Experimenting with scissors offsets

Another change
Another check
2026-09-05 03:54:32 +00:00 · 2026-09-01 21:53:15 -04:00 · 2026-09-02 02:57:14 +02:00 · 2026-09-02 02:57:14 +02:00 · 2026-09-02 02:57:14 +02:00 · 2026-09-02 02:57:14 +02:00
44 changed files with 1056 additions and 650 deletions
@@ -238,7 +238,7 @@ option(YUZU_USE_BUNDLED_SIRIT "Download bundled sirit" ${BUNDLED_SIRIT_DEFAULT})
 # FreeBSD 15+ has libusb, versions below should disable it
 cmake_dependent_option(ENABLE_LIBUSB "Enable the use of LibUSB" ON "WIN32 OR LINUX OR FREEBSD OR APPLE" OFF)

-cmake_dependent_option(ENABLE_OPENGL "Enable OpenGL" ON "NOT (WIN32 AND ARCHITECTURE_arm64) AND NOT APPLE AND NOT ANDROID" OFF)
+cmake_dependent_option(ENABLE_OPENGL "Enable OpenGL" ON "NOT (WIN32 AND ARCHITECTURE_arm64) AND NOT APPLE" OFF)
 mark_as_advanced(FORCE ENABLE_OPENGL)

 option(ENABLE_WEB_SERVICE "Enable web services (telemetry, etc.)" ON)
@@ -76,7 +76,7 @@ The following options are desktop only.

 - `ENABLE_LIBUSB` (ON) Enable the use of the libusb input backend (HIGHLY RECOMMENDED)
 - `ENABLE_OPENGL` (ON) Enable the OpenGL graphics backend
-  - Unavailable on Windows/ARM64 and on Android
+  - Unavailable on Windows/ARM64
  - You probably shouldn't turn this off.

 ### Qt
@@ -11,7 +11,6 @@
 #include <limits>
 #include <span>
 #include <array>
-#include <algorithm>
 #include <time.h>

 namespace Tz {
@@ -5,7 +5,6 @@
 // SPDX-License-Identifier: GPL-2.0-or-later

 #include <string>
-#include <cstdlib>
 #include <string_view>
 #ifdef _WIN32
 #include <llvm/Demangle/Demangle.h>
@@ -358,7 +358,7 @@ private:

        ConnectionState(boost::asio::ip::tcp::socket&& client_socket_, async_pipe signal_pipe_, Kernel::KernelCore& kernel)
            : client_socket{std::move(client_socket_)}
-            , signal_pipe{std::move(signal_pipe_)}
+            , signal_pipe{signal_pipe_}
            , active_thread{kernel, nullptr}
        {}

@@ -1021,9 +1021,6 @@ Result KProcess::Run(KernelCore& kernel, s32 priority, size_t stack_size) {

    // Suspend for debug, if we should.
    if (kernel.System().DebuggerEnabled()) {
-        LOG_INFO(Debug_GDBStub,
-                 "GDB stub enabled; suspending guest process until a debugger continues execution on port {}",
-                 Settings::values.gdbstub_port.GetValue());
        main_thread->RequestSuspend(kernel, SuspendType::Debug);
    }

@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later

@@ -32,7 +29,7 @@ struct FuncTraits<ReturnType_ (*)(Args...)> {
 };

 template <auto func, typename... Args>
-[[maybe_unused]] void SetDefinition(EmitContext& ctx, IR::Inst* inst, Args... args) {
+void SetDefinition(EmitContext& ctx, IR::Inst* inst, Args... args) {
    inst->SetDefinition<Id>(func(ctx, std::forward<Args>(args)...));
 }

@@ -492,9 +492,6 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
        if (ctx.uses_nonuniform_storage_texel_buffer) {
            ctx.AddCapability(spv::Capability::StorageTexelBufferArrayNonUniformIndexing);
        }
-        if (ctx.uses_nonuniform_storage_buffer) {
-            ctx.AddCapability(spv::Capability::StorageBufferArrayNonUniformIndexing);
-        }
    }
 }

@@ -4,6 +4,8 @@
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later

+#include <bit>
+
 #include "shader_recompiler/backend/spirv/emit_spirv.h"
 #include "shader_recompiler/backend/spirv/emit_spirv_instructions.h"
 #include "shader_recompiler/backend/spirv/spirv_emit_context.h"
@@ -21,13 +23,29 @@ Id SharedPointer(EmitContext& ctx, Id offset, u32 index_offset = 0) {
               : ctx.OpAccessChain(ctx.shared_u32, ctx.shared_memory_u32, index);
 }

+Id StorageIndex(EmitContext& ctx, const IR::Value& offset, size_t element_size) {
+    if (offset.IsImmediate()) {
+        const u32 imm_offset{static_cast<u32>(offset.U32() / element_size)};
+        return ctx.Const(imm_offset);
+    }
+    const u32 shift{static_cast<u32>(std::countr_zero(element_size))};
+    const Id index{ctx.Def(offset)};
+    if (shift == 0) {
+        return index;
+    }
+    const Id shift_id{ctx.Const(shift)};
+    return ctx.OpShiftRightLogical(ctx.U32[1], index, shift_id);
+}
+
 Id StoragePointer(EmitContext& ctx, const StorageTypeDefinition& type_def,
                  Id StorageDefinitions::*member_ptr, const IR::Value& binding,
                  const IR::Value& offset, size_t element_size) {
    if (!binding.IsImmediate()) {
        throw NotImplementedException("Dynamic storage buffer indexing");
    }
-    return ctx.StoragePointer(binding.U32(), ctx.Def(offset), type_def, static_cast<u32>(element_size), member_ptr);
+    const Id ssbo{ctx.ssbos[binding.U32()].*member_ptr};
+    const Id index{StorageIndex(ctx, offset, element_size)};
+    return ctx.OpAccessChain(type_def.element, ssbo, ctx.u32_zero_value, index);
 }

 std::pair<Id, Id> AtomicArgs(EmitContext& ctx) {
@@ -198,14 +216,16 @@ Id EmitStorageAtomicUMax32(EmitContext& ctx, const IR::Value& binding, const IR:

 Id EmitStorageAtomicInc32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                          Id value) {
-    const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
-    return ctx.OpFunctionCall(ctx.U32[1], ctx.increment_cas_ssbo, pointer, value);
+    const Id ssbo{ctx.ssbos[binding.U32()].U32};
+    const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
+    return ctx.OpFunctionCall(ctx.U32[1], ctx.increment_cas_ssbo, base_index, value, ssbo);
 }

 Id EmitStorageAtomicDec32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                          Id value) {
-    const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
-    return ctx.OpFunctionCall(ctx.U32[1], ctx.decrement_cas_ssbo, pointer, value);
+    const Id ssbo{ctx.ssbos[binding.U32()].U32};
+    const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
+    return ctx.OpFunctionCall(ctx.U32[1], ctx.decrement_cas_ssbo, base_index, value, ssbo);
 }

 Id EmitStorageAtomicAnd32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
@@ -344,49 +364,56 @@ Id EmitStorageAtomicExchange32x2(EmitContext& ctx, const IR::Value& binding,

 Id EmitStorageAtomicAddF32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                           Id value) {
-    const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
-    return ctx.OpFunctionCall(ctx.F32[1], ctx.f32_add_cas, pointer, value);
+    const Id ssbo{ctx.ssbos[binding.U32()].U32};
+    const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
+    return ctx.OpFunctionCall(ctx.F32[1], ctx.f32_add_cas, base_index, value, ssbo);
 }

 Id EmitStorageAtomicAddF16x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                             Id value) {
-    const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
-    const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_add_cas, pointer, value)};
+    const Id ssbo{ctx.ssbos[binding.U32()].U32};
+    const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
+    const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_add_cas, base_index, value, ssbo)};
    return ctx.OpBitcast(ctx.U32[1], result);
 }

 Id EmitStorageAtomicAddF32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                             Id value) {
-    const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
-    const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_add_cas, pointer, value)};
+    const Id ssbo{ctx.ssbos[binding.U32()].U32};
+    const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
+    const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_add_cas, base_index, value, ssbo)};
    return ctx.OpPackHalf2x16(ctx.U32[1], result);
 }

 Id EmitStorageAtomicMinF16x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                             Id value) {
-    const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
-    const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_min_cas, pointer, value)};
+    const Id ssbo{ctx.ssbos[binding.U32()].U32};
+    const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
+    const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_min_cas, base_index, value, ssbo)};
    return ctx.OpBitcast(ctx.U32[1], result);
 }

 Id EmitStorageAtomicMinF32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                             Id value) {
-    const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
-    const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_min_cas, pointer, value)};
+    const Id ssbo{ctx.ssbos[binding.U32()].U32};
+    const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
+    const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_min_cas, base_index, value, ssbo)};
    return ctx.OpPackHalf2x16(ctx.U32[1], result);
 }

 Id EmitStorageAtomicMaxF16x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                             Id value) {
-    const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
-    const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_max_cas, pointer, value)};
+    const Id ssbo{ctx.ssbos[binding.U32()].U32};
+    const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
+    const Id result{ctx.OpFunctionCall(ctx.F16[2], ctx.f16x2_max_cas, base_index, value, ssbo)};
    return ctx.OpBitcast(ctx.U32[1], result);
 }

 Id EmitStorageAtomicMaxF32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                             Id value) {
-    const Id pointer{StoragePointer(ctx, ctx.storage_types.U32, &StorageDefinitions::U32, binding, offset, sizeof(u32))};
-    const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_max_cas, pointer, value)};
+    const Id ssbo{ctx.ssbos[binding.U32()].U32};
+    const Id base_index{StorageIndex(ctx, offset, sizeof(u32))};
+    const Id result{ctx.OpFunctionCall(ctx.F32[2], ctx.f32x2_max_cas, base_index, value, ssbo)};
    return ctx.OpPackHalf2x16(ctx.U32[1], result);
 }

@@ -4,33 +4,46 @@
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later

+#include <bit>
+
 #include "shader_recompiler/backend/spirv/emit_spirv.h"
 #include "shader_recompiler/backend/spirv/emit_spirv_instructions.h"
 #include "shader_recompiler/backend/spirv/spirv_emit_context.h"

 namespace Shader::Backend::SPIRV {
 namespace {
-Id StorageByteOffset(EmitContext& ctx, const IR::Value& offset, size_t element_size, u32 index_offset) {
-    Id byte_offset{ctx.Def(offset)};
-    if (index_offset != 0) {
-        byte_offset = ctx.OpIAdd(ctx.U32[1], byte_offset, ctx.Const(static_cast<u32>(index_offset * element_size)));
+Id StorageIndex(EmitContext& ctx, const IR::Value& offset, size_t element_size,
+                u32 index_offset = 0) {
+    if (offset.IsImmediate()) {
+        const u32 imm_offset{static_cast<u32>(offset.U32() / element_size) + index_offset};
+        return ctx.Const(imm_offset);
    }
-    return byte_offset;
+    const u32 shift{static_cast<u32>(std::countr_zero(element_size))};
+    Id index{ctx.Def(offset)};
+    if (shift != 0) {
+        const Id shift_id{ctx.Const(shift)};
+        index = ctx.OpShiftRightLogical(ctx.U32[1], index, shift_id);
+    }
+    if (index_offset != 0) {
+        index = ctx.OpIAdd(ctx.U32[1], index, ctx.Const(index_offset));
+    }
+    return index;
 }

 Id StoragePointer(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                  const StorageTypeDefinition& type_def, size_t element_size,
-                  Id StorageDefinitions::* member_ptr, u32 index_offset = 0) {
+                  Id StorageDefinitions::*member_ptr, u32 index_offset = 0) {
    if (!binding.IsImmediate()) {
        throw NotImplementedException("Dynamic storage buffer indexing");
    }
-    const Id byte_offset{StorageByteOffset(ctx, offset, element_size, index_offset)};
-    return ctx.StoragePointer(binding.U32(), byte_offset, type_def, static_cast<u32>(element_size), member_ptr);
+    const Id ssbo{ctx.ssbos[binding.U32()].*member_ptr};
+    const Id index{StorageIndex(ctx, offset, element_size, index_offset)};
+    return ctx.OpAccessChain(type_def.element, ssbo, ctx.u32_zero_value, index);
 }

 Id LoadStorage(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset, Id result_type,
               const StorageTypeDefinition& type_def, size_t element_size,
-               Id StorageDefinitions::* member_ptr, u32 index_offset = 0) {
+               Id StorageDefinitions::*member_ptr, u32 index_offset = 0) {
    const Id pointer{
        StoragePointer(ctx, binding, offset, type_def, element_size, member_ptr, index_offset)};
    return ctx.OpLoad(result_type, pointer);
@@ -44,7 +57,7 @@ Id LoadStorage32(EmitContext& ctx, const IR::Value& binding, const IR::Value& of

 void WriteStorage(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset, Id value,
                  const StorageTypeDefinition& type_def, size_t element_size,
-                  Id StorageDefinitions::* member_ptr, u32 index_offset = 0) {
+                  Id StorageDefinitions::*member_ptr, u32 index_offset = 0) {
    const Id pointer{
        StoragePointer(ctx, binding, offset, type_def, element_size, member_ptr, index_offset)};
    ctx.OpStore(pointer, value);
@@ -299,7 +299,7 @@ void DefineConstBuffers(EmitContext& ctx, const Info& info, Id UniformDefinition
 }

 void DefineSsbos(EmitContext& ctx, StorageTypeDefinition& type_def,
-                 Id StorageDefinitions::* member_type, const Info& info, u32 binding, Id type,
+                 Id StorageDefinitions::*member_type, const Info& info, u32 binding, Id type,
                 u32 stride) {
    const Id array_type{ctx.TypeRuntimeArray(type)};
    ctx.Decorate(array_type, spv::Decoration::ArrayStride, stride);
@@ -309,27 +309,23 @@ void DefineSsbos(EmitContext& ctx, StorageTypeDefinition& type_def,
    ctx.MemberDecorate(struct_type, 0, spv::Decoration::Offset, 0U);

    const Id struct_pointer{ctx.TypePointer(spv::StorageClass::StorageBuffer, struct_type)};
+    type_def.array = struct_pointer;
    type_def.element = ctx.TypePointer(spv::StorageClass::StorageBuffer, type);

    u32 index{};
    for (const StorageBufferDescriptor& desc : info.storage_buffers_descriptors) {
-        const Id variable_type{[&] {
-            if (desc.count == 1) {
-                return struct_pointer;
-            }
-            const Id descriptor_array{ctx.TypeArray(struct_type, ctx.Const(desc.count))};
-            return ctx.TypePointer(spv::StorageClass::StorageBuffer, descriptor_array);
-        }()};
-        const Id id{ctx.AddGlobalVariable(variable_type, spv::StorageClass::StorageBuffer)};
+        const Id id{ctx.AddGlobalVariable(struct_pointer, spv::StorageClass::StorageBuffer)};
        ctx.Decorate(id, spv::Decoration::Binding, binding);
        ctx.Decorate(id, spv::Decoration::DescriptorSet, 0U);
        ctx.Name(id, fmt::format("ssbo{}", index));
        if (ctx.profile.supported_spirv >= 0x00010400) {
            ctx.interfaces.push_back(id);
        }
-        ctx.ssbos[index].*member_type = id;
-        ++index;
-        ++binding;
+        for (size_t i = 0; i < desc.count; ++i) {
+            ctx.ssbos[index + i].*member_type = id;
+        }
+        index += desc.count;
+        binding += desc.count;
    }
 }

@@ -372,7 +368,8 @@ Id CasFunction(EmitContext& ctx, Operation operation, Id value_type) {
    return func;
 }

-Id CasLoop(EmitContext& ctx, Operation operation, Id element_pointer, Id value_type, Id memory_type, spv::Scope scope) {
+Id CasLoop(EmitContext& ctx, Operation operation, Id array_pointer, Id element_pointer,
+           Id value_type, Id memory_type, spv::Scope scope) {
    const bool is_shared{scope == spv::Scope::Workgroup};
    const bool is_struct{!is_shared || ctx.uses_explicit_workgroup_layout};
    const Id cas_func{CasFunction(ctx, operation, value_type)};
@@ -382,12 +379,14 @@ Id CasLoop(EmitContext& ctx, Operation operation, Id element_pointer, Id value_t
    const Id loop_header{ctx.OpLabel()};
    const Id continue_block{ctx.OpLabel()};
    const Id merge_block{ctx.OpLabel()};
-    const Id func_type{is_shared ? ctx.TypeFunction(value_type, ctx.U32[1], value_type)
-                                 : ctx.TypeFunction(value_type, element_pointer, value_type)};
+    const Id func_type{is_shared
+                           ? ctx.TypeFunction(value_type, ctx.U32[1], value_type)
+                           : ctx.TypeFunction(value_type, ctx.U32[1], value_type, array_pointer)};

    const Id func{ctx.OpFunction(value_type, spv::FunctionControlMask::MaskNone, func_type)};
-    const Id address{ctx.OpFunctionParameter(is_shared ? ctx.U32[1] : element_pointer)};
+    const Id index{ctx.OpFunctionParameter(ctx.U32[1])};
    const Id op_b{ctx.OpFunctionParameter(value_type)};
+    const Id base{is_shared ? ctx.shared_memory_u32 : ctx.OpFunctionParameter(array_pointer)};
    ctx.AddLabel();
    ctx.OpBranch(loop_header);
    ctx.AddLabel(loop_header);
@@ -396,13 +395,8 @@ Id CasLoop(EmitContext& ctx, Operation operation, Id element_pointer, Id value_t
    ctx.OpBranch(continue_block);

    ctx.AddLabel(continue_block);
-    const Id word_pointer{[&] {
-        if (!is_shared) {
-            return address;
-        }
-        return is_struct ? ctx.OpAccessChain(element_pointer, ctx.shared_memory_u32, zero, address)
-                         : ctx.OpAccessChain(element_pointer, ctx.shared_memory_u32, address);
-    }()};
+    const Id word_pointer{is_struct ? ctx.OpAccessChain(element_pointer, base, zero, index)
+                                    : ctx.OpAccessChain(element_pointer, base, index)};
    if (value_type.value == ctx.F32[2].value) {
        const Id u32_value{ctx.OpLoad(ctx.U32[1], word_pointer)};
        const Id value{ctx.OpUnpackHalf2x16(ctx.F32[2], u32_value)};
@@ -486,7 +480,6 @@ EmitContext::EmitContext(const Profile& profile_, const RuntimeInfo& runtime_inf
    DefineSharedMemoryFunctions(program);
    DefineConstantBuffers(program.info, uniform_binding);
    DefineConstantBufferIndirectFunctions(program.info);
-    DefineStorageBufferMappings(program.info, storage_binding);
    DefineStorageBuffers(program.info, storage_binding);
    DefineTextureBuffers(program.info, texture_binding);
    DefineImageBuffers(program.info, image_binding);
@@ -539,36 +532,6 @@ Id EmitContext::BitOffset16(const IR::Value& offset) {
    return OpBitwiseAnd(U32[1], OpShiftLeftLogical(U32[1], Def(offset), Const(3u)), Const(16u));
 }

-Id EmitContext::StoragePointer(u32 binding, Id byte_offset, const StorageTypeDefinition& type_def,
-                               u32 element_size, Id StorageDefinitions::* member_ptr) {
-    const Id ssbo{ssbos[binding].*member_ptr};
-    const u32 segment_count{storage_buffer_mapping_counts[binding]};
-    if (segment_count <= 1) {
-        const Id index{
-            element_size == 1
-                ? byte_offset
-                : OpShiftRightLogical(U32[1], byte_offset, Const(static_cast<u32>(std::countr_zero(element_size))))};
-        return OpAccessChain(type_def.element, ssbo, u32_zero_value, index);
-    }
-
-    const Id mapped{OpFunctionCall(U32[2], storage_buffer_map_func,
-                                   Const(storage_buffer_mapping_bases[binding]),
-                                   Const(segment_count), byte_offset)};
-    const Id segment{OpCompositeExtract(U32[1], mapped, 0U)};
-    const Id local_offset{OpCompositeExtract(U32[1], mapped, 1U)};
-    const Id index{
-        element_size == 1
-            ? local_offset
-            : OpShiftRightLogical(U32[1], local_offset, Const(static_cast<u32>(std::countr_zero(element_size))))};
-    Decorate(segment, spv::Decoration::NonUniform);
-    non_uniform_ids.insert(segment.value);
-    const Id pointer{OpAccessChain(type_def.element, ssbo, segment, u32_zero_value, index)};
-    Decorate(pointer, spv::Decoration::NonUniform);
-    non_uniform_ids.insert(pointer.value);
-    uses_nonuniform_storage_buffer = true;
-    return pointer;
-}
-
 void EmitContext::DefineCommonTypes(const Info& info) {
    void_id = TypeVoid();

@@ -733,10 +696,12 @@ void EmitContext::DefineSharedMemory(const IR::Program& program) {

 void EmitContext::DefineSharedMemoryFunctions(const IR::Program& program) {
    if (program.info.uses_shared_increment) {
-        increment_cas_shared = CasLoop(*this, Operation::Increment, shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
+        increment_cas_shared = CasLoop(*this, Operation::Increment, shared_memory_u32_type,
+                                       shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
    }
    if (program.info.uses_shared_decrement) {
-        decrement_cas_shared = CasLoop(*this, Operation::Decrement, shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
+        decrement_cas_shared = CasLoop(*this, Operation::Decrement, shared_memory_u32_type,
+                                       shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
    }
 }

@@ -980,7 +945,8 @@ void EmitContext::DefineGlobalMemoryFunctions(const Info& info) {
    }
    using DefPtr = Id StorageDefinitions::*;
    const Id zero{u32_zero_value};
-    const auto define_body{[&](DefPtr ssbo_member, Id addr, const StorageTypeDefinition& type_def, u32 shift, auto&& callback) {
+    const auto define_body{[&](DefPtr ssbo_member, Id addr, Id element_pointer, u32 shift,
+                               auto&& callback) {
        AddLabel();
        const size_t num_buffers{info.storage_buffers_descriptors.size()};
        for (size_t index = 0; index < num_buffers; ++index) {
@@ -1007,28 +973,30 @@ void EmitContext::DefineGlobalMemoryFunctions(const Info& info) {
            OpSelectionMerge(else_label, spv::SelectionControlMask::MaskNone);
            OpBranchConditional(cond, then_label, else_label);
            AddLabel(then_label);
+            const Id ssbo_id{ssbos[index].*ssbo_member};
            const Id ssbo_offset{OpUConvert(U32[1], OpISub(U64, addr, ssbo_addr))};
-            const Id ssbo_pointer{StoragePointer(static_cast<u32>(index), ssbo_offset, type_def, 1U << shift, ssbo_member)};
+            const Id ssbo_index{OpShiftRightLogical(U32[1], ssbo_offset, Const(shift))};
+            const Id ssbo_pointer{OpAccessChain(element_pointer, ssbo_id, zero, ssbo_index)};
            callback(ssbo_pointer);
            AddLabel(else_label);
        }
    }};
-    const auto define_load{
-        [&](DefPtr ssbo_member, const StorageTypeDefinition& type_def, Id type, u32 shift) {
-            const Id function_type{TypeFunction(type, U64)};
-            const Id func_id{OpFunction(type, spv::FunctionControlMask::MaskNone, function_type)};
-            const Id addr{OpFunctionParameter(U64)};
-            define_body(ssbo_member, addr, type_def, shift, [&](Id ssbo_pointer) { OpReturnValue(OpLoad(type, ssbo_pointer)); });
-            OpReturnValue(ConstantNull(type));
-            OpFunctionEnd();
-            return func_id;
-        }};
-    const auto define_write{[&](DefPtr ssbo_member, const StorageTypeDefinition& type_def, Id type, u32 shift) {
+    const auto define_load{[&](DefPtr ssbo_member, Id element_pointer, Id type, u32 shift) {
+        const Id function_type{TypeFunction(type, U64)};
+        const Id func_id{OpFunction(type, spv::FunctionControlMask::MaskNone, function_type)};
+        const Id addr{OpFunctionParameter(U64)};
+        define_body(ssbo_member, addr, element_pointer, shift,
+                    [&](Id ssbo_pointer) { OpReturnValue(OpLoad(type, ssbo_pointer)); });
+        OpReturnValue(ConstantNull(type));
+        OpFunctionEnd();
+        return func_id;
+    }};
+    const auto define_write{[&](DefPtr ssbo_member, Id element_pointer, Id type, u32 shift) {
        const Id function_type{TypeFunction(void_id, U64, type)};
        const Id func_id{OpFunction(void_id, spv::FunctionControlMask::MaskNone, function_type)};
        const Id addr{OpFunctionParameter(U64)};
        const Id data{OpFunctionParameter(type)};
-        define_body(ssbo_member, addr, type_def, shift, [&](Id ssbo_pointer) {
+        define_body(ssbo_member, addr, element_pointer, shift, [&](Id ssbo_pointer) {
            OpStore(ssbo_pointer, data);
            OpReturn();
        });
@@ -1038,9 +1006,10 @@ void EmitContext::DefineGlobalMemoryFunctions(const Info& info) {
    }};
    const auto define{
        [&](DefPtr ssbo_member, const StorageTypeDefinition& type_def, Id type, size_t size) {
+            const Id element_type{type_def.element};
            const u32 shift{static_cast<u32>(std::countr_zero(size))};
-            const Id load_func{define_load(ssbo_member, type_def, type, shift)};
-            const Id write_func{define_write(ssbo_member, type_def, type, shift)};
+            const Id load_func{define_load(ssbo_member, element_type, type, shift)};
+            const Id write_func{define_write(ssbo_member, element_type, type, shift)};
            return std::make_pair(load_func, write_func);
        }};
    std::tie(load_global_func_u32, write_global_func_u32) =
@@ -1259,79 +1228,6 @@ void EmitContext::DefineConstantBufferIndirectFunctions(const Info& info) {
    }
 }

-void EmitContext::DefineStorageBufferMappings(const Info& info, u32& binding) {
-    if (!UsesStorageBufferMappings(info)) {
-        return;
-    }
-    ASSERT(profile.support_storage_buffer_array_nonuniform_indexing);
-    AddExtension("SPV_KHR_storage_buffer_storage_class");
-
-    const u32 num_entries{NumDescriptors(info.storage_buffers_descriptors)};
-    const Id array_type{TypeArray(U32[1], Const(num_entries))};
-    Decorate(array_type, spv::Decoration::ArrayStride, sizeof(u32));
-    const Id struct_type{TypeStruct(array_type)};
-    Decorate(struct_type, spv::Decoration::Block);
-    MemberName(struct_type, 0, "segment_sizes");
-    MemberDecorate(struct_type, 0, spv::Decoration::Offset, 0U);
-    const Id pointer_type{TypePointer(spv::StorageClass::StorageBuffer, struct_type)};
-    storage_buffer_mapping_u32 = TypePointer(spv::StorageClass::StorageBuffer, U32[1]);
-    storage_buffer_mapping = AddGlobalVariable(pointer_type, spv::StorageClass::StorageBuffer);
-    Decorate(storage_buffer_mapping, spv::Decoration::Binding, binding++);
-    Decorate(storage_buffer_mapping, spv::Decoration::DescriptorSet, 0U);
-    Name(storage_buffer_mapping, "storage_buffer_mapping");
-    //Starting with version 1.4... (https://registry.khronos.org/SPIR-V/specs/unified1/SPIRV.html)
-    if (profile.supported_spirv >= 0x00010400) {
-        interfaces.push_back(storage_buffer_mapping);
-    }
-
-    u32 mapping_base{};
-    for (u32 index = 0; index < info.storage_buffers_descriptors.size(); ++index) {
-        const StorageBufferDescriptor& desc = info.storage_buffers_descriptors[index];
-        storage_buffer_mapping_bases[index] = mapping_base;
-        storage_buffer_mapping_counts[index] = desc.count;
-        mapping_base += desc.count;
-    }
-
-    const Id function_type{TypeFunction(U32[2], U32[1], U32[1], U32[1])};
-    storage_buffer_map_func = OpFunction(U32[2], spv::FunctionControlMask::MaskNone, function_type);
-    const Id base{OpFunctionParameter(U32[1])};
-    const Id count{OpFunctionParameter(U32[1])};
-    const Id byte_offset{OpFunctionParameter(U32[1])};
-    const Id index_pointer_type{TypePointer(spv::StorageClass::Function, U32[1])};
-    const Id loop_header{OpLabel()};
-    const Id continue_block{OpLabel()};
-    const Id merge_block{OpLabel()};
-    AddLabel();
-    const Id segment_var{AddLocalVariable(index_pointer_type, spv::StorageClass::Function)};
-    const Id offset_var{AddLocalVariable(index_pointer_type, spv::StorageClass::Function)};
-    OpStore(segment_var, u32_zero_value);
-    OpStore(offset_var, byte_offset);
-    OpBranch(loop_header);
-
-    AddLabel(loop_header);
-    const Id segment{OpLoad(U32[1], segment_var)};
-    const Id local_offset{OpLoad(U32[1], offset_var)};
-    const Id mapping_index{OpIAdd(U32[1], base, segment)};
-    const Id size_pointer{OpAccessChain(storage_buffer_mapping_u32, storage_buffer_mapping, u32_zero_value, mapping_index)};
-    const Id segment_size{OpLoad(U32[1], size_pointer)};
-    const Id fits{OpULessThan(U1, local_offset, segment_size)};
-    const Id last_segment{OpISub(U32[1], count, Const(1U))};
-    const Id is_last{OpIEqual(U1, segment, last_segment)};
-    const Id found{OpLogicalOr(U1, fits, is_last)};
-    OpLoopMerge(merge_block, continue_block, spv::LoopControlMask::MaskNone);
-    OpBranchConditional(found, merge_block, continue_block);
-
-    AddLabel(continue_block);
-    OpStore(offset_var, OpISub(U32[1], local_offset, segment_size));
-    OpStore(segment_var, OpIAdd(U32[1], segment, Const(1U)));
-    OpBranch(loop_header);
-
-    AddLabel(merge_block);
-    OpReturnValue(OpCompositeConstruct(U32[2], segment, local_offset));
-    OpFunctionEnd();
-    Name(storage_buffer_map_func, "map_storage_buffer");
-}
-
 void EmitContext::DefineStorageBuffers(const Info& info, u32& binding) {
    if (info.storage_buffers_descriptors.empty()) {
        return;
@@ -1375,7 +1271,9 @@ void EmitContext::DefineStorageBuffers(const Info& info, u32& binding) {
        DefineSsbos(*this, storage_types.U32x4, &StorageDefinitions::U32x4, info, binding, U32[4],
                    sizeof(u32[4]));
    }
-    binding += static_cast<u32>(info.storage_buffers_descriptors.size());
+    for (const StorageBufferDescriptor& desc : info.storage_buffers_descriptors) {
+        binding += desc.count;
+    }
    const bool needs_function{
        info.uses_global_increment || info.uses_global_decrement || info.uses_atomic_f32_add ||
        info.uses_atomic_f16x2_add || info.uses_atomic_f16x2_min || info.uses_atomic_f16x2_max ||
@@ -1384,31 +1282,40 @@ void EmitContext::DefineStorageBuffers(const Info& info, u32& binding) {
        AddCapability(spv::Capability::VariablePointersStorageBuffer);
    }
    if (info.uses_global_increment) {
-        increment_cas_ssbo = CasLoop(*this, Operation::Increment, storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
+        increment_cas_ssbo = CasLoop(*this, Operation::Increment, storage_types.U32.array,
+                                     storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
    }
    if (info.uses_global_decrement) {
-        decrement_cas_ssbo = CasLoop(*this, Operation::Decrement, storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
+        decrement_cas_ssbo = CasLoop(*this, Operation::Decrement, storage_types.U32.array,
+                                     storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
    }
    if (info.uses_atomic_f32_add) {
-        f32_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.element, F32[1], U32[1], spv::Scope::Device);
+        f32_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.array,
+                              storage_types.U32.element, F32[1], U32[1], spv::Scope::Device);
    }
    if (info.uses_atomic_f16x2_add) {
-        f16x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
+        f16x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.array,
+                                storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
    }
    if (info.uses_atomic_f16x2_min) {
-        f16x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
+        f16x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.array,
+                                storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
    }
    if (info.uses_atomic_f16x2_max) {
-        f16x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
+        f16x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.array,
+                                storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
    }
    if (info.uses_atomic_f32x2_add) {
-        f32x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
+        f32x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.array,
+                                storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
    }
    if (info.uses_atomic_f32x2_min) {
-        f32x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
+        f32x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.array,
+                                storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
    }
    if (info.uses_atomic_f32x2_max) {
-        f32x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
+        f32x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.array,
+                                storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
    }
 }

@@ -114,6 +114,7 @@ struct UniformDefinitions {
 };

 struct StorageTypeDefinition {
+    Id array{};
    Id element{};
 };

@@ -172,8 +173,6 @@ public:
    [[nodiscard]] Id BitOffset8(const IR::Value& offset);
    [[nodiscard]] Id BitOffset16(const IR::Value& offset);

-    [[nodiscard]] Id StoragePointer(u32 binding, Id byte_offset, const StorageTypeDefinition& type_def, u32 element_size, Id StorageDefinitions::*member_ptr);
-
    Id Const(u32 value) {
        return Constant(U32[1], value);
    }
@@ -258,11 +257,6 @@ public:

    std::array<UniformDefinitions, Info::MAX_CBUFS> cbufs{};
    std::array<StorageDefinitions, Info::MAX_SSBOS> ssbos{};
-    std::array<u32, Info::MAX_SSBOS> storage_buffer_mapping_bases{};
-    std::array<u32, Info::MAX_SSBOS> storage_buffer_mapping_counts{};
-    Id storage_buffer_mapping{};
-    Id storage_buffer_mapping_u32{};
-    Id storage_buffer_map_func{};
    std::vector<TextureBufferDefinition> texture_buffers;
    std::vector<ImageBufferDefinition> image_buffers;
    std::vector<TextureDefinition> textures;
@@ -382,7 +376,6 @@ public:
    bool uses_nonuniform_storage_image{};
    bool uses_nonuniform_uniform_texel_buffer{};
    bool uses_nonuniform_storage_texel_buffer{};
-    bool uses_nonuniform_storage_buffer{};

 private:
    void DefineCommonTypes(const Info& info);
@@ -393,7 +386,6 @@ private:
    void DefineSharedMemoryFunctions(const IR::Program& program);
    void DefineConstantBuffers(const Info& info, u32& binding);
    void DefineConstantBufferIndirectFunctions(const Info& info);
-    void DefineStorageBufferMappings(const Info& info, u32& binding);
    void DefineStorageBuffers(const Info& info, u32& binding);
    void DefineTextureBuffers(const Info& info, u32& binding);
    void DefineImageBuffers(const Info& info, u32& binding);
@@ -132,14 +132,6 @@ void AddNVNStorageBuffers(IR::Program& program) {
    }
 }

-void ConfigureStorageBufferMappings(IR::Program& program, const HostTranslateInfo& host_info) {
-    // https://docs.vulkan.org/guide/latest/descriptor_arrays.html
-    // will be needed: descriptor array elements represent physical spans of one guest virtual buffer.
-    for (StorageBufferDescriptor& desc : program.info.storage_buffers_descriptors) {
-        desc.count = host_info.storage_buffer_segment_count;
-    }
-}
-
 using IR::IsLegacyAttribute; //rescoped to attribute.h to make it visible in load_store_attribute.cpp IPA

 std::map<IR::Attribute, IR::Attribute> GenerateLegacyToGenericMappings(
@@ -307,7 +299,6 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
    Optimization::PositionPass(env, program);

    Optimization::GlobalMemoryToStorageBufferPass(program, normalized_host_info);
-    ConfigureStorageBufferMappings(program, normalized_host_info);
    Optimization::TexturePass(env, program, normalized_host_info);

    if (Settings::values.resolution_info.active || Settings::values.rescale_hack.GetValue()) {
@@ -323,7 +314,6 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo

    CollectInterpolationInfo(env, program);
    AddNVNStorageBuffers(program);
-    ConfigureStorageBufferMappings(program, normalized_host_info);
    return program;
 }

@@ -6,7 +6,6 @@

 #pragma once

-#include <algorithm>
 #include "common/common_types.h"

 namespace Shader {
@@ -20,7 +19,6 @@ struct HostTranslateInfo {

    u64 min_ssbo_alignment{};            ///< Minimum alignment supported by the device for SSBOs
    u32 max_per_stage_descriptor_sampled_images{}; ///< maximum sampled descriptors per stage
-    u32 max_per_stage_descriptor_storage_buffers{}; ///< maximum storage descriptors per stage
    u32 max_per_stage_resources{};                 ///< maximum resources per stage
    u32 max_descriptor_set_samplers{};
    u32 max_descriptor_set_uniform_buffers{};
@@ -40,14 +38,12 @@ struct HostTranslateInfo {
                                                ///< passthrough shaders
    bool support_conditional_barrier{}; ///< True when the device supports barriers in conditional
                                        ///< control flow
-    u32 storage_buffer_segment_count{1}; ///< Physical ranges available to each guest SSBO

    void ApplyDescriptorLimitPolicy() noexcept {
        if (min_ssbo_alignment == 0) {
            min_ssbo_alignment = 1;
        }
        ApplyDescriptorLimitFallback(max_per_stage_descriptor_sampled_images);
-        ApplyDescriptorLimitFallback(max_per_stage_descriptor_storage_buffers);
        ApplyDescriptorLimitFallback(max_per_stage_resources);
        ApplyDescriptorLimitFallback(max_descriptor_set_samplers);
        ApplyDescriptorLimitFallback(max_descriptor_set_uniform_buffers);
@@ -57,7 +53,6 @@ struct HostTranslateInfo {
        ApplyDescriptorLimitFallback(max_descriptor_set_sampled_images);
        ApplyDescriptorLimitFallback(max_descriptor_set_storage_images);
        ApplyDescriptorLimitFallback(max_descriptor_set_input_attachements);
-        storage_buffer_segment_count = (std::max)(storage_buffer_segment_count, 1U);
    }

 private:
@@ -64,7 +64,6 @@ struct Profile {
    bool support_storage_image_array_nonuniform_indexing{};
    bool support_uniform_texel_buffer_array_nonuniform_indexing{};
    bool support_storage_texel_buffer_array_nonuniform_indexing{};
-    bool support_storage_buffer_array_nonuniform_indexing{};

    bool warp_size_potentially_larger_than_guest{};

@@ -6,7 +6,6 @@

 #pragma once

-#include <algorithm>
 #include <array>
 #include <bitset>
 #include <map>
@@ -341,10 +340,6 @@ struct Info {
    ImageDescriptors image_descriptors;
 };

-[[nodiscard]] inline bool UsesStorageBufferMappings(const Info& info) noexcept {
-    return std::ranges::any_of(info.storage_buffers_descriptors, [](const auto& desc) { return desc.count > 1; });
-}
-
 template <typename Descriptors>
 u32 NumDescriptors(const Descriptors& descriptors) {
    u32 num{};
@@ -20,8 +20,6 @@ namespace VideoCommon {

 enum class BufferFlagBits {
    Picked = 1 << 0,
-    CachedWrites = 1 << 1,
-    PreemtiveDownload = 1 << 2,
 };
 DECLARE_ENUM_FLAG_OPERATORS(BufferFlagBits)

@@ -58,15 +56,6 @@ public:
        flags |= BufferFlagBits::Picked;
    }

-    void MarkPreemtiveDownload() noexcept {
-        flags |= BufferFlagBits::PreemtiveDownload;
-    }
-
-    /// Unmark buffer as picked
-    void Unpick() noexcept {
-        flags &= ~BufferFlagBits::Picked;
-    }
-
    /// Increases the likeliness of this being a stream buffer
    void IncreaseStreamScore(int score) noexcept {
        stream_score += score;
@@ -87,15 +76,6 @@ public:
        return True(flags & BufferFlagBits::Picked);
    }

-    /// Returns true when the buffer has pending cached writes
-    [[nodiscard]] bool HasCachedWrites() const noexcept {
-        return True(flags & BufferFlagBits::CachedWrites);
-    }
-
-    bool IsPreemtiveDownload() const noexcept {
-        return True(flags & BufferFlagBits::PreemtiveDownload);
-    }
-
    /// Returns the base CPU address of the buffer
    [[nodiscard]] VAddr CpuAddr() const noexcept {
        return cpu_addr;
@@ -7,7 +7,6 @@
 #pragma once

 #include <algorithm>
-#include <cstring>
 #include <limits>
 #include <memory>
 #include <numeric>
@@ -123,25 +122,6 @@ void BufferCache<P>::WriteMemory(DAddr device_addr, u64 size) {
    memory_tracker.MarkRegionAsCpuModified(device_addr, size);
 }

-template <class P>
-void BufferCache<P>::CachedWriteMemory(DAddr device_addr, u64 size) {
-    const bool is_dirty = IsRegionRegistered(device_addr, size);
-    if (!is_dirty) {
-        return;
-    }
-    DAddr aligned_start = Common::AlignDown(device_addr, DEVICE_PAGESIZE);
-    DAddr aligned_end = Common::AlignUp(device_addr + size, DEVICE_PAGESIZE);
-    if (!IsRegionGpuModified(aligned_start, aligned_end - aligned_start)) {
-        WriteMemory(device_addr, size);
-        return;
-    }
-
-    tmp_buffer.resize_destructive(size);
-    device_memory.ReadBlockUnsafe(device_addr, tmp_buffer.data(), size);
-
-    InlineMemoryImplementation(device_addr, size, tmp_buffer);
-}
-
 template <class P>
 bool BufferCache<P>::OnCPUWrite(DAddr device_addr, u64 size) {
    const bool is_dirty = IsRegionRegistered(device_addr, size);
@@ -424,8 +404,8 @@ void BufferCache<P>::UnbindGraphicsStorageBuffers(size_t stage) {
 }

 template <class P>
-bool BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index,
-                                               u32 cbuf_offset, bool is_written, u32 descriptor_count) {
+void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index,
+                                               u32 cbuf_offset, bool is_written) {
    const bool already_enabled =
        ((channel_state->enabled_storage_buffers[stage] >> ssbo_index) & 1U) != 0;
    if constexpr (requires { runtime.ShouldLimitDynamicStorageBuffers(); }) {
@@ -435,7 +415,7 @@ bool BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index,
                LOG_WARNING(HW_GPU,
                            "Skipping graphics storage buffer {} due to driver limit {}",
                            ssbo_index, max_bindings);
-                return false;
+                return;
            }
        }
    }
@@ -450,8 +430,7 @@ bool BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index,
    const auto& cbufs = maxwell3d->state.shader_stages[stage];
    const GPUVAddr ssbo_addr = cbufs.const_buffers[cbuf_index].address + cbuf_offset;
    channel_state->storage_buffers[stage][ssbo_index] =
-        StorageBufferBinding(ssbo_addr, cbuf_index, is_written, descriptor_count);
-    return channel_state->storage_buffers[stage][ssbo_index].gpu_addr != 0;
+        StorageBufferBinding(ssbo_addr, cbuf_index, is_written);
 }

 template <class P>
@@ -489,7 +468,7 @@ void BufferCache<P>::UnbindComputeStorageBuffers() {

 template <class P>
 void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
-                                              bool is_written, u32 descriptor_count) {
+                                              bool is_written) {
    if (ssbo_index >= channel_state->compute_storage_buffers.size()) [[unlikely]] {
        LOG_ERROR(HW_GPU, "Storage buffer index {} exceeds maximum storage buffer count",
                  ssbo_index);
@@ -526,7 +505,7 @@ void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index,
    const auto& cbufs = launch_desc.const_buffer_config;
    const GPUVAddr ssbo_addr = cbufs[cbuf_index].Address() + cbuf_offset;
    channel_state->compute_storage_buffers[ssbo_index] =
-        StorageBufferBinding(ssbo_addr, cbuf_index, is_written, descriptor_count);
+        StorageBufferBinding(ssbo_addr, cbuf_index, is_written);
 }

 template <class P>
@@ -764,16 +743,6 @@ void BufferCache<P>::BindHostIndexBuffer() {
    }
 }

-template <class P>
-void BufferCache<P>::BindHostVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size,
-                                          u32 stride) {
-    if constexpr (IS_OPENGL) {
-        runtime.BindVertexBuffer(index, buffer, offset, size, stride);
-    } else {
-        runtime.BindVertexBuffer(index, buffer.Handle(), offset, size, stride);
-    }
-}
-
 template <class P>
 Binding& BufferCache<P>::VertexBufferSlot(u32 index) {
    ASSERT(index < NUM_VERTEX_BUFFERS);
@@ -1002,52 +971,27 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32

 template <class P>
 void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
-    boost::container::small_vector<u32, NUM_STORAGE_BUFFERS> segment_sizes;
-    bool uses_mapping{};
-    ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
-        const StorageBufferBindingInfo& binding = channel_state->storage_buffers[stage][index];
-        uses_mapping |= binding.descriptor_count > 1;
-        for (u32 segment = 0; segment < binding.descriptor_count; ++segment) {
-            segment_sizes.push_back(segment < binding.segments.size() ? binding.segments[segment].size : 0);
-        }
-    });
-    if (uses_mapping) {
-        const u32 mapping_size = static_cast<u32>(segment_sizes.size() * sizeof(u32));
-        if constexpr (!IS_OPENGL) {
-            const std::span<u8> mapped = runtime.BindMappedStorageBuffer(mapping_size);
-            std::memcpy(mapped.data(), segment_sizes.data(), mapping_size);
-        }
-    }
-
    u32 binding_index = 0;
    ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
-        const StorageBufferBindingInfo& storage = channel_state->storage_buffers[stage][index];
+        const Binding& binding = channel_state->storage_buffers[stage][index];
+        Buffer& buffer = slot_buffers[binding.buffer_id];
+        TouchBuffer(buffer, binding.buffer_id);
+        const u32 size = binding.size;
+        SynchronizeBuffer(buffer, binding.device_addr, size);
+
+        const u32 offset = buffer.Offset(binding.device_addr);
+        buffer.MarkUsage(offset, size);
        const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
-        for (u32 segment = 0; segment < storage.descriptor_count; ++segment) {
-            Buffer* buffer = &slot_buffers[NULL_BUFFER_ID];
-            u32 offset{};
-            u32 size{IS_OPENGL ? 0U : static_cast<u32>(sizeof(u32))};
-            const bool is_actual_segment = segment < storage.segments.size();
-            // shall be safe enough if the segment is not actual, use the last available segment or nullptr if none exist.
-            const Binding* binding = is_actual_segment ? &storage.segments[segment] : storage.segments.empty() ? nullptr : &storage.segments.back();
-            if (binding) {
-                buffer = &slot_buffers[binding->buffer_id];
-                size = binding->size;
-                offset = buffer->Offset(binding->device_addr);
-                if (is_actual_segment) {
-                    TouchBuffer(*buffer, binding->buffer_id);
-                    SynchronizeBuffer(*buffer, binding->device_addr, size);
-                    buffer->MarkUsage(offset, size);
-                    if (is_written) {
-                        MarkWrittenBuffer(binding->buffer_id, binding->device_addr, size);
-                    }
-                }
-            }
-            if constexpr (NEEDS_BIND_STORAGE_INDEX) {
-                runtime.BindStorageBuffer(stage, binding_index++, *buffer, offset, size, is_written);
-            } else {
-                runtime.BindStorageBuffer(*buffer, offset, size, is_written);
-            }
+
+        if (is_written) {
+            MarkWrittenBuffer(binding.buffer_id, binding.device_addr, size);
+        }
+
+        if constexpr (NEEDS_BIND_STORAGE_INDEX) {
+            runtime.BindStorageBuffer(stage, binding_index, buffer, offset, size, is_written);
+            ++binding_index;
+        } else {
+            runtime.BindStorageBuffer(buffer, offset, size, is_written);
        }
    });
 }
@@ -1163,53 +1107,28 @@ void BufferCache<P>::BindHostComputeUniformBuffers() {

 template <class P>
 void BufferCache<P>::BindHostComputeStorageBuffers() {
-    boost::container::small_vector<u32, NUM_STORAGE_BUFFERS> segment_sizes;
-    bool uses_mapping{};
-    ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
-        const StorageBufferBindingInfo& binding = channel_state->compute_storage_buffers[index];
-        uses_mapping |= binding.descriptor_count > 1;
-        for (u32 segment = 0; segment < binding.descriptor_count; ++segment) {
-            segment_sizes.push_back(segment < binding.segments.size() ? binding.segments[segment].size : 0);
-        }
-    });
-    if (uses_mapping) {
-        const u32 mapping_size = static_cast<u32>(segment_sizes.size() * sizeof(u32));
-        if constexpr (!IS_OPENGL) {
-            const std::span<u8> mapped = runtime.BindMappedStorageBuffer(mapping_size);
-            std::memcpy(mapped.data(), segment_sizes.data(), mapping_size);
-        }
-    }
-
    u32 binding_index = 0;
    ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
-        const StorageBufferBindingInfo& storage = channel_state->compute_storage_buffers[index];
+        const Binding& binding = channel_state->compute_storage_buffers[index];
+        Buffer& buffer = slot_buffers[binding.buffer_id];
+        TouchBuffer(buffer, binding.buffer_id);
+        const u32 size = binding.size;
+        SynchronizeBuffer(buffer, binding.device_addr, size);
+
+        const u32 offset = buffer.Offset(binding.device_addr);
+        buffer.MarkUsage(offset, size);
        const bool is_written =
            ((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
-        for (u32 segment = 0; segment < storage.descriptor_count; ++segment) {
-            Buffer* buffer = &slot_buffers[NULL_BUFFER_ID];
-            u32 offset{};
-            u32 size{IS_OPENGL ? 0U : static_cast<u32>(sizeof(u32))};
-            const bool is_actual_segment = segment < storage.segments.size();
-            //same fallback logic
-            const Binding* binding = is_actual_segment ? &storage.segments[segment] : storage.segments.empty() ? nullptr : &storage.segments.back();
-            if (binding) {
-                buffer = &slot_buffers[binding->buffer_id];
-                size = binding->size;
-                offset = buffer->Offset(binding->device_addr);
-                if (is_actual_segment) {
-                    TouchBuffer(*buffer, binding->buffer_id);
-                    SynchronizeBuffer(*buffer, binding->device_addr, size);
-                    buffer->MarkUsage(offset, size);
-                    if (is_written) {
-                        MarkWrittenBuffer(binding->buffer_id, binding->device_addr, size);
-                    }
-                }
-            }
-            if constexpr (NEEDS_BIND_STORAGE_INDEX) {
-                runtime.BindComputeStorageBuffer(binding_index++, *buffer, offset, size, is_written);
-            } else {
-                runtime.BindStorageBuffer(*buffer, offset, size, is_written);
-            }
+
+        if (is_written) {
+            MarkWrittenBuffer(binding.buffer_id, binding.device_addr, size);
+        }
+
+        if constexpr (NEEDS_BIND_STORAGE_INDEX) {
+            runtime.BindComputeStorageBuffer(binding_index, buffer, offset, size, is_written);
+            ++binding_index;
+        } else {
+            runtime.BindStorageBuffer(buffer, offset, size, is_written);
        }
    });
 }
@@ -1249,7 +1168,7 @@ void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) {
        if (is_indexed) {
            UpdateIndexBuffer();
        }
-        UpdateVertexBuffers();
+        UpdateVertexBuffers(is_indexed);
        UpdateTransformFeedbackBuffers();
        for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
            UpdateUniformBuffers(stage);
@@ -1303,9 +1222,14 @@ void BufferCache<P>::UpdateIndexBuffer() {
    const GPUVAddr gpu_addr_begin = index_buffer_ref.StartAddress();
    const GPUVAddr gpu_addr_end = index_buffer_ref.EndAddress();
    const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(gpu_addr_begin);
-    const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin);
-    const u32 draw_size = (index_buffer_ref.count + index_buffer_ref.first) * u32(index_buffer_ref.FormatSizeInBytes());
-    const u32 size = (std::min)(address_size, draw_size);
+    u64 address_size = 0;
+    if (gpu_addr_end > gpu_addr_begin) {
+        address_size = (std::min)(gpu_addr_end - gpu_addr_begin,
+                                  u64{(std::numeric_limits<u32>::max)()});
+    }
+    const u64 draw_size = (u64{index_buffer_ref.count} + u64{index_buffer_ref.first}) *
+                          u64{index_buffer_ref.FormatSizeInBytes()};
+    const u32 size = static_cast<u32>((std::min)(address_size, draw_size));
    if (size == 0 || !device_addr) {
        channel_state->index_buffer = NULL_BINDING;
        return;
@@ -1318,20 +1242,142 @@ void BufferCache<P>::UpdateIndexBuffer() {
 }

 template <class P>
-void BufferCache<P>::UpdateVertexBuffers() {
+u64 BufferCache<P>::DrawMaxIndex() {
+    if (max_index_scanned) {
+        return cached_max_index;
+    }
+    max_index_scanned = true;
+    cached_max_index = 0;
+    const auto& index_buffer_ref = maxwell3d->draw_manager.draw_state.index_buffer;
+    const u32 count = index_buffer_ref.count;
+    if (count == 0) {
+        return 0;
+    }
+    index_scan_buffer.resize_destructive(count);
+    gpu_memory->ReadBlockUnsafe(index_buffer_ref.IndexStart(), index_scan_buffer.data(),
+                                size_t{count} * sizeof(u32));
+    u32 restart_index = (std::numeric_limits<u32>::max)();
+    if (maxwell3d->regs.primitive_restart.enabled != 0) {
+        restart_index = maxwell3d->regs.primitive_restart.index;
+    }
+    u32 max_index = 0;
+    for (u32 i = 0; i < count; ++i) {
+        const u32 value = index_scan_buffer[i];
+        if (value == restart_index) {
+            continue;
+        }
+        max_index = (std::max)(max_index, value);
+    }
+    cached_max_index = max_index;
+    return cached_max_index;
+}
+
+template <class P>
+u64 BufferCache<P>::StreamAttributeExtent(u32 index) {
+    using VertexAttribute = typename Maxwell::VertexAttribute;
+    if (stream_extents_valid) {
+        return stream_extents[index];
+    }
+    stream_extents_valid = true;
+    stream_extents.fill(0);
+    for (size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) {
+        const auto& attribute = maxwell3d->regs.vertex_attrib_format[i];
+        if (attribute.constant != 0 || attribute.size == VertexAttribute::Size::Invalid) {
+            continue;
+        }
+        const u32 buffer = attribute.buffer.Value();
+        if (buffer >= NUM_VERTEX_BUFFERS) {
+            continue;
+        }
+        const u64 end = static_cast<u64>(attribute.offset.Value()) +
+                        static_cast<u64>(attribute.SizeInBytes());
+        stream_extents[buffer] = (std::max)(stream_extents[buffer], end);
+    }
+    return stream_extents[index];
+}
+
+template <class P>
+u64 BufferCache<P>::DrawVertexBound(u32 index, bool is_indexed) {
+    const auto& array = maxwell3d->regs.vertex_streams[index];
+    if (array.enable == 0) {
+        return 0;
+    }
+    const u64 extent = StreamAttributeExtent(index);
+    if (extent == 0) {
+        return 0;
+    }
+    const u64 stride = static_cast<u64>(array.stride);
+    if (stride == 0) {
+        return extent;
+    }
+    const auto& draw_state = maxwell3d->draw_manager.draw_state;
+    u64 elements = 0;
+    if (maxwell3d->regs.vertex_stream_instances.IsInstancingEnabled(index)) {
+        if (draw_instance_count == 0) {
+            return 0;
+        }
+        const u64 base_instance = static_cast<u64>(draw_state.base_instance);
+        elements = base_instance + 1;
+        if (array.frequency != 0) {
+            elements = (base_instance + static_cast<u64>(draw_instance_count) - 1) /
+                           static_cast<u64>(array.frequency) +
+                       1;
+        }
+    } else if (!is_indexed) {
+        elements = static_cast<u64>(draw_state.vertex_buffer.first) +
+                   static_cast<u64>(draw_state.vertex_buffer.count);
+    } else {
+        const auto format = draw_state.index_buffer.format;
+        u64 max_index = 0xFF;
+        if (format == Maxwell::IndexFormat::UnsignedShort) {
+            max_index = 0xFFFF;
+        } else if (format != Maxwell::IndexFormat::UnsignedByte) {
+            const auto& limit = maxwell3d->regs.vertex_stream_limits[index];
+            const GPUVAddr gpu_addr_begin = array.Address();
+            const GPUVAddr gpu_addr_end = limit.Address() + 1;
+            if (gpu_addr_end <= gpu_addr_begin) {
+                return 0;
+            }
+            const bool walks = gpu_addr_end - gpu_addr_begin >= IMPLAUSIBLE_VERTEX_SIZE ||
+                               !gpu_memory->IsWithinGPUAddressRange(gpu_addr_end);
+            if (!walks) {
+                return 0;
+            }
+            max_index = DrawMaxIndex();
+        }
+        elements = static_cast<u64>(draw_state.base_index) + max_index + 1;
+    }
+    if (elements == 0) {
+        return extent;
+    }
+    return (elements - 1) * stride + extent;
+}
+
+template <class P>
+void BufferCache<P>::UpdateVertexBuffers(bool is_indexed) {
    auto& flags = maxwell3d->dirty.flags;
+    max_index_scanned = false;
+    stream_extents_valid = false;
+    for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
+        const u64 bound = DrawVertexBound(index, is_indexed);
+        if (bound <= last_draw_bounds[index]) {
+            continue;
+        }
+        flags[Dirty::VertexBuffer0 + index] = true;
+        flags[Dirty::VertexBuffers] = true;
+    }
    if (!maxwell3d->dirty.flags[Dirty::VertexBuffers]) {
        return;
    }
    flags[Dirty::VertexBuffers] = false;

    for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
-        UpdateVertexBuffer(index);
+        UpdateVertexBuffer(index, is_indexed);
    }
 }

 template <class P>
-void BufferCache<P>::UpdateVertexBuffer(u32 index) {
+void BufferCache<P>::UpdateVertexBuffer(u32 index, bool is_indexed) {
    if (!maxwell3d->dirty.flags[Dirty::VertexBuffer0 + index]) {
        return;
    }
@@ -1340,15 +1386,31 @@ void BufferCache<P>::UpdateVertexBuffer(u32 index) {
    const GPUVAddr gpu_addr_begin = array.Address();
    const GPUVAddr gpu_addr_end = limit.Address() + 1;
    const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(gpu_addr_begin);
-    const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin);
-    u32 size = address_size; // TODO: Analyze stride and number of vertices
-    if (array.enable == 0 || size == 0 || !device_addr) {
+    if (array.enable == 0 || !device_addr || gpu_addr_end <= gpu_addr_begin) {
        channel_state->vertex_buffers[index] = NULL_BINDING;
        UpdateVertexBufferSlot(index, NULL_BINDING);
        return;
    }
-    if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end) || size >= 64_MiB) {
-        size = static_cast<u32>(gpu_memory->MaxContinuousRange(gpu_addr_begin, size));
+    // TODO: Analyze stride and number of vertices
+    constexpr u64 implausible_size = IMPLAUSIBLE_VERTEX_SIZE;
+    u64 address_size = gpu_addr_end - gpu_addr_begin;
+    if (address_size > u64{(std::numeric_limits<u32>::max)()}) {
+        address_size = implausible_size;
+    }
+    const u64 draw_bound = DrawVertexBound(index, is_indexed);
+    last_draw_bounds[index] = (std::numeric_limits<u64>::max)();
+    if (draw_bound != 0) {
+        last_draw_bounds[index] = draw_bound;
+        address_size = (std::min)(address_size, draw_bound);
+    }
+    if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end) || address_size >= implausible_size) {
+        address_size = gpu_memory->MaxContinuousRange(gpu_addr_begin, address_size);
+    }
+    const u32 size = static_cast<u32>(address_size);
+    if (size == 0) {
+        channel_state->vertex_buffers[index] = NULL_BINDING;
+        UpdateVertexBufferSlot(index, NULL_BINDING);
+        return;
    }
    const BufferId buffer_id = FindBuffer(*device_addr, size);
    const Binding binding{
@@ -1402,7 +1464,10 @@ void BufferCache<P>::UpdateUniformBuffers(size_t stage) {
 template <class P>
 void BufferCache<P>::UpdateStorageBuffers(size_t stage) {
    ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) {
-        UpdateStorageBuffer(channel_state->storage_buffers[stage][index]);
+        // Resolve buffer
+        Binding& binding = channel_state->storage_buffers[stage][index];
+        const BufferId buffer_id = FindBuffer(binding.device_addr, binding.size);
+        binding.buffer_id = buffer_id;
    });
 }

@@ -1463,54 +1528,12 @@ void BufferCache<P>::UpdateComputeUniformBuffers() {
 template <class P>
 void BufferCache<P>::UpdateComputeStorageBuffers() {
    ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
-        UpdateStorageBuffer(channel_state->compute_storage_buffers[index]);
+        // Resolve buffer
+        Binding& binding = channel_state->compute_storage_buffers[index];
+        binding.buffer_id = FindBuffer(binding.device_addr, binding.size);
    });
 }

-template <class P>
-void BufferCache<P>::UpdateStorageBuffer(StorageBufferBindingInfo& binding) {
-    binding.segments.clear();
-    if (binding.gpu_addr == 0 || binding.size == 0) { return;}
-    if (binding.descriptor_count == 1) {
-        // for safety gotta preserve the legacy path on possible hosts without storage-buffer descriptor indexing.
-        const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(binding.gpu_addr);
-        if (device_addr) {
-            binding.segments.push_back(Binding{
-                .device_addr = *device_addr,
-                .size = binding.size,
-                .buffer_id = FindBuffer(*device_addr, binding.size),
-            });
-        }
-        return;
-    }
-
-    const auto ranges = gpu_memory->GetSubmappedRange(binding.gpu_addr, binding.size);
-    const size_t mapped_size =
-        std::accumulate(ranges.begin(), ranges.end(), size_t{}, [](size_t total, const auto& range) { return total + range.second; });
-    if (mapped_size != binding.size) {
-        LOG_ERROR(HW_GPU, "Storage buffer range {:#x}+{:#x} is not fully mapped", binding.gpu_addr, binding.size);
-        return;
-    }
-    if (ranges.size() > binding.descriptor_count) {
-        LOG_ERROR(HW_GPU, "Storage buffer range {:#x}+{:#x} has {} physical segments, exceeding host capacity {}",
-                  binding.gpu_addr, binding.size, ranges.size(), binding.descriptor_count);
-        return;
-    }
-    for (const auto& [gpu_addr, size] : ranges) {
-        const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
-        if (!device_addr || size > (std::numeric_limits<u32>::max)()) {
-            binding.segments.clear();
-            return;
-        }
-        const u32 segment_size = static_cast<u32>(size);
-        binding.segments.push_back(Binding{
-            .device_addr = *device_addr,
-            .size = segment_size,
-            .buffer_id = FindBuffer(*device_addr, segment_size),
-        });
-    }
-}
-
 template <class P>
 void BufferCache<P>::UpdateComputeTextureBuffers() {
    ForEachEnabledBit(channel_state->enabled_compute_texture_buffers, [&](u32 index) {
@@ -1669,9 +1692,10 @@ template <class P>
 BufferId BufferCache<P>::CreateBuffer(DAddr device_addr, u32 wanted_size) {
    DAddr device_addr_end = Common::AlignUp(device_addr + wanted_size, CACHING_PAGESIZE);
    device_addr = Common::AlignDown(device_addr, CACHING_PAGESIZE);
-    wanted_size = static_cast<u32>(device_addr_end - device_addr);
+    constexpr u64 max_buffer_size = u64{(std::numeric_limits<u32>::max)()};
+    wanted_size = static_cast<u32>((std::min)(device_addr_end - device_addr, max_buffer_size));
    const OverlapResult overlap = ResolveOverlaps(device_addr, wanted_size);
-    const u32 size = static_cast<u32>(overlap.end - overlap.begin);
+    const u32 size = static_cast<u32>((std::min)(overlap.end - overlap.begin, max_buffer_size));
    const BufferId new_buffer_id = slot_buffers.insert(runtime, overlap.begin, size);
    auto& new_buffer = slot_buffers[new_buffer_id];
    const size_t size_bytes = new_buffer.SizeBytes();
@@ -1932,11 +1956,6 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
    const auto replace = [scalar_replace](std::span<Binding> bindings) {
        std::ranges::for_each(bindings, scalar_replace);
    };
-    const auto storage_replace = [scalar_replace](std::span<StorageBufferBindingInfo> bindings) {
-        for (StorageBufferBindingInfo& binding : bindings) {
-            std::ranges::for_each(binding.segments, scalar_replace);
-        }
-    };

    if (channel_state->index_buffer.buffer_id == buffer_id) {
        channel_state->index_buffer.buffer_id = BufferId{};
@@ -1952,10 +1971,10 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
        }
    }
    std::ranges::for_each(channel_state->uniform_buffers, replace);
-    std::ranges::for_each(channel_state->storage_buffers, storage_replace);
+    std::ranges::for_each(channel_state->storage_buffers, replace);
    replace(channel_state->transform_feedback_buffers);
    replace(channel_state->compute_uniform_buffers);
-    storage_replace(channel_state->compute_storage_buffers);
+    replace(channel_state->compute_storage_buffers);

    // Mark the whole buffer as CPU written to stop tracking CPU writes
    if (!do_not_mark) {
@@ -1992,14 +2011,12 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
 }

 template <class P>
-StorageBufferBindingInfo BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
-                                                              bool is_written, u32 descriptor_count) const {
-    // time to get rid of these null bindings
-    ASSERT(descriptor_count > 0); // shant happen
+Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
+                                             bool is_written) const {
    const GPUVAddr gpu_addr = gpu_memory->Read<u64>(ssbo_addr);

    if (gpu_addr == 0) {
-        return {.descriptor_count = descriptor_count};
+        return NULL_BINDING;
    }

    const auto size = [&]() {
@@ -2021,17 +2038,21 @@ StorageBufferBindingInfo BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr
    const GPUVAddr aligned_gpu_addr = Common::AlignDown(gpu_addr, alignment);
    const u32 aligned_size = static_cast<u32>(gpu_addr - aligned_gpu_addr) + size;

-    if (!gpu_memory->GpuToCpuAddress(aligned_gpu_addr) || size == 0) {
+    const std::optional<DAddr> aligned_device_addr = gpu_memory->GpuToCpuAddress(aligned_gpu_addr);
+    if (!aligned_device_addr || size == 0) {
        LOG_DEBUG(HW_GPU, "Failed to find storage buffer for cbuf index {}", cbuf_index);
-        return {.descriptor_count = descriptor_count};
+        return NULL_BINDING;
    }
+    const std::optional<DAddr> device_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
+    ASSERT_MSG(device_addr, "Unaligned storage buffer address not found for cbuf index {}",
+               cbuf_index);
    // The end address used for size calculation does not need to be aligned
-    const GPUVAddr gpu_end = Common::AlignUp(gpu_addr + size, Core::DEVICE_PAGESIZE);
+    const DAddr cpu_end = Common::AlignUp(*device_addr + size, Core::DEVICE_PAGESIZE);

-    const StorageBufferBindingInfo binding{
-        .gpu_addr = aligned_gpu_addr,
-        .size = is_written ? aligned_size : static_cast<u32>(gpu_end - aligned_gpu_addr),
-        .descriptor_count = descriptor_count,
+    const Binding binding{
+        .device_addr = *aligned_device_addr,
+        .size = is_written ? aligned_size : static_cast<u32>(cpu_end - *aligned_device_addr),
+        .buffer_id = BufferId{},
    };
    return binding;
 }
@@ -51,6 +51,7 @@ constexpr u32 NUM_VERTEX_BUFFERS = 16;
 #else
 constexpr u32 NUM_VERTEX_BUFFERS = 32;
 #endif
+constexpr u64 IMPLAUSIBLE_VERTEX_SIZE = 64_MiB;
 constexpr u32 NUM_TRANSFORM_FEEDBACK_BUFFERS = 4;
 constexpr u32 NUM_GRAPHICS_UNIFORM_BUFFERS = 18;
 constexpr u32 NUM_COMPUTE_UNIFORM_BUFFERS = 8;
@@ -89,15 +90,6 @@ struct TextureBufferBinding : Binding {
    PixelFormat format;
 };

-struct StorageBufferBindingInfo {
-    // another good one: guest SSBO is a virtual interval and may span discontiguous device-memory ranges.
-    // exact case of missing character frames (high sample lane)
-    GPUVAddr gpu_addr{};
-    u32 size{};
-    u32 descriptor_count{1};
-    boost::container::small_vector<Binding, 1> segments;
-};
-
 static constexpr Binding NULL_BINDING{
    .device_addr = 0,
    .size = 0,
@@ -124,15 +116,14 @@ public:
    Binding index_buffer;
    std::array<Binding, NUM_VERTEX_BUFFERS> vertex_buffers;
    std::array<std::array<Binding, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES> uniform_buffers;
-    std::array<std::array<StorageBufferBindingInfo, NUM_STORAGE_BUFFERS>, NUM_STAGES>
-        storage_buffers;
+    std::array<std::array<Binding, NUM_STORAGE_BUFFERS>, NUM_STAGES> storage_buffers;
    std::array<std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS>, NUM_STAGES> texture_buffers;
    std::array<Binding, NUM_TRANSFORM_FEEDBACK_BUFFERS> transform_feedback_buffers;
    Binding count_buffer_binding;
    Binding indirect_buffer_binding;

    std::array<Binding, NUM_COMPUTE_UNIFORM_BUFFERS> compute_uniform_buffers;
-    std::array<StorageBufferBindingInfo, NUM_STORAGE_BUFFERS> compute_storage_buffers;
+    std::array<Binding, NUM_STORAGE_BUFFERS> compute_storage_buffers;
    std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS> compute_texture_buffers;

    std::array<u32, NUM_STAGES> enabled_uniform_buffer_masks{};
@@ -227,8 +218,6 @@ public:

    void WriteMemory(DAddr device_addr, u64 size);

-    void CachedWriteMemory(DAddr device_addr, u64 size);
-
    bool OnCPUWrite(DAddr device_addr, u64 size);

    void DownloadMemory(DAddr device_addr, u64 size);
@@ -258,8 +247,8 @@ public:

    void UnbindGraphicsStorageBuffers(size_t stage);

-    bool BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
-                                   bool is_written, u32 descriptor_count = 1);
+    void BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
+                                   bool is_written);

    void UnbindGraphicsTextureBuffers(size_t stage);

@@ -269,7 +258,7 @@ public:
    void UnbindComputeStorageBuffers();

    void BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
-                                  bool is_written, u32 descriptor_count = 1);
+                                  bool is_written);

    void UnbindComputeTextureBuffers();

@@ -319,6 +308,10 @@ public:
        current_draw_indirect = current_draw_indirect_;
    }

+    void SetDrawInstanceCount(u32 draw_instance_count_) {
+        draw_instance_count = draw_instance_count_;
+    }
+
    [[nodiscard]] std::pair<Buffer*, u32> GetDrawIndirectCount();

    [[nodiscard]] std::pair<Buffer*, u32> GetDrawIndirectBuffer();
@@ -386,8 +379,6 @@ private:

    void BindHostTransformFeedbackBuffers();

-    void BindHostVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride);
-
    void BindHostComputeUniformBuffers();

    void BindHostComputeStorageBuffers();
@@ -400,9 +391,15 @@ private:

    void UpdateIndexBuffer();

-    void UpdateVertexBuffers();
+    void UpdateVertexBuffers(bool is_indexed);

-    void UpdateVertexBuffer(u32 index);
+    void UpdateVertexBuffer(u32 index, bool is_indexed);
+
+    [[nodiscard]] u64 DrawVertexBound(u32 index, bool is_indexed);
+
+    [[nodiscard]] u64 DrawMaxIndex();
+
+    [[nodiscard]] u64 StreamAttributeExtent(u32 index);

    void UpdateDrawIndirect();

@@ -410,8 +407,6 @@ private:

    void UpdateStorageBuffers(size_t stage);

-    void UpdateStorageBuffer(StorageBufferBindingInfo& binding);
-
    void UpdateTextureBuffers(size_t stage);

    void UpdateTransformFeedbackBuffers();
@@ -461,9 +456,8 @@ private:

    void DeleteBuffer(BufferId buffer_id, bool do_not_mark = false);

-    [[nodiscard]] StorageBufferBindingInfo StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
-                                                                bool is_written,
-                                                                u32 descriptor_count) const;
+    [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
+                                               bool is_written) const;

    [[nodiscard]] TextureBufferBinding GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size,
                                                               PixelFormat format);
@@ -497,6 +491,14 @@ private:

    const Tegra::Engines::Maxwell3D::DrawManager::IndirectParams* current_draw_indirect{};

+    u32 draw_instance_count = 0;
+    std::array<u64, NUM_VERTEX_BUFFERS> last_draw_bounds{};
+    Common::ScratchBuffer<u32> index_scan_buffer;
+    u64 cached_max_index = 0;
+    bool max_index_scanned = false;
+    std::array<u64, NUM_VERTEX_BUFFERS> stream_extents{};
+    bool stream_extents_valid = false;
+
    u32 last_index_count = 0;

    u32 enabled_vertex_buffers_mask = 0;
@@ -206,6 +206,40 @@ foreach(VARIANT IN ITEMS ${SHADER_TYPE_VARIANTS})
    set(SHADER_HEADERS ${SHADER_HEADERS} ${VARIANT_HEADER_FILE})
 endforeach()

+set(SHADER_DEFINE_VARIANTS
+    "block_linear_unswizzle_2d.comp|nonarrow|HAS_EXTENDED_TYPES=0"
+    "pitch_unswizzle.comp|nonarrow|HAS_EXTENDED_TYPES=0"
+    "block_linear_unswizzle_3d.comp|nonarrow|HAS_EXTENDED_TYPES=0"
+)
+
+foreach(VARIANT IN ITEMS ${SHADER_DEFINE_VARIANTS})
+    string(REPLACE "|" ";" VARIANT_PARTS ${VARIANT})
+    list(GET VARIANT_PARTS 0 VARIANT_FILENAME)
+    list(GET VARIANT_PARTS 1 VARIANT_SUFFIX)
+    list(GET VARIANT_PARTS 2 VARIANT_DEFINE)
+
+    set(VARIANT_SOURCE ${CMAKE_CURRENT_SOURCE_DIR}/${VARIANT_FILENAME})
+    get_filename_component(VARIANT_STEM ${VARIANT_FILENAME} NAME_WE)
+    get_filename_component(VARIANT_EXT ${VARIANT_FILENAME} EXT)
+    string(REPLACE "." "" VARIANT_EXT ${VARIANT_EXT})
+    set(VARIANT_NAME ${VARIANT_STEM}_${VARIANT_SUFFIX}_${VARIANT_EXT})
+
+    string(TOUPPER ${VARIANT_NAME}_SPV VARIANT_VARIABLE_NAME)
+    set(VARIANT_HEADER_FILE ${SHADER_DIR}/${VARIANT_NAME}_spv.h)
+    add_custom_command(
+        OUTPUT
+            ${VARIANT_HEADER_FILE}
+        COMMAND
+            ${GLSLANGVALIDATOR} -V ${QUIET_FLAG} -I"${FIDELITYFX_INCLUDE_DIR}" ${GLSL_FLAGS}
+            -D${VARIANT_DEFINE}
+            --variable-name ${VARIANT_VARIABLE_NAME} -o ${VARIANT_HEADER_FILE} ${VARIANT_SOURCE}
+            --target-env ${SPIR_V_VERSION}
+        MAIN_DEPENDENCY
+            ${VARIANT_SOURCE}
+    )
+    set(SHADER_HEADERS ${SHADER_HEADERS} ${VARIANT_HEADER_FILE})
+endforeach()
+
 foreach(FILEPATH IN ITEMS ${FIDELITYFX_FILES})
    get_filename_component(FILENAME ${FILEPATH} NAME)
    string(REPLACE "." "_" HEADER_NAME ${FILENAME})
@@ -5,9 +5,13 @@

 #ifdef VULKAN

+#ifndef HAS_EXTENDED_TYPES
+#define HAS_EXTENDED_TYPES 1
+#endif
+#if HAS_EXTENDED_TYPES
 #extension GL_EXT_shader_16bit_storage : require
 #extension GL_EXT_shader_8bit_storage : require
-#define HAS_EXTENDED_TYPES 1
+#endif
 #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)
@@ -5,9 +5,13 @@

 #ifdef VULKAN

+#ifndef HAS_EXTENDED_TYPES
+#define HAS_EXTENDED_TYPES 1
+#endif
+#if HAS_EXTENDED_TYPES
 #extension GL_EXT_shader_16bit_storage : require
 #extension GL_EXT_shader_8bit_storage : require
-#define HAS_EXTENDED_TYPES 1
+#endif
 #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)
@@ -5,9 +5,13 @@

 #ifdef VULKAN

+#ifndef HAS_EXTENDED_TYPES
+#define HAS_EXTENDED_TYPES 1
+#endif
+#if HAS_EXTENDED_TYPES
 #extension GL_EXT_shader_16bit_storage : require
 #extension GL_EXT_shader_8bit_storage : require
-#define HAS_EXTENDED_TYPES 1
+#endif
 #define BEGIN_PUSH_CONSTANTS layout(push_constant) uniform PushConstants {
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)
@@ -226,22 +226,6 @@ void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) {
    }
 }

-void BufferCacheRuntime::BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size,
-                                          u32 stride) {
-    if (index >= max_attributes) {
-        return;
-    }
-    if (has_unified_vertex_buffers) {
-        buffer.MakeResident(GL_READ_ONLY);
-        glBindVertexBuffer(index, 0, 0, static_cast<GLsizei>(stride));
-        glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, index,
-                               buffer.HostGpuAddr() + offset, static_cast<GLsizeiptr>(size));
-    } else {
-        glBindVertexBuffer(index, buffer.Handle(), static_cast<GLintptr>(offset),
-                           static_cast<GLsizei>(stride));
-    }
-}
-
 void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings) {
    // TODO: Should HostBindings provide the correct runtime types to avoid these transforms?
    std::array<GLuint, 32> buffer_handles;
@@ -99,8 +99,6 @@ public:

    void BindIndexBuffer(Buffer& buffer, u32 offset, u32 size);

-    void BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride);
-
    void BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings);

    void BindUniformBuffer(size_t stage, u32 binding_index, Buffer& buffer, u32 offset, u32 size);
@@ -259,6 +259,7 @@ void RasterizerOpenGL::PrepareDraw(bool is_indexed, Func&& draw_func) {
 }

 void RasterizerOpenGL::Draw(bool is_indexed, u32 instance_count) {
+    buffer_cache.SetDrawInstanceCount(instance_count);
    PrepareDraw(is_indexed, [this, is_indexed, instance_count](GLenum primitive_mode) {
        const auto& draw_state = maxwell3d->draw_manager.draw_state;
        const GLuint base_instance = GLuint(draw_state.base_instance);
@@ -304,6 +305,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, u32 instance_count) {
 void RasterizerOpenGL::DrawIndirect() {
    const auto& params = maxwell3d->draw_manager.indirect_state;
    buffer_cache.SetDrawIndirect(&params);
+    buffer_cache.SetDrawInstanceCount(0);
    PrepareDraw(params.is_indexed, [this, &params](GLenum primitive_mode) {
        if (params.is_byte_count) {
            const GPUVAddr tfb_object_base_addr = params.indirect_start_address - 4U;
@@ -6,7 +6,6 @@

 #pragma once

-#include <array>
 #include <cstddef>
 #include <optional>

@@ -137,7 +136,6 @@ inline void WriteDescriptorBuffer(const Device& device, const DescriptorBufferLa

 [[nodiscard]] inline u32 NumDescriptorEntries(const Shader::Info& info) {
    return Shader::NumDescriptors(info.constant_buffer_descriptors) +
-           static_cast<u32>(Shader::UsesStorageBufferMappings(info)) +
           Shader::NumDescriptors(info.storage_buffers_descriptors) +
           Shader::NumDescriptors(info.texture_buffer_descriptors) +
           Shader::NumDescriptors(info.image_buffer_descriptors) +
@@ -270,12 +268,6 @@ public:
        is_compute |= (stage & VK_SHADER_STAGE_COMPUTE_BIT) != 0;

        Add(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, stage, info.constant_buffer_descriptors);
-        // for extra implicit storage-buffer binding required by mapped-storage-buffer support
-        if (Shader::UsesStorageBufferMappings(info)) {
-            struct Descriptor { u32 count; };
-            const std::array descriptors{Descriptor{1}};
-            Add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, stage, descriptors);
-        }
        Add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, stage, info.storage_buffers_descriptors);
        Add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, stage, info.texture_buffer_descriptors);
        Add(VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, stage, info.image_buffer_descriptors);
@@ -585,29 +585,6 @@ void BufferCacheRuntime::BindQuadIndexBuffer(PrimitiveTopology topology, u32 fir
    }
 }

-void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, u32 stride) {
-    if (index >= device.GetMaxVertexInputBindings()) {
-        return;
-    }
-    if (device.IsExtExtendedDynamicStateSupported()) {
-        scheduler.Record([index, buffer, offset, size, stride](vk::CommandBuffer cmdbuf) {
-            const VkDeviceSize vk_offset = buffer != VK_NULL_HANDLE ? offset : 0;
-            const VkDeviceSize vk_size = buffer != VK_NULL_HANDLE ? size : VK_WHOLE_SIZE;
-            const VkDeviceSize vk_stride = stride;
-            cmdbuf.BindVertexBuffers2EXT(index, 1, &buffer, &vk_offset, &vk_size, &vk_stride);
-        });
-    } else {
-        if (!device.HasNullDescriptor() && buffer == VK_NULL_HANDLE) {
-            ReserveNullBuffer();
-            buffer = *null_buffer;
-            offset = 0;
-        }
-        scheduler.Record([index, buffer, offset](vk::CommandBuffer cmdbuf) {
-            cmdbuf.BindVertexBuffer(index, buffer, offset);
-        });
-    }
-}
-
 void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings) {
    boost::container::static_vector<VkBuffer, VideoCommon::NUM_VERTEX_BUFFERS> buffer_handles(bindings.buffers.size());
    for (u32 i = 0; i < bindings.buffers.size(); ++i) {
@@ -695,7 +672,7 @@ vk::Buffer BufferCacheRuntime::CreateNullBuffer() {
        .flags = 0,
        .size = 4,
        .usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
-                 VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
+                 VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT,
        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
        .queueFamilyIndexCount = 0,
        .pQueueFamilyIndices = nullptr,
@@ -138,8 +138,6 @@ public:

    void BindQuadIndexBuffer(PrimitiveTopology topology, u32 first, u32 count);

-    void BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, u32 stride);
-
    void BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings);

    void BindTransformFeedbackBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size);
@@ -155,13 +153,6 @@ public:
        return ref.mapped_span;
    }

-    // compute/graphics new binding
-    std::span<u8> BindMappedStorageBuffer(u32 size) {
-        const StagingBufferRef ref = staging_pool.Request(size, MemoryUsage::Upload);
-        guest_descriptor_queue.AddBuffer(ref.buffer, ref.device_address, static_cast<u32>(ref.offset), size);
-        return ref.mapped_span;
-    }
-
    void BindUniformBuffer(const Buffer& buffer, u32 offset, u32 size) {
        BindBuffer(buffer, offset, size);
    }
@@ -22,7 +22,13 @@
 #include "video_core/host_shaders/resolve_conditional_render_comp_spv.h"
 #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
 #include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
+#include "video_core/host_shaders/block_linear_unswizzle_2d_comp_spv.h"
+#include "video_core/host_shaders/block_linear_unswizzle_2d_nonarrow_comp_spv.h"
 #include "video_core/host_shaders/block_linear_unswizzle_3d_bcn_comp_spv.h"
+#include "video_core/host_shaders/block_linear_unswizzle_3d_comp_spv.h"
+#include "video_core/host_shaders/block_linear_unswizzle_3d_nonarrow_comp_spv.h"
+#include "video_core/host_shaders/pitch_unswizzle_comp_spv.h"
+#include "video_core/host_shaders/pitch_unswizzle_nonarrow_comp_spv.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/surface.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
@@ -872,4 +878,291 @@ void BlockLinearUnswizzle3DPass::UnswizzleChunk(
    });
 }

+namespace {
+
+constexpr u32 UNSWIZZLE_BINDING_INPUT_BUFFER = 0;
+constexpr u32 UNSWIZZLE_BINDING_OUTPUT_IMAGE = 1;
+constexpr size_t UNSWIZZLE_NUM_BINDINGS = 2;
+
+constexpr std::array<VkDescriptorSetLayoutBinding, UNSWIZZLE_NUM_BINDINGS>
+    UNSWIZZLE_DESCRIPTOR_SET_BINDINGS{{
+        {
+            .binding = UNSWIZZLE_BINDING_INPUT_BUFFER,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .descriptorCount = 1,
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .pImmutableSamplers = nullptr,
+        },
+        {
+            .binding = UNSWIZZLE_BINDING_OUTPUT_IMAGE,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+            .descriptorCount = 1,
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .pImmutableSamplers = nullptr,
+        },
+    }};
+
+constexpr std::array<VkDescriptorUpdateTemplateEntry, UNSWIZZLE_NUM_BINDINGS>
+    UNSWIZZLE_DESCRIPTOR_UPDATE_TEMPLATE{{
+        {
+            .dstBinding = UNSWIZZLE_BINDING_INPUT_BUFFER,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+            .offset = UNSWIZZLE_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+        {
+            .dstBinding = UNSWIZZLE_BINDING_OUTPUT_IMAGE,
+            .dstArrayElement = 0,
+            .descriptorCount = 1,
+            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+            .offset = UNSWIZZLE_BINDING_OUTPUT_IMAGE * sizeof(DescriptorUpdateEntry),
+            .stride = sizeof(DescriptorUpdateEntry),
+        },
+    }};
+
+constexpr DescriptorBankInfo UNSWIZZLE_BANK_INFO{
+    .uniform_buffers = 0,
+    .storage_buffers = 1,
+    .texture_buffers = 0,
+    .image_buffers = 0,
+    .textures = 0,
+    .images = 1,
+    .score = 2,
+};
+
+[[nodiscard]] std::span<const u32> UnswizzleSpv(const Device& device,
+                                                std::span<const u32> extended,
+                                                std::span<const u32> narrow) {
+    if (device.IsStorageBuffer8BitAccessSupported() &&
+        device.IsStorageBuffer16BitAccessSupported()) {
+        return extended;
+    }
+    return narrow;
+}
+
+struct PitchUnswizzlePushConstants {
+    alignas(8) std::array<u32, 2> origin;
+    alignas(8) std::array<s32, 2> destination;
+    u32 bytes_per_block;
+    u32 pitch;
+};
+
+void RecordUnswizzleEntryBarrier(Scheduler& scheduler, VkPipeline vk_pipeline, VkImage vk_image,
+                                 VkImageAspectFlags aspect_mask, bool is_initialized) {
+    scheduler.Record([vk_pipeline, vk_image, aspect_mask,
+                      is_initialized](vk::CommandBuffer cmdbuf) {
+        VkAccessFlags src_access = VK_ACCESS_NONE;
+        VkImageLayout old_layout = VK_IMAGE_LAYOUT_UNDEFINED;
+        if (is_initialized) {
+            src_access = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT |
+                         VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+            old_layout = VK_IMAGE_LAYOUT_GENERAL;
+        }
+        const VkImageMemoryBarrier image_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = src_access,
+            .dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+            .oldLayout = old_layout,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = vk_image,
+            .subresourceRange{
+                .aspectMask = aspect_mask,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        VkPipelineStageFlags src_stage = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
+        if (is_initialized) {
+            src_stage = vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER;
+        }
+        cmdbuf.PipelineBarrier(src_stage, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline);
+    });
+}
+
+void RecordUnswizzleExitBarrier(Scheduler& scheduler, VkImage vk_image,
+                                VkImageAspectFlags aspect_mask) {
+    scheduler.Record([vk_image, aspect_mask](vk::CommandBuffer cmdbuf) {
+        const VkImageMemoryBarrier image_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT |
+                             VK_ACCESS_COLOR_ATTACHMENT_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = vk_image,
+            .subresourceRange{
+                .aspectMask = aspect_mask,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+                               vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER, 0, image_barrier);
+    });
+}
+
+} // Anonymous namespace
+
+BlockLinearUnswizzle2DPass::BlockLinearUnswizzle2DPass(
+    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, UNSWIZZLE_DESCRIPTOR_SET_BINDINGS,
+                  UNSWIZZLE_DESCRIPTOR_UPDATE_TEMPLATE, UNSWIZZLE_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(
+                      VideoCommon::Accelerated::BlockLinearSwizzle2DParams)>,
+                  UnswizzleSpv(device_, BLOCK_LINEAR_UNSWIZZLE_2D_COMP_SPV,
+                               BLOCK_LINEAR_UNSWIZZLE_2D_NONARROW_COMP_SPV)),
+      scheduler{scheduler_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+BlockLinearUnswizzle2DPass::~BlockLinearUnswizzle2DPass() = default;
+
+void BlockLinearUnswizzle2DPass::Unswizzle(
+    Image& image, const StagingBufferRef& map,
+    std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    using namespace VideoCommon::Accelerated;
+    scheduler.RequestOutsideRenderPassOperationContext();
+    const VkPipeline vk_pipeline = *pipeline;
+    const VkImageAspectFlags aspect_mask = image.AspectMask();
+    const VkImage vk_image = image.Handle();
+    const bool is_initialized = image.ExchangeInitialization();
+    RecordUnswizzleEntryBarrier(scheduler, vk_pipeline, vk_image, aspect_mask, is_initialized);
+
+    const u32 num_layers = static_cast<u32>(image.info.resources.layers);
+    for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
+        const size_t input_offset = swizzle.buffer_offset + map.offset;
+        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
+        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
+
+        compute_pass_descriptor_queue.Acquire(scheduler, 2);
+        compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
+                                                image.guest_size_bytes - swizzle.buffer_offset);
+        compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
+        const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
+
+        const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
+        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_layers, params,
+                          descriptor_data](vk::CommandBuffer cmdbuf) {
+            const VkDescriptorSet set = descriptor_allocator.Commit();
+            device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+            cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params);
+            cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_layers);
+        });
+    }
+    RecordUnswizzleExitBarrier(scheduler, vk_image, aspect_mask);
+}
+
+BlockLinearUnswizzleImage3DPass::BlockLinearUnswizzleImage3DPass(
+    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, UNSWIZZLE_DESCRIPTOR_SET_BINDINGS,
+                  UNSWIZZLE_DESCRIPTOR_UPDATE_TEMPLATE, UNSWIZZLE_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(BlockLinearSwizzle3DParams)>,
+                  UnswizzleSpv(device_, BLOCK_LINEAR_UNSWIZZLE_3D_COMP_SPV,
+                               BLOCK_LINEAR_UNSWIZZLE_3D_NONARROW_COMP_SPV)),
+      scheduler{scheduler_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+BlockLinearUnswizzleImage3DPass::~BlockLinearUnswizzleImage3DPass() = default;
+
+void BlockLinearUnswizzleImage3DPass::Unswizzle(
+    Image& image, const StagingBufferRef& map,
+    std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    using namespace VideoCommon::Accelerated;
+    scheduler.RequestOutsideRenderPassOperationContext();
+    const VkPipeline vk_pipeline = *pipeline;
+    const VkImageAspectFlags aspect_mask = image.AspectMask();
+    const VkImage vk_image = image.Handle();
+    const bool is_initialized = image.ExchangeInitialization();
+    RecordUnswizzleEntryBarrier(scheduler, vk_pipeline, vk_image, aspect_mask, is_initialized);
+
+    for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
+        const size_t input_offset = swizzle.buffer_offset + map.offset;
+        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 16U);
+        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U);
+        const u32 num_dispatches_z = Common::DivCeil(swizzle.num_tiles.depth, 8U);
+
+        compute_pass_descriptor_queue.Acquire(scheduler, 2);
+        compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
+                                                image.guest_size_bytes - swizzle.buffer_offset);
+        compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
+        const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
+
+        const auto params = MakeBlockLinearSwizzle3DParams(swizzle, image.info);
+        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, params,
+                          descriptor_data](vk::CommandBuffer cmdbuf) {
+            const VkDescriptorSet set = descriptor_allocator.Commit();
+            device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+            cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params);
+            cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z);
+        });
+    }
+    RecordUnswizzleExitBarrier(scheduler, vk_image, aspect_mask);
+}
+
+PitchUnswizzlePass::PitchUnswizzlePass(
+    const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+    : ComputePass(device_, scheduler_, descriptor_pool_, UNSWIZZLE_DESCRIPTOR_SET_BINDINGS,
+                  UNSWIZZLE_DESCRIPTOR_UPDATE_TEMPLATE, UNSWIZZLE_BANK_INFO,
+                  COMPUTE_PUSH_CONSTANT_RANGE<sizeof(PitchUnswizzlePushConstants)>,
+                  UnswizzleSpv(device_, PITCH_UNSWIZZLE_COMP_SPV,
+                               PITCH_UNSWIZZLE_NONARROW_COMP_SPV)),
+      scheduler{scheduler_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+PitchUnswizzlePass::~PitchUnswizzlePass() = default;
+
+void PitchUnswizzlePass::Unswizzle(Image& image, const StagingBufferRef& map,
+                                   std::span<const VideoCommon::SwizzleParameters> swizzles) {
+    scheduler.RequestOutsideRenderPassOperationContext();
+    const VkPipeline vk_pipeline = *pipeline;
+    const VkImageAspectFlags aspect_mask = image.AspectMask();
+    const VkImage vk_image = image.Handle();
+    const bool is_initialized = image.ExchangeInitialization();
+    RecordUnswizzleEntryBarrier(scheduler, vk_pipeline, vk_image, aspect_mask, is_initialized);
+
+    const u32 bytes_per_block = VideoCore::Surface::BytesPerBlock(image.info.format);
+    const u32 pitch = image.info.pitch;
+    for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
+        const size_t input_offset = swizzle.buffer_offset + map.offset;
+        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
+        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
+
+        compute_pass_descriptor_queue.Acquire(scheduler, 2);
+        compute_pass_descriptor_queue.AddBuffer(map.buffer, input_offset,
+                                                image.guest_size_bytes - swizzle.buffer_offset);
+        compute_pass_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
+        const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
+
+        const PitchUnswizzlePushConstants params{
+            .origin{0, 0},
+            .destination{0, 0},
+            .bytes_per_block = bytes_per_block,
+            .pitch = pitch,
+        };
+        scheduler.Record([this, num_dispatches_x, num_dispatches_y, params,
+                          descriptor_data](vk::CommandBuffer cmdbuf) {
+            const VkDescriptorSet set = descriptor_allocator.Commit();
+            device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+            cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, params);
+            cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, 1);
+        });
+    }
+    RecordUnswizzleExitBarrier(scheduler, vk_image, aspect_mask);
+}
+
 } // namespace Vulkan
@@ -164,4 +164,49 @@ private:
    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
 };

+class BlockLinearUnswizzle2DPass final : public ComputePass {
+public:
+    explicit BlockLinearUnswizzle2DPass(
+        const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+        ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~BlockLinearUnswizzle2DPass();
+
+    void Unswizzle(Image& image, const StagingBufferRef& map,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
+class BlockLinearUnswizzleImage3DPass final : public ComputePass {
+public:
+    explicit BlockLinearUnswizzleImage3DPass(
+        const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+        ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~BlockLinearUnswizzleImage3DPass();
+
+    void Unswizzle(Image& image, const StagingBufferRef& map,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
+class PitchUnswizzlePass final : public ComputePass {
+public:
+    explicit PitchUnswizzlePass(const Device& device_, Scheduler& scheduler_,
+                                DescriptorPool& descriptor_pool_,
+                                ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+    ~PitchUnswizzlePass();
+
+    void Unswizzle(Image& image, const StagingBufferRef& map,
+                   std::span<const VideoCommon::SwizzleParameters> swizzles);
+
+private:
+    Scheduler& scheduler;
+    ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
 } // namespace Vulkan
@@ -157,7 +157,9 @@ bool ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
    buffer_cache.UnbindComputeStorageBuffers();
    size_t ssbo_index{};
    for (const auto& desc : info.storage_buffers_descriptors) {
-        buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset, desc.is_written, desc.count);
+        ASSERT(desc.count == 1);
+        buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset,
+                                              desc.is_written);
        ++ssbo_index;
    }

@@ -47,7 +47,7 @@ static DescriptorBankInfo MakeBankInfo(std::span<const Shader::Info> infos) {
    DescriptorBankInfo bank;
    for (const Shader::Info& info : infos) {
        bank.uniform_buffers += Accumulate(info.constant_buffer_descriptors);
-        bank.storage_buffers += Accumulate(info.storage_buffers_descriptors) + static_cast<u32>(Shader::UsesStorageBufferMappings(info));
+        bank.storage_buffers += Accumulate(info.storage_buffers_descriptors);
        bank.texture_buffers += Accumulate(info.texture_buffer_descriptors);
        bank.image_buffers += Accumulate(info.image_buffer_descriptors);
        bank.textures += Accumulate(info.texture_descriptors);
@@ -367,8 +367,9 @@ bool GraphicsPipeline::ConfigureImpl(bool is_indexed) {
        if constexpr (Spec::has_storage_buffers) {
            size_t ssbo_index{};
            for (const auto& desc : info.storage_buffers_descriptors) {
+                ASSERT(desc.count == 1);
                buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, desc.cbuf_index,
-                                                       desc.cbuf_offset, desc.is_written, desc.count);
+                                                       desc.cbuf_offset, desc.is_written);
                ++ssbo_index;
            }
        }
@@ -693,9 +694,11 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
        const size_t num_vertex_arrays = (std::min)(
            Maxwell::NumVertexArrays, static_cast<size_t>(device.GetMaxVertexInputBindings()));
        for (size_t index = 0; index < num_vertex_arrays; ++index) {
-            const bool instanced = key.state.binding_divisors[index] != 0;
-            const auto rate =
-                instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX;
+            const bool instanced = ((key.state.enabled_divisors >> index) & 1) != 0;
+            auto rate = VK_VERTEX_INPUT_RATE_VERTEX;
+            if (instanced) {
+                rate = VK_VERTEX_INPUT_RATE_INSTANCE;
+            }
            vertex_bindings.push_back({
                .binding = static_cast<u32>(index),
                .stride = key.state.vertex_strides[index],
@@ -704,7 +707,7 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
            if (instanced) {
                vertex_binding_divisors.push_back({
                    .binding = static_cast<u32>(index),
-                    .divisor = key.state.binding_divisors[index],
+                    .divisor = device.GetVertexAttribDivisor(key.state.binding_divisors[index]),
                });
            }
        }
@@ -62,12 +62,7 @@ using VideoCommon::FileEnvironment;
 using VideoCommon::GenericEnvironment;
 using VideoCommon::GraphicsEnvironment;

-// SPIR-V descriptor arrays require a fixed pipeline-layout count.
-// Exploration ceiling; buffer-cache telemetry records the actual physical-range demand.
-// Keep this modest because every mapped SSBO binds the full fixed array on each update.
-constexpr u32 MAX_MAPPED_STORAGE_BUFFER_DESCRIPTORS = 32;
-
-constexpr u32 CACHE_VERSION = 19;
+constexpr u32 CACHE_VERSION = 18;
 constexpr size_t VULKAN_CACHE_FLUSH_PIPELINES = 128;
 constexpr size_t VULKAN_CACHE_FLUSH_MIN_SECONDS = 30;
 constexpr std::array<char, 8> VULKAN_CACHE_MAGIC_NUMBER{'y', 'u', 'z', 'u', 'v', 'k', 'c', 'h'};
@@ -437,8 +432,6 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
            device.IsUniformTexelBufferArrayNonUniformIndexingSupported(),
        .support_storage_texel_buffer_array_nonuniform_indexing =
            device.IsStorageTexelBufferArrayNonUniformIndexingSupported(),
-        .support_storage_buffer_array_nonuniform_indexing =
-            device.IsStorageBufferArrayNonUniformIndexingSupported(),

        .warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyBiggerThanGuest(),

@@ -467,7 +460,6 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
    host_info = Shader::HostTranslateInfo{
        .min_ssbo_alignment = device.GetStorageBufferAlignment(),
        .max_per_stage_descriptor_sampled_images = device.GetMaxPerStageDescriptorSampledImages(),
-        .max_per_stage_descriptor_storage_buffers = device.GetMaxPerStageDescriptorStorageBuffers(),
        .max_per_stage_resources = device.GetMaxPerStageResources(),
        .max_descriptor_set_samplers = device.GetMaxDescriptorSetSamplers(),
        .max_descriptor_set_uniform_buffers = device.GetMaxDescriptorSetUniformBuffers(),
@@ -487,20 +479,6 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
        .support_viewport_index_layer = device.IsExtShaderViewportIndexLayerSupported(),
        .support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(),
        .support_conditional_barrier = device.SupportsConditionalBarriers(),
-        .storage_buffer_segment_count = [&] {
-            if (!device.IsStorageBufferArrayNonUniformIndexingSupported()) {
-                return 1U;
-            }
-            constexpr u32 MaxGraphicsStages = static_cast<u32>(Maxwell::MaxShaderStage);
-            const auto reserve = [](u32 limit, u32 count) {
-                return limit > count ? limit - count : 0U;
-            };
-            const u32 per_stage = reserve(device.GetMaxPerStageDescriptorStorageBuffers(), 1) / static_cast<u32>(Shader::Info::MAX_SSBOS);
-            const u32 per_set = reserve(device.GetMaxDescriptorSetStorageBuffers(), MaxGraphicsStages) / (static_cast<u32>(Shader::Info::MAX_SSBOS) * MaxGraphicsStages);
-            const u32 resources = reserve(device.GetMaxPerStageResources(), 1) / (static_cast<u32>(Shader::Info::MAX_SSBOS) * 2);
-            // ensure at least one storage buffer segment is available per stage. max is still arbitrary
-            return (std::max)(1U, (std::min)({MAX_MAPPED_STORAGE_BUFFER_DESCRIPTORS, per_stage, per_set, resources}));
-        }(),
    };
    host_info.ApplyDescriptorLimitPolicy();

@@ -137,10 +137,12 @@ VkRect2D GetScissorState(const Maxwell& regs, size_t index, u32 up_scale = 1, u3
    max_y = (std::max)(max_y, 0);

    if (src.enable) {
-        scissor.offset.x = scale_up(src.min_x);
+        const s32 min_x = static_cast<s32>(src.min_x.Value());
+        const s32 max_x = static_cast<s32>(src.max_x.Value());
+        scissor.offset.x = scale_up(min_x);
        scissor.offset.y = scale_up(min_y);
-        scissor.extent.width = scale_up(src.max_x - src.min_x);
-        scissor.extent.height = scale_up(max_y - min_y);
+        scissor.extent.width = scale_up((std::max)(max_x - min_x, 0));
+        scissor.extent.height = scale_up((std::max)(max_y - min_y, 0));
    } else {
        scissor.offset.x = 0;
        scissor.offset.y = 0;
@@ -260,6 +262,7 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
 }

 void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
+    buffer_cache.SetDrawInstanceCount(instance_count);
    PrepareDraw(is_indexed, [this, is_indexed, instance_count] {
        const auto& draw_state = maxwell3d->draw_manager.draw_state;
        const u32 num_instances{instance_count};
@@ -295,6 +298,7 @@ void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
 void RasterizerVulkan::DrawIndirect() {
    const auto& params = maxwell3d->draw_manager.indirect_state;
    buffer_cache.SetDrawIndirect(&params);
+    buffer_cache.SetDrawInstanceCount(0);
    PrepareDraw(params.is_indexed, [this, &params] {
        const auto indirect_buffer = buffer_cache.GetDrawIndirectBuffer();
        const auto& buffer = indirect_buffer.first;
@@ -1917,13 +1921,19 @@ void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs)
    for (u32 binding = 0; binding < max_bindings; ++binding) {
        const auto& input_binding{regs.vertex_streams[binding]};
        const bool is_instanced{regs.vertex_stream_instances.IsInstancingEnabled(binding)};
+        auto input_rate = VK_VERTEX_INPUT_RATE_VERTEX;
+        u32 divisor = 1;
+        if (is_instanced) {
+            input_rate = VK_VERTEX_INPUT_RATE_INSTANCE;
+            divisor = device.GetVertexAttribDivisor(input_binding.frequency);
+        }
        bindings.push_back({
            .sType = VK_STRUCTURE_TYPE_VERTEX_INPUT_BINDING_DESCRIPTION_2_EXT,
            .pNext = nullptr,
            .binding = binding,
            .stride = input_binding.stride,
-            .inputRate = is_instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX,
-            .divisor = is_instanced ? input_binding.frequency : 1,
+            .inputRate = input_rate,
+            .divisor = divisor,
        });
    }

@@ -449,7 +449,9 @@ void Scheduler::EndRenderPass()
                                         | VK_ACCESS_COLOR_ATTACHMENT_READ_BIT
                                         | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT
                                         | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT
-                                         | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
+                                         | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT
+                                         | VK_ACCESS_TRANSFER_READ_BIT
+                                         | VK_ACCESS_TRANSFER_WRITE_BIT,
                        .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                        .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                        .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
@@ -460,7 +462,7 @@ void Scheduler::EndRenderPass()
            }
            cmdbuf.EndRenderPass();
            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-                                   VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
+                                   VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE_TRANSFER,
                                   0, nullptr, nullptr, vk::Span(barriers.data(), num_images));
            if (has_transform_feedback) {
                static constexpr VkMemoryBarrier XFB_OUTPUT_BARRIER{
@@ -160,6 +160,55 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
          info.size.depth == 1;
 }

+[[nodiscard]] PixelFormat UnswizzleViewFormat(u32 bytes_per_block) {
+    switch (bytes_per_block) {
+    case 1:
+        return PixelFormat::R8_UINT;
+    case 2:
+        return PixelFormat::R16_UINT;
+    case 4:
+        return PixelFormat::R32_UINT;
+    case 8:
+        return PixelFormat::R32G32_UINT;
+    case 16:
+        return PixelFormat::R32G32B32A32_UINT;
+    default:
+        return PixelFormat::Invalid;
+    }
+}
+
+constexpr u32 UNSWIZZLE_WORKGROUP_INVOCATIONS = 32 * 32;
+
+[[nodiscard]] bool SupportsAcceleratedUnswizzleDevice(const Device& device) {
+    return device.IsKhrImageFormatListSupported() &&
+           device.GetMaxComputeWorkGroupInvocations() >= UNSWIZZLE_WORKGROUP_INVOCATIONS;
+}
+
+[[nodiscard]] bool SupportsAcceleratedUnswizzle(const Device& device, const ImageInfo& info) {
+    if (!SupportsAcceleratedUnswizzleDevice(device)) {
+        return false;
+    }
+    if (info.num_samples > 1) {
+        return false;
+    }
+    if (info.type != ImageType::e2D && info.type != ImageType::e3D &&
+        info.type != ImageType::Linear) {
+        return false;
+    }
+    const PixelFormat view_format =
+        UnswizzleViewFormat(VideoCore::Surface::BytesPerBlock(info.format));
+    if (view_format == PixelFormat::Invalid) {
+        return false;
+    }
+    if (!VideoCore::Surface::IsViewCompatible(info.format, view_format, false, true)) {
+        return false;
+    }
+    const auto host_format =
+        MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, view_format);
+    return device.IsFormatSupported(host_format.format, VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT,
+                                    FormatType::Optimal);
+}
+
 [[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info,
                                                    std::optional<VkFormat> format_override = {}) {
    auto format_info =
@@ -248,8 +297,18 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
    return allocator.CreateImage(image_ci);
 }

+[[nodiscard]] VkImageViewType StorageViewType(ImageType type) {
+    if (type == ImageType::e3D) {
+        return VK_IMAGE_VIEW_TYPE_3D;
+    }
+    if (type == ImageType::Linear) {
+        return VK_IMAGE_VIEW_TYPE_2D;
+    }
+    return VK_IMAGE_VIEW_TYPE_2D_ARRAY;
+}
+
 [[nodiscard]] vk::ImageView MakeStorageView(const vk::Device& device, u32 level, VkImage image,
-                                            VkFormat format) {
+                                            VkFormat format, VkImageViewType view_type) {
    static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
        .pNext = nullptr,
@@ -260,7 +319,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
        .pNext = &storage_image_view_usage_create_info,
        .flags = 0,
        .image = image,
-        .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY,
+        .viewType = view_type,
        .format = format,
        .components{
            .r = VK_COMPONENT_SWIZZLE_IDENTITY,
@@ -658,18 +717,11 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im
            .subresourceRange = subresource_range,
    };

-    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-                           VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                           VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
+    cmdbuf.PipelineBarrier(vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
                           read_barrier);
    cmdbuf.CopyBufferToImage(src_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copies);
-    // TODO: Move this to another API
-    cmdbuf.PipelineBarrier(
-            VK_PIPELINE_STAGE_TRANSFER_BIT,
-            VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
-            VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
-            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
-            0, nullptr, nullptr, write_barrier);
+    cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE, 0,
+                           nullptr, nullptr, write_barrier);
 }

 [[nodiscard]] VkImageBlit MakeImageBlit(const Region2D& dst_region, const Region2D& src_region,
@@ -968,6 +1020,14 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
        bl3d_unswizzle_pass.emplace(device, scheduler, descriptor_pool,
                                   staging_buffer_pool, compute_pass_descriptor_queue);
    }
+    if (SupportsAcceleratedUnswizzleDevice(device)) {
+        bl_unswizzle_2d_pass.emplace(device, scheduler, descriptor_pool,
+                                     compute_pass_descriptor_queue);
+        bl_unswizzle_image_3d_pass.emplace(device, scheduler, descriptor_pool,
+                                           compute_pass_descriptor_queue);
+        pitch_unswizzle_pass.emplace(device, scheduler, descriptor_pool,
+                                     compute_pass_descriptor_queue);
+    }
 }

 void TextureCacheRuntime::Finish() {
@@ -1902,16 +1962,12 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
    if (runtime->device.HasDebuggingToolAttached()) {
        original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
    }
+    if (False(flags & VideoCommon::ImageFlagBits::Converted) &&
+        SupportsAcceleratedUnswizzle(runtime->device, info)) {
+        flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
+    }
    current_image = &Image::original_image;
    storage_image_views.resize(info.resources.levels);
-    if (WillUseAcceleratedAstcDecode(runtime->device, info)) {
-        const auto& device = runtime->device.GetLogical();
-        const VkFormat storage_format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
-        for (s32 level = 0; level < info.resources.levels; ++level) {
-            storage_image_views[level] =
-                MakeStorageView(device, level, *original_image, storage_format);
-        }
-    }
 }

 Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBase{params} {}
@@ -2035,7 +2091,7 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
                                                temp_vk_image, info.format, info.num_samples,
                                                {image_copies.data(), image_copies.size()}, false);
        }
-        initialized = true;
+        InitializationFor(current_image) = true;
        runtime->ReleaseMsaaScratchImage(temp_vk_image);

        if (is_rescaled) {
@@ -2056,7 +2112,7 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
    const VkBuffer src_buffer = buffer;
    const VkImage vk_image = *original_image;
    const VkImageAspectFlags vk_aspect_mask = aspect_mask;
-    const bool was_initialized = std::exchange(initialized, true);
+    const bool was_initialized = std::exchange(InitializationFor(&Image::original_image), true);

    scheduler->Record([src_buffer, vk_image, vk_aspect_mask, was_initialized,
                       vk_copies](vk::CommandBuffer cmdbuf) {
@@ -2321,16 +2377,46 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm
    DownloadMemory(buffers, offsets, copies);
 }

+std::vector<vk::ImageView>& Image::StorageViewsFor(vk::Image Image::*image) {
+    if (image == &Image::scaled_image) {
+        if (scaled_storage_image_views.empty()) {
+            scaled_storage_image_views.resize(info.resources.levels);
+        }
+        return scaled_storage_image_views;
+    }
+    return storage_image_views;
+}
+
+bool& Image::InitializationFor(vk::Image Image::*image) noexcept {
+    if (image == &Image::scaled_image) {
+        return scaled_initialized;
+    }
+    return original_initialized;
+}
+
 VkImageView Image::StorageImageView(s32 level) noexcept {
-    auto& view = storage_image_views[level];
+    const bool astc_decode = WillUseAcceleratedAstcDecode(runtime->device, info);
+    const bool unswizzle_upload =
+        !astc_decode && True(flags & ImageFlagBits::AcceleratedUpload);
+    vk::Image Image::*target = current_image;
+    if (astc_decode || unswizzle_upload) {
+        target = &Image::original_image;
+    }
+    auto& view = StorageViewsFor(target)[level];
    if (!view) {
        auto format_info =
            MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format);
-        if (WillUseAcceleratedAstcDecode(runtime->device, info)) {
+        if (astc_decode) {
            format_info.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
        }
-        view = MakeStorageView(runtime->device.GetLogical(), level, *(this->*current_image),
-                               format_info.format);
+        if (unswizzle_upload) {
+            const PixelFormat view_format =
+                UnswizzleViewFormat(VideoCore::Surface::BytesPerBlock(info.format));
+            format_info = MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, false,
+                                                     view_format);
+        }
+        view = MakeStorageView(runtime->device.GetLogical(), level, *(this->*target),
+                               format_info.format, StorageViewType(info.type));
    }
    return *view;
 }
@@ -2370,6 +2456,7 @@ bool Image::ScaleUp(bool ignore) {
    }
    if (NeedsScaleHelper()) {
        if (!BlitScaleHelper(true)) {
+            flags &= ~ImageFlagBits::Rescaled;
            current_image = &Image::original_image;
            return false;
        }
@@ -2559,6 +2646,10 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
    if (device->IsExtAstcDecodeModeSupported() && IsLdrAstcFormat(format_info.format)) {
        view_next = &astc_decode_mode;
    }
+    auto subresource_range = MakeSubresourceRange(aspect_mask, info.range);
+    if (True(flags & VideoCommon::ImageViewFlagBits::Slice)) {
+        subresource_range.levelCount = 1;
+    }
    const VkImageViewCreateInfo create_info{
        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
        .pNext = view_next,
@@ -2567,7 +2658,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
        .viewType = VkImageViewType{},
        .format = format_info.format,
        .components = swizzle_mapping,
-        .subresourceRange = MakeSubresourceRange(aspect_mask, info.range),
+        .subresourceRange = subresource_range,
    };
    const auto create = [&](TextureType tex_type, std::optional<u32> num_layers) {
        VkImageViewCreateInfo ci{create_info};
@@ -3141,6 +3232,19 @@ void TextureCacheRuntime::AccelerateImageUpload(
        return astc_decoder_pass->Assemble(image, map, swizzles);
    }

+    if (bl_unswizzle_2d_pass && image.info.type == ImageType::e2D) {
+        return bl_unswizzle_2d_pass->Unswizzle(image, map, swizzles);
+    }
+
+    if (bl_unswizzle_image_3d_pass && image.info.type == ImageType::e3D &&
+        !IsPixelFormatBCn(image.info.format)) {
+        return bl_unswizzle_image_3d_pass->Unswizzle(image, map, swizzles);
+    }
+
+    if (pitch_unswizzle_pass && image.info.type == ImageType::Linear) {
+        return pitch_unswizzle_pass->Unswizzle(image, map, swizzles);
+    }
+
    if (!Settings::values.gpu_unswizzle_enabled.GetValue() || !bl3d_unswizzle_pass) {
        if (IsPixelFormatBCn(image.info.format) && image.info.type == ImageType::e3D) {
            ASSERT(false && "GPU unswizzle is disabled for BCn 3D texture");
@@ -159,6 +159,9 @@ public:
    std::optional<ASTCDecoderPass> astc_decoder_pass;

    std::optional<BlockLinearUnswizzle3DPass> bl3d_unswizzle_pass;
+    std::optional<BlockLinearUnswizzle2DPass> bl_unswizzle_2d_pass;
+    std::optional<BlockLinearUnswizzleImage3DPass> bl_unswizzle_image_3d_pass;
+    std::optional<PitchUnswizzlePass> pitch_unswizzle_pass;
    const Settings::ResolutionScalingInfo& resolution;
    std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats;

@@ -350,7 +353,7 @@ public:

    /// Returns true when the image is already initialized and mark it as initialized
    [[nodiscard]] bool ExchangeInitialization() noexcept {
-        return std::exchange(initialized, true);
+        return std::exchange(InitializationFor(current_image), true);
    }

    VkImageView StorageImageView(s32 level) noexcept;
@@ -370,6 +373,10 @@ private:

    bool NeedsScaleHelper() const;

+    std::vector<vk::ImageView>& StorageViewsFor(vk::Image Image::*image);
+
+    bool& InitializationFor(vk::Image Image::*image) noexcept;
+
    Scheduler* scheduler{};
    TextureCacheRuntime* runtime{};

@@ -387,8 +394,10 @@ private:
    vk::Image Image::*current_image{};

    std::vector<vk::ImageView> storage_image_views;
+    std::vector<vk::ImageView> scaled_storage_image_views;
    VkImageAspectFlags aspect_mask = 0;
-    bool initialized = false;
+    bool original_initialized = false;
+    bool scaled_initialized = false;

    std::optional<Framebuffer> scale_framebuffer;
    std::optional<Framebuffer> normal_framebuffer;
@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later

@@ -23,8 +26,8 @@ struct BlockLinearSwizzle2DParams {
 };

 struct BlockLinearSwizzle3DParams {
-    std::array<u32, 3> origin;
-    std::array<s32, 3> destination;
+    alignas(16) std::array<u32, 3> origin;
+    alignas(16) std::array<s32, 3> destination;
    u32 bytes_per_block_log2;
    u32 slice_size;
    u32 block_size;
@@ -20,6 +20,7 @@
 #include "video_core/engines/kepler_compute.h"
 #include "video_core/guest_memory.h"
 #include "video_core/host1x/gpu_device_memory_manager.h"
+#include "video_core/texture_cache/accelerated_swizzle.h"
 #include "video_core/texture_cache/image_view_base.h"
 #include "video_core/texture_cache/samples_helper.h"
 #include "video_core/texture_cache/texture_cache_base.h"
@@ -279,11 +280,11 @@ void TextureCache<P>::CheckFeedbackLoop(std::span<const ImageViewInOut> views) {

            const ImageId view_image_id = slot_image_views[view.id].image_id;
            {
-                bool is_continue = false;
+                bool is_feedback = false;
                for (size_t i = 0; i < 8; ++i)
-                    is_continue |= (rt_active_mask & (1u << i)) && view_image_id == rt_image_id[i];
-                if (is_continue)
-                    continue;
+                    is_feedback |= (rt_active_mask & (1u << i)) && view_image_id == rt_image_id[i];
+                if (is_feedback)
+                    return true;
            }
            if (depth_active && view_image_id == rt_depth_image_id) {
                return true;
@@ -626,14 +627,25 @@ void TextureCache<P>::DownloadMemory(DAddr cpu_addr, size_t size) {
    std::ranges::sort(images, [this](ImageId lhs, ImageId rhs) {
        return slot_images[lhs].modification_tick < slot_images[rhs].modification_tick;
    });
+    size_t total_size_bytes = 0;
+    for (const ImageId image_id : images) {
+        total_size_bytes += slot_images[image_id].unswizzled_size_bytes;
+    }
+    auto download_map = runtime.DownloadStagingBuffer(total_size_bytes);
    for (const ImageId image_id : images) {
        Image& image = slot_images[image_id];
-        auto map = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes);
        const auto copies = FixSmallVectorADL(FullDownloadCopies(image.info));
-        image.DownloadMemory(map, copies);
-        runtime.Finish();
-        SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span,
+        image.DownloadMemory(download_map, copies);
+        download_map.offset += image.unswizzled_size_bytes;
+    }
+    runtime.Finish();
+    std::span<u8> download_span = download_map.mapped_span;
+    for (const ImageId image_id : images) {
+        const ImageBase& image = slot_images[image_id];
+        const auto copies = FixSmallVectorADL(FullDownloadCopies(image.info));
+        SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, download_span,
                     swizzle_data_buffer);
+        download_span = download_span.subspan(image.unswizzled_size_bytes);
    }
 }

@@ -1122,6 +1134,12 @@ void TextureCache<P>::RefreshContents(Image& image, ImageId image_id) {

    TrackImage(image, image_id);

+    if (image.info.rescaleable &&
+        IsRegionGpuModified(image.cpu_addr, image.guest_size_bytes)) {
+        runtime.TransitionImageLayout(image);
+        return;
+    }
+
    if (image.info.num_samples > 1 && !runtime.CanUploadMSAA()) {
        LOG_WARNING(HW_GPU, "MSAA image uploads are not implemented");
        runtime.TransitionImageLayout(image);
@@ -1157,8 +1175,7 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging)
    const GPUVAddr gpu_addr = image.gpu_addr;

    if (True(image.flags & ImageFlagBits::AcceleratedUpload)) {
-        gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(),
-                              VideoCommon::CacheType::NoTextureCache);
+        gpu_memory->ReadBlockUnsafe(gpu_addr, mapped_span.data(), image.guest_size_bytes);
        const auto uploads = FullUploadSwizzles(image.info);
        runtime.AccelerateImageUpload(image, staging, FixSmallVectorADL(uploads), 0, 0);
        return;
@@ -1265,6 +1282,9 @@ ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,

 template <class P>
 bool TextureCache<P>::ImageCanRescale(ImageBase& image) {
+    if (!Settings::values.resolution_info.active) {
+        return false;
+    }
    if (!image.info.rescaleable) {
        return false;
    }
@@ -1352,12 +1372,12 @@ void TextureCache<P>::QueueAsyncDecode(Image& image, ImageId image_id) {
    decode->image_id = image_id;
    async_decodes.push_back(std::move(decode));

-    std::vector<u8> local_unswizzle_data_buffer(image.unswizzled_size_bytes, 0);
+    Common::ScratchBuffer<u8> local_unswizzle_data_buffer(image.unswizzled_size_bytes);
    Tegra::Memory::GpuGuestMemory<u8, Tegra::Memory::GuestMemoryFlags::UnsafeRead> swizzle_data(*gpu_memory, image.gpu_addr, image.guest_size_bytes, &swizzle_data_buffer);
    auto copies = UnswizzleImage(*gpu_memory, image.gpu_addr, image.info, swizzle_data, local_unswizzle_data_buffer);
    const size_t out_size = MapSizeBytes(image);

-    auto func = [out_size, copies, info = image.info,
+    auto func = [out_size, copies = std::move(copies), info = image.info,
                 input = std::move(local_unswizzle_data_buffer),
                 async_decode = decode_ptr]() mutable {
        async_decode->decoded_data.resize_destructive(out_size);
@@ -1426,18 +1446,16 @@ void TextureCache<P>::TickAsyncUnswizzle() {
    Image& image = slot_images[task.image_id];

    if (!task.initialized) {
-        task.total_size = MapSizeBytes(image);
+        task.total_size = image.guest_size_bytes;
        task.staging_buffer = runtime.UploadStagingBuffer(task.total_size, true);

-        const auto& info = image.info;
-        const u32 bytes_per_block = BytesPerBlock(info.format);
-        const u32 width_blocks = Common::DivCeil(info.size.width, 4u);
-        const u32 height_blocks = Common::DivCeil(info.size.height, 4u);
-
-        const u32 stride = width_blocks * bytes_per_block;
-        const u32 aligned_height = height_blocks;
-        task.bytes_per_slice = static_cast<size_t>(stride) * aligned_height;
+        const auto layout = FullUploadSwizzles(task.info);
+        const auto params =
+            VideoCommon::Accelerated::MakeBlockLinearSwizzle3DParams(layout.front(), task.info);
+        task.bytes_per_slice = params.slice_size;
+        task.chunked = task.info.block.depth == 0;
        task.last_submitted_offset = 0;
+        task.slices_submitted = 0;
        task.initialized = true;
    }

@@ -1452,31 +1470,39 @@ void TextureCache<P>::TickAsyncUnswizzle() {
            if (copy_amount == 0) copy_amount = task.bytes_per_slice;
        }

-        gpu_memory->ReadBlock(image.gpu_addr + task.current_offset,
-                              task.staging_buffer.mapped_span.data() + task.current_offset,
-                              copy_amount);
+        gpu_memory->ReadBlockUnsafe(image.gpu_addr + task.current_offset,
+                                    task.staging_buffer.mapped_span.data() + task.current_offset,
+                                    copy_amount);
        task.current_offset += copy_amount;
    }

    const bool is_final_batch = task.current_offset >= task.total_size;
-    const size_t bytes_ready = task.current_offset - task.last_submitted_offset;
-    const u32 complete_slices = static_cast<u32>(bytes_ready / task.bytes_per_slice);

-    if (complete_slices >= swizzle_slices_per_batch || (is_final_batch && complete_slices > 0)) {
-        const u32 z_start = static_cast<u32>(task.last_submitted_offset / task.bytes_per_slice);
-        const u32 slices_to_process = (std::min)(complete_slices, swizzle_slices_per_batch);
-        const u32 z_count = (std::min)(slices_to_process, image.info.size.depth - z_start);
+    if (task.chunked) {
+        const size_t bytes_ready = task.current_offset - task.last_submitted_offset;
+        const u32 complete_slices = static_cast<u32>(bytes_ready / task.bytes_per_slice);

-        if (z_count > 0) {
-            const auto uploads = FullUploadSwizzles(task.info);
-            runtime.AccelerateImageUpload(image, task.staging_buffer, FixSmallVectorADL(uploads), z_start, z_count);
-            task.last_submitted_offset += (static_cast<size_t>(z_count) * task.bytes_per_slice);
+        if (complete_slices >= swizzle_slices_per_batch || (is_final_batch && complete_slices > 0)) {
+            const u32 z_start = task.slices_submitted;
+            const u32 slices_to_process = (std::min)(complete_slices, swizzle_slices_per_batch);
+            const u32 z_count = (std::min)(slices_to_process, image.info.size.depth - z_start);
+
+            if (z_count > 0) {
+                const auto uploads = FullUploadSwizzles(task.info);
+                runtime.AccelerateImageUpload(image, task.staging_buffer,
+                                              FixSmallVectorADL(uploads), z_start, z_count);
+                task.last_submitted_offset += static_cast<size_t>(z_count) * task.bytes_per_slice;
+                task.slices_submitted += z_count;
+            }
        }
+    } else if (is_final_batch && task.slices_submitted == 0) {
+        const auto uploads = FullUploadSwizzles(task.info);
+        runtime.AccelerateImageUpload(image, task.staging_buffer, FixSmallVectorADL(uploads), 0,
+                                      image.info.size.depth);
+        task.slices_submitted = image.info.size.depth;
    }

-    // Check if complete
-    const u32 slices_submitted = static_cast<u32>(task.last_submitted_offset / task.bytes_per_slice);
-    const bool all_slices_submitted = slices_submitted >= image.info.size.depth;
+    const bool all_slices_submitted = task.slices_submitted >= image.info.size.depth;

    if (is_final_batch && all_slices_submitted) {
        runtime.FreeDeferredStagingBuffer(task.staging_buffer);
@@ -139,6 +139,8 @@ class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelI
        AsyncBuffer staging_buffer;
        size_t last_submitted_offset = 0;
        size_t bytes_per_slice;
+        u32 slices_submitted = 0;
+        bool chunked = false;
        bool initialized = false;
    };

@@ -708,6 +708,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
        descriptor_indexing.shaderUniformTexelBufferArrayDynamicIndexing = false;
        descriptor_indexing.shaderStorageTexelBufferArrayDynamicIndexing = false;
        descriptor_indexing.shaderUniformBufferArrayNonUniformIndexing = false;
+        descriptor_indexing.shaderStorageBufferArrayNonUniformIndexing = false;
        descriptor_indexing.shaderInputAttachmentArrayNonUniformIndexing = false;
        descriptor_indexing.descriptorBindingUniformBufferUpdateAfterBind = false;
        descriptor_indexing.descriptorBindingSampledImageUpdateAfterBind = false;
@@ -1218,6 +1219,11 @@ bool Device::GetSuitability(bool requires_swapchain) {
            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT;
        SetNext(next, properties.transform_feedback);
    }
+    if (extensions.vertex_attribute_divisor) {
+        properties.vertex_attribute_divisor.sType =
+            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT;
+        SetNext(next, properties.vertex_attribute_divisor);
+    }
    if (extensions.maintenance5) {
        properties.maintenance5.sType =
            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_5_PROPERTIES_KHR;
@@ -70,6 +70,7 @@ VK_DEFINE_HANDLE(VmaAllocator)
    FEATURE(EXT, ProvokingVertex, PROVOKING_VERTEX, provoking_vertex)                              \
    FEATURE(EXT, Robustness2, ROBUSTNESS_2, robustness2)                                           \
    FEATURE(EXT, TransformFeedback, TRANSFORM_FEEDBACK, transform_feedback)                        \
+    FEATURE(EXT, VertexAttributeDivisor, VERTEX_ATTRIBUTE_DIVISOR, vertex_attribute_divisor)       \
    FEATURE(EXT, VertexInputDynamicState, VERTEX_INPUT_DYNAMIC_STATE, vertex_input_dynamic_state)  \
    FEATURE(KHR, Maintenance5, MAINTENANCE_5, maintenance5)                                        \
    FEATURE(KHR, Maintenance6, MAINTENANCE_6, maintenance6)                                        \
@@ -92,7 +93,6 @@ VK_DEFINE_HANDLE(VmaAllocator)
    EXTENSION(EXT, SHADER_STENCIL_EXPORT, shader_stencil_export)                                   \
    EXTENSION(EXT, SHADER_VIEWPORT_INDEX_LAYER, shader_viewport_index_layer)                       \
    EXTENSION(EXT, TOOLING_INFO, tooling_info)                                                     \
-    EXTENSION(EXT, VERTEX_ATTRIBUTE_DIVISOR, vertex_attribute_divisor)                             \
    EXTENSION(KHR, CREATE_RENDERPASS_2, create_renderpass2)                                        \
    EXTENSION(KHR, DEPTH_STENCIL_RESOLVE, depth_stencil_resolve)                                   \
    EXTENSION(KHR, DRAW_INDIRECT_COUNT, draw_indirect_count)                                       \
@@ -359,8 +359,8 @@ public:

 #define FN_MAX_LIMIT_LIST \
    FN_MAX_LIMIT_ELEM(ComputeSharedMemorySize) \
+    FN_MAX_LIMIT_ELEM(ComputeWorkGroupInvocations) \
    FN_MAX_LIMIT_ELEM(PerStageDescriptorSampledImages) \
-    FN_MAX_LIMIT_ELEM(PerStageDescriptorStorageBuffers) \
    FN_MAX_LIMIT_ELEM(PerStageResources) \
    FN_MAX_LIMIT_ELEM(DescriptorSetSamplers) \
    FN_MAX_LIMIT_ELEM(DescriptorSetUniformBuffers) \
@@ -416,10 +416,6 @@ FN_MAX_LIMIT_LIST
        return features.descriptor_indexing.shaderStorageTexelBufferArrayNonUniformIndexing;
    }

-    bool IsStorageBufferArrayNonUniformIndexingSupported() const {
-        return features.descriptor_indexing.shaderStorageBufferArrayNonUniformIndexing;
-    }
-
    /// Returns true if the device supports float64 natively.
    bool IsFloat64Supported() const {
        return features.features.shaderFloat64;
@@ -694,6 +690,32 @@ FN_MAX_LIMIT_LIST
        return features.host_query_reset.hostQueryReset != VK_FALSE;
    }

+    u32 GetMaxVertexAttribDivisor() const {
+        const u32 reported = properties.vertex_attribute_divisor.maxVertexAttribDivisor;
+        if (reported == 0) {
+            return 1;
+        }
+        return reported;
+    }
+
+    bool IsVertexAttributeInstanceRateZeroDivisorSupported() const {
+        return features.vertex_attribute_divisor.vertexAttributeInstanceRateZeroDivisor == VK_TRUE;
+    }
+
+    u32 GetVertexAttribDivisor(u32 frequency) const {
+        const u32 max_divisor = GetMaxVertexAttribDivisor();
+        if (frequency == 0) {
+            if (IsVertexAttributeInstanceRateZeroDivisorSupported()) {
+                return 0;
+            }
+            return max_divisor;
+        }
+        if (frequency > max_divisor) {
+            return max_divisor;
+        }
+        return frequency;
+    }
+
    /// Returns true if the device supports VK_EXT_transform_feedback.
    bool IsExtTransformFeedbackSupported() const {
        return extensions.transform_feedback;
@@ -1194,6 +1216,7 @@ private:
        VkPhysicalDeviceDescriptorBufferPropertiesEXT descriptor_buffer{};
        VkPhysicalDeviceSubgroupSizeControlProperties subgroup_size_control{};
        VkPhysicalDeviceTransformFeedbackPropertiesEXT transform_feedback{};
+        VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT vertex_attribute_divisor{};
        VkPhysicalDeviceMaintenance5PropertiesKHR maintenance5{};
        VkPhysicalDeviceDepthStencilResolveProperties depth_stencil_resolve{};
        VkPhysicalDeviceCustomBorderColorPropertiesEXT custom_border_color{};
Author	SHA1	Message	Date
CamilleLaVey	5f0f6e6306	Experimenting with scissors offsets	2026-09-01 21:53:15 -04:00
CamilleLaVey	d43238b9af	Another change	2026-09-02 02:57:14 +02:00
CamilleLaVey	2addf1556e	Another check	2026-09-02 02:57:14 +02:00
CamilleLaVey	883cdead8b	Some sanity on the vertex stream	2026-09-02 02:57:14 +02:00
CamilleLaVey	104bfa3a8f	Another take for indexing guards	2026-09-02 02:57:14 +02:00
CamilleLaVey	023c4758e9	Unify vertex entries from the indexed and non-indexed resolves into a single path	2026-09-02 02:57:14 +02:00
CamilleLaVey	b9239c4944	Add instance draw counts to vertex buffers	2026-09-02 02:57:14 +02:00
CamilleLaVey	f229172d3b	Some cleanups on the vertex buffer path	2026-09-02 02:57:14 +02:00
CamilleLaVey	eb18c5b8bd	Removing redundant stagging per downloaded and swizzled image	2026-09-02 02:57:14 +02:00
CamilleLaVey	0c7d068112	Blit corrections within the image initialization	2026-09-02 02:57:14 +02:00
CamilleLaVey	f29a23738a	A quick change on the e3D pass	2026-09-02 02:57:14 +02:00
CamilleLaVey	7e5e326842	Another meow	2026-09-02 02:57:14 +02:00
CamilleLaVey	f93e4031ba	tie gpu modified regions to rescalable textures	2026-09-02 02:57:14 +02:00
CamilleLaVey	7e55b4b563	Dale a tu cuerpo alegría macarena	2026-09-02 02:57:14 +02:00
CamilleLaVey	ee5049f0d3	Mercy please	2026-09-02 02:57:14 +02:00
CamilleLaVey	15de6e0e08	AAAAAAH	2026-09-02 02:57:14 +02:00
CamilleLaVey	75519010d2	Refining fix	2026-09-02 02:57:14 +02:00
CamilleLaVey	3e61e6da87	AQUI LLEGO TU LOBITO, MI LOBA.	2026-09-02 02:57:14 +02:00
CamilleLaVey	8242e54e42	This is so fucking tiring	2026-09-02 02:57:14 +02:00
CamilleLaVey	47b310266e	Another try to fix QCOM driver, dear god.	2026-09-02 02:57:14 +02:00
CamilleLaVey	3cfac4abcd	Just a small touch on the 3d swizzle params	2026-09-02 02:57:14 +02:00
CamilleLaVey	92c52342b8	Another try on the 3D pass	2026-09-02 02:57:14 +02:00
CamilleLaVey	db2bdde9b3	Revert "Just another change" This reverts commit `2ea449f71a`.	2026-09-02 02:57:14 +02:00
CamilleLaVey	ce0b15dcf7	Just another change	2026-09-02 02:57:14 +02:00
CamilleLaVey	db922e50e3	Another check on feedback loop	2026-09-02 02:57:14 +02:00
CamilleLaVey	ca52cf679a	Another try on unswizzling	2026-09-02 02:57:14 +02:00