13 files changed, 264 insertions, 1 deletions

diff --git a/src/shader_recompiler/CMakeLists.txt b/src/shader_recompiler/CMakeLists.txt
index 525b2363c..07e75f9d8 100644
--- a/src/shader_recompiler/CMakeLists.txt
+++ b/src/shader_recompiler/CMakeLists.txt
@@ -216,6 +216,7 @@ add_library(shader_recompiler STATIC
     frontend/maxwell/translate_program.h
     host_translate_info.h
     ir_opt/collect_shader_info_pass.cpp
+    ir_opt/conditional_barrier_pass.cpp
     ir_opt/constant_propagation_pass.cpp
     ir_opt/dead_code_elimination_pass.cpp
     ir_opt/dual_vertex_pass.cpp
@@ -223,6 +224,7 @@ add_library(shader_recompiler STATIC
     ir_opt/identity_removal_pass.cpp
     ir_opt/layer_pass.cpp
     ir_opt/lower_fp16_to_fp32.cpp
+    ir_opt/lower_fp64_to_fp32.cpp
     ir_opt/lower_int64_to_int32.cpp
     ir_opt/passes.h
     ir_opt/position_pass.cpp
diff --git a/src/shader_recompiler/frontend/maxwell/translate_program.cpp b/src/shader_recompiler/frontend/maxwell/translate_program.cpp
index 17a6d4888..928b35561 100644
--- a/src/shader_recompiler/frontend/maxwell/translate_program.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate_program.cpp
@@ -280,12 +280,18 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
     RemoveUnreachableBlocks(program);
 
     // Replace instructions before the SSA rewrite
+    if (!host_info.support_float64) {
+        Optimization::LowerFp64ToFp32(program);
+    }
     if (!host_info.support_float16) {
         Optimization::LowerFp16ToFp32(program);
     }
     if (!host_info.support_int64) {
         Optimization::LowerInt64ToInt32(program);
     }
+    if (!host_info.support_conditional_barrier) {
+        Optimization::ConditionalBarrierPass(program);
+    }
     Optimization::SsaRewritePass(program);
 
     Optimization::ConstantPropagationPass(env, program);
diff --git a/src/shader_recompiler/host_translate_info.h b/src/shader_recompiler/host_translate_info.h
index 2aaa6c5ea..7d2ded907 100644
--- a/src/shader_recompiler/host_translate_info.h
+++ b/src/shader_recompiler/host_translate_info.h
@@ -10,6 +10,7 @@ namespace Shader {
 
 /// Misc information about the host
 struct HostTranslateInfo {
+    bool support_float64{};      ///< True when the device supports 64-bit floats
     bool support_float16{};      ///< True when the device supports 16-bit floats
     bool support_int64{};        ///< True when the device supports 64-bit integers
     bool needs_demote_reorder{}; ///< True when the device needs DemoteToHelperInvocation reordered
@@ -17,6 +18,8 @@ struct HostTranslateInfo {
     bool support_viewport_index_layer{}; ///< True when the device supports gl_Layer in VS
     bool support_geometry_shader_passthrough{}; ///< True when the device supports geometry
                                                 ///< passthrough shaders
+    bool support_conditional_barrier{}; ///< True when the device supports barriers in conditional
+                                        ///< control flow
 };
 
 } // namespace Shader
diff --git a/src/shader_recompiler/ir_opt/conditional_barrier_pass.cpp b/src/shader_recompiler/ir_opt/conditional_barrier_pass.cpp
new file mode 100644
index 000000000..c3ed27f4f
--- /dev/null
+++ b/src/shader_recompiler/ir_opt/conditional_barrier_pass.cpp
@@ -0,0 +1,44 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "shader_recompiler/frontend/ir/program.h"
+#include "shader_recompiler/ir_opt/passes.h"
+
+namespace Shader::Optimization {
+
+void ConditionalBarrierPass(IR::Program& program) {
+    s32 conditional_control_flow_count{0};
+    s32 conditional_return_count{0};
+    for (IR::AbstractSyntaxNode& node : program.syntax_list) {
+        switch (node.type) {
+        case IR::AbstractSyntaxNode::Type::If:
+        case IR::AbstractSyntaxNode::Type::Loop:
+            conditional_control_flow_count++;
+            break;
+        case IR::AbstractSyntaxNode::Type::EndIf:
+        case IR::AbstractSyntaxNode::Type::Repeat:
+            conditional_control_flow_count--;
+            break;
+        case IR::AbstractSyntaxNode::Type::Unreachable:
+        case IR::AbstractSyntaxNode::Type::Return:
+            if (conditional_control_flow_count > 0) {
+                conditional_return_count++;
+            }
+            break;
+        case IR::AbstractSyntaxNode::Type::Block:
+            for (IR::Inst& inst : node.data.block->Instructions()) {
+                if ((conditional_control_flow_count > 0 || conditional_return_count > 0) &&
+                    inst.GetOpcode() == IR::Opcode::Barrier) {
+                    LOG_WARNING(Shader, "Barrier within conditional control flow");
+                    inst.ReplaceOpcode(IR::Opcode::Identity);
+                }
+            }
+            break;
+        default:
+            break;
+        }
+    }
+    ASSERT(conditional_control_flow_count == 0);
+}
+
+} // namespace Shader::Optimization
diff --git a/src/shader_recompiler/ir_opt/lower_fp64_to_fp32.cpp b/src/shader_recompiler/ir_opt/lower_fp64_to_fp32.cpp
new file mode 100644
index 000000000..5db7a38ad
--- /dev/null
+++ b/src/shader_recompiler/ir_opt/lower_fp64_to_fp32.cpp
@@ -0,0 +1,185 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "shader_recompiler/frontend/ir/ir_emitter.h"
+#include "shader_recompiler/frontend/ir/opcodes.h"
+#include "shader_recompiler/frontend/ir/value.h"
+#include "shader_recompiler/ir_opt/passes.h"
+
+namespace Shader::Optimization {
+namespace {
+
+constexpr s32 F64ToF32Exp = +1023 - 127;
+constexpr s32 F32ToF64Exp = +127 - 1023;
+
+IR::F32 PackedF64ToF32(IR::IREmitter& ir, const IR::Value& packed) {
+    const IR::U32 lo{ir.CompositeExtract(packed, 0)};
+    const IR::U32 hi{ir.CompositeExtract(packed, 1)};
+    const IR::U32 sign{ir.BitFieldExtract(hi, ir.Imm32(31), ir.Imm32(1))};
+    const IR::U32 exp{ir.BitFieldExtract(hi, ir.Imm32(20), ir.Imm32(11))};
+    const IR::U32 mantissa_hi{ir.BitFieldExtract(hi, ir.Imm32(0), ir.Imm32(20))};
+    const IR::U32 mantissa_lo{ir.BitFieldExtract(lo, ir.Imm32(29), ir.Imm32(3))};
+    const IR::U32 mantissa{
+        ir.BitwiseOr(ir.ShiftLeftLogical(mantissa_hi, ir.Imm32(3)), mantissa_lo)};
+    const IR::U32 exp_if_subnorm{
+        ir.Select(ir.IEqual(exp, ir.Imm32(0)), ir.Imm32(0), ir.IAdd(exp, ir.Imm32(F64ToF32Exp)))};
+    const IR::U32 exp_if_infnan{
+        ir.Select(ir.IEqual(exp, ir.Imm32(0x7ff)), ir.Imm32(0xff), exp_if_subnorm)};
+    const IR::U32 result{
+        ir.BitwiseOr(ir.ShiftLeftLogical(sign, ir.Imm32(31)),
+                     ir.BitwiseOr(ir.ShiftLeftLogical(exp_if_infnan, ir.Imm32(23)), mantissa))};
+    return ir.BitCast<IR::F32>(result);
+}
+
+IR::Value F32ToPackedF64(IR::IREmitter& ir, const IR::Value& raw) {
+    const IR::U32 value{ir.BitCast<IR::U32>(IR::F32(raw))};
+    const IR::U32 sign{ir.BitFieldExtract(value, ir.Imm32(31), ir.Imm32(1))};
+    const IR::U32 exp{ir.BitFieldExtract(value, ir.Imm32(23), ir.Imm32(8))};
+    const IR::U32 mantissa{ir.BitFieldExtract(value, ir.Imm32(0), ir.Imm32(23))};
+    const IR::U32 mantissa_hi{ir.BitFieldExtract(mantissa, ir.Imm32(3), ir.Imm32(20))};
+    const IR::U32 mantissa_lo{ir.BitFieldExtract(mantissa, ir.Imm32(0), ir.Imm32(3))};
+    const IR::U32 exp_if_subnorm{
+        ir.Select(ir.IEqual(exp, ir.Imm32(0)), ir.Imm32(0), ir.IAdd(exp, ir.Imm32(F32ToF64Exp)))};
+    const IR::U32 exp_if_infnan{
+        ir.Select(ir.IEqual(exp, ir.Imm32(0xff)), ir.Imm32(0x7ff), exp_if_subnorm)};
+    const IR::U32 lo{ir.ShiftLeftLogical(mantissa_lo, ir.Imm32(29))};
+    const IR::U32 hi{
+        ir.BitwiseOr(ir.ShiftLeftLogical(sign, ir.Imm32(31)),
+                     ir.BitwiseOr(ir.ShiftLeftLogical(exp_if_infnan, ir.Imm32(20)), mantissa_hi))};
+    return ir.CompositeConstruct(lo, hi);
+}
+
+IR::Opcode Replace(IR::Opcode op) {
+    switch (op) {
+    case IR::Opcode::FPAbs64:
+        return IR::Opcode::FPAbs32;
+    case IR::Opcode::FPAdd64:
+        return IR::Opcode::FPAdd32;
+    case IR::Opcode::FPCeil64:
+        return IR::Opcode::FPCeil32;
+    case IR::Opcode::FPFloor64:
+        return IR::Opcode::FPFloor32;
+    case IR::Opcode::FPFma64:
+        return IR::Opcode::FPFma32;
+    case IR::Opcode::FPMul64:
+        return IR::Opcode::FPMul32;
+    case IR::Opcode::FPNeg64:
+        return IR::Opcode::FPNeg32;
+    case IR::Opcode::FPRoundEven64:
+        return IR::Opcode::FPRoundEven32;
+    case IR::Opcode::FPSaturate64:
+        return IR::Opcode::FPSaturate32;
+    case IR::Opcode::FPClamp64:
+        return IR::Opcode::FPClamp32;
+    case IR::Opcode::FPTrunc64:
+        return IR::Opcode::FPTrunc32;
+    case IR::Opcode::CompositeConstructF64x2:
+        return IR::Opcode::CompositeConstructF32x2;
+    case IR::Opcode::CompositeConstructF64x3:
+        return IR::Opcode::CompositeConstructF32x3;
+    case IR::Opcode::CompositeConstructF64x4:
+        return IR::Opcode::CompositeConstructF32x4;
+    case IR::Opcode::CompositeExtractF64x2:
+        return IR::Opcode::CompositeExtractF32x2;
+    case IR::Opcode::CompositeExtractF64x3:
+        return IR::Opcode::CompositeExtractF32x3;
+    case IR::Opcode::CompositeExtractF64x4:
+        return IR::Opcode::CompositeExtractF32x4;
+    case IR::Opcode::CompositeInsertF64x2:
+        return IR::Opcode::CompositeInsertF32x2;
+    case IR::Opcode::CompositeInsertF64x3:
+        return IR::Opcode::CompositeInsertF32x3;
+    case IR::Opcode::CompositeInsertF64x4:
+        return IR::Opcode::CompositeInsertF32x4;
+    case IR::Opcode::FPOrdEqual64:
+        return IR::Opcode::FPOrdEqual32;
+    case IR::Opcode::FPUnordEqual64:
+        return IR::Opcode::FPUnordEqual32;
+    case IR::Opcode::FPOrdNotEqual64:
+        return IR::Opcode::FPOrdNotEqual32;
+    case IR::Opcode::FPUnordNotEqual64:
+        return IR::Opcode::FPUnordNotEqual32;
+    case IR::Opcode::FPOrdLessThan64:
+        return IR::Opcode::FPOrdLessThan32;
+    case IR::Opcode::FPUnordLessThan64:
+        return IR::Opcode::FPUnordLessThan32;
+    case IR::Opcode::FPOrdGreaterThan64:
+        return IR::Opcode::FPOrdGreaterThan32;
+    case IR::Opcode::FPUnordGreaterThan64:
+        return IR::Opcode::FPUnordGreaterThan32;
+    case IR::Opcode::FPOrdLessThanEqual64:
+        return IR::Opcode::FPOrdLessThanEqual32;
+    case IR::Opcode::FPUnordLessThanEqual64:
+        return IR::Opcode::FPUnordLessThanEqual32;
+    case IR::Opcode::FPOrdGreaterThanEqual64:
+        return IR::Opcode::FPOrdGreaterThanEqual32;
+    case IR::Opcode::FPUnordGreaterThanEqual64:
+        return IR::Opcode::FPUnordGreaterThanEqual32;
+    case IR::Opcode::FPIsNan64:
+        return IR::Opcode::FPIsNan32;
+    case IR::Opcode::ConvertS16F64:
+        return IR::Opcode::ConvertS16F32;
+    case IR::Opcode::ConvertS32F64:
+        return IR::Opcode::ConvertS32F32;
+    case IR::Opcode::ConvertS64F64:
+        return IR::Opcode::ConvertS64F32;
+    case IR::Opcode::ConvertU16F64:
+        return IR::Opcode::ConvertU16F32;
+    case IR::Opcode::ConvertU32F64:
+        return IR::Opcode::ConvertU32F32;
+    case IR::Opcode::ConvertU64F64:
+        return IR::Opcode::ConvertU64F32;
+    case IR::Opcode::ConvertF32F64:
+        return IR::Opcode::Identity;
+    case IR::Opcode::ConvertF64F32:
+        return IR::Opcode::Identity;
+    case IR::Opcode::ConvertF64S8:
+        return IR::Opcode::ConvertF32S8;
+    case IR::Opcode::ConvertF64S16:
+        return IR::Opcode::ConvertF32S16;
+    case IR::Opcode::ConvertF64S32:
+        return IR::Opcode::ConvertF32S32;
+    case IR::Opcode::ConvertF64S64:
+        return IR::Opcode::ConvertF32S64;
+    case IR::Opcode::ConvertF64U8:
+        return IR::Opcode::ConvertF32U8;
+    case IR::Opcode::ConvertF64U16:
+        return IR::Opcode::ConvertF32U16;
+    case IR::Opcode::ConvertF64U32:
+        return IR::Opcode::ConvertF32U32;
+    case IR::Opcode::ConvertF64U64:
+        return IR::Opcode::ConvertF32U64;
+    default:
+        return op;
+    }
+}
+
+void Lower(IR::Block& block, IR::Inst& inst) {
+    switch (inst.GetOpcode()) {
+    case IR::Opcode::PackDouble2x32: {
+        IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
+        inst.ReplaceUsesWith(PackedF64ToF32(ir, inst.Arg(0)));
+        break;
+    }
+    case IR::Opcode::UnpackDouble2x32: {
+        IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
+        inst.ReplaceUsesWith(F32ToPackedF64(ir, inst.Arg(0)));
+        break;
+    }
+    default:
+        inst.ReplaceOpcode(Replace(inst.GetOpcode()));
+        break;
+    }
+}
+
+} // Anonymous namespace
+
+void LowerFp64ToFp32(IR::Program& program) {
+    for (IR::Block* const block : program.blocks) {
+        for (IR::Inst& inst : block->Instructions()) {
+            Lower(*block, inst);
+        }
+    }
+}
+
+} // namespace Shader::Optimization
diff --git a/src/shader_recompiler/ir_opt/passes.h b/src/shader_recompiler/ir_opt/passes.h
index 1f8f2ba95..629d18fa1 100644
--- a/src/shader_recompiler/ir_opt/passes.h
+++ b/src/shader_recompiler/ir_opt/passes.h
@@ -13,10 +13,12 @@ struct HostTranslateInfo;
 namespace Shader::Optimization {
 
 void CollectShaderInfoPass(Environment& env, IR::Program& program);
+void ConditionalBarrierPass(IR::Program& program);
 void ConstantPropagationPass(Environment& env, IR::Program& program);
 void DeadCodeEliminationPass(IR::Program& program);
 void GlobalMemoryToStorageBufferPass(IR::Program& program);
 void IdentityRemovalPass(IR::Program& program);
+void LowerFp64ToFp32(IR::Program& program);
 void LowerFp16ToFp32(IR::Program& program);
 void LowerInt64ToInt32(IR::Program& program);
 void RescalingPass(IR::Program& program);
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index 400c21981..03d234f2f 100644
--- a/src/video_core/renderer_opengl/gl_device.cpp
+++ b/src/video_core/renderer_opengl/gl_device.cpp
@@ -201,6 +201,7 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) {
     use_asynchronous_shaders = Settings::values.use_asynchronous_shaders.GetValue() &&
                                !(is_amd || (is_intel && !is_linux)) && !strict_context_required;
     use_driver_cache = is_nvidia;
+    supports_conditional_barriers = !is_intel;
 
     LOG_INFO(Render_OpenGL, "Renderer_VariableAOFFI: {}", has_variable_aoffi);
     LOG_INFO(Render_OpenGL, "Renderer_ComponentIndexingBug: {}", has_component_indexing_bug);
diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h
index cc0b95f1a..ad27264e5 100644
--- a/src/video_core/renderer_opengl/gl_device.h
+++ b/src/video_core/renderer_opengl/gl_device.h
@@ -188,6 +188,10 @@ public:
         return strict_context_required;
     }
 
+    bool SupportsConditionalBarriers() const {
+        return supports_conditional_barriers;
+    }
+
 private:
     static bool TestVariableAoffi();
     static bool TestPreciseBug();
@@ -233,6 +237,7 @@ private:
     bool has_bool_ref_bug{};
     bool can_report_memory{};
     bool strict_context_required{};
+    bool supports_conditional_barriers{};
 
     std::string vendor_name;
 };
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 6ecda2984..3f077311e 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -232,12 +232,14 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
           .gl_max_compute_smem_size = device.GetMaxComputeSharedMemorySize(),
       },
       host_info{
+          .support_float64 = true,
           .support_float16 = false,
           .support_int64 = device.HasShaderInt64(),
           .needs_demote_reorder = device.IsAmd(),
           .support_snorm_render_buffer = false,
           .support_viewport_index_layer = device.HasVertexViewportLayer(),
           .support_geometry_shader_passthrough = device.HasGeometryShaderPassthrough(),
+          .support_conditional_barrier = device.SupportsConditionalBarriers(),
       } {
     if (use_asynchronous_shaders) {
         workers = CreateWorkers();
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index 9482e91b0..5734f51e5 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -350,6 +350,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device
         .has_broken_spirv_subgroup_mask_vector_extract_dynamic =
             driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY};
     host_info = Shader::HostTranslateInfo{
+        .support_float64 = device.IsFloat64Supported(),
         .support_float16 = device.IsFloat16Supported(),
         .support_int64 = device.IsShaderInt64Supported(),
         .needs_demote_reorder =
diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp
index 0158b6b0d..a46f9beed 100644
--- a/src/video_core/vulkan_common/vulkan_device.cpp
+++ b/src/video_core/vulkan_common/vulkan_device.cpp
@@ -386,6 +386,8 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
         IsFormatSupported(VK_FORMAT_D24_UNORM_S8_UINT,
                           VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT, FormatType::Optimal);
 
+    supports_conditional_barriers = !(is_intel_anv || is_intel_windows);
+
     CollectPhysicalMemoryInfo();
     CollectToolingInfo();
 
diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h
index d62a103a1..f314d0ffe 100644
--- a/src/video_core/vulkan_common/vulkan_device.h
+++ b/src/video_core/vulkan_common/vulkan_device.h
@@ -300,6 +300,11 @@ public:
         return GetDriverID() != VK_DRIVER_ID_QUALCOMM_PROPRIETARY;
     }
 
+    /// Returns true if the device suppors float64 natively.
+    bool IsFloat64Supported() const {
+        return features.features.shaderFloat64;
+    }
+
     /// Returns true if the device supports float16 natively.
     bool IsFloat16Supported() const {
         return features.shader_float16_int8.shaderFloat16;
@@ -580,6 +585,10 @@ public:
         return properties.properties.limits.maxVertexInputBindings;
     }
 
+    bool SupportsConditionalBarriers() const {
+        return supports_conditional_barriers;
+    }
+
 private:
     /// Checks if the physical device is suitable and configures the object state
     /// with all necessary info about its properties.
@@ -683,6 +692,7 @@ private:
     bool must_emulate_bgr565{};             ///< Emulates BGR565 by swizzling RGB565 format.
     bool dynamic_state3_blending{};         ///< Has all blending features of dynamic_state3.
     bool dynamic_state3_enables{};          ///< Has all enables features of dynamic_state3.
+    bool supports_conditional_barriers{};   ///< Allows barriers in conditional control flow.
     u64 device_access_memory{};             ///< Total size of device local memory in bytes.
     u32 sets_per_pool{};                    ///< Sets per Description Pool
 
diff --git a/src/yuzu/main.cpp b/src/yuzu/main.cpp
index 9d06b21b6..013715b44 100644
--- a/src/yuzu/main.cpp
+++ b/src/yuzu/main.cpp
@@ -3067,7 +3067,7 @@ InstallResult GMainWindow::InstallNSPXCI(const QString& filename) {
             return false;
         }
 
-        std::array<u8, 0x1000> buffer{};
+        std::vector<u8> buffer(1_MiB);
 
         for (std::size_t i = 0; i < src->GetSize(); i += buffer.size()) {
             if (install_progress->wasCanceled()) {