shader_ir/memory: Implement patch stores

src/video_core/engines/shader_bytecode.h

@@ -98,10 +98,11 @@ union Attribute {
         BitField<20, 10, u64> immediate;
         BitField<22, 2, u64> element;
         BitField<24, 6, Index> index;
+        BitField<31, 1, u64> patch;
         BitField<47, 3, AttributeSize> size;
 
         bool IsPhysical() const {
-            return element == 0 && static_cast<u64>(index.Value()) == 0;
+            return patch == 0 && element == 0 && static_cast<u64>(index.Value()) == 0;
         }
     } fmt20;
 

src/video_core/shader/decode/memory.cpp

@@ -21,6 +21,7 @@ using Tegra::Shader::OpCode;
 using Tegra::Shader::Register;
 
 namespace {
+
 u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) {
     switch (uniform_type) {
     case Tegra::Shader::UniformType::Single:
@@ -35,6 +36,7 @@ u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) {
         return 1;
     }
 }
+
 } // Anonymous namespace
 
 u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
@@ -196,28 +198,28 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         UNIMPLEMENTED_IF_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) != 0,
                              "Unaligned attribute loads are not supported");
 
-        u64 next_element = instr.attribute.fmt20.element;
+        u64 element = instr.attribute.fmt20.element;
-        auto next_index = static_cast<u64>(instr.attribute.fmt20.index.Value());
+        auto index = static_cast<u64>(instr.attribute.fmt20.index.Value());
 
-        const auto StoreNextElement = [&](u32 reg_offset) {
+        const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1;
-            const auto dest = GetOutputAttribute(static_cast<Attribute::Index>(next_index),
+        for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) {
-                                                 next_element, GetRegister(instr.gpr39));
+            Node dest;
+            if (instr.attribute.fmt20.patch) {
+                const u32 offset = static_cast<u32>(index) * 4 + static_cast<u32>(element);
+                dest = MakeNode<PatchNode>(offset);
+            } else {
+                dest = GetOutputAttribute(static_cast<Attribute::Index>(index), element,
+                                          GetRegister(instr.gpr39));
+            }
             const auto src = GetRegister(instr.gpr0.Value() + reg_offset);
 
             bb.push_back(Operation(OperationCode::Assign, dest, src));
 
-            // Load the next attribute element into the following register. If the element
+            // Load the next attribute element into the following register. If the element to load
-            // to load goes beyond the vec4 size, load the first element of the next
+            // goes beyond the vec4 size, load the first element of the next attribute.
-            // attribute.
+            element = (element + 1) % 4;
-            next_element = (next_element + 1) % 4;
+            index = index + (element == 0 ? 1 : 0);
-            next_index = next_index + (next_element == 0 ? 1 : 0);
-        };
-
-        const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1;
-        for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) {
-            StoreNextElement(reg_offset);
         }
-
         break;
     }
     case OpCode::Id::ST_L:

src/video_core/shader/node.h

@@ -213,13 +213,14 @@ class PredicateNode;
 class AbufNode;
 class CbufNode;
 class LmemNode;
+class PatchNode;
 class SmemNode;
 class GmemNode;
 class CommentNode;
 
-using NodeData =
+using NodeData = std::variant<OperationNode, ConditionalNode, GprNode, ImmediateNode,
-    std::variant<OperationNode, ConditionalNode, GprNode, ImmediateNode, InternalFlagNode,
+                              InternalFlagNode, PredicateNode, AbufNode, PatchNode, CbufNode,
-                 PredicateNode, AbufNode, CbufNode, LmemNode, SmemNode, GmemNode, CommentNode>;
+                              LmemNode, SmemNode, GmemNode, CommentNode>;
 using Node = std::shared_ptr<NodeData>;
 using Node4 = std::array<Node, 4>;
 using NodeBlock = std::vector<Node>;
@@ -542,6 +543,19 @@ private:
     u32 element{};
 };
 
+/// Patch memory (used to communicate tessellation stages).
+class PatchNode final {
+public:
+    explicit PatchNode(u32 offset) : offset{offset} {}
+
+    u32 GetOffset() const {
+        return offset;
+    }
+
+private:
+    u32 offset{};
+};
+
 /// Constant buffer node, usually mapped to uniform buffers in GLSL
 class CbufNode final {
 public:

src/video_core/shader/track.cpp

@@ -7,6 +7,7 @@
 #include <variant>
 
 #include "common/common_types.h"
+#include "video_core/shader/node.h"
 #include "video_core/shader/shader_ir.h"
 
 namespace VideoCommon::Shader {