From e61046fb232c2396c28a61937d67613d2fad4733 Mon Sep 17 00:00:00 2001
From: "Anna (navi) Figueiredo Gomes" <navi@vlhl.dev>
Date: Fri, 6 Sep 2024 22:03:05 +0200
Subject: [PATCH] vkd3d-shader/hlsl: Implement the "earlydepthstencil"
 attribute.

---
 libs/vkd3d-shader/hlsl.h                   |  2 ++
 libs/vkd3d-shader/hlsl_codegen.c           |  4 +++-
 libs/vkd3d-shader/tpf.c                    | 14 ++++++++++++++
 tests/hlsl/depth-bounds.shader_test        |  4 ++--
 tests/hlsl/early-depth-stencil.shader_test |  2 +-
 5 files changed, 22 insertions(+), 4 deletions(-)

diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h
index d40181a6..66c43826 100644
--- a/libs/vkd3d-shader/hlsl.h
+++ b/libs/vkd3d-shader/hlsl.h
@@ -603,6 +603,8 @@ struct hlsl_ir_function_decl
     unsigned int attr_count;
     const struct hlsl_attribute *const *attrs;
 
+    bool early_depth_test;
+
     /* Synthetic boolean variable marking whether a return statement has been
      * executed. Needed to deal with return statements in non-uniform control
      * flow, since some backends can't handle them. */
diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index 397eb2d7..93f19360 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -6027,7 +6027,7 @@ static void parse_patchconstantfunc_attribute(struct hlsl_ctx *ctx, const struct
                 "Patch constant function \"%s\" is not defined.", name);
 }
 
-static void parse_entry_function_attributes(struct hlsl_ctx *ctx, const struct hlsl_ir_function_decl *entry_func)
+static void parse_entry_function_attributes(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func)
 {
     const struct hlsl_profile_info *profile = ctx->profile;
     unsigned int i;
@@ -6049,6 +6049,8 @@ static void parse_entry_function_attributes(struct hlsl_ctx *ctx, const struct h
             parse_partitioning_attribute(ctx, attr);
         else if (!strcmp(attr->name, "patchconstantfunc") && profile->type == VKD3D_SHADER_TYPE_HULL)
             parse_patchconstantfunc_attribute(ctx, attr);
+        else if (!strcmp(attr->name, "earlydepthstencil") && profile->type == VKD3D_SHADER_TYPE_PIXEL)
+            entry_func->early_depth_test = true;
         else
             hlsl_warning(ctx, &entry_func->attrs[i]->loc, VKD3D_SHADER_WARNING_HLSL_UNKNOWN_ATTRIBUTE,
                     "Ignoring unknown attribute \"%s\".", entry_func->attrs[i]->name);
diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c
index 34d32272..78b94c3f 100644
--- a/libs/vkd3d-shader/tpf.c
+++ b/libs/vkd3d-shader/tpf.c
@@ -4630,6 +4630,17 @@ static void write_sm4_dcl_thread_group(const struct tpf_writer *tpf, const uint3
     write_sm4_instruction(tpf, &instr);
 }
 
+static void write_sm4_dcl_global_flags(const struct tpf_writer *tpf, uint32_t flags)
+{
+    struct sm4_instruction instr =
+    {
+        .opcode = VKD3D_SM4_OP_DCL_GLOBAL_FLAGS,
+        .extra_bits = flags << VKD3D_SM4_GLOBAL_FLAGS_SHIFT,
+    };
+
+    write_sm4_instruction(tpf, &instr);
+}
+
 static void write_sm4_ret(const struct tpf_writer *tpf)
 {
     struct sm4_instruction instr =
@@ -6119,6 +6130,9 @@ static void write_sm4_shdr(struct hlsl_ctx *ctx, const struct hlsl_ir_function_d
             write_sm4_dcl_textures(&tpf, resource, true);
     }
 
+    if (entry_func->early_depth_test && profile->major_version >= 5)
+        write_sm4_dcl_global_flags(&tpf, VKD3DSGF_FORCE_EARLY_DEPTH_STENCIL);
+
     LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry)
     {
         if ((var->is_input_semantic && var->last_read) || (var->is_output_semantic && var->first_write))
diff --git a/tests/hlsl/depth-bounds.shader_test b/tests/hlsl/depth-bounds.shader_test
index e808b117..9ab1c332 100644
--- a/tests/hlsl/depth-bounds.shader_test
+++ b/tests/hlsl/depth-bounds.shader_test
@@ -38,7 +38,7 @@ clear rtv 0 0.0 0.0 0.0 0.0
 clear dsv 0.0
 todo(glsl) draw quad
 probe (0, 0) rgba(0, 0, 0, 0)
-todo(sm<6) probe uav 1 (0) r(0.0)
+probe uav 1 (0) r(0.0)
 
 clear rtv 0 0.0 0.0 0.0 0.0
 clear dsv 0.3
@@ -62,4 +62,4 @@ clear rtv 0 0.0 0.0 0.0 0.0
 clear dsv 1.0
 todo(glsl) draw quad
 probe (0, 0) rgba(0, 0, 0, 0)
-todo(sm<6) probe uav 1 (0) r(0.0)
+probe uav 1 (0) r(0.0)
diff --git a/tests/hlsl/early-depth-stencil.shader_test b/tests/hlsl/early-depth-stencil.shader_test
index 07d56a39..810bd4a7 100644
--- a/tests/hlsl/early-depth-stencil.shader_test
+++ b/tests/hlsl/early-depth-stencil.shader_test
@@ -28,4 +28,4 @@ clear rtv 0 0.0 0.0 0.0 0.0
 clear dsv 0.5
 todo(glsl) draw quad
 probe (0, 0) rgba(0, 0, 0, 0)
-todo(sm<6) probe uav 1 (0) r(0.0)
+probe uav 1 (0) r(0.0)