radeonsi: implement polygon stippling

Reviewed-by: Michel Dänzer <[email protected]>

7 files changed, 79 insertions, 5 deletions

diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index eff8d690959..aac3363b106 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -53,6 +53,8 @@ static void si_destroy_context(struct pipe_context *context)
 	si_pm4_delete_state(sctx, gs_onoff, sctx->gs_on);
 	si_pm4_delete_state(sctx, gs_onoff, sctx->gs_off);
 
+	if (sctx->pstipple_sampler_state)
+		sctx->b.b.delete_sampler_state(&sctx->b.b, sctx->pstipple_sampler_state);
 	if (sctx->dummy_pixel_shader) {
 		sctx->b.b.delete_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
 	}
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 02820a1dd2d..059fe0dc99b 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -133,6 +133,7 @@ struct si_context {
 	void				*custom_blend_resolve;
 	void				*custom_blend_decompress;
 	void				*custom_blend_fastclear;
+	void				*pstipple_sampler_state;
 	struct si_screen		*screen;
 	struct si_pm4_state		*init_config;
 
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 2c1dac928a2..e6849ad22c7 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -37,6 +37,7 @@
 #include "radeon/radeon_elf_util.h"
 #include "radeon/radeon_llvm_emit.h"
 #include "util/u_memory.h"
+#include "util/u_pstipple.h"
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_util.h"
 #include "tgsi/tgsi_dump.h"
@@ -2742,16 +2743,26 @@ static int si_generate_gs_copy_shader(struct si_screen *sscreen,
 int si_shader_create(struct si_screen *sscreen, struct si_shader *shader)
 {
 	struct si_shader_selector *sel = shader->selector;
+	struct tgsi_token *tokens = sel->tokens;
 	struct si_shader_context si_shader_ctx;
 	struct lp_build_tgsi_context * bld_base;
+	struct tgsi_shader_info stipple_shader_info;
 	LLVMModuleRef mod;
 	int r = 0;
+	bool poly_stipple = sel->type == PIPE_SHADER_FRAGMENT &&
+			    shader->key.ps.poly_stipple;
 	bool dump = r600_can_dump_shader(&sscreen->b, sel->tokens);
 
+	if (poly_stipple) {
+		tokens = util_pstipple_create_fragment_shader(tokens, NULL,
+						SI_POLY_STIPPLE_SAMPLER);
+		tgsi_scan_shader(tokens, &stipple_shader_info);
+	}
+
 	/* Dump TGSI code before doing TGSI->LLVM conversion in case the
 	 * conversion fails. */
 	if (dump) {
-		tgsi_dump(sel->tokens, 0);
+		tgsi_dump(tokens, 0);
 		si_dump_streamout(&sel->so);
 	}
 
@@ -2768,7 +2779,7 @@ int si_shader_create(struct si_screen *sscreen, struct si_shader *shader)
 		shader->db_shader_control |= S_02880C_KILL_ENABLE(1);
 
 	shader->uses_instanceid = sel->info.uses_instanceid;
-	bld_base->info = &sel->info;
+	bld_base->info = poly_stipple ? &stipple_shader_info : &sel->info;
 	bld_base->emit_fetch_funcs[TGSI_FILE_CONSTANT] = fetch_constant;
 
 	bld_base->op_actions[TGSI_OPCODE_TEX] = tex_action;
@@ -2799,7 +2810,7 @@ int si_shader_create(struct si_screen *sscreen, struct si_shader *shader)
 
 	si_shader_ctx.radeon_bld.load_system_value = declare_system_value;
 	si_shader_ctx.shader = shader;
-	si_shader_ctx.type = tgsi_get_processor_type(sel->tokens);
+	si_shader_ctx.type = tgsi_get_processor_type(tokens);
 	si_shader_ctx.screen = sscreen;
 
 	switch (si_shader_ctx.type) {
@@ -2848,7 +2859,7 @@ int si_shader_create(struct si_screen *sscreen, struct si_shader *shader)
 					bld_base->uint_bld.elem_type, "");
 	}
 
-	if (!lp_build_tgsi_llvm(bld_base, sel->tokens)) {
+	if (!lp_build_tgsi_llvm(bld_base, tokens)) {
 		fprintf(stderr, "Failed to translate shader from TGSI to LLVM\n");
 		goto out;
 	}
@@ -2880,7 +2891,8 @@ int si_shader_create(struct si_screen *sscreen, struct si_shader *shader)
 out:
 	for (int i = 0; i < SI_NUM_CONST_BUFFERS; i++)
 		FREE(si_shader_ctx.constants[i]);
-
+	if (poly_stipple)
+		tgsi_free_tokens(tokens);
 	return r;
 }
 
diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 1d7efc23f59..551c7dc9f23 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -125,6 +125,7 @@ union si_shader_key {
 		unsigned	color_two_side:1;
 		unsigned	alpha_func:3;
 		unsigned	alpha_to_one:1;
+		unsigned	poly_stipple:1;
 	} ps;
 	struct {
 		unsigned	instance_divisors[SI_NUM_VERTEX_BUFFERS];
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 2deee4516d9..fb353ad419d 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -32,6 +32,7 @@
 #include "util/u_format.h"
 #include "util/u_format_s3tc.h"
 #include "util/u_memory.h"
+#include "util/u_pstipple.h"
 
 static void si_init_atom(struct r600_atom *atom, struct r600_atom **list_elem,
 			 void (*emit)(struct si_context *ctx, struct r600_atom *state),
@@ -616,6 +617,7 @@ static void *si_create_rs_state(struct pipe_context *ctx,
 	rs->multisample_enable = state->multisample;
 	rs->clip_plane_enable = state->clip_plane_enable;
 	rs->line_stipple_enable = state->line_stipple_enable;
+	rs->poly_stipple_enable = state->poly_stipple_enable;
 
 	polygon_dual_mode = (state->fill_front != PIPE_POLYGON_MODE_FILL ||
 				state->fill_back != PIPE_POLYGON_MODE_FILL);
@@ -2760,6 +2762,56 @@ static void si_set_index_buffer(struct pipe_context *ctx,
 static void si_set_polygon_stipple(struct pipe_context *ctx,
 				   const struct pipe_poly_stipple *state)
 {
+	struct si_context *sctx = (struct si_context *)ctx;
+	struct pipe_resource *tex;
+	struct pipe_sampler_view *view;
+	bool is_zero = true;
+	bool is_one = true;
+	int i;
+
+	/* The hardware obeys 0 and 1 swizzles in the descriptor even if
+	 * the resource is NULL/invalid. Take advantage of this fact and skip
+	 * texture allocation if the stipple pattern is constant.
+	 *
+	 * This is an optimization for the common case when stippling isn't
+	 * used but set_polygon_stipple is still called by st/mesa.
+	 */
+	for (i = 0; i < Elements(state->stipple); i++) {
+		is_zero = is_zero && state->stipple[i] == 0;
+		is_one = is_one && state->stipple[i] == 0xffffffff;
+	}
+
+	if (is_zero || is_one) {
+		struct pipe_sampler_view templ = {{0}};
+
+		templ.swizzle_r = PIPE_SWIZZLE_ZERO;
+		templ.swizzle_g = PIPE_SWIZZLE_ZERO;
+		templ.swizzle_b = PIPE_SWIZZLE_ZERO;
+		/* The pattern should be inverted in the texture. */
+		templ.swizzle_a = is_zero ? PIPE_SWIZZLE_ONE : PIPE_SWIZZLE_ZERO;
+
+		view = ctx->create_sampler_view(ctx, NULL, &templ);
+	} else {
+		/* Create a new texture. */
+		tex = util_pstipple_create_stipple_texture(ctx, state->stipple);
+		if (!tex)
+			return;
+
+		view = util_pstipple_create_sampler_view(ctx, tex);
+		pipe_resource_reference(&tex, NULL);
+	}
+
+	ctx->set_sampler_views(ctx, PIPE_SHADER_FRAGMENT,
+			       SI_POLY_STIPPLE_SAMPLER, 1, &view);
+	pipe_sampler_view_reference(&view, NULL);
+
+	/* Bind the sampler state if needed. */
+	if (!sctx->pstipple_sampler_state) {
+		sctx->pstipple_sampler_state = util_pstipple_create_sampler(ctx);
+		ctx->bind_sampler_states(ctx, PIPE_SHADER_FRAGMENT,
+					 SI_POLY_STIPPLE_SAMPLER, 1,
+					 &sctx->pstipple_sampler_state);
+	}
 }
 
 static void si_texture_barrier(struct pipe_context *ctx)
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 263728963d4..d1ed53065ca 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -67,6 +67,7 @@ struct si_state_rasterizer {
 	unsigned		clip_plane_enable;
 	float			offset_units;
 	float			offset_scale;
+	bool			poly_stipple_enable;
 };
 
 struct si_state_dsa {
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index a40926800dd..27ccc8e5034 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -376,6 +376,11 @@ static INLINE void si_shader_selector_key(struct pipe_context *ctx,
 						       sctx->queued.named.rasterizer->multisample_enable &&
 						       !sctx->framebuffer.cb0_is_integer;
 			}
+
+			key->ps.poly_stipple = sctx->queued.named.rasterizer->poly_stipple_enable &&
+					       ((sctx->current_rast_prim >= PIPE_PRIM_TRIANGLES &&
+						 sctx->current_rast_prim <= PIPE_PRIM_POLYGON) ||
+						sctx->current_rast_prim >= PIPE_PRIM_TRIANGLES_ADJACENCY);
 		}
 		if (sctx->queued.named.dsa) {
 			key->ps.alpha_func = sctx->queued.named.dsa->alpha_func;