8 files changed, 298 insertions, 164 deletions

diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 75ef8d541..f31d960c7 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -345,7 +345,7 @@ void RasterizerOpenGL::ConfigureFramebuffers() {
 
     texture_cache.GuardRenderTargets(true);
 
-    View depth_surface = texture_cache.GetDepthBufferSurface(true);
+    View depth_surface = texture_cache.GetDepthBufferSurface();
 
     const auto& regs = gpu.regs;
     UNIMPLEMENTED_IF(regs.rt_separate_frag_data == 0);
@@ -354,7 +354,7 @@ void RasterizerOpenGL::ConfigureFramebuffers() {
     FramebufferCacheKey key;
     const auto colors_count = static_cast<std::size_t>(regs.rt_control.count);
     for (std::size_t index = 0; index < colors_count; ++index) {
-        View color_surface{texture_cache.GetColorBufferSurface(index, true)};
+        View color_surface{texture_cache.GetColorBufferSurface(index)};
         if (!color_surface) {
             continue;
         }
@@ -387,12 +387,12 @@ void RasterizerOpenGL::ConfigureClearFramebuffer(bool using_color_fb, bool using
     View color_surface;
     if (using_color_fb) {
         const std::size_t index = regs.clear_buffers.RT;
-        color_surface = texture_cache.GetColorBufferSurface(index, true);
+        color_surface = texture_cache.GetColorBufferSurface(index);
         texture_cache.MarkColorBufferInUse(index);
     }
     View depth_surface;
     if (using_depth_fb || using_stencil_fb) {
-        depth_surface = texture_cache.GetDepthBufferSurface(true);
+        depth_surface = texture_cache.GetDepthBufferSurface();
         texture_cache.MarkDepthBufferInUse();
     }
     texture_cache.GuardRenderTargets(false);
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 160ae4340..1f1f01313 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -1819,15 +1819,15 @@ private:
     }
 
     Expression HMergeH0(Operation operation) {
-        std::string dest = VisitOperand(operation, 0).AsUint();
-        std::string src = VisitOperand(operation, 1).AsUint();
-        return {fmt::format("(({} & 0x0000FFFFU) | ({} & 0xFFFF0000U))", src, dest), Type::Uint};
+        const std::string dest = VisitOperand(operation, 0).AsUint();
+        const std::string src = VisitOperand(operation, 1).AsUint();
+        return {fmt::format("bitfieldInsert({}, {}, 0, 16)", dest, src), Type::Uint};
     }
 
     Expression HMergeH1(Operation operation) {
-        std::string dest = VisitOperand(operation, 0).AsUint();
-        std::string src = VisitOperand(operation, 1).AsUint();
-        return {fmt::format("(({} & 0x0000FFFFU) | ({} & 0xFFFF0000U))", dest, src), Type::Uint};
+        const std::string dest = VisitOperand(operation, 0).AsUint();
+        const std::string src = VisitOperand(operation, 1).AsUint();
+        return {fmt::format("bitfieldInsert({}, {}, 16, 16)", dest, src), Type::Uint};
     }
 
     Expression HPack2(Operation operation) {
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 36590a6d0..0b4d999d7 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -411,14 +411,13 @@ CachedSurfaceView::~CachedSurfaceView() = default;
 void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const {
     ASSERT(params.num_levels == 1);
 
-    const GLuint texture = surface.GetTexture();
     if (params.num_layers > 1) {
         // Layered framebuffer attachments
         UNIMPLEMENTED_IF(params.base_layer != 0);
 
         switch (params.target) {
         case SurfaceTarget::Texture2DArray:
-            glFramebufferTexture(target, attachment, texture, params.base_level);
+            glFramebufferTexture(target, attachment, GetTexture(), params.base_level);
             break;
         default:
             UNIMPLEMENTED();
@@ -427,6 +426,7 @@ void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const {
     }
 
     const GLenum view_target = surface.GetTarget();
+    const GLuint texture = surface.GetTexture();
     switch (surface.GetSurfaceParams().target) {
     case SurfaceTarget::Texture1D:
         glFramebufferTexture1D(target, attachment, view_target, texture, params.base_level);
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 0a2ea4fd4..6b99cbbbc 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -599,7 +599,7 @@ RasterizerVulkan::Texceptions RasterizerVulkan::UpdateAttachments() {
     Texceptions texceptions;
     for (std::size_t rt = 0; rt < Maxwell::NumRenderTargets; ++rt) {
         if (update_rendertargets) {
-            color_attachments[rt] = texture_cache.GetColorBufferSurface(rt, true);
+            color_attachments[rt] = texture_cache.GetColorBufferSurface(rt);
         }
         if (color_attachments[rt] && WalkAttachmentOverlaps(*color_attachments[rt])) {
             texceptions[rt] = true;
@@ -607,7 +607,7 @@ RasterizerVulkan::Texceptions RasterizerVulkan::UpdateAttachments() {
     }
 
     if (update_rendertargets) {
-        zeta_attachment = texture_cache.GetDepthBufferSurface(true);
+        zeta_attachment = texture_cache.GetDepthBufferSurface();
     }
     if (zeta_attachment && WalkAttachmentOverlaps(*zeta_attachment)) {
         texceptions[ZETA_TEXCEPTION_INDEX] = true;
diff --git a/src/video_core/shader/decode/conversion.cpp b/src/video_core/shader/decode/conversion.cpp
index c72690b2b..b9989c88c 100644
--- a/src/video_core/shader/decode/conversion.cpp
+++ b/src/video_core/shader/decode/conversion.cpp
@@ -2,6 +2,10 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <limits>
+#include <optional>
+#include <utility>
+
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "video_core/engines/shader_bytecode.h"
@@ -15,9 +19,49 @@ using Tegra::Shader::OpCode;
 using Tegra::Shader::Register;
 
 namespace {
+
 constexpr OperationCode GetFloatSelector(u64 selector) {
     return selector == 0 ? OperationCode::FCastHalf0 : OperationCode::FCastHalf1;
 }
+
+constexpr u32 SizeInBits(Register::Size size) {
+    switch (size) {
+    case Register::Size::Byte:
+        return 8;
+    case Register::Size::Short:
+        return 16;
+    case Register::Size::Word:
+        return 32;
+    case Register::Size::Long:
+        return 64;
+    }
+    return 0;
+}
+
+constexpr std::optional<std::pair<s32, s32>> IntegerSaturateBounds(Register::Size src_size,
+                                                                   Register::Size dst_size,
+                                                                   bool src_signed,
+                                                                   bool dst_signed) {
+    const u32 dst_bits = SizeInBits(dst_size);
+    if (src_size == Register::Size::Word && dst_size == Register::Size::Word) {
+        if (src_signed == dst_signed) {
+            return std::nullopt;
+        }
+        return std::make_pair(0, std::numeric_limits<s32>::max());
+    }
+    if (dst_signed) {
+        // Signed destination, clamp to [-128, 127] for instance
+        return std::make_pair(-(1 << (dst_bits - 1)), (1 << (dst_bits - 1)) - 1);
+    } else {
+        // Unsigned destination
+        if (dst_bits == 32) {
+            // Avoid shifting by 32, that is undefined behavior
+            return std::make_pair(0, s32(std::numeric_limits<u32>::max()));
+        }
+        return std::make_pair(0, (1 << dst_bits) - 1);
+    }
+}
+
 } // Anonymous namespace
 
 u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
@@ -28,14 +72,13 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
     case OpCode::Id::I2I_R:
     case OpCode::Id::I2I_C:
     case OpCode::Id::I2I_IMM: {
-        UNIMPLEMENTED_IF(instr.conversion.int_src.selector != 0);
-        UNIMPLEMENTED_IF(instr.conversion.dst_size != Register::Size::Word);
-        UNIMPLEMENTED_IF(instr.alu.saturate_d);
+        const bool src_signed = instr.conversion.is_input_signed;
+        const bool dst_signed = instr.conversion.is_output_signed;
+        const Register::Size src_size = instr.conversion.src_size;
+        const Register::Size dst_size = instr.conversion.dst_size;
+        const u32 selector = static_cast<u32>(instr.conversion.int_src.selector);
 
-        const bool input_signed = instr.conversion.is_input_signed;
-        const bool output_signed = instr.conversion.is_output_signed;
-
-        Node value = [&]() {
+        Node value = [this, instr, opcode] {
             switch (opcode->get().GetId()) {
             case OpCode::Id::I2I_R:
                 return GetRegister(instr.gpr20);
@@ -48,16 +91,60 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
                 return Immediate(0);
             }
         }();
-        value = ConvertIntegerSize(value, instr.conversion.src_size, input_signed);
 
-        value = GetOperandAbsNegInteger(value, instr.conversion.abs_a, instr.conversion.negate_a,
-                                        input_signed);
-        if (input_signed != output_signed) {
-            value = SignedOperation(OperationCode::ICastUnsigned, output_signed, NO_PRECISE, value);
+        // Ensure the source selector is valid
+        switch (instr.conversion.src_size) {
+        case Register::Size::Byte:
+            break;
+        case Register::Size::Short:
+            ASSERT(selector == 0 || selector == 2);
+            break;
+        default:
+            ASSERT(selector == 0);
+            break;
+        }
+
+        if (src_size != Register::Size::Word || selector != 0) {
+            value = SignedOperation(OperationCode::IBitfieldExtract, src_signed, std::move(value),
+                                    Immediate(selector * 8), Immediate(SizeInBits(src_size)));
+        }
+
+        value = GetOperandAbsNegInteger(std::move(value), instr.conversion.abs_a,
+                                        instr.conversion.negate_a, src_signed);
+
+        if (instr.alu.saturate_d) {
+            if (src_signed && !dst_signed) {
+                Node is_negative = Operation(OperationCode::LogicalUGreaterEqual, value,
+                                             Immediate(1 << (SizeInBits(src_size) - 1)));
+                value = Operation(OperationCode::Select, std::move(is_negative), Immediate(0),
+                                  std::move(value));
+
+                // Simplify generated expressions, this can be removed without semantic impact
+                SetTemporary(bb, 0, std::move(value));
+                value = GetTemporary(0);
+
+                if (dst_size != Register::Size::Word) {
+                    const Node limit = Immediate((1 << SizeInBits(dst_size)) - 1);
+                    Node is_large =
+                        Operation(OperationCode::LogicalUGreaterThan, std::move(value), limit);
+                    value = Operation(OperationCode::Select, std::move(is_large), limit,
+                                      std::move(value));
+                }
+            } else if (const std::optional bounds =
+                           IntegerSaturateBounds(src_size, dst_size, src_signed, dst_signed)) {
+                value = SignedOperation(OperationCode::IMax, src_signed, std::move(value),
+                                        Immediate(bounds->first));
+                value = SignedOperation(OperationCode::IMin, src_signed, std::move(value),
+                                        Immediate(bounds->second));
+            }
+        } else if (dst_size != Register::Size::Word) {
+            // No saturation, we only have to mask the result
+            Node mask = Immediate((1 << SizeInBits(dst_size)) - 1);
+            value = Operation(OperationCode::UBitwiseAnd, std::move(value), std::move(mask));
         }
 
         SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
-        SetRegister(bb, instr.gpr0, value);
+        SetRegister(bb, instr.gpr0, std::move(value));
         break;
     }
     case OpCode::Id::I2F_R:
diff --git a/src/video_core/shader/decode/texture.cpp b/src/video_core/shader/decode/texture.cpp
index 48350e042..6c4a1358b 100644
--- a/src/video_core/shader/decode/texture.cpp
+++ b/src/video_core/shader/decode/texture.cpp
@@ -780,20 +780,6 @@ Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is
     // When lod is used always is in gpr20
     const Node lod = lod_enabled ? GetRegister(instr.gpr20) : Immediate(0);
 
-    // Fill empty entries from the guest sampler
-    const std::size_t entry_coord_count = GetCoordCount(sampler.GetType());
-    if (type_coord_count != entry_coord_count) {
-        LOG_WARNING(HW_GPU, "Bound and built texture types mismatch");
-
-        // When the size is higher we insert zeroes
-        for (std::size_t i = type_coord_count; i < entry_coord_count; ++i) {
-            coords.push_back(GetRegister(Register::ZeroIndex));
-        }
-
-        // Then we ensure the size matches the number of entries (dropping unused values)
-        coords.resize(entry_coord_count);
-    }
-
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
         auto coords_copy = coords;
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 88fe3e25f..cfc7fe6e9 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -108,7 +108,7 @@ public:
         }
 
         const auto params{SurfaceParams::CreateForTexture(format_lookup_table, tic, entry)};
-        const auto [surface, view] = GetSurface(gpu_addr, *cpu_addr, params, true, false);
+        const auto [surface, view] = GetSurface(gpu_addr, *cpu_addr, params, false);
         if (guard_samplers) {
             sampled_textures.push_back(surface);
         }
@@ -128,7 +128,7 @@ public:
             return GetNullSurface(SurfaceParams::ExpectedTarget(entry));
         }
         const auto params{SurfaceParams::CreateForImage(format_lookup_table, tic, entry)};
-        const auto [surface, view] = GetSurface(gpu_addr, *cpu_addr, params, true, false);
+        const auto [surface, view] = GetSurface(gpu_addr, *cpu_addr, params, false);
         if (guard_samplers) {
             sampled_textures.push_back(surface);
         }
@@ -143,7 +143,7 @@ public:
         return any_rt;
     }
 
-    TView GetDepthBufferSurface(bool preserve_contents) {
+    TView GetDepthBufferSurface() {
         std::lock_guard lock{mutex};
         auto& maxwell3d = system.GPU().Maxwell3D();
         if (!maxwell3d.dirty.flags[VideoCommon::Dirty::ZetaBuffer]) {
@@ -164,7 +164,7 @@ public:
             return {};
         }
         const auto depth_params{SurfaceParams::CreateForDepthBuffer(system)};
-        auto surface_view = GetSurface(gpu_addr, *cpu_addr, depth_params, preserve_contents, true);
+        auto surface_view = GetSurface(gpu_addr, *cpu_addr, depth_params, true);
         if (depth_buffer.target)
             depth_buffer.target->MarkAsRenderTarget(false, NO_RT);
         depth_buffer.target = surface_view.first;
@@ -174,7 +174,7 @@ public:
         return surface_view.second;
     }
 
-    TView GetColorBufferSurface(std::size_t index, bool preserve_contents) {
+    TView GetColorBufferSurface(std::size_t index) {
         std::lock_guard lock{mutex};
         ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets);
         auto& maxwell3d = system.GPU().Maxwell3D();
@@ -204,9 +204,8 @@ public:
             return {};
         }
 
-        auto surface_view =
-            GetSurface(gpu_addr, *cpu_addr, SurfaceParams::CreateForFramebuffer(system, index),
-                       preserve_contents, true);
+        auto surface_view = GetSurface(gpu_addr, *cpu_addr,
+                                       SurfaceParams::CreateForFramebuffer(system, index), true);
         if (render_targets[index].target)
             render_targets[index].target->MarkAsRenderTarget(false, NO_RT);
         render_targets[index].target = surface_view.first;
@@ -260,9 +259,9 @@ public:
         const std::optional<VAddr> src_cpu_addr =
             system.GPU().MemoryManager().GpuToCpuAddress(src_gpu_addr);
         std::pair<TSurface, TView> dst_surface =
-            GetSurface(dst_gpu_addr, *dst_cpu_addr, dst_params, true, false);
+            GetSurface(dst_gpu_addr, *dst_cpu_addr, dst_params, false);
         std::pair<TSurface, TView> src_surface =
-            GetSurface(src_gpu_addr, *src_cpu_addr, src_params, true, false);
+            GetSurface(src_gpu_addr, *src_cpu_addr, src_params, false);
         ImageBlit(src_surface.second, dst_surface.second, copy_config);
         dst_surface.first->MarkAsModified(true, Tick());
     }
@@ -451,22 +450,18 @@ private:
      * @param overlaps          The overlapping surfaces registered in the cache.
      * @param params            The parameters for the new surface.
      * @param gpu_addr          The starting address of the new surface.
-     * @param preserve_contents Indicates that the new surface should be loaded from memory or left
-     *                          blank.
      * @param untopological     Indicates to the recycler that the texture has no way to match the
      *                          overlaps due to topological reasons.
      **/
     std::pair<TSurface, TView> RecycleSurface(std::vector<TSurface>& overlaps,
                                               const SurfaceParams& params, const GPUVAddr gpu_addr,
-                                              const bool preserve_contents,
                                               const MatchTopologyResult untopological) {
-        const bool do_load = preserve_contents && Settings::values.use_accurate_gpu_emulation;
         for (auto& surface : overlaps) {
             Unregister(surface);
         }
         switch (PickStrategy(overlaps, params, gpu_addr, untopological)) {
         case RecycleStrategy::Ignore: {
-            return InitializeSurface(gpu_addr, params, do_load);
+            return InitializeSurface(gpu_addr, params, Settings::values.use_accurate_gpu_emulation);
         }
         case RecycleStrategy::Flush: {
             std::sort(overlaps.begin(), overlaps.end(),
@@ -476,7 +471,7 @@ private:
             for (auto& surface : overlaps) {
                 FlushSurface(surface);
             }
-            return InitializeSurface(gpu_addr, params, preserve_contents);
+            return InitializeSurface(gpu_addr, params);
         }
         case RecycleStrategy::BufferCopy: {
             auto new_surface = GetUncachedSurface(gpu_addr, params);
@@ -485,7 +480,7 @@ private:
         }
         default: {
             UNIMPLEMENTED_MSG("Unimplemented Texture Cache Recycling Strategy!");
-            return InitializeSurface(gpu_addr, params, do_load);
+            return InitializeSurface(gpu_addr, params);
         }
         }
     }
@@ -621,14 +616,11 @@ private:
      * @param params            The parameters on the new surface.
      * @param gpu_addr          The starting address of the new surface.
      * @param cache_addr        The starting address of the new surface on physical memory.
-     * @param preserve_contents Indicates that the new surface should be loaded from memory or
-     *                          left blank.
      */
     std::optional<std::pair<TSurface, TView>> Manage3DSurfaces(std::vector<TSurface>& overlaps,
                                                                const SurfaceParams& params,
                                                                const GPUVAddr gpu_addr,
-                                                               const VAddr cpu_addr,
-                                                               bool preserve_contents) {
+                                                               const VAddr cpu_addr) {
         if (params.target == SurfaceTarget::Texture3D) {
             bool failed = false;
             if (params.num_levels > 1) {
@@ -677,7 +669,7 @@ private:
                             return std::nullopt;
                         }
                         Unregister(surface);
-                        return InitializeSurface(gpu_addr, params, preserve_contents);
+                        return InitializeSurface(gpu_addr, params);
                     }
                     return std::nullopt;
                 }
@@ -688,7 +680,7 @@ private:
                     return {{surface, surface->GetMainView()}};
                 }
             }
-            return InitializeSurface(gpu_addr, params, preserve_contents);
+            return InitializeSurface(gpu_addr, params);
         }
     }
 
@@ -711,13 +703,10 @@ private:
      *
      * @param gpu_addr          The starting address of the candidate surface.
      * @param params            The parameters on the candidate surface.
-     * @param preserve_contents Indicates that the new surface should be loaded from memory or
-     *                          left blank.
      * @param is_render         Whether or not the surface is a render target.
      **/
     std::pair<TSurface, TView> GetSurface(const GPUVAddr gpu_addr, const VAddr cpu_addr,
-                                          const SurfaceParams& params, bool preserve_contents,
-                                          bool is_render) {
+                                          const SurfaceParams& params, bool is_render) {
         // Step 1
         // Check Level 1 Cache for a fast structural match. If candidate surface
         // matches at certain level we are pretty much done.
@@ -726,8 +715,7 @@ private:
             const auto topological_result = current_surface->MatchesTopology(params);
             if (topological_result != MatchTopologyResult::FullMatch) {
                 std::vector<TSurface> overlaps{current_surface};
-                return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
-                                      topological_result);
+                return RecycleSurface(overlaps, params, gpu_addr, topological_result);
             }
 
             const auto struct_result = current_surface->MatchesStructure(params);
@@ -752,7 +740,7 @@ private:
 
         // If none are found, we are done. we just load the surface and create it.
         if (overlaps.empty()) {
-            return InitializeSurface(gpu_addr, params, preserve_contents);
+            return InitializeSurface(gpu_addr, params);
         }
 
         // Step 3
@@ -762,15 +750,13 @@ private:
         for (const auto& surface : overlaps) {
             const auto topological_result = surface->MatchesTopology(params);
             if (topological_result != MatchTopologyResult::FullMatch) {
-                return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
-                                      topological_result);
+                return RecycleSurface(overlaps, params, gpu_addr, topological_result);
             }
         }
 
         // Check if it's a 3D texture
         if (params.block_depth > 0) {
-            auto surface =
-                Manage3DSurfaces(overlaps, params, gpu_addr, cpu_addr, preserve_contents);
+            auto surface = Manage3DSurfaces(overlaps, params, gpu_addr, cpu_addr);
             if (surface) {
                 return *surface;
             }
@@ -790,8 +776,7 @@ private:
                         return *view;
                     }
                 }
-                return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
-                                      MatchTopologyResult::FullMatch);
+                return RecycleSurface(overlaps, params, gpu_addr, MatchTopologyResult::FullMatch);
             }
             // Now we check if the candidate is a mipmap/layer of the overlap
             std::optional<TView> view =
@@ -815,7 +800,7 @@ private:
                         pair.first->EmplaceView(params, gpu_addr, candidate_size);
                     if (mirage_view)
                         return {pair.first, *mirage_view};
-                    return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
+                    return RecycleSurface(overlaps, params, gpu_addr,
                                           MatchTopologyResult::FullMatch);
                 }
                 return {current_surface, *view};
@@ -831,8 +816,7 @@ private:
             }
         }
         // We failed all the tests, recycle the overlaps into a new texture.
-        return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
-                              MatchTopologyResult::FullMatch);
+        return RecycleSurface(overlaps, params, gpu_addr, MatchTopologyResult::FullMatch);
     }
 
     /**
@@ -990,10 +974,10 @@ private:
     }
 
     std::pair<TSurface, TView> InitializeSurface(GPUVAddr gpu_addr, const SurfaceParams& params,
-                                                 bool preserve_contents) {
+                                                 bool do_load = true) {
         auto new_surface{GetUncachedSurface(gpu_addr, params)};
         Register(new_surface);
-        if (preserve_contents) {
+        if (do_load) {
             LoadSurface(new_surface);
         }
         return {new_surface, new_surface->GetMainView()};
diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp
index 062b4f252..365bde2f1 100644
--- a/src/video_core/textures/astc.cpp
+++ b/src/video_core/textures/astc.cpp
@@ -20,6 +20,8 @@
 #include <cstring>
 #include <vector>
 
+#include <boost/container/static_vector.hpp>
+
 #include "common/common_types.h"
 
 #include "video_core/textures/astc.h"
@@ -39,25 +41,25 @@ constexpr u32 Popcnt(u32 n) {
 
 class InputBitStream {
 public:
-    explicit InputBitStream(const u8* ptr, std::size_t start_offset = 0)
-        : m_CurByte(ptr), m_NextBit(start_offset % 8) {}
+    constexpr explicit InputBitStream(const u8* ptr, std::size_t start_offset = 0)
+        : cur_byte{ptr}, next_bit{start_offset % 8} {}
 
-    std::size_t GetBitsRead() const {
-        return m_BitsRead;
+    constexpr std::size_t GetBitsRead() const {
+        return bits_read;
     }
 
-    u32 ReadBit() {
-        u32 bit = *m_CurByte >> m_NextBit++;
-        while (m_NextBit >= 8) {
-            m_NextBit -= 8;
-            m_CurByte++;
+    constexpr bool ReadBit() {
+        const bool bit = (*cur_byte >> next_bit++) & 1;
+        while (next_bit >= 8) {
+            next_bit -= 8;
+            cur_byte++;
         }
 
-        m_BitsRead++;
-        return bit & 1;
+        bits_read++;
+        return bit;
     }
 
-    u32 ReadBits(std::size_t nBits) {
+    constexpr u32 ReadBits(std::size_t nBits) {
         u32 ret = 0;
         for (std::size_t i = 0; i < nBits; ++i) {
             ret |= (ReadBit() & 1) << i;
@@ -66,7 +68,7 @@ public:
     }
 
     template <std::size_t nBits>
-    u32 ReadBits() {
+    constexpr u32 ReadBits() {
         u32 ret = 0;
         for (std::size_t i = 0; i < nBits; ++i) {
             ret |= (ReadBit() & 1) << i;
@@ -75,64 +77,58 @@ public:
     }
 
 private:
-    const u8* m_CurByte;
-    std::size_t m_NextBit = 0;
-    std::size_t m_BitsRead = 0;
+    const u8* cur_byte;
+    std::size_t next_bit = 0;
+    std::size_t bits_read = 0;
 };
 
 class OutputBitStream {
 public:
-    explicit OutputBitStream(u8* ptr, s32 nBits = 0, s32 start_offset = 0)
-        : m_NumBits(nBits), m_CurByte(ptr), m_NextBit(start_offset % 8) {}
-
-    ~OutputBitStream() = default;
+    constexpr explicit OutputBitStream(u8* ptr, std::size_t bits = 0, std::size_t start_offset = 0)
+        : cur_byte{ptr}, num_bits{bits}, next_bit{start_offset % 8} {}
 
-    s32 GetBitsWritten() const {
-        return m_BitsWritten;
+    constexpr std::size_t GetBitsWritten() const {
+        return bits_written;
     }
 
-    void WriteBitsR(u32 val, u32 nBits) {
+    constexpr void WriteBitsR(u32 val, u32 nBits) {
         for (u32 i = 0; i < nBits; i++) {
             WriteBit((val >> (nBits - i - 1)) & 1);
         }
     }
 
-    void WriteBits(u32 val, u32 nBits) {
+    constexpr void WriteBits(u32 val, u32 nBits) {
         for (u32 i = 0; i < nBits; i++) {
             WriteBit((val >> i) & 1);
         }
     }
 
 private:
-    void WriteBit(s32 b) {
-
-        if (done)
+    constexpr void WriteBit(bool b) {
+        if (bits_written >= num_bits) {
             return;
+        }
 
-        const u32 mask = 1 << m_NextBit++;
+        const u32 mask = 1 << next_bit++;
 
         // clear the bit
-        *m_CurByte &= static_cast<u8>(~mask);
+        *cur_byte &= static_cast<u8>(~mask);
 
         // Write the bit, if necessary
         if (b)
-            *m_CurByte |= static_cast<u8>(mask);
+            *cur_byte |= static_cast<u8>(mask);
 
         // Next byte?
-        if (m_NextBit >= 8) {
-            m_CurByte += 1;
-            m_NextBit = 0;
+        if (next_bit >= 8) {
+            cur_byte += 1;
+            next_bit = 0;
         }
-
-        done = done || ++m_BitsWritten >= m_NumBits;
     }
 
-    s32 m_BitsWritten = 0;
-    const s32 m_NumBits;
-    u8* m_CurByte;
-    s32 m_NextBit = 0;
-
-    bool done = false;
+    u8* cur_byte;
+    std::size_t num_bits;
+    std::size_t bits_written = 0;
+    std::size_t next_bit = 0;
 };
 
 template <typename IntType>
@@ -195,9 +191,13 @@ struct IntegerEncodedValue {
         u32 trit_value;
     };
 };
+using IntegerEncodedVector = boost::container::static_vector<
+    IntegerEncodedValue, 64,
+    boost::container::static_vector_options<
+        boost::container::inplace_alignment<alignof(IntegerEncodedValue)>,
+        boost::container::throw_on_overflow<false>>::type>;
 
-static void DecodeTritBlock(InputBitStream& bits, std::vector<IntegerEncodedValue>& result,
-                            u32 nBitsPerValue) {
+static void DecodeTritBlock(InputBitStream& bits, IntegerEncodedVector& result, u32 nBitsPerValue) {
     // Implement the algorithm in section C.2.12
     u32 m[5];
     u32 t[5];
@@ -255,7 +255,7 @@ static void DecodeTritBlock(InputBitStream& bits, std::vector<IntegerEncodedValu
     }
 }
 
-static void DecodeQus32Block(InputBitStream& bits, std::vector<IntegerEncodedValue>& result,
+static void DecodeQus32Block(InputBitStream& bits, IntegerEncodedVector& result,
                              u32 nBitsPerValue) {
     // Implement the algorithm in section C.2.12
     u32 m[3];
@@ -343,8 +343,8 @@ static constexpr std::array EncodingsValues = MakeEncodedValues();
 // Fills result with the values that are encoded in the given
 // bitstream. We must know beforehand what the maximum possible
 // value is, and how many values we're decoding.
-static void DecodeIntegerSequence(std::vector<IntegerEncodedValue>& result, InputBitStream& bits,
-                                  u32 maxRange, u32 nValues) {
+static void DecodeIntegerSequence(IntegerEncodedVector& result, InputBitStream& bits, u32 maxRange,
+                                  u32 nValues) {
     // Determine encoding parameters
     IntegerEncodedValue val = EncodingsValues[maxRange];
 
@@ -634,12 +634,14 @@ static void FillError(u32* outBuf, u32 blockWidth, u32 blockHeight) {
 // Replicates low numBits such that [(toBit - 1):(toBit - 1 - fromBit)]
 // is the same as [(numBits - 1):0] and repeats all the way down.
 template <typename IntType>
-static IntType Replicate(IntType val, u32 numBits, u32 toBit) {
-    if (numBits == 0)
+static constexpr IntType Replicate(IntType val, u32 numBits, u32 toBit) {
+    if (numBits == 0) {
         return 0;
-    if (toBit == 0)
+    }
+    if (toBit == 0) {
         return 0;
-    IntType v = val & static_cast<IntType>((1 << numBits) - 1);
+    }
+    const IntType v = val & static_cast<IntType>((1 << numBits) - 1);
     IntType res = v;
     u32 reslen = numBits;
     while (reslen < toBit) {
@@ -656,6 +658,89 @@ static IntType Replicate(IntType val, u32 numBits, u32 toBit) {
     return res;
 }
 
+static constexpr std::size_t NumReplicateEntries(u32 num_bits) {
+    return std::size_t(1) << num_bits;
+}
+
+template <typename IntType, u32 num_bits, u32 to_bit>
+static constexpr auto MakeReplicateTable() {
+    std::array<IntType, NumReplicateEntries(num_bits)> table{};
+    for (IntType value = 0; value < static_cast<IntType>(std::size(table)); ++value) {
+        table[value] = Replicate(value, num_bits, to_bit);
+    }
+    return table;
+}
+
+static constexpr auto REPLICATE_BYTE_TO_16_TABLE = MakeReplicateTable<u32, 8, 16>();
+static constexpr u32 ReplicateByteTo16(std::size_t value) {
+    return REPLICATE_BYTE_TO_16_TABLE[value];
+}
+
+static constexpr auto REPLICATE_BIT_TO_7_TABLE = MakeReplicateTable<u32, 1, 7>();
+static constexpr u32 ReplicateBitTo7(std::size_t value) {
+    return REPLICATE_BIT_TO_7_TABLE[value];
+}
+
+static constexpr auto REPLICATE_BIT_TO_9_TABLE = MakeReplicateTable<u32, 1, 9>();
+static constexpr u32 ReplicateBitTo9(std::size_t value) {
+    return REPLICATE_BIT_TO_9_TABLE[value];
+}
+
+static constexpr auto REPLICATE_1_BIT_TO_8_TABLE = MakeReplicateTable<u32, 1, 8>();
+static constexpr auto REPLICATE_2_BIT_TO_8_TABLE = MakeReplicateTable<u32, 2, 8>();
+static constexpr auto REPLICATE_3_BIT_TO_8_TABLE = MakeReplicateTable<u32, 3, 8>();
+static constexpr auto REPLICATE_4_BIT_TO_8_TABLE = MakeReplicateTable<u32, 4, 8>();
+static constexpr auto REPLICATE_5_BIT_TO_8_TABLE = MakeReplicateTable<u32, 5, 8>();
+static constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
+static constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
+static constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
+/// Use a precompiled table with the most common usages, if it's not in the expected range, fallback
+/// to the runtime implementation
+static constexpr u32 FastReplicateTo8(u32 value, u32 num_bits) {
+    switch (num_bits) {
+    case 1:
+        return REPLICATE_1_BIT_TO_8_TABLE[value];
+    case 2:
+        return REPLICATE_2_BIT_TO_8_TABLE[value];
+    case 3:
+        return REPLICATE_3_BIT_TO_8_TABLE[value];
+    case 4:
+        return REPLICATE_4_BIT_TO_8_TABLE[value];
+    case 5:
+        return REPLICATE_5_BIT_TO_8_TABLE[value];
+    case 6:
+        return REPLICATE_6_BIT_TO_8_TABLE[value];
+    case 7:
+        return REPLICATE_7_BIT_TO_8_TABLE[value];
+    case 8:
+        return REPLICATE_8_BIT_TO_8_TABLE[value];
+    default:
+        return Replicate(value, num_bits, 8);
+    }
+}
+
+static constexpr auto REPLICATE_1_BIT_TO_6_TABLE = MakeReplicateTable<u32, 1, 6>();
+static constexpr auto REPLICATE_2_BIT_TO_6_TABLE = MakeReplicateTable<u32, 2, 6>();
+static constexpr auto REPLICATE_3_BIT_TO_6_TABLE = MakeReplicateTable<u32, 3, 6>();
+static constexpr auto REPLICATE_4_BIT_TO_6_TABLE = MakeReplicateTable<u32, 4, 6>();
+static constexpr auto REPLICATE_5_BIT_TO_6_TABLE = MakeReplicateTable<u32, 5, 6>();
+static constexpr u32 FastReplicateTo6(u32 value, u32 num_bits) {
+    switch (num_bits) {
+    case 1:
+        return REPLICATE_1_BIT_TO_6_TABLE[value];
+    case 2:
+        return REPLICATE_2_BIT_TO_6_TABLE[value];
+    case 3:
+        return REPLICATE_3_BIT_TO_6_TABLE[value];
+    case 4:
+        return REPLICATE_4_BIT_TO_6_TABLE[value];
+    case 5:
+        return REPLICATE_5_BIT_TO_6_TABLE[value];
+    default:
+        return Replicate(value, num_bits, 6);
+    }
+}
+
 class Pixel {
 protected:
     using ChannelType = s16;
@@ -674,10 +759,10 @@ public:
     // significant bits when going from larger to smaller bit depth
     // or by repeating the most significant bits when going from
     // smaller to larger bit depths.
-    void ChangeBitDepth(const u8 (&depth)[4]) {
+    void ChangeBitDepth() {
         for (u32 i = 0; i < 4; i++) {
-            Component(i) = ChangeBitDepth(Component(i), m_BitDepth[i], depth[i]);
-            m_BitDepth[i] = depth[i];
+            Component(i) = ChangeBitDepth(Component(i), m_BitDepth[i]);
+            m_BitDepth[i] = 8;
         }
     }
 
@@ -689,28 +774,23 @@ public:
 
     // Changes the bit depth of a single component. See the comment
     // above for how we do this.
-    static ChannelType ChangeBitDepth(Pixel::ChannelType val, u8 oldDepth, u8 newDepth) {
-        assert(newDepth <= 8);
+    static ChannelType ChangeBitDepth(Pixel::ChannelType val, u8 oldDepth) {
         assert(oldDepth <= 8);
 
-        if (oldDepth == newDepth) {
+        if (oldDepth == 8) {
             // Do nothing
             return val;
-        } else if (oldDepth == 0 && newDepth != 0) {
-            return static_cast<ChannelType>((1 << newDepth) - 1);
-        } else if (newDepth > oldDepth) {
-            return Replicate(val, oldDepth, newDepth);
+        } else if (oldDepth == 0) {
+            return static_cast<ChannelType>((1 << 8) - 1);
+        } else if (8 > oldDepth) {
+            return static_cast<ChannelType>(FastReplicateTo8(static_cast<u32>(val), oldDepth));
         } else {
             // oldDepth > newDepth
-            if (newDepth == 0) {
-                return 0xFF;
-            } else {
-                u8 bitsWasted = static_cast<u8>(oldDepth - newDepth);
-                u16 v = static_cast<u16>(val);
-                v = static_cast<u16>((v + (1 << (bitsWasted - 1))) >> bitsWasted);
-                v = ::std::min<u16>(::std::max<u16>(0, v), static_cast<u16>((1 << newDepth) - 1));
-                return static_cast<u8>(v);
-            }
+            const u8 bitsWasted = static_cast<u8>(oldDepth - 8);
+            u16 v = static_cast<u16>(val);
+            v = static_cast<u16>((v + (1 << (bitsWasted - 1))) >> bitsWasted);
+            v = ::std::min<u16>(::std::max<u16>(0, v), static_cast<u16>((1 << 8) - 1));
+            return static_cast<u8>(v);
         }
 
         assert(false && "We shouldn't get here.");
@@ -760,8 +840,7 @@ public:
     // up in the most-significant byte.
     u32 Pack() const {
         Pixel eightBit(*this);
-        const u8 eightBitDepth[4] = {8, 8, 8, 8};
-        eightBit.ChangeBitDepth(eightBitDepth);
+        eightBit.ChangeBitDepth();
 
         u32 r = 0;
         r |= eightBit.A();
@@ -816,8 +895,7 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP
     }
 
     // We now have enough to decode our integer sequence.
-    std::vector<IntegerEncodedValue> decodedColorValues;
-    decodedColorValues.reserve(32);
+    IntegerEncodedVector decodedColorValues;
 
     InputBitStream colorStream(data);
     DecodeIntegerSequence(decodedColorValues, colorStream, range, nValues);
@@ -839,12 +917,12 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP
 
         u32 A = 0, B = 0, C = 0, D = 0;
         // A is just the lsb replicated 9 times.
-        A = Replicate(bitval & 1, 1, 9);
+        A = ReplicateBitTo9(bitval & 1);
 
         switch (val.encoding) {
         // Replicate bits
         case IntegerEncoding::JustBits:
-            out[outIdx++] = Replicate(bitval, bitlen, 8);
+            out[outIdx++] = FastReplicateTo8(bitval, bitlen);
             break;
 
         // Use algorithm in C.2.13
@@ -962,13 +1040,13 @@ static u32 UnquantizeTexelWeight(const IntegerEncodedValue& val) {
     u32 bitval = val.bit_value;
     u32 bitlen = val.num_bits;
 
-    u32 A = Replicate(bitval & 1, 1, 7);
+    u32 A = ReplicateBitTo7(bitval & 1);
     u32 B = 0, C = 0, D = 0;
 
     u32 result = 0;
     switch (val.encoding) {
     case IntegerEncoding::JustBits:
-        result = Replicate(bitval, bitlen, 6);
+        result = FastReplicateTo6(bitval, bitlen);
         break;
 
     case IntegerEncoding::Trit: {
@@ -1047,7 +1125,7 @@ static u32 UnquantizeTexelWeight(const IntegerEncodedValue& val) {
     return result;
 }
 
-static void UnquantizeTexelWeights(u32 out[2][144], const std::vector<IntegerEncodedValue>& weights,
+static void UnquantizeTexelWeights(u32 out[2][144], const IntegerEncodedVector& weights,
                                    const TexelWeightParams& params, const u32 blockWidth,
                                    const u32 blockHeight) {
     u32 weightIdx = 0;
@@ -1545,8 +1623,7 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32
         static_cast<u8>((1 << (weightParams.GetPackedBitSize() % 8)) - 1);
     memset(texelWeightData + clearByteStart, 0, 16 - clearByteStart);
 
-    std::vector<IntegerEncodedValue> texelWeightValues;
-    texelWeightValues.reserve(64);
+    IntegerEncodedVector texelWeightValues;
 
     InputBitStream weightStream(texelWeightData);
 
@@ -1568,9 +1645,9 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32
             Pixel p;
             for (u32 c = 0; c < 4; c++) {
                 u32 C0 = endpos32s[partition][0].Component(c);
-                C0 = Replicate(C0, 8, 16);
+                C0 = ReplicateByteTo16(C0);
                 u32 C1 = endpos32s[partition][1].Component(c);
-                C1 = Replicate(C1, 8, 16);
+                C1 = ReplicateByteTo16(C1);
 
                 u32 plane = 0;
                 if (weightParams.m_bDualPlane && (((planeIdx + 1) & 3) == c)) {