Update src/shader_recompiler/backend/glsl/emit_glsl.cpp

unrelated. merely to avoid build error.

CONTRIBUTING.md

@@ -22,9 +22,9 @@ Eden is free, open-source, copyleft software, licensed under the terms of the [G
 
 - No LLM or AI usage, *period*, for patches, pull requests, issues, comments, debugging, brainstorming, etc.
   - For details on why, see the [detailed AI policy](docs/policies/AI.md).
-- Usage of any form of profanity or otherwise unsavory language is prohibited.
+- Usage of any form of profanity or otherwise unsavory language is generally discouraged.
   - This is primarily because it rarely helps to actually understand what's going on.
-  - Remember that your comments should be focused and actually address what's happening.
+  - Remember that your comments should be focused and actually address what's happening. With very few exceptions, expletives are actively detrimental at best.
 - New code must follow the same general style as the surrounding codebase. Exceptions may be granted in certain cases.
 - Maintainers reserve the right to change your patches and pull requests at will. We will try to avoid this.
   - You should generally respect all decisions made by the [code owners](docs/CODEOWNERS) in your particular subsystem.

docs/Caveats.md

@@ -239,6 +239,6 @@ find ./*/ -name "*.dll" | while read -r dll; do deps "$dll"; done
 
 ## RedoxOS
 
-Package installs randomly hang at times, you may need to restart it about 3-5 times to complete. Always do `sudo pkg update` or it will hang. Try installing each package individually.
+The package install may randomly hang at times, in which case it has to be restarted. ALWAYS do a `sudo pkg update` or the chances of it hanging will be close to 90%. If "multiple" installs fail at once, try installing 1 by 1 the packages.
 
-When CMake invokes certain file syscalls - most of the time it causes crashes, and corruptions on the address space. Reboot if there's a freeze over 5 minutes.
+When CMake invokes certain file syscalls - it may sometimes cause crashes or corruptions on the (kernel?) address space - so reboot the system if there is a "hang" in CMake.

docs/Development.md

@@ -38,7 +38,7 @@ Pull requests are only to be merged by core developers when properly tested and
 
 - The level of namespacing is generally left to the committer's choice.
 - However, we never recommend going more than two levels *except* in `hle`, in which case you may go as many as four levels depending on the specificity of your changes.
-- Occasionally, up to two additional namespaces may be provided for more clarity.
+- Ocassionally, up to two additional namespaces may be provided for more clarity.
   - Changes that affect the entire project (sans CMake changes) should be namespaced as `meta`.
 - Maintainers are permitted to change namespaces at will.
 - Commits within PRs are not required to be namespaced, but it is highly recommended.
@@ -51,8 +51,8 @@ When adding new settings, use `tr("Setting:")` if the setting is meant to be a f
 - Debug settings must never be turned on by default.
 - Provide reasonable bounds (for example, a setting controlling the amount of VRAM should never be 0).
 - The description of the setting must be short and concise, if the setting "does a lot of things" consider splitting the setting into multiple if possible.
-- Try to avoid excessive/redundant explanations "recommended for most users and games" can just be "(recommended)".
-- Try to not write "slow/fast" options unless it clearly degrades/increases performance for a given case, as most options may modify behavior that result in different metrics across different systems. If for example the option is an "accuracy" option, writing "High" is sufficient to imply "Slow". No need to write "High (Slow)".
+- Try to avoid excessive/redundant explainations "recommended for most users and games" can just be "(recommended)".
+- Try to not write "slow/fast" options unless it clearly degrades/increases performance for a given case, as most options may modify behaviour that result in different metrics accross different systems. If for example the option is an "accuracy" option, writing "High" is sufficient to imply "Slow". No need to write "High (Slow)".
 
 Some examples:
 
@@ -91,7 +91,7 @@ You may additionally need the `Qt Extension Pack` extension if building Qt.
 
 # Build speedup
 
-If you have an HDD, use ramdisk (build in RAM); you need about 4GB for a `RelWithDebInfo` build:
+If you have an HDD, use ramdisk (build in RAM), approximately you need 4GB for a full build with debug symbols:
 
 ```sh
 mkdir /tmp/ramdisk

docs/Settings.md

@@ -3,9 +3,9 @@
 > [!WARNING]
 > This guide is intended for developers ONLY. If you're looking for configuring the emulator itself, read **[the user handbook](./user/README.md)**.
 
-Settings have their own dedicated subsystem. Toggles can be added to gate features and fixes, as not all of them apply to every game.
+Settings on the emulator are very important, toggles and such can be used to guard and/or add branches to paths where some games may crash while others won't, and viceversa.
 
-However, this process can be tedious for those unfamiliar; this document serves as an outline for the settings subsystem.
+However, this process can be tedious for those unfamiliar; this document serves as an outline/documentation for the settings subsystem.
 
 ## Index
 

docs/user/Architectures.md

@@ -12,7 +12,7 @@ AMD64, aka x86_64, is the most tested and supported architecture for desktop tar
 
 ### Caveats
 
-AMD64 systems are almost always limited by the CPU. For example, a Zen 5/RX 6600 system will often hit max CPU usage before the GPU ever reaches 70% usage, with minimal exceptions (that tend to pop up only at >200fps). JIT is slow.
+AMD64 systems are almost always limited by the CPU. For example, a Zen 5/RX 6600 system will often hit max CPU usage before the GPU ever reaches 70% usage, with minimal exceptions (that tend to pop up only at >200fps). JIT is slow!
 
 Computers on Linux will almost always run Eden strictly better than an equivalent machine on Windows. This is largely due to the way the Linux kernel handles memory management (and the lack of Microsoft spyware).
 
@@ -24,7 +24,7 @@ ARM64, aka aarch64, is the only supported architecture for Android, with limited
 
 ### Caveats
 
-NCE (Native Code Execution) is currently only available on Android and (experimentally) Linux. Support for macOS is in the works, but Windows is extremely unlikely to ever happen (if you want it -- submit patches). Generally, if NCE is available, you should pretty much always use it due to the massive performance hit JIT has.
+NCE (Native Code Execution) is currently only available on Android and (experimentally) Linux. Support for macOS is in the works, but Windows is extremely unlikely to ever happen (if you want it--submit patches!). Generally, if NCE is available, you should pretty much always use it due to the massive performance hit JIT has.
 
 When NCE is enabled, do note that the GPU will almost always be the limiting factor. This is especially the case for Android, as well as desktops that lack dedicated GPUs; Adreno, Mali, PowerVR, etc. GPUs are generally significantly weaker relative to their respective CPUs.
 
@@ -44,7 +44,7 @@ Only Fedora/riscv64 has been tested, but in theory, every riscv64 distribution t
 
 ## Other
 
-Other architectures, such as SPARC, MIPS, PowerPC, Loong, and all 32-bit architectures are completely unsupported, as there is no JIT backend or emitter thereof. If you want support for it -- submit patches.
+Other architectures, such as SPARC, MIPS, PowerPC, Loong, and all 32-bit architectures are completely unsupported, as there is no JIT backend or emitter thereof. If you want support for it -- submit patches!
 
 IA-64 (Itanium) support is completely unknown. Existing amd64 packages will not run on IA-64 (assuming you can even find a supported Windows/Linux distribution)
 
@@ -106,14 +106,14 @@ Qualcomm Snapdragon SoCs are generally the most well supported.
   * A good base to compare to is the Snapdragon 865--e.g. [Tensor vs SD865](https://archive.is/M1P58)
   * Some benchmarks may be misleading due to thermal throttling OR RAM requirements.
     - For example, a Pixel 6a (Tensor G1) performs about 1/3 as well as an 865 due to its lack of RAM and poor thermals.
-  * Remember--always use a cooler if you can, and you MUST have *at least* 8GB of RAM.
+  * Remember--always use a cooler if you can, and you MUST have *at least* 8GB of RAM!
 - If you're not sure what SoC you have, check [GSMArena](https://www.gsmarena.com) - e.g. [Pixel 9 Pro](https://archive.ph/91VhA)
 
 Custom ROMs are recommended, *as long as* you know what you're doing.
 - For most devices, [LineageOS](https://lineageos.org/) is preferred.
 - [CalyxOS](https://calyxos.org/) is available as well.
 - For Google Pixel devices ONLY... and [soon another OEM](https://archive.ph/cPpMd)... [GrapheneOS](https://grapheneos.org/) is highly recommended.
-  * As of October 5, 2025, the Pixel 10 line is unsupported, however, [it will be](https://archive.is/viAUl) in the very near future.
+  * As of October 5, 2025, the Pixel 10 line is unsupported, however, [it will be](https://archive.is/viAUl) in the very near future!
   * Keep checking the [FAQ page](https://grapheneos.org/faq#supported-devices) for news.
 - Custom ROMs will likely be exclusively recommended in the future due to Google's upcoming [draconian](https://archive.is/hGIjZ), [anti-privacy, anti-user](https://archive.is/mc1CJ) verification requirements.
 

externals/tz/tz/tz.h

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

src/common/demangle.cpp

@@ -5,6 +5,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <string>
+#include <cstdlib>
 #include <string_view>
 #ifdef _WIN32
 #include <llvm/Demangle/Demangle.h>

src/shader_recompiler/backend/glsl/emit_glsl.cpp

@@ -1,3 +1,6 @@
+// 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
 
@@ -29,7 +32,7 @@ struct FuncTraits<ReturnType_ (*)(Args...)> {
 };
 
 template <auto func, typename... Args>
-void SetDefinition(EmitContext& ctx, IR::Inst* inst, Args... args) {
+[[maybe_unused]] void SetDefinition(EmitContext& ctx, IR::Inst* inst, Args... args) {
     inst->SetDefinition<Id>(func(ctx, std::forward<Args>(args)...));
 }
 

src/shader_recompiler/backend/spirv/emit_spirv.cpp

@@ -492,6 +492,9 @@ 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);
+        }
     }
 }
 

src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp

@@ -4,8 +4,6 @@
 // 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"
@@ -23,29 +21,13 @@ 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");
     }
-    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);
+    return ctx.StoragePointer(binding.U32(), ctx.Def(offset), type_def, static_cast<u32>(element_size), member_ptr);
 }
 
 std::pair<Id, Id> AtomicArgs(EmitContext& ctx) {
@@ -216,16 +198,14 @@ 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 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);
+    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);
 }
 
 Id EmitStorageAtomicDec32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                           Id 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);
+    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);
 }
 
 Id EmitStorageAtomicAnd32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
@@ -364,56 +344,49 @@ Id EmitStorageAtomicExchange32x2(EmitContext& ctx, const IR::Value& binding,
 
 Id EmitStorageAtomicAddF32(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                            Id 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);
+    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);
 }
 
 Id EmitStorageAtomicAddF16x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                              Id 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)};
+    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)};
     return ctx.OpBitcast(ctx.U32[1], result);
 }
 
 Id EmitStorageAtomicAddF32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                              Id 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)};
+    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)};
     return ctx.OpPackHalf2x16(ctx.U32[1], result);
 }
 
 Id EmitStorageAtomicMinF16x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                              Id 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)};
+    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)};
     return ctx.OpBitcast(ctx.U32[1], result);
 }
 
 Id EmitStorageAtomicMinF32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                              Id 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)};
+    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)};
     return ctx.OpPackHalf2x16(ctx.U32[1], result);
 }
 
 Id EmitStorageAtomicMaxF16x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                              Id 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)};
+    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)};
     return ctx.OpBitcast(ctx.U32[1], result);
 }
 
 Id EmitStorageAtomicMaxF32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset,
                              Id 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)};
+    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)};
     return ctx.OpPackHalf2x16(ctx.U32[1], result);
 }
 

src/shader_recompiler/backend/spirv/emit_spirv_memory.cpp

@@ -4,46 +4,33 @@
 // 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 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);
-    }
-    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);
-    }
+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) {
-        index = ctx.OpIAdd(ctx.U32[1], index, ctx.Const(index_offset));
+        byte_offset = ctx.OpIAdd(ctx.U32[1], byte_offset, ctx.Const(static_cast<u32>(index_offset * element_size)));
     }
-    return index;
+    return byte_offset;
 }
 
 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 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);
+    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);
 }
 
 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);
@@ -57,7 +44,7 @@ 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);

src/shader_recompiler/backend/spirv/spirv_emit_context.cpp

@@ -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);
@@ -309,23 +309,27 @@ 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 id{ctx.AddGlobalVariable(struct_pointer, spv::StorageClass::StorageBuffer)};
+        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)};
         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);
         }
-        for (size_t i = 0; i < desc.count; ++i) {
-            ctx.ssbos[index + i].*member_type = id;
-        }
-        index += desc.count;
-        binding += desc.count;
+        ctx.ssbos[index].*member_type = id;
+        ++index;
+        ++binding;
     }
 }
 
@@ -368,8 +372,7 @@ Id CasFunction(EmitContext& ctx, Operation operation, Id value_type) {
     return func;
 }
 
-Id CasLoop(EmitContext& ctx, Operation operation, Id array_pointer, Id element_pointer,
-           Id value_type, Id memory_type, spv::Scope scope) {
+Id CasLoop(EmitContext& ctx, Operation operation, 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)};
@@ -379,14 +382,12 @@ Id CasLoop(EmitContext& ctx, Operation operation, Id array_pointer, Id element_p
     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, ctx.U32[1], value_type, array_pointer)};
+    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{ctx.OpFunction(value_type, spv::FunctionControlMask::MaskNone, func_type)};
-    const Id index{ctx.OpFunctionParameter(ctx.U32[1])};
+    const Id address{ctx.OpFunctionParameter(is_shared ? ctx.U32[1] : element_pointer)};
     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);
@@ -395,8 +396,13 @@ Id CasLoop(EmitContext& ctx, Operation operation, Id array_pointer, Id element_p
     ctx.OpBranch(continue_block);
 
     ctx.AddLabel(continue_block);
-    const Id word_pointer{is_struct ? ctx.OpAccessChain(element_pointer, base, zero, index)
-                                    : ctx.OpAccessChain(element_pointer, base, index)};
+    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);
+    }()};
     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)};
@@ -480,6 +486,7 @@ 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);
@@ -532,6 +539,36 @@ 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();
 
@@ -696,12 +733,10 @@ 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_memory_u32_type,
-                                       shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
+        increment_cas_shared = CasLoop(*this, Operation::Increment, shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
     }
     if (program.info.uses_shared_decrement) {
-        decrement_cas_shared = CasLoop(*this, Operation::Decrement, shared_memory_u32_type,
-                                       shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
+        decrement_cas_shared = CasLoop(*this, Operation::Decrement, shared_u32, U32[1], U32[1], spv::Scope::Workgroup);
     }
 }
 
@@ -945,8 +980,7 @@ 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, Id element_pointer, u32 shift,
-                               auto&& callback) {
+    const auto define_body{[&](DefPtr ssbo_member, Id addr, const StorageTypeDefinition& type_def, u32 shift, auto&& callback) {
         AddLabel();
         const size_t num_buffers{info.storage_buffers_descriptors.size()};
         for (size_t index = 0; index < num_buffers; ++index) {
@@ -973,30 +1007,28 @@ 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_index{OpShiftRightLogical(U32[1], ssbo_offset, Const(shift))};
-            const Id ssbo_pointer{OpAccessChain(element_pointer, ssbo_id, zero, ssbo_index)};
+            const Id ssbo_pointer{StoragePointer(static_cast<u32>(index), ssbo_offset, type_def, 1U << shift, ssbo_member)};
             callback(ssbo_pointer);
             AddLabel(else_label);
         }
     }};
-    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 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 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, element_pointer, shift, [&](Id ssbo_pointer) {
+        define_body(ssbo_member, addr, type_def, shift, [&](Id ssbo_pointer) {
             OpStore(ssbo_pointer, data);
             OpReturn();
         });
@@ -1006,10 +1038,9 @@ 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, element_type, type, shift)};
-            const Id write_func{define_write(ssbo_member, element_type, type, shift)};
+            const Id load_func{define_load(ssbo_member, type_def, type, shift)};
+            const Id write_func{define_write(ssbo_member, type_def, type, shift)};
             return std::make_pair(load_func, write_func);
         }};
     std::tie(load_global_func_u32, write_global_func_u32) =
@@ -1228,6 +1259,79 @@ 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;
@@ -1271,9 +1375,7 @@ void EmitContext::DefineStorageBuffers(const Info& info, u32& binding) {
         DefineSsbos(*this, storage_types.U32x4, &StorageDefinitions::U32x4, info, binding, U32[4],
                     sizeof(u32[4]));
     }
-    for (const StorageBufferDescriptor& desc : info.storage_buffers_descriptors) {
-        binding += desc.count;
-    }
+    binding += static_cast<u32>(info.storage_buffers_descriptors.size());
     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 ||
@@ -1282,40 +1384,31 @@ 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.array,
-                                     storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
+        increment_cas_ssbo = CasLoop(*this, Operation::Increment, 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.array,
-                                     storage_types.U32.element, U32[1], U32[1], spv::Scope::Device);
+        decrement_cas_ssbo = CasLoop(*this, Operation::Decrement, 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.array,
-                              storage_types.U32.element, F32[1], U32[1], spv::Scope::Device);
+        f32_add_cas = CasLoop(*this, Operation::FPAdd, 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.array,
-                                storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
+        f16x2_add_cas = CasLoop(*this, Operation::FPAdd, 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.array,
-                                storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
+        f16x2_min_cas = CasLoop(*this, Operation::FPMin, 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.array,
-                                storage_types.U32.element, F16[2], F16[2], spv::Scope::Device);
+        f16x2_max_cas = CasLoop(*this, Operation::FPMax, 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.array,
-                                storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
+        f32x2_add_cas = CasLoop(*this, Operation::FPAdd, 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.array,
-                                storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
+        f32x2_min_cas = CasLoop(*this, Operation::FPMin, 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.array,
-                                storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
+        f32x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.element, F32[2], F32[2], spv::Scope::Device);
     }
 }
 

src/shader_recompiler/backend/spirv/spirv_emit_context.h

@@ -114,7 +114,6 @@ struct UniformDefinitions {
 };
 
 struct StorageTypeDefinition {
-    Id array{};
     Id element{};
 };
 
@@ -173,6 +172,8 @@ 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);
     }
@@ -257,6 +258,11 @@ 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;
@@ -376,6 +382,7 @@ 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);
@@ -386,6 +393,7 @@ 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);

src/shader_recompiler/frontend/maxwell/translate_program.cpp

@@ -132,6 +132,14 @@ 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(
@@ -299,6 +307,7 @@ 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()) {
@@ -314,6 +323,7 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
 
     CollectInterpolationInfo(env, program);
     AddNVNStorageBuffers(program);
+    ConfigureStorageBufferMappings(program, normalized_host_info);
     return program;
 }
 

src/shader_recompiler/host_translate_info.h

@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <algorithm>
 #include "common/common_types.h"
 
 namespace Shader {
@@ -19,6 +20,7 @@ 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{};
@@ -38,12 +40,14 @@ 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);
@@ -53,6 +57,7 @@ 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:

src/shader_recompiler/profile.h

@@ -64,6 +64,7 @@ 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{};
 

src/shader_recompiler/shader_info.h

@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <algorithm>
 #include <array>
 #include <bitset>
 #include <map>
@@ -340,6 +341,10 @@ 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{};

src/video_core/buffer_cache/buffer_cache.h

@@ -7,6 +7,8 @@
 #pragma once
 
 #include <algorithm>
+#include <cstring>
+#include <limits>
 #include <memory>
 #include <numeric>
 
@@ -423,7 +425,7 @@ 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 cbuf_offset, bool is_written, u32 descriptor_count) {
     const bool already_enabled =
         ((channel_state->enabled_storage_buffers[stage] >> ssbo_index) & 1U) != 0;
     if constexpr (requires { runtime.ShouldLimitDynamicStorageBuffers(); }) {
@@ -448,8 +450,8 @@ 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);
-    return (channel_state->storage_buffers[stage][ssbo_index].buffer_id != NULL_BUFFER_ID);
+        StorageBufferBinding(ssbo_addr, cbuf_index, is_written, descriptor_count);
+    return channel_state->storage_buffers[stage][ssbo_index].gpu_addr != 0;
 }
 
 template <class P>
@@ -487,7 +489,7 @@ void BufferCache<P>::UnbindComputeStorageBuffers() {
 
 template <class P>
 void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
-                                              bool is_written) {
+                                              bool is_written, u32 descriptor_count) {
     if (ssbo_index >= channel_state->compute_storage_buffers.size()) [[unlikely]] {
         LOG_ERROR(HW_GPU, "Storage buffer index {} exceeds maximum storage buffer count",
                   ssbo_index);
@@ -524,7 +526,7 @@ 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);
+        StorageBufferBinding(ssbo_addr, cbuf_index, is_written, descriptor_count);
 }
 
 template <class P>
@@ -1000,27 +1002,52 @@ 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 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 StorageBufferBindingInfo& storage = channel_state->storage_buffers[stage][index];
         const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
-
-        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);
+        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);
+            }
         }
     });
 }
@@ -1136,28 +1163,53 @@ 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 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 StorageBufferBindingInfo& storage = channel_state->compute_storage_buffers[index];
         const bool is_written =
             ((channel_state->written_compute_storage_buffers >> index) & 1) != 0;
-
-        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);
+        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);
+            }
         }
     });
 }
@@ -1350,10 +1402,7 @@ 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) {
-        // 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;
+        UpdateStorageBuffer(channel_state->storage_buffers[stage][index]);
     });
 }
 
@@ -1414,12 +1463,54 @@ void BufferCache<P>::UpdateComputeUniformBuffers() {
 template <class P>
 void BufferCache<P>::UpdateComputeStorageBuffers() {
     ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) {
-        // Resolve buffer
-        Binding& binding = channel_state->compute_storage_buffers[index];
-        binding.buffer_id = FindBuffer(binding.device_addr, binding.size);
+        UpdateStorageBuffer(channel_state->compute_storage_buffers[index]);
     });
 }
 
+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) {
@@ -1841,6 +1932,11 @@ 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{};
@@ -1856,10 +1952,10 @@ 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, replace);
+    std::ranges::for_each(channel_state->storage_buffers, storage_replace);
     replace(channel_state->transform_feedback_buffers);
     replace(channel_state->compute_uniform_buffers);
-    replace(channel_state->compute_storage_buffers);
+    storage_replace(channel_state->compute_storage_buffers);
 
     // Mark the whole buffer as CPU written to stop tracking CPU writes
     if (!do_not_mark) {
@@ -1896,12 +1992,14 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
 }
 
 template <class P>
-Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
-                                             bool is_written) const {
+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
     const GPUVAddr gpu_addr = gpu_memory->Read<u64>(ssbo_addr);
 
     if (gpu_addr == 0) {
-        return NULL_BINDING;
+        return {.descriptor_count = descriptor_count};
     }
 
     const auto size = [&]() {
@@ -1923,21 +2021,17 @@ Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
     const GPUVAddr aligned_gpu_addr = Common::AlignDown(gpu_addr, alignment);
     const u32 aligned_size = static_cast<u32>(gpu_addr - aligned_gpu_addr) + size;
 
-    const std::optional<DAddr> aligned_device_addr = gpu_memory->GpuToCpuAddress(aligned_gpu_addr);
-    if (!aligned_device_addr || size == 0) {
+    if (!gpu_memory->GpuToCpuAddress(aligned_gpu_addr) || size == 0) {
         LOG_DEBUG(HW_GPU, "Failed to find storage buffer for cbuf index {}", cbuf_index);
-        return NULL_BINDING;
+        return {.descriptor_count = descriptor_count};
     }
-    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 DAddr cpu_end = Common::AlignUp(*device_addr + size, Core::DEVICE_PAGESIZE);
+    const GPUVAddr gpu_end = Common::AlignUp(gpu_addr + size, Core::DEVICE_PAGESIZE);
 
-    const Binding binding{
-        .device_addr = *aligned_device_addr,
-        .size = is_written ? aligned_size : static_cast<u32>(cpu_end - *aligned_device_addr),
-        .buffer_id = BufferId{},
+    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,
     };
     return binding;
 }

src/video_core/buffer_cache/buffer_cache_base.h

@@ -89,6 +89,15 @@ 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,
@@ -115,14 +124,15 @@ 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<Binding, NUM_STORAGE_BUFFERS>, NUM_STAGES> storage_buffers;
+    std::array<std::array<StorageBufferBindingInfo, 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<Binding, NUM_STORAGE_BUFFERS> compute_storage_buffers;
+    std::array<StorageBufferBindingInfo, NUM_STORAGE_BUFFERS> compute_storage_buffers;
     std::array<TextureBufferBinding, NUM_TEXTURE_BUFFERS> compute_texture_buffers;
 
     std::array<u32, NUM_STAGES> enabled_uniform_buffer_masks{};
@@ -249,7 +259,7 @@ public:
     void UnbindGraphicsStorageBuffers(size_t stage);
 
     bool BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
-                                   bool is_written);
+                                   bool is_written, u32 descriptor_count = 1);
 
     void UnbindGraphicsTextureBuffers(size_t stage);
 
@@ -259,7 +269,7 @@ public:
     void UnbindComputeStorageBuffers();
 
     void BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
-                                  bool is_written);
+                                  bool is_written, u32 descriptor_count = 1);
 
     void UnbindComputeTextureBuffers();
 
@@ -400,6 +410,8 @@ private:
 
     void UpdateStorageBuffers(size_t stage);
 
+    void UpdateStorageBuffer(StorageBufferBindingInfo& binding);
+
     void UpdateTextureBuffers(size_t stage);
 
     void UpdateTransformFeedbackBuffers();
@@ -449,8 +461,9 @@ private:
 
     void DeleteBuffer(BufferId buffer_id, bool do_not_mark = false);
 
-    [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
-                                               bool is_written) const;
+    [[nodiscard]] StorageBufferBindingInfo StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index,
+                                                                bool is_written,
+                                                                u32 descriptor_count) const;
 
     [[nodiscard]] TextureBufferBinding GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size,
                                                                PixelFormat format);

src/video_core/renderer_vulkan/pipeline_helper.h

@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <array>
 #include <cstddef>
 #include <optional>
 
@@ -136,6 +137,7 @@ 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) +
@@ -268,6 +270,12 @@ 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);

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -695,7 +695,7 @@ 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_TRANSFER_DST_BIT | VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
         .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
         .queueFamilyIndexCount = 0,
         .pQueueFamilyIndices = nullptr,

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -155,6 +155,13 @@ 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);
     }

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -157,9 +157,7 @@ bool ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute,
     buffer_cache.UnbindComputeStorageBuffers();
     size_t ssbo_index{};
     for (const auto& desc : info.storage_buffers_descriptors) {
-        ASSERT(desc.count == 1);
-        buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset,
-                                              desc.is_written);
+        buffer_cache.BindComputeStorageBuffer(ssbo_index, desc.cbuf_index, desc.cbuf_offset, desc.is_written, desc.count);
         ++ssbo_index;
     }
 

src/video_core/renderer_vulkan/vk_descriptor_pool.cpp

@@ -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);
+        bank.storage_buffers += Accumulate(info.storage_buffers_descriptors) + static_cast<u32>(Shader::UsesStorageBufferMappings(info));
         bank.texture_buffers += Accumulate(info.texture_buffer_descriptors);
         bank.image_buffers += Accumulate(info.image_buffer_descriptors);
         bank.textures += Accumulate(info.texture_descriptors);

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -367,9 +367,8 @@ 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.cbuf_offset, desc.is_written, desc.count);
                 ++ssbo_index;
             }
         }

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -62,7 +62,12 @@ using VideoCommon::FileEnvironment;
 using VideoCommon::GenericEnvironment;
 using VideoCommon::GraphicsEnvironment;
 
-constexpr u32 CACHE_VERSION = 18;
+// 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 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'};
@@ -432,6 +437,8 @@ 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(),
 
@@ -460,6 +467,7 @@ 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(),
@@ -479,6 +487,20 @@ 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();
 

src/video_core/vulkan_common/vulkan_device.cpp

@@ -708,7 +708,6 @@ 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;

src/video_core/vulkan_common/vulkan_device.h

@@ -360,6 +360,7 @@ public:
 #define FN_MAX_LIMIT_LIST \
     FN_MAX_LIMIT_ELEM(ComputeSharedMemorySize) \
     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) \
@@ -415,6 +416,10 @@ 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;