[spv, qcom] Implement warp intrinsics support

2025-12-24 16:41:49 -03:00
26 changed files with 513 additions and 115 deletions
--- a/cpmfile.json
+++ b/cpmfile.json
@@ -96,8 +96,8 @@
        "package": "VVL",
        "repo": "KhronosGroup/Vulkan-ValidationLayers",
        "tag": "vulkan-sdk-%VERSION%",
-        "git_version": "1.4.335.0",
+        "git_version": "1.4.328.1",
        "artifact": "android-binaries-%VERSION%.zip",
-        "hash": "48167c4a17736301bd08f9290f41830443e1f18cce8ad867fc6f289b49e18b40e93c9850b377951af82f51b5b6d7313aa6a884fc5df79f5ce3df82696c1c1244"
+        "hash": "5ec895a453cb7c2f156830b9766953a0c2bd44dea99e6a3dac4160305041ccd3e87534b4ce0bd102392178d2a8eca48411856298f9395e60117cdfe89f72137e"
    }
 }
--- a/externals/cpmfile.json
+++ b/externals/cpmfile.json
@@ -119,10 +119,10 @@
        "package": "VulkanUtilityLibraries",
        "repo": "scripts/VulkanUtilityHeaders",
        "tag": "%VERSION%",
-        "git_version": "1.4.335",
+        "git_version": "1.4.328",
        "artifact": "VulkanUtilityHeaders.tar.zst",
        "git_host": "git.crueter.xyz",
-        "hash": "16dac0e6586702580c4279e4cd37ffe3cf909c93eb31b5069da7af36436d47b270a9cbaac953bb66c22ed12ed67ffa096688599267f307dfb62be1bc09f79833"
+        "hash": "9922217b39faf73cd4fc1510f2fdba14a49aa5c0d77f9ee24ee0512cef16b234d0cabc83c1fec861fa5df1d43e7f086ca9b6501753899119f39c5ca530cb0dae"
    },
    "spirv-tools": {
        "package": "SPIRV-Tools",
--- a/src/shader_recompiler/backend/spirv/emit_spirv.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.cpp
@@ -11,6 +11,7 @@
 #include <vector>
 #include <spirv-tools/optimizer.hpp>

+#include "common/logging/log.h"
 #include "common/settings.h"
 #include "shader_recompiler/backend/spirv/emit_spirv.h"
 #include "shader_recompiler/backend/spirv/emit_spirv_instructions.h"
@@ -439,15 +440,23 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
        ctx.AddExtension("SPV_KHR_shader_draw_parameters");
        ctx.AddCapability(spv::Capability::DrawParameters);
    }
-    if ((info.uses_subgroup_vote || info.uses_subgroup_invocation_id ||
-         info.uses_subgroup_shuffles) &&
-        profile.support_vote) {
+    const bool stage_supports_warp = profile.SupportsWarpIntrinsics(ctx.stage);
+    const bool needs_warp_intrinsics = info.uses_subgroup_vote ||
+                                       info.uses_subgroup_invocation_id ||
+                                       info.uses_subgroup_shuffles;
+
+    if (needs_warp_intrinsics && profile.support_vote && stage_supports_warp) {
        ctx.AddCapability(spv::Capability::GroupNonUniformBallot);
        ctx.AddCapability(spv::Capability::GroupNonUniformShuffle);
        if (!profile.warp_size_potentially_larger_than_guest) {
            // vote ops are only used when not taking the long path
            ctx.AddCapability(spv::Capability::GroupNonUniformVote);
        }
+    } else if (needs_warp_intrinsics && !stage_supports_warp) {
+        LOG_WARNING(Shader,
+                    "Warp intrinsics requested in stage {} but the device does not report subgroup "
+                    "support; falling back to scalar approximations",
+                    static_cast<u32>(ctx.stage));
    }
    if (info.uses_int64_bit_atomics && profile.support_int64_atomics) {
        ctx.AddCapability(spv::Capability::Int64Atomics);
--- a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2025 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

@@ -78,9 +81,25 @@ Id AddPartitionBase(EmitContext& ctx, Id thread_id) {
    const Id partition_base{ctx.OpShiftLeftLogical(ctx.U32[1], partition_idx, ctx.Const(5u))};
    return ctx.OpIAdd(ctx.U32[1], thread_id, partition_base);
 }
+
+bool SupportsWarpIntrinsics(const EmitContext& ctx) {
+    return ctx.profile.SupportsWarpIntrinsics(ctx.stage);
+}
+
+void SetAlwaysInBounds(EmitContext& ctx, IR::Inst* inst) {
+    SetInBoundsFlag(inst, ctx.true_value);
+}
+
+Id FallbackBallotMask(EmitContext& ctx, Id pred) {
+    const Id full_mask{ctx.Const(0xFFFFFFFFu)};
+    return ctx.OpSelect(ctx.U32[1], pred, full_mask, ctx.u32_zero_value);
+}
 } // Anonymous namespace

 Id EmitLaneId(EmitContext& ctx) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return ctx.u32_zero_value;
+    }
    const Id id{GetThreadId(ctx)};
    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
        return id;
@@ -89,6 +108,9 @@ Id EmitLaneId(EmitContext& ctx) {
 }

 Id EmitVoteAll(EmitContext& ctx, Id pred) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return pred;
+    }
    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
        return ctx.OpGroupNonUniformAll(ctx.U1, SubgroupScope(ctx), pred);
    }
@@ -102,6 +124,9 @@ Id EmitVoteAll(EmitContext& ctx, Id pred) {
 }

 Id EmitVoteAny(EmitContext& ctx, Id pred) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return pred;
+    }
    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
        return ctx.OpGroupNonUniformAny(ctx.U1, SubgroupScope(ctx), pred);
    }
@@ -115,6 +140,9 @@ Id EmitVoteAny(EmitContext& ctx, Id pred) {
 }

 Id EmitVoteEqual(EmitContext& ctx, Id pred) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return pred;
+    }
    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
        return ctx.OpGroupNonUniformAllEqual(ctx.U1, SubgroupScope(ctx), pred);
    }
@@ -129,6 +157,9 @@ Id EmitVoteEqual(EmitContext& ctx, Id pred) {
 }

 Id EmitSubgroupBallot(EmitContext& ctx, Id pred) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return FallbackBallotMask(ctx, pred);
+    }
    const Id ballot{ctx.OpGroupNonUniformBallot(ctx.U32[4], SubgroupScope(ctx), pred)};
    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
        return ctx.OpCompositeExtract(ctx.U32[1], ballot, 0U);
@@ -137,27 +168,46 @@ Id EmitSubgroupBallot(EmitContext& ctx, Id pred) {
 }

 Id EmitSubgroupEqMask(EmitContext& ctx) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return ctx.u32_zero_value;
+    }
    return LoadMask(ctx, ctx.subgroup_mask_eq);
 }

 Id EmitSubgroupLtMask(EmitContext& ctx) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return ctx.u32_zero_value;
+    }
    return LoadMask(ctx, ctx.subgroup_mask_lt);
 }

 Id EmitSubgroupLeMask(EmitContext& ctx) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return ctx.u32_zero_value;
+    }
    return LoadMask(ctx, ctx.subgroup_mask_le);
 }

 Id EmitSubgroupGtMask(EmitContext& ctx) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return ctx.u32_zero_value;
+    }
    return LoadMask(ctx, ctx.subgroup_mask_gt);
 }

 Id EmitSubgroupGeMask(EmitContext& ctx) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        return ctx.u32_zero_value;
+    }
    return LoadMask(ctx, ctx.subgroup_mask_ge);
 }

 Id EmitShuffleIndex(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                    Id segmentation_mask) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        SetAlwaysInBounds(ctx, inst);
+        return value;
+    }
    const Id not_seg_mask{ctx.OpNot(ctx.U32[1], segmentation_mask)};
    const Id thread_id{EmitLaneId(ctx)};
    const Id min_thread_id{ComputeMinThreadId(ctx, thread_id, segmentation_mask)};
@@ -177,6 +227,10 @@ Id EmitShuffleIndex(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id cla

 Id EmitShuffleUp(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                 Id segmentation_mask) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        SetAlwaysInBounds(ctx, inst);
+        return value;
+    }
    const Id thread_id{EmitLaneId(ctx)};
    const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)};
    Id src_thread_id{ctx.OpISub(ctx.U32[1], thread_id, index)};
@@ -192,6 +246,10 @@ Id EmitShuffleUp(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,

 Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                   Id segmentation_mask) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        SetAlwaysInBounds(ctx, inst);
+        return value;
+    }
    const Id thread_id{EmitLaneId(ctx)};
    const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)};
    Id src_thread_id{ctx.OpIAdd(ctx.U32[1], thread_id, index)};
@@ -207,6 +265,10 @@ Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clam

 Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                        Id segmentation_mask) {
+    if (!SupportsWarpIntrinsics(ctx)) {
+        SetAlwaysInBounds(ctx, inst);
+        return value;
+    }
    const Id thread_id{EmitLaneId(ctx)};
    const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)};
    Id src_thread_id{ctx.OpBitwiseXor(ctx.U32[1], thread_id, index)};
--- a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
+++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
@@ -1416,6 +1416,7 @@ void EmitContext::DefineImages(const Info& info, u32& binding, u32& scaling_inde
 void EmitContext::DefineInputs(const IR::Program& program) {
    const Info& info{program.info};
    const VaryingState loads{info.loads.mask | info.passthrough.mask};
+    const bool stage_supports_warp = profile.SupportsWarpIntrinsics(stage);

    if (info.uses_workgroup_id) {
        workgroup_id = DefineInput(*this, U32[3], false, spv::BuiltIn::WorkgroupId);
@@ -1436,16 +1437,17 @@ void EmitContext::DefineInputs(const IR::Program& program) {
    if (info.uses_is_helper_invocation) {
        is_helper_invocation = DefineInput(*this, U1, false, spv::BuiltIn::HelperInvocation);
    }
-    if (info.uses_subgroup_mask) {
+    if (info.uses_subgroup_mask && stage_supports_warp) {
        subgroup_mask_eq = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupEqMaskKHR);
        subgroup_mask_lt = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupLtMaskKHR);
        subgroup_mask_le = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupLeMaskKHR);
        subgroup_mask_gt = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupGtMaskKHR);
        subgroup_mask_ge = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupGeMaskKHR);
    }
-    if (info.uses_fswzadd || info.uses_subgroup_invocation_id || info.uses_subgroup_shuffles ||
-        (profile.warp_size_potentially_larger_than_guest &&
-         (info.uses_subgroup_vote || info.uses_subgroup_mask))) {
+    if (stage_supports_warp &&
+        (info.uses_fswzadd || info.uses_subgroup_invocation_id || info.uses_subgroup_shuffles ||
+         (profile.warp_size_potentially_larger_than_guest &&
+          (info.uses_subgroup_vote || info.uses_subgroup_mask)))) {
        AddCapability(spv::Capability::GroupNonUniform);
        subgroup_local_invocation_id =
            DefineInput(*this, U32[1], false, spv::BuiltIn::SubgroupLocalInvocationId);
--- a/src/shader_recompiler/profile.h
+++ b/src/shader_recompiler/profile.h
@@ -1,9 +1,15 @@
+// SPDX-FileCopyrightText: Copyright 2025 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

 #pragma once

+#include <limits>
+
 #include "common/common_types.h"
+#include "shader_recompiler/stage.h"

 namespace Shader {

@@ -46,6 +52,8 @@ struct Profile {
    bool support_multi_viewport{};
    bool support_geometry_streams{};

+    u32 warp_stage_support_mask{std::numeric_limits<u32>::max()};
+
    bool warp_size_potentially_larger_than_guest{};

    bool lower_left_origin_mode{};
@@ -90,6 +98,11 @@ struct Profile {
    u64 min_ssbo_alignment{};

    u32 max_user_clip_distances{};
+
+    bool SupportsWarpIntrinsics(Stage stage) const {
+        const u32 bit = 1u << static_cast<u32>(stage);
+        return (warp_stage_support_mask & bit) != 0;
+    }
 };

 } // namespace Shader
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -7,6 +7,7 @@
 #pragma once

 #include <algorithm>
+#include <cstring>
 #include <memory>
 #include <numeric>

@@ -15,6 +16,8 @@
 #include "video_core/guest_memory.h"
 #include "video_core/host1x/gpu_device_memory_manager.h"
 #include "video_core/texture_cache/util.h"
+#include "video_core/polygon_mode_utils.h"
+#include "video_core/renderer_vulkan/line_loop_utils.h"

 namespace VideoCommon {

@@ -353,14 +356,37 @@ void BufferCache<P>::UpdateComputeBuffers() {

 template <class P>
 void BufferCache<P>::BindHostGeometryBuffers(bool is_indexed) {
+    const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
    if (is_indexed) {
        BindHostIndexBuffer();
-    } else if constexpr (!HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) {
-        const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
-        if (draw_state.topology == Maxwell::PrimitiveTopology::Quads ||
-            draw_state.topology == Maxwell::PrimitiveTopology::QuadStrip) {
-            runtime.BindQuadIndexBuffer(draw_state.topology, draw_state.vertex_buffer.first,
-                                        draw_state.vertex_buffer.count);
+    } else {
+        if constexpr (!P::IS_OPENGL) {
+            const auto polygon_mode = VideoCore::EffectivePolygonMode(maxwell3d->regs);
+            if (draw_state.topology == Maxwell::PrimitiveTopology::Polygon &&
+                polygon_mode == Maxwell::PolygonMode::Line && draw_state.vertex_buffer.count > 1) {
+                const u32 vertex_count = draw_state.vertex_buffer.count;
+                const u32 generated_count = vertex_count + 1;
+                const bool use_u16 = vertex_count <= 0x10000;
+                const u32 element_size = use_u16 ? sizeof(u16) : sizeof(u32);
+                auto staging = runtime.UploadStagingBuffer(
+                    static_cast<size_t>(generated_count) * element_size);
+                std::span<u8> dst_span{staging.mapped_span.data(),
+                                       generated_count * static_cast<size_t>(element_size)};
+                Vulkan::LineLoop::GenerateSequentialWithClosureRaw(dst_span, element_size);
+                const auto synthetic_format = use_u16 ? Maxwell::IndexFormat::UnsignedShort
+                                                      : Maxwell::IndexFormat::UnsignedInt;
+                runtime.BindIndexBuffer(draw_state.topology, synthetic_format,
+                                        draw_state.vertex_buffer.first, generated_count,
+                                        staging.buffer, static_cast<u32>(staging.offset),
+                                        generated_count * element_size);
+            }
+        }
+        if constexpr (!HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) {
+            if (draw_state.topology == Maxwell::PrimitiveTopology::Quads ||
+                draw_state.topology == Maxwell::PrimitiveTopology::QuadStrip) {
+                runtime.BindQuadIndexBuffer(draw_state.topology, draw_state.vertex_buffer.first,
+                                            draw_state.vertex_buffer.count);
+            }
        }
    }
    BindHostVertexBuffers();
@@ -689,6 +715,44 @@ void BufferCache<P>::BindHostIndexBuffer() {
    const u32 offset = buffer.Offset(channel_state->index_buffer.device_addr);
    const u32 size = channel_state->index_buffer.size;
    const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
+    if constexpr (!P::IS_OPENGL) {
+        const auto polygon_mode = VideoCore::EffectivePolygonMode(maxwell3d->regs);
+        const bool polygon_line =
+            draw_state.topology == Maxwell::PrimitiveTopology::Polygon &&
+            polygon_mode == Maxwell::PolygonMode::Line;
+        if (polygon_line && draw_state.index_buffer.count > 1) {
+            const u32 element_size = draw_state.index_buffer.FormatSizeInBytes();
+            const size_t src_bytes = static_cast<size_t>(draw_state.index_buffer.count) * element_size;
+            const size_t total_bytes = src_bytes + element_size;
+            auto staging = runtime.UploadStagingBuffer(total_bytes);
+            std::span<u8> dst_span{staging.mapped_span.data(), total_bytes};
+            std::span<const u8> src_span;
+            if (!draw_state.inline_index_draw_indexes.empty()) {
+                const u8* const src =
+                    draw_state.inline_index_draw_indexes.data() +
+                    static_cast<size_t>(draw_state.index_buffer.first) * element_size;
+                src_span = {src, src_bytes};
+            } else if (const u8* const cpu_base =
+                           device_memory.GetPointer<u8>(channel_state->index_buffer.device_addr)) {
+                const u8* const src = cpu_base +
+                                      static_cast<size_t>(draw_state.index_buffer.first) * element_size;
+                src_span = {src, src_bytes};
+            } else {
+                const DAddr src_addr =
+                    channel_state->index_buffer.device_addr +
+                    static_cast<DAddr>(draw_state.index_buffer.first) * element_size;
+                device_memory.ReadBlockUnsafe(src_addr, dst_span.data(), src_bytes);
+                src_span = {dst_span.data(), src_bytes};
+            }
+            Vulkan::LineLoop::CopyWithClosureRaw(dst_span, src_span, element_size);
+            buffer.MarkUsage(offset, size);
+            runtime.BindIndexBuffer(draw_state.topology, draw_state.index_buffer.format,
+                                    draw_state.index_buffer.first, draw_state.index_buffer.count + 1,
+                                    staging.buffer, static_cast<u32>(staging.offset),
+                                    static_cast<u32>(total_bytes));
+            return;
+        }
+    }
    if (!draw_state.inline_index_draw_indexes.empty()) [[unlikely]] {
        if constexpr (USE_MEMORY_MAPS_FOR_UPLOADS) {
            auto upload_staging = runtime.UploadStagingBuffer(size);
--- a/src/video_core/polygon_mode_utils.h
+++ b/src/video_core/polygon_mode_utils.h
@@ -0,0 +1,46 @@
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include "video_core/engines/maxwell_3d.h"
+
+namespace VideoCore {
+
+inline Tegra::Engines::Maxwell3D::Regs::PolygonMode EffectivePolygonMode(
+    const Tegra::Engines::Maxwell3D::Regs& regs) {
+    using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+
+    const bool cull_enabled = regs.gl_cull_test_enabled != 0;
+    const auto cull_face = regs.gl_cull_face;
+    const bool cull_front = cull_enabled && (cull_face == Maxwell::CullFace::Front ||
+                                             cull_face == Maxwell::CullFace::FrontAndBack);
+    const bool cull_back = cull_enabled && (cull_face == Maxwell::CullFace::Back ||
+                                            cull_face == Maxwell::CullFace::FrontAndBack);
+
+    const bool render_front = !cull_front;
+    const bool render_back = !cull_back;
+
+    const auto front_mode = regs.polygon_mode_front;
+    const auto back_mode = regs.polygon_mode_back;
+
+    if (render_front && render_back && front_mode != back_mode) {
+        if (front_mode == Maxwell::PolygonMode::Line || back_mode == Maxwell::PolygonMode::Line) {
+            return Maxwell::PolygonMode::Line;
+        }
+        if (front_mode == Maxwell::PolygonMode::Point || back_mode == Maxwell::PolygonMode::Point) {
+            return Maxwell::PolygonMode::Point;
+        }
+    }
+
+    if (render_front) {
+        return front_mode;
+    }
+    if (render_back) {
+        return back_mode;
+    }
+    return front_mode;
+}
+
+} // namespace VideoCore
+
--- a/src/video_core/query_cache/query_cache.h
+++ b/src/video_core/query_cache/query_cache.h
@@ -27,7 +27,7 @@
 #include "video_core/query_cache/query_cache_base.h"
 #include "video_core/query_cache/query_stream.h"
 #include "video_core/query_cache/types.h"
-namespace Vulkan { class Scheduler; }
+
 namespace VideoCommon {

 using Maxwell = Tegra::Engines::Maxwell3D;
@@ -222,12 +222,6 @@ void QueryCacheBase<Traits>::CounterReset(QueryType counter_type) {
    streamer->ResetCounter();
 }

-// Called at frame start to batch vkCmdResetQueryPool outside render passes.
-template <typename Traits>
-void QueryCacheBase<Traits>::FramePrologueResets(Vulkan::Scheduler& scheduler) {
-        impl->runtime.FramePrologueResets(scheduler);
-}
-
 template <typename Traits>
 void QueryCacheBase<Traits>::BindToChannel(s32 id) {
    VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo>::BindToChannel(id);
--- a/src/video_core/query_cache/query_cache_base.h
+++ b/src/video_core/query_cache/query_cache_base.h
@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later

@@ -24,9 +21,7 @@
 namespace VideoCore {
 class RasterizerInterface;
 }
-namespace Vulkan {
-    class Scheduler;
-}
+
 namespace Tegra {
 class GPU;
 }
@@ -58,7 +53,7 @@ public:
                            RuntimeType& runtime_);

    ~QueryCacheBase();
-    void FramePrologueResets(Vulkan::Scheduler& scheduler);
+
    void InvalidateRegion(VAddr addr, std::size_t size) {
        IterateCache<true>(addr, size,
                           [this](QueryLocation location) { InvalidateQuery(location); });
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -220,6 +220,7 @@ ShaderCache::ShaderCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
          .support_gl_sparse_textures = device.HasSparseTexture2(),
          .support_gl_derivative_control = device.HasDerivativeControl(),
          .support_geometry_streams = true,
+          .warp_stage_support_mask = 0xFFFFFFFFu,

          .warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyLargerThanGuest(),

--- a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
+++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp
@@ -15,6 +15,7 @@
 #include "video_core/engines/draw_manager.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
 #include "video_core/renderer_vulkan/vk_state_tracker.h"
+#include "video_core/polygon_mode_utils.h"

 namespace Vulkan {
 namespace {
@@ -65,7 +66,7 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d, DynamicFe
    dynamic_vertex_input.Assign(features.has_dynamic_vertex_input ? 1 : 0);
    xfb_enabled.Assign(regs.transform_feedback_enabled != 0);
    ndc_minus_one_to_one.Assign(regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1 : 0);
-    polygon_mode.Assign(PackPolygonMode(regs.polygon_mode_front));
+    polygon_mode.Assign(PackPolygonMode(VideoCore::EffectivePolygonMode(regs)));
    tessellation_primitive.Assign(static_cast<u32>(regs.tessellation.params.domain_type.Value()));
    tessellation_spacing.Assign(static_cast<u32>(regs.tessellation.params.spacing.Value()));
    tessellation_clockwise.Assign(regs.tessellation.params.output_primitives.Value() ==
--- a/src/video_core/renderer_vulkan/line_loop_utils.h
+++ b/src/video_core/renderer_vulkan/line_loop_utils.h
@@ -0,0 +1,68 @@
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <algorithm>
+#include <cstring>
+#include <span>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+
+namespace Vulkan::LineLoop {
+
+inline void CopyWithClosureRaw(std::span<u8> dst, std::span<const u8> src, size_t element_size) {
+    ASSERT_MSG(dst.size() == src.size() + element_size, "Invalid line loop copy sizes");
+    if (src.empty()) {
+        if (!dst.empty()) {
+            std::fill(dst.begin(), dst.end(), u8{0});
+        }
+        return;
+    }
+    std::memcpy(dst.data(), src.data(), src.size());
+    std::memcpy(dst.data() + src.size(), src.data(), element_size);
+}
+
+inline void GenerateSequentialWithClosureRaw(std::span<u8> dst, size_t element_size,
+                                             u64 start_value = 0) {
+    if (dst.empty()) {
+        return;
+    }
+    const size_t last = dst.size() - element_size;
+    size_t offset = 0;
+    u64 value = start_value;
+    while (offset < last) {
+        std::memcpy(dst.data() + offset, &value, element_size);
+        offset += element_size;
+        ++value;
+    }
+    std::memcpy(dst.data() + offset, &start_value, element_size);
+}
+
+template <typename T>
+inline void CopyWithClosure(std::span<T> dst, std::span<const T> src) {
+    ASSERT_MSG(dst.size() == src.size() + 1, "Invalid destination size for line loop copy");
+    if (src.empty()) {
+        if (!dst.empty()) {
+            dst.front() = {};
+        }
+        return;
+    }
+    std::copy(src.begin(), src.end(), dst.begin());
+    dst.back() = src.front();
+}
+
+template <typename T>
+inline void GenerateSequentialWithClosure(std::span<T> dst, T start_value = {}) {
+    if (dst.empty()) {
+        return;
+    }
+    const size_t last = dst.size() - 1;
+    for (size_t i = 0; i < last; ++i) {
+        dst[i] = static_cast<T>(start_value + static_cast<T>(i));
+    }
+    dst.back() = start_value;
+}
+
+} // namespace Vulkan::LineLoop
--- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
+++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later

@@ -323,44 +326,9 @@ VkShaderStageFlagBits ShaderStage(Shader::Stage stage) {
 }

 VkPrimitiveTopology PrimitiveTopology([[maybe_unused]] const Device& device,
-                                      Maxwell::PrimitiveTopology topology) {
-    switch (topology) {
-    case Maxwell::PrimitiveTopology::Points:
-        return VK_PRIMITIVE_TOPOLOGY_POINT_LIST;
-    case Maxwell::PrimitiveTopology::Lines:
-        return VK_PRIMITIVE_TOPOLOGY_LINE_LIST;
-    case Maxwell::PrimitiveTopology::LineLoop:
-        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
-    case Maxwell::PrimitiveTopology::LineStrip:
-        return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP;
-    case Maxwell::PrimitiveTopology::Triangles:
-        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
-    case Maxwell::PrimitiveTopology::TriangleStrip:
-        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP;
-    case Maxwell::PrimitiveTopology::TriangleFan:
-        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN;
-    case Maxwell::PrimitiveTopology::LinesAdjacency:
-        return VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY;
-    case Maxwell::PrimitiveTopology::LineStripAdjacency:
-        return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY;
-    case Maxwell::PrimitiveTopology::TrianglesAdjacency:
-        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY;
-    case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
-        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY;
-    case Maxwell::PrimitiveTopology::Quads:
-    case Maxwell::PrimitiveTopology::QuadStrip:
-        // TODO: Use VK_PRIMITIVE_TOPOLOGY_QUAD_LIST_EXT/VK_PRIMITIVE_TOPOLOGY_QUAD_STRIP_EXT
-        // whenever it releases
-        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
-    case Maxwell::PrimitiveTopology::Patches:
-        return VK_PRIMITIVE_TOPOLOGY_PATCH_LIST;
-    case Maxwell::PrimitiveTopology::Polygon:
-        LOG_WARNING(Render_Vulkan, "Draw mode is Polygon with a polygon mode of lines should be a "
-                                   "single body and not a bunch of triangles.");
-        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN;
-    }
-    UNIMPLEMENTED_MSG("Unimplemented topology={}", topology);
-    return {};
+                                      Maxwell::PrimitiveTopology topology,
+                                      Maxwell::PolygonMode polygon_mode) {
+    return detail::PrimitiveTopologyNoDevice(topology, polygon_mode);
 }

 VkFormat VertexFormat(const Device& device, Maxwell::VertexAttribute::Type type,
--- a/src/video_core/renderer_vulkan/maxwell_to_vk.h
+++ b/src/video_core/renderer_vulkan/maxwell_to_vk.h
@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later

@@ -15,6 +18,52 @@ namespace Vulkan::MaxwellToVK {
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 using PixelFormat = VideoCore::Surface::PixelFormat;

+namespace detail {
+constexpr VkPrimitiveTopology PrimitiveTopologyNoDevice(Maxwell::PrimitiveTopology topology,
+                                                        Maxwell::PolygonMode polygon_mode) {
+    switch (topology) {
+    case Maxwell::PrimitiveTopology::Points:
+        return VK_PRIMITIVE_TOPOLOGY_POINT_LIST;
+    case Maxwell::PrimitiveTopology::Lines:
+        return VK_PRIMITIVE_TOPOLOGY_LINE_LIST;
+    case Maxwell::PrimitiveTopology::LineLoop:
+        return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP;
+    case Maxwell::PrimitiveTopology::LineStrip:
+        return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP;
+    case Maxwell::PrimitiveTopology::Triangles:
+        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
+    case Maxwell::PrimitiveTopology::TriangleStrip:
+        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP;
+    case Maxwell::PrimitiveTopology::TriangleFan:
+        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN;
+    case Maxwell::PrimitiveTopology::LinesAdjacency:
+        return VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY;
+    case Maxwell::PrimitiveTopology::LineStripAdjacency:
+        return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY;
+    case Maxwell::PrimitiveTopology::TrianglesAdjacency:
+        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY;
+    case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
+        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY;
+    case Maxwell::PrimitiveTopology::Quads:
+    case Maxwell::PrimitiveTopology::QuadStrip:
+        return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
+    case Maxwell::PrimitiveTopology::Patches:
+        return VK_PRIMITIVE_TOPOLOGY_PATCH_LIST;
+    case Maxwell::PrimitiveTopology::Polygon:
+        switch (polygon_mode) {
+        case Maxwell::PolygonMode::Fill:
+            return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN;
+        case Maxwell::PolygonMode::Line:
+            return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP;
+        case Maxwell::PolygonMode::Point:
+            return VK_PRIMITIVE_TOPOLOGY_POINT_LIST;
+        }
+        break;
+    }
+    return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
+}
+} // namespace detail
+
 namespace Sampler {

 VkFilter Filter(Tegra::Texture::TextureFilter filter);
@@ -46,7 +95,8 @@ struct FormatInfo {

 VkShaderStageFlagBits ShaderStage(Shader::Stage stage);

-VkPrimitiveTopology PrimitiveTopology(const Device& device, Maxwell::PrimitiveTopology topology);
+VkPrimitiveTopology PrimitiveTopology(const Device& device, Maxwell::PrimitiveTopology topology,
+                                      Maxwell::PolygonMode polygon_mode);

 VkFormat VertexFormat(const Device& device, Maxwell::VertexAttribute::Type type,
                      Maxwell::VertexAttribute::Size size);
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@@ -459,10 +459,14 @@ QueriesPrefixScanPass::QueriesPrefixScanPass(
          device_, descriptor_pool_, QUERIES_SCAN_DESCRIPTOR_SET_BINDINGS,
          QUERIES_SCAN_DESCRIPTOR_UPDATE_TEMPLATE, QUERIES_SCAN_BANK_INFO,
          COMPUTE_PUSH_CONSTANT_RANGE<sizeof(QueriesPrefixScanPushConstants)>,
-          device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_BASIC_BIT) &&
-                  device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_ARITHMETIC_BIT) &&
-                  device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_SHUFFLE_BIT) &&
-                  device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT)
+          device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_BASIC_BIT,
+                                             VK_SHADER_STAGE_COMPUTE_BIT) &&
+                  device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_ARITHMETIC_BIT,
+                                                      VK_SHADER_STAGE_COMPUTE_BIT) &&
+                  device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_SHUFFLE_BIT,
+                                                      VK_SHADER_STAGE_COMPUTE_BIT) &&
+                  device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT,
+                                                      VK_SHADER_STAGE_COMPUTE_BIT)
              ? std::span<const u32>(QUERIES_PREFIX_SCAN_SUM_COMP_SPV)
              : std::span<const u32>(QUERIES_PREFIX_SCAN_SUM_NOSUBGROUPS_COMP_SPV)),
      scheduler{scheduler_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
--- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
+++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
@@ -7,6 +7,7 @@
 #include <algorithm>
 #include <iostream>
 #include <span>
+#include <string_view>

 #include <boost/container/small_vector.hpp>
 #include <boost/container/static_vector.hpp>
@@ -22,6 +23,7 @@
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
+#include "video_core/polygon_mode_utils.h"
 #include "video_core/shader_notify.h"
 #include "video_core/texture_cache/texture_cache.h"
 #include "video_core/vulkan_common/vulkan_device.h"
@@ -614,7 +616,10 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
        vertex_input_ci.pNext = &input_divisor_ci;
    }
    const bool has_tess_stages = spv_modules[1] || spv_modules[2];
-    auto input_assembly_topology = MaxwellToVK::PrimitiveTopology(device, key.state.topology);
+    const auto polygon_mode =
+        FixedPipelineState::UnpackPolygonMode(key.state.polygon_mode.Value());
+    auto input_assembly_topology =
+        MaxwellToVK::PrimitiveTopology(device, key.state.topology, polygon_mode);
    if (input_assembly_topology == VK_PRIMITIVE_TOPOLOGY_PATCH_LIST) {
        if (!has_tess_stages) {
            LOG_WARNING(Render_Vulkan, "Patch topology used without tessellation, using points");
@@ -629,6 +634,33 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
            input_assembly_topology = VK_PRIMITIVE_TOPOLOGY_PATCH_LIST;
        }
    }
+    if (key.state.topology == Maxwell::PrimitiveTopology::Polygon) {
+        const auto polygon_mode_name = [polygon_mode]() -> std::string_view {
+            switch (polygon_mode) {
+            case Maxwell::PolygonMode::Fill:
+                return "Fill";
+            case Maxwell::PolygonMode::Line:
+                return "Line";
+            case Maxwell::PolygonMode::Point:
+                return "Point";
+            }
+            return "Unknown";
+        }();
+        const auto vk_topology_name = [input_assembly_topology]() -> std::string_view {
+            switch (input_assembly_topology) {
+            case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN:
+                return "TriangleFan";
+            case VK_PRIMITIVE_TOPOLOGY_LINE_STRIP:
+                return "LineStrip";
+            case VK_PRIMITIVE_TOPOLOGY_POINT_LIST:
+                return "PointList";
+            default:
+                return "Unexpected";
+            }
+        }();
+        LOG_DEBUG(Render_Vulkan, "Polygon primitive in {} mode mapped to {}", polygon_mode_name,
+                  vk_topology_name);
+    }
    const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci{
        .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
        .pNext = nullptr,
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -105,6 +105,26 @@ Shader::CompareFunction MaxwellToCompareFunction(Maxwell::ComparisonOp compariso
    return {};
 }

+VkShaderStageFlagBits StageToVkStage(Shader::Stage stage) {
+    switch (stage) {
+    case Shader::Stage::VertexA:
+    case Shader::Stage::VertexB:
+        return VK_SHADER_STAGE_VERTEX_BIT;
+    case Shader::Stage::TessellationControl:
+        return VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT;
+    case Shader::Stage::TessellationEval:
+        return VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT;
+    case Shader::Stage::Geometry:
+        return VK_SHADER_STAGE_GEOMETRY_BIT;
+    case Shader::Stage::Fragment:
+        return VK_SHADER_STAGE_FRAGMENT_BIT;
+    case Shader::Stage::Compute:
+        return VK_SHADER_STAGE_COMPUTE_BIT;
+    default:
+        return VK_SHADER_STAGE_VERTEX_BIT;
+    }
+}
+
 Shader::AttributeType CastAttributeType(const FixedPipelineState::VertexAttribute& attr) {
    if (attr.enabled == 0) {
        return Shader::AttributeType::Disabled;
@@ -395,6 +415,27 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
        .support_conditional_barrier = device.SupportsConditionalBarriers(),
    };

+    profile.warp_stage_support_mask = 0;
+    static constexpr std::array kAllStages{
+        Shader::Stage::VertexA,          Shader::Stage::VertexB,
+        Shader::Stage::TessellationControl, Shader::Stage::TessellationEval,
+        Shader::Stage::Geometry,         Shader::Stage::Fragment,
+        Shader::Stage::Compute,
+    };
+    for (const auto stage : kAllStages) {
+        const auto vk_stage = StageToVkStage(stage);
+        if (device.SupportsWarpIntrinsics(vk_stage)) {
+            profile.warp_stage_support_mask |= 1u << static_cast<u32>(stage);
+        }
+    }
+    profile.support_vote = profile.warp_stage_support_mask != 0;
+
+    if (!profile.SupportsWarpIntrinsics(Shader::Stage::Fragment)) {
+        LOG_WARNING(Render_Vulkan,
+                    "Fragment shaders lack subgroup support on this driver; warp intrinsics will be "
+                    "approximated and visual artifacts may remain");
+    }
+
    if (device.GetMaxVertexInputAttributes() < Maxwell::NumVertexAttributes) {
        LOG_WARNING(Render_Vulkan, "maxVertexInputAttributes is too low: {} < {}",
                    device.GetMaxVertexInputAttributes(), Maxwell::NumVertexAttributes);
--- a/src/video_core/renderer_vulkan/vk_query_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp
@@ -156,7 +156,7 @@ public:

        ReserveHostQuery();

-        /* Ensure outside render pass
+        // Ensure outside render pass
        scheduler.RequestOutsideRenderPassOperationContext();

        // Reset query pool outside render pass
@@ -167,7 +167,7 @@ public:

        // Manually restart the render pass (required for vkCmdClearAttachments, etc.)
        scheduler.RequestRenderpass(texture_cache.GetFramebuffer());
-        */
+
        // Begin query inside the newly started render pass
        scheduler.Record([query_pool = current_query_pool,
                                 query_index = current_bank_slot](vk::CommandBuffer cmdbuf) {
@@ -402,13 +402,6 @@ public:
        }
    }

-    VkQueryPool GetOrCreateCurrentPoolForPrologue() {
-               if (!current_bank || current_bank->IsClosed()) {
-                        ReserveBank();
-                    }
-                return current_query_pool;
-    }
-
 private:
    template <typename Func>
    void ApplyBankOp(VideoCommon::HostQueryBase* query, Func&& func) {
@@ -1312,19 +1305,6 @@ void QueryCacheRuntime::ResumeHostConditionalRendering() {
    impl->is_hcr_running = true;
 }

-void QueryCacheRuntime::FramePrologueResets(Scheduler& scheduler) {
-            // Reset the occlusion queries we plan to use this frame in one go.
-                    // Ensure this is recorded OUTSIDE any render pass.
-                            const VkQueryPool pool = impl->sample_streamer.GetOrCreateCurrentPoolForPrologue();
-           scheduler.RequestOutsideRenderPassOperationContext();
-           scheduler.Record([pool](vk::CommandBuffer cmdbuf) {
-            // Reset the whole bank so subsequent BeginQuery calls don't need per-slot resets.
-            cmdbuf.ResetQueryPool(pool,
-                                  /*first*/ 0,
-                                  /*count*/ SamplesQueryBank::BANK_SIZE);
-            });
-}
-
 void QueryCacheRuntime::HostConditionalRenderingCompareValueImpl(VideoCommon::LookupData object,
                                                                 bool is_equal) {
    {
--- a/src/video_core/renderer_vulkan/vk_query_cache.h
+++ b/src/video_core/renderer_vulkan/vk_query_cache.h
@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later

@@ -23,7 +20,7 @@ namespace Vulkan {
 class Device;
 class Scheduler;
 class StagingBufferPool;
-class Scheduler;
+
 struct QueryCacheRuntimeImpl;

 class QueryCacheRuntime {
@@ -36,7 +33,7 @@ public:
                               ComputePassDescriptorQueue& compute_pass_descriptor_queue,
                               DescriptorPool& descriptor_pool, TextureCache& texture_cache_);
    ~QueryCacheRuntime();
-    void FramePrologueResets(Scheduler& scheduler);
+
    template <typename SyncValuesType>
    void SyncValues(std::span<SyncValuesType> values, VkBuffer base_src_buffer = nullptr);

--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -37,6 +37,7 @@
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/shader_cache.h"
 #include "video_core/texture_cache/texture_cache_base.h"
+#include "video_core/polygon_mode_utils.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"

@@ -148,7 +149,8 @@ VkRect2D GetScissorState(const Maxwell& regs, size_t index, u32 up_scale = 1, u3
    return scissor;
 }

-DrawParams MakeDrawParams(const MaxwellDrawState& draw_state, u32 num_instances, bool is_indexed) {
+DrawParams MakeDrawParams(const MaxwellDrawState& draw_state, u32 num_instances, bool is_indexed,
+                          Maxwell::PolygonMode polygon_mode) {
    DrawParams params{
        .base_instance = draw_state.base_instance,
        .num_instances = num_instances,
@@ -168,6 +170,21 @@ DrawParams MakeDrawParams(const MaxwellDrawState& draw_state, u32 num_instances,
        params.base_vertex = 0;
        params.is_indexed = true;
    }
+    const bool polygon_line =
+        draw_state.topology == Maxwell::PrimitiveTopology::Polygon &&
+        polygon_mode == Maxwell::PolygonMode::Line;
+    if (polygon_line) {
+        if (params.is_indexed) {
+            if (draw_state.index_buffer.count > 1) {
+                params.num_vertices = draw_state.index_buffer.count + 1;
+            }
+        } else if (draw_state.vertex_buffer.count > 1) {
+            params.num_vertices = draw_state.vertex_buffer.count + 1;
+            params.is_indexed = true;
+            params.first_index = 0;
+            params.base_vertex = draw_state.vertex_buffer.first;
+        }
+    }
    return params;
 }
 } // Anonymous namespace
@@ -233,7 +250,8 @@ void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
    PrepareDraw(is_indexed, [this, is_indexed, instance_count] {
        const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
        const u32 num_instances{instance_count};
-        const DrawParams draw_params{MakeDrawParams(draw_state, num_instances, is_indexed)};
+        const auto polygon_mode = VideoCore::EffectivePolygonMode(maxwell3d->regs);
+        const DrawParams draw_params{MakeDrawParams(draw_state, num_instances, is_indexed, polygon_mode)};
        scheduler.Record([draw_params](vk::CommandBuffer cmdbuf) {
            if (draw_params.is_indexed) {
                cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances,
--- a/src/video_core/renderer_vulkan/vk_scheduler.cpp
+++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp
@@ -257,11 +257,6 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se

 void Scheduler::AllocateNewContext() {
    // Enable counters once again. These are disabled when a command buffer is finished.
-    // Record per frame query resets outside any render pass, before the first draw.
-    if (query_cache) {
-        query_cache->FramePrologueResets(*this);
-    }
-
 }

 void Scheduler::InvalidateState() {
--- a/src/video_core/vulkan_common/vulkan_device.cpp
+++ b/src/video_core/vulkan_common/vulkan_device.cpp
@@ -92,6 +92,11 @@ constexpr std::array VK_FORMAT_A4B4G4R4_UNORM_PACK16{

 } // namespace Alternatives

+constexpr VkShaderStageFlags GraphicsStageMask =
+    VK_SHADER_STAGE_VERTEX_BIT | VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT |
+    VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT | VK_SHADER_STAGE_GEOMETRY_BIT |
+    VK_SHADER_STAGE_FRAGMENT_BIT;
+
 template <typename T>
 void SetNext(void**& next, T& data) {
    *next = &data;
@@ -1238,6 +1243,44 @@ void Device::RemoveUnsuitableExtensions() {
                                       VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME);
 }

+bool Device::SupportsSubgroupStage(VkShaderStageFlags stage_mask) const {
+    if (stage_mask == 0) {
+        return true;
+    }
+    const VkShaderStageFlags supported = properties.subgroup_properties.supportedStages;
+    if ((supported & stage_mask) == stage_mask) {
+        return true;
+    }
+    if ((stage_mask & GraphicsStageMask) != 0 &&
+        (supported & (VK_SHADER_STAGE_ALL_GRAPHICS | VK_SHADER_STAGE_ALL)) != 0) {
+        return true;
+    }
+    if ((stage_mask & VK_SHADER_STAGE_COMPUTE_BIT) != 0 &&
+        (supported & (VK_SHADER_STAGE_COMPUTE_BIT | VK_SHADER_STAGE_ALL)) != 0) {
+        return true;
+    }
+    return (supported & VK_SHADER_STAGE_ALL) != 0;
+}
+
+bool Device::IsSubgroupFeatureSupported(VkSubgroupFeatureFlagBits feature,
+                                        VkShaderStageFlags stage_mask) const {
+    if ((properties.subgroup_properties.supportedOperations & feature) == 0) {
+        return false;
+    }
+    return SupportsSubgroupStage(stage_mask);
+}
+
+bool Device::SupportsWarpIntrinsics(VkShaderStageFlagBits stage) const {
+    constexpr VkSubgroupFeatureFlags required_ops =
+        VK_SUBGROUP_FEATURE_BASIC_BIT | VK_SUBGROUP_FEATURE_VOTE_BIT |
+        VK_SUBGROUP_FEATURE_ARITHMETIC_BIT | VK_SUBGROUP_FEATURE_BALLOT_BIT |
+        VK_SUBGROUP_FEATURE_SHUFFLE_BIT | VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT;
+    if ((properties.subgroup_properties.supportedOperations & required_ops) != required_ops) {
+        return false;
+    }
+    return SupportsSubgroupStage(stage);
+}
+
 void Device::SetupFamilies(VkSurfaceKHR surface) {
    const std::vector queue_family_properties = physical.GetQueueFamilyProperties();
    std::optional<u32> graphics;
--- a/src/video_core/vulkan_common/vulkan_device.h
+++ b/src/video_core/vulkan_common/vulkan_device.h
@@ -360,11 +360,26 @@ public:
        return properties.subgroup_size_control.requiredSubgroupSizeStages & stage;
    }

-    /// Returns true if the device supports the provided subgroup feature.
-    bool IsSubgroupFeatureSupported(VkSubgroupFeatureFlagBits feature) const {
-        return properties.subgroup_properties.supportedOperations & feature;
+    /// Returns true if the device supports the provided subgroup feature for the given stages.
+    bool IsSubgroupFeatureSupported(VkSubgroupFeatureFlagBits feature,
+                                    VkShaderStageFlags stage_mask = 0) const;
+
+    /// Returns true if the device reports subgroup support for the provided shader stages.
+    bool SupportsSubgroupStage(VkShaderStageFlags stage_mask) const;
+
+    /// Returns the set of stages that report subgroup support.
+    VkShaderStageFlags GetSubgroupSupportedStages() const {
+        return properties.subgroup_properties.supportedStages;
    }

+    /// Returns the set of subgroup operations reported by the driver.
+    VkSubgroupFeatureFlags GetSubgroupSupportedOperations() const {
+        return properties.subgroup_properties.supportedOperations;
+    }
+
+    /// Returns true if the driver can execute all warp intrinsics for the given shader stage.
+    bool SupportsWarpIntrinsics(VkShaderStageFlagBits stage) const;
+
    /// Returns the maximum number of push descriptors.
    u32 MaxPushDescriptors() const {
        return properties.push_descriptor.maxPushDescriptors;
--- a/tools/windows/install-vulkan-sdk.ps1
+++ b/tools/windows/install-vulkan-sdk.ps1
@@ -14,7 +14,7 @@ try {
    Exit 1
 }

-$VulkanSDKVer = "1.4.335.0"
+$VulkanSDKVer = "1.4.328.1"
 $VULKAN_SDK = "C:/VulkanSDK/$VulkanSDKVer"
 $ExeFile = "vulkansdk-windows-X64-$VulkanSDKVer.exe"
 $Uri = "https://sdk.lunarg.com/sdk/download/$VulkanSDKVer/windows/$ExeFile"
--- a/tools/windows/install-vulkan-sdk.sh
+++ b/tools/windows/install-vulkan-sdk.sh
@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
 # SPDX-License-Identifier: GPL-3.0-or-later

-: "${VULKAN_SDK_VER:=1.4.335.0}"
+: "${VULKAN_SDK_VER:=1.4.328.1}"
 : "${VULKAN_ROOT:=C:/VulkanSDK/$VULKAN_SDK_VER}"
 EXE_FILE="vulkansdk-windows-X64-$VULKAN_SDK_VER.exe"
 URI="https://sdk.lunarg.com/sdk/download/$VULKAN_SDK_VER/windows/$EXE_FILE"