Merge pull request #2565 from ReinUsesLisp/track-indirect

shader/track: Track indirect buffers

src/video_core/shader/decode/image.cpp

@@ -95,12 +95,8 @@ const Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::Image
 const Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg,
                                         Tegra::Shader::ImageType type) {
     const Node image_register{GetRegister(reg)};
-    const Node base_image{
+    const auto [base_image, cbuf_index, cbuf_offset]{
         TrackCbuf(image_register, global_code, static_cast<s64>(global_code.size()))};
-    const auto cbuf{std::get_if<CbufNode>(&*base_image)};
-    const auto cbuf_offset_imm{std::get_if<ImmediateNode>(&*cbuf->GetOffset())};
-    const auto cbuf_offset{cbuf_offset_imm->GetValue()};
-    const auto cbuf_index{cbuf->GetIndex()};
     const auto cbuf_key{(static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset)};
 
     // If this image has already been used, return the existing mapping.

src/video_core/shader/decode/memory.cpp

@@ -297,18 +297,13 @@ std::tuple<Node, Node, GlobalMemoryBase> ShaderIR::TrackAndGetGlobalMemory(NodeB
     const auto addr_register{GetRegister(instr.gmem.gpr)};
     const auto immediate_offset{static_cast<u32>(instr.gmem.offset)};
 
-    const Node base_address{
-        TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size()))};
-    const auto cbuf = std::get_if<CbufNode>(&*base_address);
-    ASSERT(cbuf != nullptr);
-    const auto cbuf_offset_imm = std::get_if<ImmediateNode>(&*cbuf->GetOffset());
-    ASSERT(cbuf_offset_imm != nullptr);
-    const auto cbuf_offset = cbuf_offset_imm->GetValue();
-
-    bb.push_back(
-        Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]", cbuf->GetIndex(), cbuf_offset)));
-
-    const GlobalMemoryBase descriptor{cbuf->GetIndex(), cbuf_offset};
+    const auto [base_address, index, offset] =
+        TrackCbuf(addr_register, global_code, static_cast<s64>(global_code.size()));
+    ASSERT(base_address != nullptr);
+
+    bb.push_back(Comment(fmt::format("Base address is c[0x{:x}][0x{:x}]", index, offset)));
+
+    const GlobalMemoryBase descriptor{index, offset};
     const auto& [entry, is_new] = used_global_memory.try_emplace(descriptor);
     auto& usage = entry->second;
     if (is_write) {

src/video_core/shader/decode/texture.cpp

@@ -308,13 +308,9 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler, Textu
 const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg, TextureType type,
                                             bool is_array, bool is_shadow) {
     const Node sampler_register = GetRegister(reg);
-    const Node base_sampler =
+    const auto [base_sampler, cbuf_index, cbuf_offset] =
         TrackCbuf(sampler_register, global_code, static_cast<s64>(global_code.size()));
-    const auto cbuf = std::get_if<CbufNode>(&*base_sampler);
-    const auto cbuf_offset_imm = std::get_if<ImmediateNode>(&*cbuf->GetOffset());
-    ASSERT(cbuf_offset_imm != nullptr);
-    const auto cbuf_offset = cbuf_offset_imm->GetValue();
-    const auto cbuf_index = cbuf->GetIndex();
+    ASSERT(base_sampler != nullptr);
     const auto cbuf_key = (static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset);
 
     // If this sampler has already been used, return the existing mapping.

src/video_core/shader/shader_ir.cpp

@@ -61,7 +61,16 @@ Node ShaderIR::GetConstBufferIndirect(u64 index_, u64 offset_, Node node) {
     const auto [entry, is_new] = used_cbufs.try_emplace(index);
     entry->second.MarkAsUsedIndirect();
 
-    const Node final_offset = Operation(OperationCode::UAdd, NO_PRECISE, node, Immediate(offset));
+    const Node final_offset = [&]() {
+        // Attempt to inline constant buffer without a variable offset. This is done to allow
+        // tracking LDC calls.
+        if (const auto gpr = std::get_if<GprNode>(&*node)) {
+            if (gpr->GetIndex() == Register::ZeroIndex) {
+                return Immediate(offset);
+            }
+        }
+        return Operation(OperationCode::UAdd, NO_PRECISE, node, Immediate(offset));
+    }();
     return MakeNode<CbufNode>(index, final_offset);
 }
 

src/video_core/shader/shader_ir.h

@@ -328,7 +328,7 @@ private:
     void WriteLop3Instruction(NodeBlock& bb, Tegra::Shader::Register dest, Node op_a, Node op_b,
                               Node op_c, Node imm_lut, bool sets_cc);
 
-    Node TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const;
+    std::tuple<Node, u32, u32> TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const;
 
     std::optional<u32> TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const;
 

src/video_core/shader/track.cpp

@@ -32,39 +32,44 @@ std::pair<Node, s64> FindOperation(const NodeBlock& code, s64 cursor,
     }
     return {};
 }
-} // namespace
+} // Anonymous namespace
 
-Node ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const {
+std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code,
+                                               s64 cursor) const {
     if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
-        // Cbuf found, but it has to be immediate
-        return std::holds_alternative<ImmediateNode>(*cbuf->GetOffset()) ? tracked : nullptr;
+        // Constant buffer found, test if it's an immediate
+        const auto offset = cbuf->GetOffset();
+        if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
+            return {tracked, cbuf->GetIndex(), immediate->GetValue()};
+        }
+        return {};
     }
     if (const auto gpr = std::get_if<GprNode>(&*tracked)) {
         if (gpr->GetIndex() == Tegra::Shader::Register::ZeroIndex) {
-            return nullptr;
+            return {};
         }
         // Reduce the cursor in one to avoid infinite loops when the instruction sets the same
         // register that it uses as operand
         const auto [source, new_cursor] = TrackRegister(gpr, code, cursor - 1);
         if (!source) {
-            return nullptr;
+            return {};
         }
         return TrackCbuf(source, code, new_cursor);
     }
     if (const auto operation = std::get_if<OperationNode>(&*tracked)) {
         for (std::size_t i = 0; i < operation->GetOperandsCount(); ++i) {
-            if (const auto found = TrackCbuf((*operation)[i], code, cursor)) {
-                // Cbuf found in operand
+            if (auto found = TrackCbuf((*operation)[i], code, cursor); std::get<0>(found)) {
+                // Cbuf found in operand.
                 return found;
             }
         }
-        return nullptr;
+        return {};
     }
     if (const auto conditional = std::get_if<ConditionalNode>(&*tracked)) {
         const auto& conditional_code = conditional->GetCode();
         return TrackCbuf(tracked, conditional_code, static_cast<s64>(conditional_code.size()));
     }
-    return nullptr;
+    return {};
 }
 
 std::optional<u32> ShaderIR::TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const {