Mesa (main): radeonsi: add si_create_passthrough_tcs

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: 3ab9c42b43d907696d2e18c5020a5abcaae47096
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=3ab9c42b43d907696d2e18c5020a5abcaae47096

Author: Qiang Yu <yuq825 at gmail.com>
Date:   Wed May 18 17:12:58 2022 +0800

radeonsi: add si_create_passthrough_tcs

For replacing si_create_fixed_func_tcs.

Reviewed-by: Marek Olšák <marek.olsak at amd.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer at amd.com>
Signed-off-by: Qiang Yu <yuq825 at gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16705>

---

 src/gallium/drivers/radeonsi/si_pipe.h          |  1 +
 src/gallium/drivers/radeonsi/si_shaderlib_nir.c | 79 +++++++++++++++++++++++++
 2 files changed, 80 insertions(+)

diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 2b18772f79b..80ec9ae4a79 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -1552,6 +1552,7 @@ void si_resume_queries(struct si_context *sctx);
 void *si_create_copy_image_cs(struct si_context *sctx, bool src_is_1d_array, bool dst_is_1d_array);
 void *si_create_dcc_retile_cs(struct si_context *sctx, struct radeon_surf *surf);
 void *gfx9_create_clear_dcc_msaa_cs(struct si_context *sctx, struct si_texture *tex);
+void *si_create_passthrough_tcs(struct si_context *sctx);
 
 /* si_shaderlib_tgsi.c */
 void *si_get_blitter_vs(struct si_context *sctx, enum blitter_attrib_type type,
diff --git a/src/gallium/drivers/radeonsi/si_shaderlib_nir.c b/src/gallium/drivers/radeonsi/si_shaderlib_nir.c
index a62b246f6b2..f92218b7c8b 100644
--- a/src/gallium/drivers/radeonsi/si_shaderlib_nir.c
+++ b/src/gallium/drivers/radeonsi/si_shaderlib_nir.c
@@ -268,3 +268,82 @@ void *si_create_clear_buffer_rmw_cs(struct si_context *sctx)
    return create_shader_state(sctx, b.shader);
 }
 
+/* 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.
+ */
+void *si_create_passthrough_tcs(struct si_context *sctx)
+{
+   const nir_shader_compiler_options *options =
+      sctx->b.screen->get_compiler_options(sctx->b.screen, PIPE_SHADER_IR_NIR,
+                                           PIPE_SHADER_TESS_CTRL);
+
+   nir_builder b = nir_builder_init_simple_shader(MESA_SHADER_TESS_CTRL, options,
+                                                  "tcs passthrough");
+
+   unsigned num_inputs = 0;
+   unsigned num_outputs = 0;
+
+   nir_variable *in_inner =
+      nir_variable_create(b.shader, nir_var_system_value, glsl_vec_type(2),
+                          "tess inner default");
+   in_inner->data.location = SYSTEM_VALUE_TESS_LEVEL_INNER_DEFAULT;
+
+   nir_variable *out_inner =
+      nir_variable_create(b.shader, nir_var_shader_out, glsl_vec_type(2),
+                          "tess inner");
+   out_inner->data.location = VARYING_SLOT_TESS_LEVEL_INNER;
+   out_inner->data.driver_location = num_outputs++;
+
+   nir_ssa_def *inner = nir_load_var(&b, in_inner);
+   nir_store_var(&b, out_inner, inner, 0x3);
+
+   nir_variable *in_outer =
+      nir_variable_create(b.shader, nir_var_system_value, glsl_vec4_type(),
+                          "tess outer default");
+   in_outer->data.location = SYSTEM_VALUE_TESS_LEVEL_OUTER_DEFAULT;
+
+   nir_variable *out_outer =
+      nir_variable_create(b.shader, nir_var_shader_out, glsl_vec4_type(),
+                          "tess outer");
+   out_outer->data.location = VARYING_SLOT_TESS_LEVEL_OUTER;
+   out_outer->data.driver_location = num_outputs++;
+
+   nir_ssa_def *outer = nir_load_var(&b, in_outer);
+   nir_store_var(&b, out_outer, outer, 0xf);
+
+   nir_ssa_def *id = nir_load_invocation_id(&b);
+   struct si_shader_info *info = &sctx->shader.vs.cso->info;
+   for (unsigned i = 0; i < info->num_outputs; i++) {
+      const struct glsl_type *type;
+      unsigned semantic = info->output_semantic[i];
+      if (semantic < VARYING_SLOT_VAR31 && semantic != VARYING_SLOT_EDGE)
+         type = glsl_array_type(glsl_vec4_type(), 0, 0);
+      else if (semantic >= VARYING_SLOT_VAR0_16BIT)
+         type = glsl_array_type(glsl_vector_type(GLSL_TYPE_FLOAT16, 4), 0, 0);
+      else
+         continue;
+
+      char name[10];
+      snprintf(name, sizeof(name), "in_%u", i);
+      nir_variable *in = nir_variable_create(b.shader, nir_var_shader_in, type, name);
+      in->data.location = semantic;
+      in->data.driver_location = num_inputs++;
+
+      snprintf(name, sizeof(name), "out_%u", i);
+      nir_variable *out = nir_variable_create(b.shader, nir_var_shader_out, type, name);
+      out->data.location = semantic;
+      out->data.driver_location = num_outputs++;
+
+      /* no need to use copy_var to save a lower pass */
+      nir_ssa_def *value = nir_load_array_var(&b, in, id);
+      nir_store_array_var(&b, out, id, value, 0xf);
+   }
+
+   b.shader->num_inputs = num_inputs;
+   b.shader->num_outputs = num_outputs;
+
+   b.shader->info.tess.tcs_vertices_out = sctx->patch_vertices;
+
+   return create_shader_state(sctx, b.shader);
+}