From cd1689b74b69aefae9cbeabefc6621a1e12877e2 Mon Sep 17 00:00:00 2001
From: psucien <bad_cast@protonmail.com>
Date: Fri, 21 Jun 2024 22:32:10 +0200
Subject: [PATCH] shader_recompiler: added support for fetch instance id

---
 src/shader_recompiler/frontend/fetch_shader.cpp       |  3 +++
 src/shader_recompiler/frontend/fetch_shader.h         | 11 ++++++-----
 .../frontend/translate/translate.cpp                  |  6 ++++++
 src/shader_recompiler/runtime_info.h                  |  8 ++++++++
 .../renderer_vulkan/vk_graphics_pipeline.cpp          |  4 +++-
 5 files changed, 26 insertions(+), 6 deletions(-)

diff --git a/src/shader_recompiler/frontend/fetch_shader.cpp b/src/shader_recompiler/frontend/fetch_shader.cpp
index 7f4f50e94..80917e0a3 100644
--- a/src/shader_recompiler/frontend/fetch_shader.cpp
+++ b/src/shader_recompiler/frontend/fetch_shader.cpp
@@ -72,6 +72,9 @@ std::vector<VertexAttribute> ParseFetchShader(const u32* code) {
             attrib.sgpr_base = it->base_sgpr;
             attrib.dword_offset = it->dword_offset;
 
+            // Store instance id rate
+            attrib.instance_data = inst.src[0].code;
+
             // Mark load as used.
             it->dst_reg = -1;
         }
diff --git a/src/shader_recompiler/frontend/fetch_shader.h b/src/shader_recompiler/frontend/fetch_shader.h
index 2f8eae12c..14f2bf4d4 100644
--- a/src/shader_recompiler/frontend/fetch_shader.h
+++ b/src/shader_recompiler/frontend/fetch_shader.h
@@ -9,11 +9,12 @@
 namespace Shader::Gcn {
 
 struct VertexAttribute {
-    u8 semantic;     ///< Semantic index of the attribute
-    u8 dest_vgpr;    ///< Destination VGPR to load first component.
-    u8 num_elements; ///< Number of components to load
-    u8 sgpr_base;    ///< SGPR that contains the pointer to the list of vertex V#
-    u8 dword_offset; ///< The dword offset of the V# that describes this attribute.
+    u8 semantic;      ///< Semantic index of the attribute
+    u8 dest_vgpr;     ///< Destination VGPR to load first component.
+    u8 num_elements;  ///< Number of components to load
+    u8 sgpr_base;     ///< SGPR that contains the pointer to the list of vertex V#
+    u8 dword_offset;  ///< The dword offset of the V# that describes this attribute.
+    u8 instance_data; ///< Indicates that the buffer will be accessed in instance rate
 };
 
 std::vector<VertexAttribute> ParseFetchShader(const u32* code);
diff --git a/src/shader_recompiler/frontend/translate/translate.cpp b/src/shader_recompiler/frontend/translate/translate.cpp
index 344fdc8bb..2d1679f37 100644
--- a/src/shader_recompiler/frontend/translate/translate.cpp
+++ b/src/shader_recompiler/frontend/translate/translate.cpp
@@ -194,6 +194,11 @@ void Translator::EmitFetch(const GcnInst& inst) {
             ir.SetVectorReg(dst_reg++, ir.GetAttribute(attr, i));
         }
 
+        if (attrib.instance_data == 2 || attrib.instance_data == 3) {
+            LOG_WARNING(Render_Recompiler, "Unsupported instance step rate = {}",
+                        attrib.instance_data);
+        }
+
         // Read the V# of the attribute to figure out component number and type.
         const auto buffer = info.ReadUd<AmdGpu::Buffer>(attrib.sgpr_base, attrib.dword_offset);
         const u32 num_components = AmdGpu::NumComponents(buffer.data_format);
@@ -203,6 +208,7 @@ void Translator::EmitFetch(const GcnInst& inst) {
             .num_components = std::min<u16>(attrib.num_elements, num_components),
             .sgpr_base = attrib.sgpr_base,
             .dword_offset = attrib.dword_offset,
+            .instance_step_rate = static_cast<Info::VsInput::InstanceIdType>(attrib.instance_data),
         });
     }
 }
diff --git a/src/shader_recompiler/runtime_info.h b/src/shader_recompiler/runtime_info.h
index 634749585..196fad013 100644
--- a/src/shader_recompiler/runtime_info.h
+++ b/src/shader_recompiler/runtime_info.h
@@ -72,11 +72,19 @@ using SamplerResourceList = boost::container::static_vector<SamplerResource, 8>;
 
 struct Info {
     struct VsInput {
+        enum InstanceIdType : u8 {
+            None = 0,
+            OverStepRate0 = 1,
+            OverStepRate1 = 2,
+            Plain = 3,
+        };
+
         AmdGpu::NumberFormat fmt;
         u16 binding;
         u16 num_components;
         u8 sgpr_base;
         u8 dword_offset;
+        InstanceIdType instance_step_rate;
     };
     boost::container::static_vector<VsInput, 32> vs_inputs{};
 
diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
index 787fb774a..89f3c086d 100644
--- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
+++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
@@ -53,7 +53,9 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
         bindings.push_back({
             .binding = input.binding,
             .stride = buffer.GetStride(),
-            .inputRate = vk::VertexInputRate::eVertex,
+            .inputRate = input.instance_step_rate == Shader::Info::VsInput::None
+                             ? vk::VertexInputRate::eVertex
+                             : vk::VertexInputRate::eInstance,
         });
     }