tmp |= S_028644_PT_SPRITE_TEX(1);
@@ -1568,7 +1584,8 @@ void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader
 	exports_ps = 0;
 	num_cout = 0;
 	for (i = 0; i < rshader->noutput; i++) {
-		if (rshader->output[i].name == TGSI_SEMANTIC_POSITION ||			    rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
+		if (rshader->output[i].name == TGSI_SEMANTIC_POSITION ||
+		    rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
 			exports_ps |= 1;
 		else if (rshader->output[i].name == TGSI_SEMANTIC_COLOR) {
 			num_cout++;
@@ -1580,19 +1597,49 @@ void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader
 		exports_ps = 2;
 	}
 
-	spi_ps_in_control_0 = S_0286CC_NUM_INTERP(rshader->ninput) |
-				S_0286CC_PERSP_GRADIENT_ENA(1);
+	if (ninterp == 0) {
+		ninterp = 1;
+		have_perspective = TRUE;
+	}
+
+	spi_ps_in_control_0 = S_0286CC_NUM_INTERP(ninterp) |
+		              S_0286CC_PERSP_GRADIENT_ENA(have_perspective) |
+		              S_0286CC_LINEAR_GRADIENT_ENA(have_linear);
 	spi_input_z = 0;
-	if (have_pos) {
-		spi_ps_in_control_0 |=  S_0286CC_POSITION_ENA(1);
+	if (pos_index != -1) {
+		spi_ps_in_control_0 |=  S_0286CC_POSITION_ENA(1) |
+			S_0286CC_POSITION_CENTROID(rshader->input[pos_index].centroid) |
+			S_0286CC_POSITION_ADDR(rshader->input[pos_index].gpr);
 		spi_input_z |= 1;
 	}
+
+	spi_ps_in_control_1 = 0;
+	if (face_index != -1) {
+		spi_ps_in_control_1 |= S_0286D0_FRONT_FACE_ENA(1) |
+			S_0286D0_FRONT_FACE_ADDR(rshader->input[face_index].gpr);
+	}
+
+	spi_baryc_cntl = 0;
+	if (have_perspective)
+		spi_baryc_cntl |= S_0286E0_PERSP_CENTER_ENA(1) |
+				  S_0286E0_PERSP_CENTROID_ENA(have_centroid);
+	if (have_linear)
+		spi_baryc_cntl |= S_0286E0_LINEAR_CENTER_ENA(1) |
+			          S_0286E0_LINEAR_CENTROID_ENA(have_centroid);
+				
 	r600_pipe_state_add_reg(rstate, R_0286CC_SPI_PS_IN_CONTROL_0,
 				spi_ps_in_control_0, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_0286D0_SPI_PS_IN_CONTROL_1,
-				S_0286D0_FRONT_FACE_ENA(have_face), 0xFFFFFFFF, NULL);
+				spi_ps_in_control_1, 0xFFFFFFFF, NULL);
+	r600_pipe_state_add_reg(rstate, R_0286E4_SPI_PS_IN_CONTROL_2,
+				0, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_0286D8_SPI_INPUT_Z, spi_input_z, 0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate,
+				R_0286E0_SPI_BARYC_CNTL,
+				spi_baryc_cntl,
+				0xFFFFFFFF, NULL);
+
+	r600_pipe_state_add_reg(rstate,
 				R_028840_SQ_PGM_START_PS,
 				(r600_bo_offset(shader->bo)) >> 8, 0xFFFFFFFF, shader->bo);
 	r600_pipe_state_add_reg(rstate,
@@ -1607,11 +1654,6 @@ void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader
 	r600_pipe_state_add_reg(rstate,
 				R_02884C_SQ_PGM_EXPORTS_PS,
 				exports_ps, 0xFFFFFFFF, NULL);
-	r600_pipe_state_add_reg(rstate,
-				R_0286E0_SPI_BARYC_CNTL,
-				S_0286E0_PERSP_CENTROID_ENA(1) |
-				S_0286E0_LINEAR_CENTROID_ENA(1),
-				0xFFFFFFFF, NULL);
 
 	if (rshader->uses_kill) {
 		/* only set some bits here, the other bits are set in the dsa state */
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index b53d478071..94c9cbd923 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -357,6 +357,11 @@ struct r600_shader_ctx {
 	u32					*literals;
 	u32					nliterals;
 	u32					max_driver_temp_used;
+	/* needed for evergreen interpolation */
+	boolean                                 input_centroid;
+	boolean                                 input_linear;
+	boolean                                 input_perspective;
+	int					num_interp_gpr;
 };
 
 struct r600_shader_tgsi_instruction {
@@ -404,10 +409,33 @@ static int tgsi_is_supported(struct r600_shader_ctx *ctx)
 	return 0;
 }
 
-static int evergreen_interp_alu(struct r600_shader_ctx *ctx, int gpr)
+static int evergreen_interp_alu(struct r600_shader_ctx *ctx, int input)
 {
 	int i, r;
 	struct r600_bc_alu alu;
+	int gpr = 0, base_chan = 0;
+	int ij_index = 0;
+
+	if (ctx->shader->input[input].interpolate == TGSI_INTERPOLATE_PERSPECTIVE) {
+		ij_index = 0;
+		if (ctx->shader->input[input].centroid)
+			ij_index++;
+	} else if (ctx->shader->input[input].interpolate == TGSI_INTERPOLATE_LINEAR) {
+		ij_index = 0;
+		/* if we have perspective add one */
+		if (ctx->input_perspective)  {
+			ij_index++;
+			/* if we have perspective centroid */
+			if (ctx->input_centroid)
+				ij_index++;
+		}
+		if (ctx->shader->input[input].centroid)
+			ij_index++;
+	}
+		
+	/* work out gpr and base_chan from index */
+	gpr = ij_index / 2;
+	base_chan = (2 * (ij_index % 2)) + 1;
 
 	for (i = 0; i < 8; i++) {
 		memset(&alu, 0, sizeof(struct r600_bc_alu));
@@ -418,13 +446,16 @@ static int evergreen_interp_alu(struct r600_shader_ctx *ctx, int gpr)
 			alu.inst = EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INTERP_XY;
 
 		if ((i > 1) && (i < 6)) {
-			alu.dst.sel = ctx->shader->input[gpr].gpr;
+			alu.dst.sel = ctx->shader->input[input].gpr;
 			alu.dst.write = 1;
 		}
 
 		alu.dst.chan = i % 4;
-		alu.src[0].chan = (1 - (i % 2));
-		alu.src[1].sel = V_SQ_ALU_SRC_PARAM_BASE + gpr;
+
+		alu.src[0].sel = gpr;
+		alu.src[0].chan = (base_chan - (i % 2));
+
+		alu.src[1].sel = V_SQ_ALU_SRC_PARAM_BASE + ctx->shader->input[input].lds_pos;
 
 		alu.bank_swizzle_force = SQ_ALU_VEC_210;
 		if ((i % 4) == 3)
@@ -474,7 +505,12 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx)
 		}
 		if (ctx->type == TGSI_PROCESSOR_FRAGMENT && ctx->bc->chiprev == 2) {
 			/* turn input into interpolate on EG */
-			evergreen_interp_alu(ctx, i);
+			if (ctx->shader->input[i].name != TGSI_SEMANTIC_POSITION) {
+				if (ctx->shader->input[i].interpolate > 0) {
+					ctx->shader->input[i].lds_pos = ctx->shader->nlds++;
+					evergreen_interp_alu(ctx, i);
+				}
+			}
 		}
 		break;
 	case TGSI_FILE_OUTPUT:
@@ -501,6 +537,53 @@ static int r600_get_temp(struct r600_shader_ctx *ctx)
 	return ctx->temp_reg + ctx->max_driver_temp_used++;
 }
 
+/* 
+ * for evergreen we need to scan the shader to find the number of GPRs we need to
+ * reserve for interpolation.
+ *
+ * we need to know if we are going to emit
+ * any centroid inputs
+ * if perspective and linear are required
+*/
+static int evergreen_gpr_count(struct r600_shader_ctx *ctx)
+{
+	int i;
+	int num_baryc;
+
+	ctx->input_linear = FALSE;
+	ctx->input_perspective = FALSE;
+	ctx->input_centroid = FALSE;
+	ctx->num_interp_gpr = 1;
+
+	/* any centroid inputs */
+	for (i = 0; i < ctx->info.num_inputs; i++) {
+		/* skip position/face */
+		if (ctx->info.input_semantic_name[i] == TGSI_SEMANTIC_POSITION ||
+		    ctx->info.input_semantic_name[i] == TGSI_SEMANTIC_FACE)
+			continue;
+		if (ctx->info.input_interpolate[i] == TGSI_INTERPOLATE_LINEAR)
+			ctx->input_linear = TRUE;
+		if (ctx->info.input_interpolate[i] == TGSI_INTERPOLATE_PERSPECTIVE)
+			ctx->input_perspective = TRUE;
+		if (ctx->info.input_centroid[i])
+			ctx->input_centroid = TRUE;
+	}
+
+	num_baryc = 0;
+	/* ignoring sample for now */
+	if (ctx->input_perspective)
+		num_baryc++;
+	if (ctx->input_linear)
+		num_baryc++;
+	if (ctx->input_centroid)
+		num_baryc *= 2;
+
+	ctx->num_interp_gpr += (num_baryc + 1) >> 1;
+
+	/* TODO PULL MODEL and LINE STIPPLE, FIXED PT POS */
+	return ctx->num_interp_gpr;
+}
+
 int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader)
 {
 	struct tgsi_full_immediate *immediate;
@@ -547,7 +630,7 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
 		ctx.file_offset[TGSI_FILE_INPUT] = 1;
 	}
 	if (ctx.type == TGSI_PROCESSOR_FRAGMENT && ctx.bc->chiprev == 2) {
-		ctx.file_offset[TGSI_FILE_INPUT] = 1;
+		ctx.file_offset[TGSI_FILE_INPUT] = evergreen_gpr_count(&ctx);
 	}
 	ctx.file_offset[TGSI_FILE_OUTPUT] = ctx.file_offset[TGSI_FILE_INPUT] +
 						ctx.info.file_count[TGSI_FILE_INPUT];
diff --git a/src/gallium/drivers/r600/r600_shader.h b/src/gallium/drivers/r600/r600_shader.h
index a341cca083..f8bc595139 100644
--- a/src/gallium/drivers/r600/r600_shader.h
+++ b/src/gallium/drivers/r600/r600_shader.h
@@ -32,6 +32,7 @@ struct r600_shader_io {
 	int			sid;
 	unsigned		interpolate;
 	boolean                 centroid;
+	unsigned		lds_pos; /* for evergreen */
 };
 
 struct r600_shader {
@@ -40,6 +41,7 @@ struct r600_shader {
 	boolean			flat_shade;
 	unsigned		ninput;
 	unsigned		noutput;
+	unsigned		nlds;
 	struct r600_shader_io	input[32];
 	struct r600_shader_io	output[32];
 	enum radeon_family	family;