Shader Recomnpiler: implement textuzreGrad 3D emulation constant propagation

src/core/hle/kernel/k_process.cpp

@@ -38,7 +38,7 @@ namespace {
  */
 void SetupMainThread(Core::System& system, KProcess& owner_process, u32 priority,
                      KProcessAddress stack_top) {
-    const KProcessAddress entry_point = owner_process.GetEntryPoint();
+    const KProcessAddress entry_point = owner_process.GetPageTable().GetCodeRegionStart();
     ASSERT(owner_process.GetResourceLimit()->Reserve(LimitableResource::ThreadCountMax, 1));
 
     KThread* thread = KThread::Create(system.Kernel());
@@ -358,21 +358,6 @@ Result KProcess::LoadFromMetadata(const FileSys::ProgramMetadata& metadata, std:
     m_system_resource_size = metadata.GetSystemResourceSize();
     m_image_size = code_size;
 
-    if (metadata.GetAddressSpaceType() == FileSys::ProgramAddressSpaceType::Is39Bit) {
-        // For 39-bit processes, the ASLR region starts at 0x800'0000 and is ~512GiB large.
-        // However, some (buggy) programs/libraries like skyline incorrectly depend on the
-        // existence of ASLR pages before the entry point, so we will adjust the load address
-        // to point to about 2GiB into the ASLR region.
-        m_code_address = 0x8000'0000;
-    } else {
-        // All other processes can be mapped at the beginning of the code region.
-        if (metadata.GetAddressSpaceType() == FileSys::ProgramAddressSpaceType::Is36Bit) {
-            m_code_address = 0x800'0000;
-        } else {
-            m_code_address = 0x20'0000;
-        }
-    }
-
     KScopedResourceReservation memory_reservation(
         m_resource_limit, LimitableResource::PhysicalMemoryMax, code_size + m_system_resource_size);
     if (!memory_reservation.Succeeded()) {
@@ -383,15 +368,15 @@ Result KProcess::LoadFromMetadata(const FileSys::ProgramMetadata& metadata, std:
     // Initialize process address space
     if (const Result result{m_page_table.InitializeForProcess(
             metadata.GetAddressSpaceType(), false, false, false, KMemoryManager::Pool::Application,
-            this->GetEntryPoint(), code_size, std::addressof(m_kernel.GetAppSystemResource()),
-            m_resource_limit, m_kernel.System().ApplicationMemory())};
+            0x8000000, code_size, std::addressof(m_kernel.GetAppSystemResource()), m_resource_limit,
+            m_kernel.System().ApplicationMemory())};
         result.IsError()) {
         R_RETURN(result);
     }
 
     // Map process code region
-    if (const Result result{m_page_table.MapProcessCode(this->GetEntryPoint(), code_size / PageSize,
-                                                        KMemoryState::Code,
+    if (const Result result{m_page_table.MapProcessCode(m_page_table.GetCodeRegionStart(),
+                                                        code_size / PageSize, KMemoryState::Code,
                                                         KMemoryPermission::None)};
         result.IsError()) {
         R_RETURN(result);

src/core/hle/kernel/k_process.h

@@ -177,10 +177,6 @@ public:
         return m_program_id;
     }
 
-    KProcessAddress GetEntryPoint() const {
-        return m_code_address;
-    }
-
     /// Gets the resource limit descriptor for this process
     KResourceLimit* GetResourceLimit() const;
 
@@ -489,9 +485,6 @@ private:
     /// Address indicating the location of the process' dedicated TLS region.
     KProcessAddress m_plr_address = 0;
 
-    /// Address indicating the location of the process's entry point.
-    KProcessAddress m_code_address = 0;
-
     /// Random values for svcGetInfo RandomEntropy
     std::array<u64, RANDOM_ENTROPY_SIZE> m_random_entropy{};
 

src/core/loader/deconstructed_rom_directory.cpp

@@ -153,7 +153,7 @@ AppLoader_DeconstructedRomDirectory::LoadResult AppLoader_DeconstructedRomDirect
 
     // Load NSO modules
     modules.clear();
-    const VAddr base_address{GetInteger(process.GetEntryPoint())};
+    const VAddr base_address{GetInteger(process.GetPageTable().GetCodeRegionStart())};
     VAddr next_load_addr{base_address};
     const FileSys::PatchManager pm{metadata.GetTitleID(), system.GetFileSystemController(),
                                    system.GetContentProvider()};

src/core/loader/kip.cpp

@@ -96,7 +96,7 @@ AppLoader::LoadResult AppLoader_KIP::Load(Kernel::KProcess& process,
     }
 
     codeset.memory = std::move(program_image);
-    const VAddr base_address = GetInteger(process.GetEntryPoint());
+    const VAddr base_address = GetInteger(process.GetPageTable().GetCodeRegionStart());
     process.LoadModule(std::move(codeset), base_address);
 
     LOG_DEBUG(Loader, "loaded module {} @ 0x{:X}", kip->GetName(), base_address);

src/core/loader/nro.cpp

@@ -203,7 +203,7 @@ static bool LoadNroImpl(Kernel::KProcess& process, const std::vector<u8>& data)
 
     // Load codeset for current process
     codeset.memory = std::move(program_image);
-    process.LoadModule(std::move(codeset), process.GetEntryPoint());
+    process.LoadModule(std::move(codeset), process.GetPageTable().GetCodeRegionStart());
 
     return true;
 }

src/core/loader/nso.cpp

@@ -167,7 +167,7 @@ AppLoader_NSO::LoadResult AppLoader_NSO::Load(Kernel::KProcess& process, Core::S
     modules.clear();
 
     // Load module
-    const VAddr base_address = GetInteger(process.GetEntryPoint());
+    const VAddr base_address = GetInteger(process.GetPageTable().GetCodeRegionStart());
     if (!LoadModule(process, system, *file, base_address, true, true)) {
         return {ResultStatus::ErrorLoadingNSO, {}};
     }

src/core/reporter.cpp

@@ -117,8 +117,8 @@ json GetProcessorStateDataAuto(Core::System& system) {
     arm.SaveContext(context);
 
     return GetProcessorStateData(process->Is64BitProcess() ? "AArch64" : "AArch32",
-                                 GetInteger(process->GetEntryPoint()), context.sp, context.pc,
-                                 context.pstate, context.cpu_registers);
+                                 GetInteger(process->GetPageTable().GetCodeRegionStart()),
+                                 context.sp, context.pc, context.pstate, context.cpu_registers);
 }
 
 json GetBacktraceData(Core::System& system) {

src/shader_recompiler/backend/glasm/emit_glasm_image.cpp

@@ -558,12 +558,15 @@ void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
                        const IR::Value& coord, const IR::Value& derivatives,
                        const IR::Value& offset, const IR::Value& lod_clamp) {
     const auto info{inst.Flags<IR::TextureInstInfo>()};
-    ScopedRegister dpdx, dpdy;
+    ScopedRegister dpdx, dpdy, coords;
     const bool multi_component{info.num_derivates > 1 || info.has_lod_clamp};
     if (multi_component) {
         // Allocate this early to avoid aliasing other registers
         dpdx = ScopedRegister{ctx.reg_alloc};
         dpdy = ScopedRegister{ctx.reg_alloc};
+        if (info.num_derivates >= 3) {
+            coords = ScopedRegister{ctx.reg_alloc};
+        }
     }
     const auto sparse_inst{PrepareSparse(inst)};
     const std::string_view sparse_mod{sparse_inst ? ".SPARSE" : ""};
@@ -580,15 +583,27 @@ void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
                 "MOV.F {}.y,{}.w;",
                 dpdx.reg, derivatives_vec, dpdx.reg, derivatives_vec, dpdy.reg, derivatives_vec,
                 dpdy.reg, derivatives_vec);
+        Register final_coord;
+        if (info.num_derivates >= 3) {
+            ctx.Add("MOV.F {}.z,{}.x;"
+                    "MOV.F {}.z,{}.y;",
+                    dpdx.reg, coord_vec, dpdy.reg, coord_vec);
+            ctx.Add("MOV.F {}.x,0;"
+                    "MOV.F {}.y,0;",
+                    "MOV.F {}.z,0;", coords.reg, coords.reg, coords.reg);
+            final_coord = coords.reg;
+        } else {
+            final_coord = coord_vec;
+        }
         if (info.has_lod_clamp) {
             const ScalarF32 lod_clamp_value{ctx.reg_alloc.Consume(lod_clamp)};
             ctx.Add("MOV.F {}.w,{};"
                     "TXD.F.LODCLAMP{} {},{},{},{},{},{}{};",
-                    dpdy.reg, lod_clamp_value, sparse_mod, ret, coord_vec, dpdx.reg, dpdy.reg,
+                    dpdy.reg, lod_clamp_value, sparse_mod, ret, final_coord, dpdx.reg, dpdy.reg,
                     texture, type, offset_vec);
         } else {
-            ctx.Add("TXD.F{} {},{},{},{},{},{}{};", sparse_mod, ret, coord_vec, dpdx.reg, dpdy.reg,
-                    texture, type, offset_vec);
+            ctx.Add("TXD.F{} {},{},{},{},{},{}{};", sparse_mod, ret, final_coord, dpdx.reg,
+                    dpdy.reg, texture, type, offset_vec);
         }
     } else {
         ctx.Add("TXD.F{} {},{},{}.x,{}.y,{},{}{};", sparse_mod, ret, coord_vec, derivatives_vec,

src/shader_recompiler/backend/glsl/emit_glsl_image.cpp

@@ -548,7 +548,7 @@ void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
     if (sparse_inst) {
         throw NotImplementedException("EmitImageGradient Sparse");
     }
-    if (!offset.IsEmpty()) {
+    if (!offset.IsEmpty() && info.num_derivates <= 2) {
         throw NotImplementedException("EmitImageGradient offset");
     }
     const auto texture{Texture(ctx, info, index)};
@@ -556,6 +556,12 @@ void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
     const bool multi_component{info.num_derivates > 1 || info.has_lod_clamp};
     const auto derivatives_vec{ctx.var_alloc.Consume(derivatives)};
     if (multi_component) {
+        if (info.num_derivates >= 3) {
+            const auto offset_vec{ctx.var_alloc.Consume(offset)};
+            ctx.Add("{}=textureGrad({},{},vec3({}.xz, {}.x),vec3({}.yz, {}.y));", texel, texture,
+                    coords, derivatives_vec, offset_vec, derivatives_vec, offset_vec);
+            return;
+        }
         ctx.Add("{}=textureGrad({},{},vec2({}.xz),vec2({}.yz));", texel, texture, coords,
                 derivatives_vec, derivatives_vec);
     } else {

src/shader_recompiler/frontend/ir/modifiers.h

@@ -42,6 +42,7 @@ union TextureInstInfo {
     BitField<23, 2, u32> gather_component;
     BitField<25, 2, u32> num_derivates;
     BitField<27, 3, ImageFormat> image_format;
+    BitField<30, 1, u32> ndv_is_active;
 };
 static_assert(sizeof(TextureInstInfo) <= sizeof(u32));
 

src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_swizzled_add.cpp

@@ -19,7 +19,7 @@ void TranslatorVisitor::FSWZADD(u64 insn) {
     } const fswzadd{insn};
 
     if (fswzadd.ndv != 0) {
-        throw NotImplementedException("FSWZADD NDV");
+        LOG_WARNING(Shader, "(STUBBED) FSWZADD - NDV mode");
     }
 
     const IR::F32 src_a{GetFloatReg8(insn)};

src/shader_recompiler/frontend/maxwell/translate/impl/move_register.cpp

@@ -16,8 +16,10 @@ void MOV(TranslatorVisitor& v, u64 insn, const IR::U32& src, bool is_mov32i = fa
         BitField<12, 4, u64> mov32i_mask;
     } const mov{insn};
 
-    if ((is_mov32i ? mov.mov32i_mask : mov.mask) != 0xf) {
-        throw NotImplementedException("Non-full move mask");
+    u64 mask = is_mov32i ? mov.mov32i_mask : mov.mask;
+    if (mask != 0xf && mask != 0x1) {
+        LOG_WARNING(Shader, "(STUBBED) Masked Mov");
+        return;
     }
     v.X(mov.dest_reg, src);
 }

src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp

@@ -209,7 +209,7 @@ void TranslatorVisitor::R2B(u64) {
 }
 
 void TranslatorVisitor::RAM(u64) {
-    ThrowNotImplemented(Opcode::RAM);
+    LOG_WARNING(Shader, "(STUBBED) RAM Instruction");
 }
 
 void TranslatorVisitor::RET(u64) {
@@ -221,7 +221,7 @@ void TranslatorVisitor::RTT(u64) {
 }
 
 void TranslatorVisitor::SAM(u64) {
-    ThrowNotImplemented(Opcode::SAM);
+    LOG_WARNING(Shader, "(STUBBED) SAM Instruction");
 }
 
 void TranslatorVisitor::SETCRSPTR(u64) {

src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch.cpp

@@ -172,6 +172,7 @@ void Impl(TranslatorVisitor& v, u64 insn, bool aoffi, Blod blod, bool lc,
     info.is_depth.Assign(tex.dc != 0 ? 1 : 0);
     info.has_bias.Assign(blod == Blod::LB || blod == Blod::LBA ? 1 : 0);
     info.has_lod_clamp.Assign(lc ? 1 : 0);
+    info.ndv_is_active.Assign(tex.ndv != 0 ? 1 : 0);
 
     const IR::Value sample{[&]() -> IR::Value {
         if (tex.dc == 0) {

src/shader_recompiler/ir_opt/constant_propagation_pass.cpp

@@ -10,6 +10,7 @@
 #include "shader_recompiler/environment.h"
 #include "shader_recompiler/exception.h"
 #include "shader_recompiler/frontend/ir/ir_emitter.h"
+#include "shader_recompiler/frontend/ir/modifiers.h"
 #include "shader_recompiler/frontend/ir/value.h"
 #include "shader_recompiler/ir_opt/passes.h"
 
@@ -410,7 +411,49 @@ void FoldSelect(IR::Inst& inst) {
     }
 }
 
+void FoldFPAdd32(IR::Inst& inst) {
+    if (FoldWhenAllImmediates(inst, [](f32 a, f32 b) { return a + b; })) {
+        return;
+    }
+    const IR::Value lhs_value{inst.Arg(0)};
+    const IR::Value rhs_value{inst.Arg(1)};
+    const auto check_neutral = [](const IR::Value& one_operand) {
+        return one_operand.IsImmediate() && std::abs(one_operand.F32()) == 0.0f;
+    };
+    if (check_neutral(lhs_value)) {
+        inst.ReplaceUsesWith(rhs_value);
+    }
+    if (check_neutral(rhs_value)) {
+        inst.ReplaceUsesWith(lhs_value);
+    }
+}
+
+bool FoldDerivateYFromCorrection(IR::Inst& inst) {
+    const IR::Value lhs_value{inst.Arg(0)};
+    const IR::Value rhs_value{inst.Arg(1)};
+    IR::Inst* const lhs_op{lhs_value.InstRecursive()};
+    IR::Inst* const rhs_op{rhs_value.InstRecursive()};
+    if (lhs_op->GetOpcode() == IR::Opcode::YDirection) {
+        if (rhs_op->GetOpcode() != IR::Opcode::DPdyFine) {
+            return false;
+        }
+        inst.ReplaceUsesWith(rhs_value);
+        return true;
+    }
+    if (rhs_op->GetOpcode() != IR::Opcode::YDirection) {
+        return false;
+    }
+    if (lhs_op->GetOpcode() != IR::Opcode::DPdyFine) {
+        return false;
+    }
+    inst.ReplaceUsesWith(lhs_value);
+    return true;
+}
+
 void FoldFPMul32(IR::Inst& inst) {
+    if (FoldWhenAllImmediates(inst, [](f32 a, f32 b) { return a * b; })) {
+        return;
+    }
     const auto control{inst.Flags<IR::FpControl>()};
     if (control.no_contraction) {
         return;
@@ -421,6 +464,9 @@ void FoldFPMul32(IR::Inst& inst) {
     if (lhs_value.IsImmediate() || rhs_value.IsImmediate()) {
         return;
     }
+    if (FoldDerivateYFromCorrection(inst)) {
+        return;
+    }
     IR::Inst* const lhs_op{lhs_value.InstRecursive()};
     IR::Inst* const rhs_op{rhs_value.InstRecursive()};
     if (lhs_op->GetOpcode() != IR::Opcode::FPMul32 ||
@@ -622,7 +668,12 @@ void FoldFSwizzleAdd(IR::Block& block, IR::Inst& inst) {
     }
     const IR::Value value_3{GetThroughCast(inst2->Arg(0).Resolve(), IR::Opcode::BitCastU32F32)};
     if (value_2 != value_3) {
-        return;
+        if (!value_2.IsImmediate() || !value_3.IsImmediate()) {
+            return;
+        }
+        if (Common::BitCast<u32>(value_2.F32()) != value_3.U32()) {
+            return;
+        }
     }
     const IR::Value index{inst2->Arg(1)};
     const IR::Value clamp{inst2->Arg(2)};
@@ -648,6 +699,169 @@ void FoldFSwizzleAdd(IR::Block& block, IR::Inst& inst) {
     }
 }
 
+bool FindGradient3DDerivates(std::array<IR::Value, 3>& results, IR::Value coord) {
+    if (coord.IsImmediate()) {
+        return false;
+    }
+    const auto check_through_shuffle = [](IR::Value input, IR::Value& result) {
+        const IR::Value value_1{GetThroughCast(input.Resolve(), IR::Opcode::BitCastF32U32)};
+        IR::Inst* const inst2{value_1.InstRecursive()};
+        if (inst2->GetOpcode() != IR::Opcode::ShuffleIndex) {
+            return false;
+        }
+        const IR::Value index{inst2->Arg(1).Resolve()};
+        const IR::Value clamp{inst2->Arg(2).Resolve()};
+        const IR::Value segmentation_mask{inst2->Arg(3).Resolve()};
+        if (!index.IsImmediate() || !clamp.IsImmediate() || !segmentation_mask.IsImmediate()) {
+            return false;
+        }
+        if (index.U32() != 3 && clamp.U32() != 3) {
+            return false;
+        }
+        result = GetThroughCast(inst2->Arg(0).Resolve(), IR::Opcode::BitCastU32F32);
+        return true;
+    };
+    IR::Inst* const inst = coord.InstRecursive();
+    if (inst->GetOpcode() != IR::Opcode::FSwizzleAdd) {
+        return false;
+    }
+    std::array<IR::Value, 3> temporary_values;
+    IR::Value value_1 = inst->Arg(0).Resolve();
+    IR::Value value_2 = inst->Arg(1).Resolve();
+    IR::Value value_3 = inst->Arg(2).Resolve();
+    std::array<u32, 4> swizzles_mask_a{};
+    std::array<u32, 4> swizzles_mask_b{};
+    const auto resolve_mask = [](std::array<u32, 4>& mask_results, IR::Value mask) {
+        u32 value = mask.U32();
+        for (size_t i = 0; i < 4; i++) {
+            mask_results[i] = (value >> (i * 2)) & 0x3;
+        }
+    };
+    resolve_mask(swizzles_mask_a, value_3);
+    size_t coordinate_index = 0;
+    const auto resolve_pending = [&](IR::Value resolve_v) {
+        IR::Inst* const inst_r = resolve_v.InstRecursive();
+        if (inst_r->GetOpcode() != IR::Opcode::FSwizzleAdd) {
+            return false;
+        }
+        if (!check_through_shuffle(inst_r->Arg(0).Resolve(), temporary_values[1])) {
+            return false;
+        }
+        if (!check_through_shuffle(inst_r->Arg(1).Resolve(), temporary_values[2])) {
+            return false;
+        }
+        resolve_mask(swizzles_mask_b, inst_r->Arg(2).Resolve());
+        return true;
+    };
+    if (value_1.IsImmediate() || value_2.IsImmediate()) {
+        return false;
+    }
+    bool should_continue = false;
+    if (resolve_pending(value_1)) {
+        should_continue = check_through_shuffle(value_2, temporary_values[0]);
+        coordinate_index = 0;
+    }
+    if (resolve_pending(value_2)) {
+        should_continue = check_through_shuffle(value_1, temporary_values[0]);
+        coordinate_index = 2;
+    }
+    if (!should_continue) {
+        return false;
+    }
+    // figure which is which
+    size_t zero_mask_a = 0;
+    size_t zero_mask_b = 0;
+    for (size_t i = 0; i < 4; i++) {
+        if (swizzles_mask_a[i] == 2 || swizzles_mask_b[i] == 2) {
+            // last operand can be inversed, we cannot determine a result.
+            return false;
+        }
+        zero_mask_a |= static_cast<size_t>(swizzles_mask_a[i] == 3 ? 1 : 0) << i;
+        zero_mask_b |= static_cast<size_t>(swizzles_mask_b[i] == 3 ? 1 : 0) << i;
+    }
+    static constexpr size_t ddx_pattern = 0b1010;
+    static constexpr size_t ddx_pattern_inv = ~ddx_pattern & 0b00001111;
+    if (std::popcount(zero_mask_a) != 2) {
+        return false;
+    }
+    if (std::popcount(zero_mask_b) != 2) {
+        return false;
+    }
+    if (zero_mask_a == zero_mask_b) {
+        return false;
+    }
+    results[0] = temporary_values[coordinate_index];
+
+    if (coordinate_index == 0) {
+        if (zero_mask_b == ddx_pattern || zero_mask_b == ddx_pattern_inv) {
+            results[1] = temporary_values[1];
+            results[2] = temporary_values[2];
+            return true;
+        }
+        results[2] = temporary_values[1];
+        results[1] = temporary_values[2];
+    } else {
+        const auto assign_result = [&results](IR::Value temporary_value, size_t mask) {
+            if (mask == ddx_pattern || mask == ddx_pattern_inv) {
+                results[1] = temporary_value;
+                return;
+            }
+            results[2] = temporary_value;
+        };
+        assign_result(temporary_values[1], zero_mask_b);
+        assign_result(temporary_values[0], zero_mask_a);
+    }
+
+    return true;
+}
+
+void FoldImageSampleImplicitLod(IR::Block& block, IR::Inst& inst) {
+    IR::TextureInstInfo info = inst.Flags<IR::TextureInstInfo>();
+    auto orig_opcode = inst.GetOpcode();
+    if (info.ndv_is_active == 0) {
+        return;
+    }
+    if (info.type != TextureType::Color3D) {
+        return;
+    }
+    const IR::Value handle{inst.Arg(0)};
+    const IR::Value coords{inst.Arg(1)};
+    const IR::Value bias_lc{inst.Arg(2)};
+    const IR::Value offset{inst.Arg(3)};
+    if (!offset.IsImmediate()) {
+        return;
+    }
+    IR::Inst* const inst2 = coords.InstRecursive();
+    std::array<std::array<IR::Value, 3>, 3> results_matrix;
+    for (size_t i = 0; i < 3; i++) {
+        if (!FindGradient3DDerivates(results_matrix[i], inst2->Arg(i).Resolve())) {
+            return;
+        }
+    }
+    IR::F32 lod_clamp{};
+    if (info.has_lod_clamp != 0) {
+        if (!bias_lc.IsImmediate()) {
+            lod_clamp = IR::F32{bias_lc.InstRecursive()->Arg(1).Resolve()};
+        } else {
+            lod_clamp = IR::F32{bias_lc};
+        }
+    }
+    IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
+    IR::Value new_coords =
+        ir.CompositeConstruct(results_matrix[0][0], results_matrix[1][0], results_matrix[2][0]);
+    IR::Value derivatives_1 = ir.CompositeConstruct(results_matrix[0][1], results_matrix[0][2],
+                                                    results_matrix[1][1], results_matrix[1][2]);
+    IR::Value derivatives_2 = ir.CompositeConstruct(results_matrix[2][1], results_matrix[2][2]);
+    info.num_derivates.Assign(3);
+    IR::Value new_gradient_instruction =
+        ir.ImageGradient(handle, new_coords, derivatives_1, derivatives_2, lod_clamp, info);
+    IR::Inst* const new_inst = new_gradient_instruction.InstRecursive();
+    if (orig_opcode == IR::Opcode::ImageSampleImplicitLod) {
+        new_inst->ReplaceOpcode(IR::Opcode::ImageGradient);
+    }
+    inst.ReplaceUsesWith(new_gradient_instruction);
+}
+
 void FoldConstBuffer(Environment& env, IR::Block& block, IR::Inst& inst) {
     const IR::Value bank{inst.Arg(0)};
     const IR::Value offset{inst.Arg(1)};
@@ -743,6 +957,12 @@ void ConstantPropagation(Environment& env, IR::Block& block, IR::Inst& inst) {
     case IR::Opcode::SelectF32:
     case IR::Opcode::SelectF64:
         return FoldSelect(inst);
+    case IR::Opcode::FPNeg32:
+        FoldWhenAllImmediates(inst, [](f32 a) { return -a; });
+        return;
+    case IR::Opcode::FPAdd32:
+        FoldFPAdd32(inst);
+        return;
     case IR::Opcode::FPMul32:
         return FoldFPMul32(inst);
     case IR::Opcode::LogicalAnd:
@@ -858,6 +1078,11 @@ void ConstantPropagation(Environment& env, IR::Block& block, IR::Inst& inst) {
             FoldDriverConstBuffer(env, block, inst, 1);
         }
         break;
+    case IR::Opcode::BindlessImageSampleImplicitLod:
+    case IR::Opcode::BoundImageSampleImplicitLod:
+    case IR::Opcode::ImageSampleImplicitLod:
+        FoldImageSampleImplicitLod(block, inst);
+        break;
     default:
         break;
     }

src/video_core/host_shaders/CMakeLists.txt

@@ -50,7 +50,6 @@ set(SHADER_FILES
     vulkan_blit_depth_stencil.frag
     vulkan_color_clear.frag
     vulkan_color_clear.vert
-    vulkan_depthstencil_clear.frag
     vulkan_fidelityfx_fsr_easu_fp16.comp
     vulkan_fidelityfx_fsr_easu_fp32.comp
     vulkan_fidelityfx_fsr_rcas_fp16.comp

src/video_core/host_shaders/vulkan_depthstencil_clear.frag

@@ -1,12 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#version 460 core
-
-layout (push_constant) uniform PushConstants {
-    vec4 clear_depth;
-};
-
-void main() {
-    gl_FragDepth = clear_depth.x;
-}

src/video_core/renderer_vulkan/blit_image.cpp

@@ -16,7 +16,6 @@
 #include "video_core/host_shaders/vulkan_blit_depth_stencil_frag_spv.h"
 #include "video_core/host_shaders/vulkan_color_clear_frag_spv.h"
 #include "video_core/host_shaders/vulkan_color_clear_vert_spv.h"
-#include "video_core/host_shaders/vulkan_depthstencil_clear_frag_spv.h"
 #include "video_core/renderer_vulkan/blit_image.h"
 #include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
@@ -429,7 +428,6 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
       blit_depth_stencil_frag(BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_FRAG_SPV)),
       clear_color_vert(BuildShader(device, VULKAN_COLOR_CLEAR_VERT_SPV)),
       clear_color_frag(BuildShader(device, VULKAN_COLOR_CLEAR_FRAG_SPV)),
-      clear_stencil_frag(BuildShader(device, VULKAN_DEPTHSTENCIL_CLEAR_FRAG_SPV)),
       convert_depth_to_float_frag(BuildShader(device, CONVERT_DEPTH_TO_FLOAT_FRAG_SPV)),
       convert_float_to_depth_frag(BuildShader(device, CONVERT_FLOAT_TO_DEPTH_FRAG_SPV)),
       convert_abgr8_to_d24s8_frag(BuildShader(device, CONVERT_ABGR8_TO_D24S8_FRAG_SPV)),
@@ -595,28 +593,6 @@ void BlitImageHelper::ClearColor(const Framebuffer* dst_framebuffer, u8 color_ma
     scheduler.InvalidateState();
 }
 
-void BlitImageHelper::ClearDepthStencil(const Framebuffer* dst_framebuffer, bool depth_clear,
-                                        f32 clear_depth, u8 stencil_mask, u32 stencil_ref,
-                                        u32 stencil_compare_mask, const Region2D& dst_region) {
-    const BlitDepthStencilPipelineKey key{
-        .renderpass = dst_framebuffer->RenderPass(),
-        .depth_clear = depth_clear,
-        .stencil_mask = stencil_mask,
-        .stencil_compare_mask = stencil_compare_mask,
-        .stencil_ref = stencil_ref,
-    };
-    const VkPipeline pipeline = FindOrEmplaceClearStencilPipeline(key);
-    const VkPipelineLayout layout = *clear_color_pipeline_layout;
-    scheduler.RequestRenderpass(dst_framebuffer);
-    scheduler.Record([pipeline, layout, clear_depth, dst_region](vk::CommandBuffer cmdbuf) {
-        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
-        BindBlitState(cmdbuf, dst_region);
-        cmdbuf.PushConstants(layout, VK_SHADER_STAGE_FRAGMENT_BIT, clear_depth);
-        cmdbuf.Draw(3, 1, 0, 0);
-    });
-    scheduler.InvalidateState();
-}
-
 void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
                               const ImageView& src_image_view) {
     const VkPipelineLayout layout = *one_texture_pipeline_layout;
@@ -844,61 +820,6 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipel
     return *clear_color_pipelines.back();
 }
 
-VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
-    const BlitDepthStencilPipelineKey& key) {
-    const auto it = std::ranges::find(clear_stencil_keys, key);
-    if (it != clear_stencil_keys.end()) {
-        return *clear_stencil_pipelines[std::distance(clear_stencil_keys.begin(), it)];
-    }
-    clear_stencil_keys.push_back(key);
-    const std::array stages = MakeStages(*clear_color_vert, *clear_stencil_frag);
-    const auto stencil = VkStencilOpState{
-        .failOp = VK_STENCIL_OP_KEEP,
-        .passOp = VK_STENCIL_OP_REPLACE,
-        .depthFailOp = VK_STENCIL_OP_KEEP,
-        .compareOp = VK_COMPARE_OP_ALWAYS,
-        .compareMask = key.stencil_compare_mask,
-        .writeMask = key.stencil_mask,
-        .reference = key.stencil_ref,
-    };
-    const VkPipelineDepthStencilStateCreateInfo depth_stencil_ci{
-        .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .depthTestEnable = VK_FALSE,
-        .depthWriteEnable = key.depth_clear,
-        .depthCompareOp = VK_COMPARE_OP_ALWAYS,
-        .depthBoundsTestEnable = VK_FALSE,
-        .stencilTestEnable = VK_TRUE,
-        .front = stencil,
-        .back = stencil,
-        .minDepthBounds = 0.0f,
-        .maxDepthBounds = 0.0f,
-    };
-    clear_stencil_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
-        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .stageCount = static_cast<u32>(stages.size()),
-        .pStages = stages.data(),
-        .pVertexInputState = &PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
-        .pInputAssemblyState = &PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
-        .pTessellationState = nullptr,
-        .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
-        .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
-        .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
-        .pDepthStencilState = &depth_stencil_ci,
-        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
-        .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
-        .layout = *clear_color_pipeline_layout,
-        .renderPass = key.renderpass,
-        .subpass = 0,
-        .basePipelineHandle = VK_NULL_HANDLE,
-        .basePipelineIndex = 0,
-    }));
-    return *clear_stencil_pipelines.back();
-}
-
 void BlitImageHelper::ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass,
                                       bool is_target_depth) {
     if (pipeline) {

src/video_core/renderer_vulkan/blit_image.h

@@ -27,16 +27,6 @@ struct BlitImagePipelineKey {
     Tegra::Engines::Fermi2D::Operation operation;
 };
 
-struct BlitDepthStencilPipelineKey {
-    constexpr auto operator<=>(const BlitDepthStencilPipelineKey&) const noexcept = default;
-
-    VkRenderPass renderpass;
-    bool depth_clear;
-    u8 stencil_mask;
-    u32 stencil_compare_mask;
-    u32 stencil_ref;
-};
-
 class BlitImageHelper {
 public:
     explicit BlitImageHelper(const Device& device, Scheduler& scheduler,
@@ -74,10 +64,6 @@ public:
     void ClearColor(const Framebuffer* dst_framebuffer, u8 color_mask,
                     const std::array<f32, 4>& clear_color, const Region2D& dst_region);
 
-    void ClearDepthStencil(const Framebuffer* dst_framebuffer, bool depth_clear, f32 clear_depth,
-                           u8 stencil_mask, u32 stencil_ref, u32 stencil_compare_mask,
-                           const Region2D& dst_region);
-
 private:
     void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
                  const ImageView& src_image_view);
@@ -90,8 +76,6 @@ private:
     [[nodiscard]] VkPipeline FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key);
 
     [[nodiscard]] VkPipeline FindOrEmplaceClearColorPipeline(const BlitImagePipelineKey& key);
-    [[nodiscard]] VkPipeline FindOrEmplaceClearStencilPipeline(
-        const BlitDepthStencilPipelineKey& key);
 
     void ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass, bool is_target_depth);
 
@@ -124,7 +108,6 @@ private:
     vk::ShaderModule blit_depth_stencil_frag;
     vk::ShaderModule clear_color_vert;
     vk::ShaderModule clear_color_frag;
-    vk::ShaderModule clear_stencil_frag;
     vk::ShaderModule convert_depth_to_float_frag;
     vk::ShaderModule convert_float_to_depth_frag;
     vk::ShaderModule convert_abgr8_to_d24s8_frag;
@@ -139,8 +122,6 @@ private:
     std::vector<vk::Pipeline> blit_depth_stencil_pipelines;
     std::vector<BlitImagePipelineKey> clear_color_keys;
     std::vector<vk::Pipeline> clear_color_pipelines;
-    std::vector<BlitDepthStencilPipelineKey> clear_stencil_keys;
-    std::vector<vk::Pipeline> clear_stencil_pipelines;
     vk::Pipeline convert_d32_to_r32_pipeline;
     vk::Pipeline convert_r32_to_d32_pipeline;
     vk::Pipeline convert_d16_to_r16_pipeline;

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -428,27 +428,15 @@ void RasterizerVulkan::Clear(u32 layer_count) {
     if (aspect_flags == 0) {
         return;
     }
-
-    if (use_stencil && regs.stencil_front_mask != 0xFF && regs.stencil_front_mask != 0) {
-        Region2D dst_region = {
-            Offset2D{.x = clear_rect.rect.offset.x, .y = clear_rect.rect.offset.y},
-            Offset2D{.x = clear_rect.rect.offset.x + static_cast<s32>(clear_rect.rect.extent.width),
-                     .y = clear_rect.rect.offset.y +
-                          static_cast<s32>(clear_rect.rect.extent.height)}};
-        blit_image.ClearDepthStencil(framebuffer, use_depth, regs.clear_depth,
-                                     static_cast<u8>(regs.stencil_front_mask), regs.clear_stencil,
-                                     regs.stencil_front_func_mask, dst_region);
-    } else {
-        scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil,
-                          clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) {
-            VkClearAttachment attachment;
-            attachment.aspectMask = aspect_flags;
-            attachment.colorAttachment = 0;
-            attachment.clearValue.depthStencil.depth = clear_depth;
-            attachment.clearValue.depthStencil.stencil = clear_stencil;
-            cmdbuf.ClearAttachments(attachment, clear_rect);
-        });
-    }
+    scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil,
+                      clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) {
+        VkClearAttachment attachment;
+        attachment.aspectMask = aspect_flags;
+        attachment.colorAttachment = 0;
+        attachment.clearValue.depthStencil.depth = clear_depth;
+        attachment.clearValue.depthStencil.stencil = clear_stencil;
+        cmdbuf.ClearAttachments(attachment, clear_rect);
+    });
 }
 
 void RasterizerVulkan::DispatchCompute() {