From 37a61bf41a7f96e4ddb8ead2f1d6c1875d6a16ad Mon Sep 17 00:00:00 2001
From: Francisco Casas <fcasas@codeweavers.com>
Date: Mon, 11 Nov 2024 21:46:43 -0300
Subject: [PATCH] vkd3d-shader/hlsl: Store the global flags in the vsir
 program.

---
 libs/vkd3d-shader/hlsl_codegen.c | 31 +++++++++++++++++++++++++++++++
 libs/vkd3d-shader/tpf.c          | 23 ++---------------------
 2 files changed, 33 insertions(+), 21 deletions(-)

diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c
index 7d0a9aa5..2a625bc0 100644
--- a/libs/vkd3d-shader/hlsl_codegen.c
+++ b/libs/vkd3d-shader/hlsl_codegen.c
@@ -9490,6 +9490,36 @@ static void generate_vsir_scan_required_features(struct hlsl_ctx *ctx, struct vs
      * STENCIL_REF, and TYPED_UAV_LOAD_ADDITIONAL_FORMATS. */
 }
 
+static void generate_vsir_scan_global_flags(struct hlsl_ctx *ctx,
+        struct vsir_program *program, const struct hlsl_ir_function_decl *entry_func)
+{
+    const struct vkd3d_shader_version *version = &program->shader_version;
+    struct extern_resource *extern_resources;
+    unsigned int extern_resources_count, i;
+
+    extern_resources = sm4_get_extern_resources(ctx, &extern_resources_count);
+
+    if (version->major == 4)
+    {
+        for (i = 0; i < extern_resources_count; ++i)
+        {
+            const struct extern_resource *resource = &extern_resources[i];
+            const struct hlsl_type *type = resource->component_type;
+
+            if (type && type->class == HLSL_CLASS_TEXTURE && type->sampler_dim == HLSL_SAMPLER_DIM_RAW_BUFFER)
+            {
+                program->global_flags |= VKD3DSGF_ENABLE_RAW_AND_STRUCTURED_BUFFERS;
+                break;
+            }
+        }
+    }
+
+    sm4_free_extern_resources(extern_resources, extern_resources_count);
+
+    if (entry_func->early_depth_test && vkd3d_shader_ver_ge(version, 5, 0))
+        program->global_flags |= VKD3DSGF_FORCE_EARLY_DEPTH_STENCIL;
+}
+
 /* OBJECTIVE: Translate all the information from ctx and entry_func to the
  * vsir_program, so it can be used as input to tpf_compile() without relying
  * on ctx and entry_func. */
@@ -9531,6 +9561,7 @@ static void sm4_generate_vsir(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl
     }
 
     generate_vsir_scan_required_features(ctx, program);
+    generate_vsir_scan_global_flags(ctx, program, func);
 }
 
 static struct hlsl_ir_jump *loop_unrolling_find_jump(struct hlsl_block *block, struct hlsl_ir_node *stop_point,
diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c
index c9de4e47..d8763052 100644
--- a/libs/vkd3d-shader/tpf.c
+++ b/libs/vkd3d-shader/tpf.c
@@ -4870,7 +4870,6 @@ static void tpf_write_shdr(struct tpf_compiler *tpf, struct hlsl_ir_function_dec
     const struct hlsl_buffer *cbuffer;
     struct hlsl_ctx *ctx = tpf->ctx;
     size_t token_count_position;
-    uint32_t global_flags = 0;
 
     static const uint16_t shader_types[VKD3D_SHADER_TYPE_COUNT] =
     {
@@ -4892,26 +4891,8 @@ static void tpf_write_shdr(struct tpf_compiler *tpf, struct hlsl_ir_function_dec
     put_u32(&buffer, vkd3d_make_u32((version->major << 4) | version->minor, shader_types[version->type]));
     token_count_position = put_u32(&buffer, 0);
 
-    if (version->major == 4)
-    {
-        for (i = 0; i < extern_resources_count; ++i)
-        {
-            const struct extern_resource *resource = &extern_resources[i];
-            const struct hlsl_type *type = resource->component_type;
-
-            if (type && type->class == HLSL_CLASS_TEXTURE && type->sampler_dim == HLSL_SAMPLER_DIM_RAW_BUFFER)
-            {
-                global_flags |= VKD3DSGF_ENABLE_RAW_AND_STRUCTURED_BUFFERS;
-                break;
-            }
-        }
-    }
-
-    if (entry_func->early_depth_test && vkd3d_shader_ver_ge(version, 5, 0))
-        global_flags |= VKD3DSGF_FORCE_EARLY_DEPTH_STENCIL;
-
-    if (global_flags)
-        write_sm4_dcl_global_flags(tpf, global_flags);
+    if (tpf->program->global_flags)
+        write_sm4_dcl_global_flags(tpf, tpf->program->global_flags);
 
     if (version->type == VKD3D_SHADER_TYPE_HULL)
     {