radeonsi: implement a fixed-function tessellation control shader and its state

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

src/gallium/drivers/radeonsi/si_pipe.c

@@ -59,6 +59,8 @@ static void si_destroy_context(struct pipe_context *context)
 	if (sctx->dummy_pixel_shader) {
 		sctx->b.b.delete_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
 	}
+	if (sctx->fixed_func_tcs_shader)
+		sctx->b.b.delete_tcs_state(&sctx->b.b, sctx->fixed_func_tcs_shader);
 	sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush);
 	sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_resolve);
 	sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_decompress);

src/gallium/drivers/radeonsi/si_pipe.h

@@ -136,6 +136,7 @@ struct si_context {
 	struct si_screen		*screen;
 	struct si_pm4_state		*init_config;
 	struct pipe_fence_handle	*last_gfx_fence;
+	struct si_shader_selector	*fixed_func_tcs_shader;
 
 	union {
 		struct {

src/gallium/drivers/radeonsi/si_state.c

@@ -2895,6 +2895,30 @@ static void si_set_polygon_stipple(struct pipe_context *ctx,
 	}
 }
 
+static void si_set_tess_state(struct pipe_context *ctx,
+			      const float default_outer_level[4],
+			      const float default_inner_level[2])
+{
+	struct si_context *sctx = (struct si_context *)ctx;
+	struct pipe_constant_buffer cb;
+	float array[8];
+
+	memcpy(array, default_outer_level, sizeof(float) * 4);
+	memcpy(array+4, default_inner_level, sizeof(float) * 2);
+
+	cb.buffer = NULL;
+	cb.user_buffer = NULL;
+	cb.buffer_size = sizeof(array);
+
+	si_upload_const_buffer(sctx, (struct r600_resource**)&cb.buffer,
+			       (void*)array, sizeof(array),
+			       &cb.buffer_offset);
+
+	ctx->set_constant_buffer(ctx, PIPE_SHADER_TESS_CTRL,
+				 SI_DRIVER_STATE_CONST_BUF, &cb);
+	pipe_resource_reference(&cb.buffer, NULL);
+}
+
 static void si_texture_barrier(struct pipe_context *ctx)
 {
 	struct si_context *sctx = (struct si_context *)ctx;
@@ -2972,6 +2996,7 @@ void si_init_state_functions(struct si_context *sctx)
 	sctx->b.b.texture_barrier = si_texture_barrier;
 	sctx->b.b.set_polygon_stipple = si_set_polygon_stipple;
 	sctx->b.b.set_min_samples = si_set_min_samples;
+	sctx->b.b.set_tess_state = si_set_tess_state;
 
 	sctx->b.set_occlusion_query_state = si_set_occlusion_query_state;
 	sctx->b.need_gfx_cs_space = si_need_gfx_cs_space;

src/gallium/drivers/radeonsi/si_state_shaders.c

@@ -30,6 +30,7 @@
 #include "sid.h"
 
 #include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_ureg.h"
 #include "util/u_memory.h"
 #include "util/u_simple_shaders.h"
 
@@ -1149,6 +1150,40 @@ static void si_init_tess_factor_ring(struct si_context *sctx)
 	sctx->b.flags |= SI_CONTEXT_VGT_FLUSH;
 }
 
+/**
+ * This is used when TCS is NULL in the VS->TCS->TES chain. In this case,
+ * VS passes its outputs to TES directly, so the fixed-function shader only
+ * has to write TESSOUTER and TESSINNER.
+ */
+static void si_generate_fixed_func_tcs(struct si_context *sctx)
+{
+	struct ureg_src const0, const1;
+	struct ureg_dst tessouter, tessinner;
+	struct ureg_program *ureg = ureg_create(TGSI_PROCESSOR_TESS_CTRL);
+
+	if (!ureg)
+		return; /* if we get here, we're screwed */
+
+	assert(!sctx->fixed_func_tcs_shader);
+
+	ureg_DECL_constant2D(ureg, 0, 1, SI_DRIVER_STATE_CONST_BUF);
+	const0 = ureg_src_dimension(ureg_src_register(TGSI_FILE_CONSTANT, 0),
+				    SI_DRIVER_STATE_CONST_BUF);
+	const1 = ureg_src_dimension(ureg_src_register(TGSI_FILE_CONSTANT, 1),
+				    SI_DRIVER_STATE_CONST_BUF);
+
+	tessouter = ureg_DECL_output(ureg, TGSI_SEMANTIC_TESSOUTER, 0);
+	tessinner = ureg_DECL_output(ureg, TGSI_SEMANTIC_TESSINNER, 0);
+
+	ureg_MOV(ureg, tessouter, const0);
+	ureg_MOV(ureg, tessinner, const1);
+	ureg_END(ureg);
+
+	sctx->fixed_func_tcs_shader =
+		ureg_create_shader_and_destroy(ureg, &sctx->b.b);
+	assert(sctx->fixed_func_tcs_shader);
+}
+
 static void si_update_vgt_shader_config(struct si_context *sctx)
 {
 	/* Calculate the index of the config.
@@ -1202,7 +1237,11 @@ void si_update_shaders(struct si_context *sctx)
 			si_shader_select(ctx, sctx->tcs_shader);
 			si_pm4_bind_state(sctx, hs, sctx->tcs_shader->current->pm4);
 		} else {
-			assert(!"generate TCS shader");
+			if (!sctx->fixed_func_tcs_shader)
+				si_generate_fixed_func_tcs(sctx);
+			si_shader_select(ctx, sctx->fixed_func_tcs_shader);
+			si_pm4_bind_state(sctx, hs,
+					  sctx->fixed_func_tcs_shader->current->pm4);
 		}
 
 		si_shader_select(ctx, sctx->tes_shader);