fixed overflow

GPU: Implemented MUFU suboperation 8, sqrt.

src/core/hle/service/audio/audren_u.cpp

@@ -47,6 +47,7 @@ public:
 
         // Start the audio event
         CoreTiming::ScheduleEvent(audio_ticks, audio_event);
+        voice_status_list.reserve(worker_params.voice_count);
     }
     ~IAudioRenderer() {
         CoreTiming::UnscheduleEvent(audio_event, 0);
@@ -68,6 +69,12 @@ private:
                     buf.data() + sizeof(UpdateDataHeader) + config.behavior_size,
                     memory_pool_count * sizeof(MemoryPoolInfo));
 
+        std::vector<VoiceInfo> voice_info(worker_params.voice_count);
+        std::memcpy(voice_info.data(),
+                    buf.data() + sizeof(UpdateDataHeader) + config.behavior_size +
+                        config.memory_pools_size + config.voice_resource_size,
+                    worker_params.voice_count * sizeof(VoiceInfo));
+
         UpdateDataHeader response_data{worker_params};
 
         ASSERT(ctx.GetWriteBufferSize() == response_data.total_size);
@@ -86,6 +93,23 @@ private:
         std::memcpy(output.data() + sizeof(UpdateDataHeader), memory_pool.data(),
                     response_data.memory_pools_size);
 
+        for (unsigned i = 0; i < voice_info.size(); i++) {
+            if (voice_info[i].is_new) {
+                voice_status_list[i].played_sample_count = 0;
+                voice_status_list[i].wave_buffer_consumed = 0;
+            } else if (voice_info[i].play_state == (u8)PlayStates::Started) {
+                for (u32 buff_idx = 0; buff_idx < voice_info[i].wave_buffer_count; buff_idx++) {
+                    voice_status_list[i].played_sample_count +=
+                        (voice_info[i].wave_buffer[buff_idx].end_sample_offset -
+                         voice_info[i].wave_buffer[buff_idx].start_sample_offset) /
+                        2;
+                    voice_status_list[i].wave_buffer_consumed++;
+                }
+            }
+        }
+        std::memcpy(output.data() + sizeof(UpdateDataHeader) + response_data.memory_pools_size,
+                    voice_status_list.data(), response_data.voices_size);
+
         ctx.WriteBuffer(output);
 
         IPC::ResponseBuilder rb{ctx, 2};
@@ -130,6 +154,11 @@ private:
         Released = 0x6,
     };
 
+    enum class PlayStates : u8 {
+        Started = 0,
+        Stopped = 1,
+    };
+
     struct MemoryPoolEntry {
         MemoryPoolStates state;
         u32_le unknown_4;
@@ -175,11 +204,69 @@ private:
     };
     static_assert(sizeof(UpdateDataHeader) == 0x40, "UpdateDataHeader has wrong size");
 
+    struct BiquadFilter {
+        u8 enable;
+        INSERT_PADDING_BYTES(1);
+        s16_le numerator[3];
+        s16_le denominator[2];
+    };
+    static_assert(sizeof(BiquadFilter) == 0xc, "BiquadFilter has wrong size");
+
+    struct WaveBuffer {
+        u64_le buffer_addr;
+        u64_le buffer_sz;
+        s32_le start_sample_offset;
+        s32_le end_sample_offset;
+        u8 loop;
+        u8 end_of_stream;
+        u8 sent_to_server;
+        INSERT_PADDING_BYTES(5);
+        u64 context_addr;
+        u64 context_sz;
+        INSERT_PADDING_BYTES(8);
+    };
+    static_assert(sizeof(WaveBuffer) == 0x38, "WaveBuffer has wrong size");
+
+    struct VoiceInfo {
+        u32_le id;
+        u32_le node_id;
+        u8 is_new;
+        u8 is_in_use;
+        u8 play_state;
+        u8 sample_format;
+        u32_le sample_rate;
+        u32_le priority;
+        u32_le sorting_order;
+        u32_le channel_count;
+        float_le pitch;
+        float_le volume;
+        BiquadFilter biquad_filter[2];
+        u32_le wave_buffer_count;
+        u16_le wave_buffer_head;
+        INSERT_PADDING_BYTES(6);
+        u64_le additional_params_addr;
+        u64_le additional_params_sz;
+        u32_le mix_id;
+        u32_le splitter_info_id;
+        WaveBuffer wave_buffer[4];
+        u32_le voice_channel_resource_ids[6];
+        INSERT_PADDING_BYTES(24);
+    };
+    static_assert(sizeof(VoiceInfo) == 0x170, "VoiceInfo is wrong size");
+
+    struct VoiceOutStatus {
+        u64_le played_sample_count;
+        u32_le wave_buffer_consumed;
+        INSERT_PADDING_WORDS(1);
+    };
+    static_assert(sizeof(VoiceOutStatus) == 0x10, "VoiceOutStatus has wrong size");
+
     /// This is used to trigger the audio event callback.
     CoreTiming::EventType* audio_event;
 
     Kernel::SharedPtr<Kernel::Event> system_event;
     AudioRendererParameter worker_params;
+    std::vector<VoiceOutStatus> voice_status_list;
 };
 
 class IAudioDevice final : public ServiceFramework<IAudioDevice> {

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp

@@ -8,6 +8,8 @@
 #include "core/core.h"
 #include "core/hle/service/nvdrv/devices/nvhost_as_gpu.h"
 #include "core/hle/service/nvdrv/devices/nvmap.h"
+#include "video_core/renderer_base.h"
+#include "video_core/video_core.h"
 
 namespace Service::Nvidia::Devices {
 
@@ -154,6 +156,9 @@ u32 nvhost_as_gpu::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& ou
 
     ASSERT_MSG(itr != buffer_mappings.end(), "Tried to unmap invalid mapping");
 
+    // Remove this memory region from the rasterizer cache.
+    VideoCore::g_renderer->Rasterizer()->FlushAndInvalidateRegion(params.offset, itr->second.size);
+
     params.offset = gpu.memory_manager->UnmapBuffer(params.offset, itr->second.size);
 
     buffer_mappings.erase(itr->second.offset);

src/core/hle/service/nvdrv/devices/nvmap.cpp

@@ -148,6 +148,7 @@ u32 nvmap::IocParam(const std::vector<u8>& input, std::vector<u8>& output) {
 }
 
 u32 nvmap::IocFree(const std::vector<u8>& input, std::vector<u8>& output) {
+    // TODO(Subv): These flags are unconfirmed.
     enum FreeFlags {
         Freed = 0,
         NotFreedYet = 1,
@@ -161,15 +162,21 @@ u32 nvmap::IocFree(const std::vector<u8>& input, std::vector<u8>& output) {
     auto itr = handles.find(params.handle);
     ASSERT(itr != handles.end());
 
+    ASSERT(itr->second->refcount > 0);
+
     itr->second->refcount--;
 
-    params.refcount = itr->second->refcount;
     params.size = itr->second->size;
 
-    if (itr->second->refcount == 0)
+    if (itr->second->refcount == 0) {
         params.flags = Freed;
-    else
+        // The address of the nvmap is written to the output if we're finally freeing it, otherwise
+        // 0 is written.
+        params.address = itr->second->addr;
+    } else {
         params.flags = NotFreedYet;
+        params.address = 0;
+    }
 
     handles.erase(params.handle);
 

src/core/hle/service/nvdrv/devices/nvmap.h

@@ -94,7 +94,7 @@ private:
     struct IocFreeParams {
         u32_le handle;
         INSERT_PADDING_BYTES(4);
-        u64_le refcount;
+        u64_le address;
         u32_le size;
         u32_le flags;
     };

src/video_core/engines/maxwell_3d.h

@@ -280,6 +280,34 @@ public:
             UnsignedInt = 0x2,
         };
 
+        enum class ComparisonOp : u32 {
+            Never = 0x200,
+            Less = 0x201,
+            Equal = 0x202,
+            LessEqual = 0x203,
+            Greater = 0x204,
+            NotEqual = 0x205,
+            GreaterEqual = 0x206,
+            Always = 0x207,
+        };
+
+        struct Cull {
+            enum class FrontFace : u32 {
+                ClockWise = 0x0900,
+                CounterClockWise = 0x0901,
+            };
+
+            enum class CullFace : u32 {
+                Front = 0x0404,
+                Back = 0x0405,
+                FrontAndBack = 0x0408,
+            };
+
+            u32 enabled;
+            FrontFace front_face;
+            CullFace cull_face;
+        };
+
         struct Blend {
             enum class Equation : u32 {
                 Add = 1,
@@ -413,7 +441,7 @@ public:
                 struct {
                     u32 address_high;
                     u32 address_low;
-                    u32 format;
+                    Tegra::DepthFormat format;
                     u32 block_dimensions;
                     u32 layer_stride;
 
@@ -435,11 +463,21 @@ public:
                     };
                 } rt_control;
 
-                INSERT_PADDING_WORDS(0x31);
+                INSERT_PADDING_WORDS(0x2B);
+
+                u32 depth_test_enable;
+
+                INSERT_PADDING_WORDS(0x5);
 
                 u32 independent_blend_enable;
 
-                INSERT_PADDING_WORDS(0x15);
+                u32 depth_write_enabled;
+
+                INSERT_PADDING_WORDS(0x8);
+
+                ComparisonOp depth_test_func;
+
+                INSERT_PADDING_WORDS(0xB);
 
                 struct {
                     u32 separate_alpha;
@@ -455,7 +493,11 @@ public:
                     u32 enable[NumRenderTargets];
                 } blend;
 
-                INSERT_PADDING_WORDS(0x77);
+                INSERT_PADDING_WORDS(0x2D);
+
+                u32 vb_element_base;
+
+                INSERT_PADDING_WORDS(0x49);
 
                 struct {
                     u32 tsc_address_high;
@@ -536,7 +578,13 @@ public:
                     }
                 } index_array;
 
-                INSERT_PADDING_WORDS(0xC7);
+                INSERT_PADDING_WORDS(0x7);
+
+                INSERT_PADDING_WORDS(0x46);
+
+                Cull cull;
+
+                INSERT_PADDING_WORDS(0x77);
 
                 struct {
                     u32 query_address_high;
@@ -743,13 +791,18 @@ ASSERT_REG_POSITION(vertex_buffer, 0x35D);
 ASSERT_REG_POSITION(zeta, 0x3F8);
 ASSERT_REG_POSITION(vertex_attrib_format[0], 0x458);
 ASSERT_REG_POSITION(rt_control, 0x487);
+ASSERT_REG_POSITION(depth_test_enable, 0x4B3);
 ASSERT_REG_POSITION(independent_blend_enable, 0x4B9);
+ASSERT_REG_POSITION(depth_write_enabled, 0x4BA);
+ASSERT_REG_POSITION(depth_test_func, 0x4C3);
 ASSERT_REG_POSITION(blend, 0x4CF);
+ASSERT_REG_POSITION(vb_element_base, 0x50D);
 ASSERT_REG_POSITION(tsc, 0x557);
 ASSERT_REG_POSITION(tic, 0x55D);
 ASSERT_REG_POSITION(code_address, 0x582);
 ASSERT_REG_POSITION(draw, 0x585);
 ASSERT_REG_POSITION(index_array, 0x5F2);
+ASSERT_REG_POSITION(cull, 0x646);
 ASSERT_REG_POSITION(query, 0x6C0);
 ASSERT_REG_POSITION(vertex_array[0], 0x700);
 ASSERT_REG_POSITION(independent_blend, 0x780);

src/video_core/engines/maxwell_dma.cpp

@@ -49,7 +49,11 @@ void MaxwellDMA::HandleCopy() {
     ASSERT(regs.src_params.pos_y == 0);
     ASSERT(regs.dst_params.pos_x == 0);
     ASSERT(regs.dst_params.pos_y == 0);
-    ASSERT(regs.exec.is_dst_linear != regs.exec.is_src_linear);
+
+    if (regs.exec.is_dst_linear == regs.exec.is_src_linear) {
+        Memory::CopyBlock(dest_cpu, source_cpu, regs.x_count * regs.y_count);
+        return;
+    }
 
     u8* src_buffer = Memory::GetPointer(source_cpu);
     u8* dst_buffer = Memory::GetPointer(dest_cpu);

src/video_core/engines/shader_bytecode.h

@@ -142,6 +142,7 @@ enum class PredCondition : u64 {
     GreaterThan = 4,
     NotEqual = 5,
     GreaterEqual = 6,
+    NotEqualWithNan = 13,
     // TODO(Subv): Other condition types
 };
 
@@ -165,7 +166,7 @@ enum class SubOp : u64 {
     Lg2 = 0x3,
     Rcp = 0x4,
     Rsq = 0x5,
-    Min = 0x8,
+    Sqrt = 0x8,
 };
 
 enum class F2iRoundingOp : u64 {
@@ -209,7 +210,7 @@ union Instruction {
     } pred;
     BitField<19, 1, u64> negate_pred;
     BitField<20, 8, Register> gpr20;
-    BitField<20, 7, SubOp> sub_op;
+    BitField<20, 4, SubOp> sub_op;
     BitField<28, 8, Register> gpr28;
     BitField<39, 8, Register> gpr39;
     BitField<48, 16, u64> opcode;

src/video_core/gpu.h

@@ -16,6 +16,7 @@ namespace Tegra {
 enum class RenderTargetFormat : u32 {
     NONE = 0x0,
     RGBA32_FLOAT = 0xC0,
+    RGBA32_UINT = 0xC2,
     RGBA16_FLOAT = 0xCA,
     RGB10_A2_UNORM = 0xD1,
     RGBA8_UNORM = 0xD5,
@@ -23,6 +24,15 @@ enum class RenderTargetFormat : u32 {
     R11G11B10_FLOAT = 0xE0,
 };
 
+enum class DepthFormat : u32 {
+    Z32_FLOAT = 0xA,
+    Z16_UNORM = 0x13,
+    S8_Z24_UNORM = 0x14,
+    Z24_X8_UNORM = 0x15,
+    Z24_S8_UNORM = 0x16,
+    Z24_C8_UNORM = 0x18,
+};
+
 /// Returns the number of bytes per pixel of each rendertarget format.
 u32 RenderTargetBytesPerPixel(RenderTargetFormat format);
 

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -304,10 +304,15 @@ void RasterizerOpenGL::DrawArrays() {
     MICROPROFILE_SCOPE(OpenGL_Drawing);
     const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs;
 
-    // TODO(bunnei): Implement these
+    // Sync the depth test state before configuring the framebuffer surfaces.
+    SyncDepthTestState();
+
+    // TODO(bunnei): Implement this
     const bool has_stencil = false;
+
     const bool using_color_fb = true;
-    const bool using_depth_fb = false;
+    const bool using_depth_fb = regs.zeta.Address() != 0;
+
     const MathUtil::Rectangle<s32> viewport_rect{regs.viewport_transform[0].GetRect()};
 
     const bool write_color_fb =
@@ -338,11 +343,9 @@ void RasterizerOpenGL::DrawArrays() {
     // Bind the framebuffer surfaces
     BindFramebufferSurfaces(color_surface, depth_surface, has_stencil);
 
-    // Sync the viewport
     SyncViewport(surfaces_rect);
-
-    // Sync the blend state registers
     SyncBlendState();
+    SyncCullMode();
 
     // TODO(bunnei): Sync framebuffer_scale uniform here
     // TODO(bunnei): Sync scissorbox uniform(s) here
@@ -412,14 +415,16 @@ void RasterizerOpenGL::DrawArrays() {
 
     const GLenum primitive_mode{MaxwellToGL::PrimitiveTopology(regs.draw.topology)};
     if (is_indexed) {
-        const GLint index_min{static_cast<GLint>(regs.index_array.first)};
-        const GLint index_max{static_cast<GLint>(regs.index_array.first + regs.index_array.count)};
-        glDrawRangeElementsBaseVertex(primitive_mode, index_min, index_max, regs.index_array.count,
-                                      MaxwellToGL::IndexFormat(regs.index_array.format),
-                                      reinterpret_cast<const void*>(index_buffer_offset),
-                                      -index_min);
+        const GLint base_vertex{static_cast<GLint>(regs.vb_element_base)};
+
+        // Adjust the index buffer offset so it points to the first desired index.
+        index_buffer_offset += regs.index_array.first * regs.index_array.FormatSizeInBytes();
+
+        glDrawElementsBaseVertex(primitive_mode, regs.index_array.count,
+                                 MaxwellToGL::IndexFormat(regs.index_array.format),
+                                 reinterpret_cast<const void*>(index_buffer_offset), base_vertex);
     } else {
-        glDrawArrays(primitive_mode, 0, regs.vertex_buffer.count);
+        glDrawArrays(primitive_mode, regs.vertex_buffer.first, regs.vertex_buffer.count);
     }
 
     // Disable scissor test
@@ -710,7 +715,11 @@ void RasterizerOpenGL::SyncClipCoef() {
 }
 
 void RasterizerOpenGL::SyncCullMode() {
-    UNREACHABLE();
+    const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs;
+
+    state.cull.enabled = regs.cull.enabled != 0;
+    state.cull.front_face = MaxwellToGL::FrontFace(regs.cull.front_face);
+    state.cull.mode = MaxwellToGL::CullFace(regs.cull.cull_face);
 }
 
 void RasterizerOpenGL::SyncDepthScale() {
@@ -721,6 +730,14 @@ void RasterizerOpenGL::SyncDepthOffset() {
     UNREACHABLE();
 }
 
+void RasterizerOpenGL::SyncDepthTestState() {
+    const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs;
+
+    state.depth.test_enabled = regs.depth_test_enable != 0;
+    state.depth.write_mask = regs.depth_write_enabled ? GL_TRUE : GL_FALSE;
+    state.depth.test_func = MaxwellToGL::ComparisonOp(regs.depth_test_func);
+}
+
 void RasterizerOpenGL::SyncBlendState() {
     const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs;
 

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -126,6 +126,9 @@ private:
     /// Syncs the depth offset to match the guest state
     void SyncDepthOffset();
 
+    /// Syncs the depth test state to match the guest state
+    void SyncDepthTestState();
+
     /// Syncs the blend state to match the guest state
     void SyncBlendState();
 

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -27,6 +27,7 @@ struct FormatTuple {
     GLint internal_format;
     GLenum format;
     GLenum type;
+    ComponentType component_type;
     bool compressed;
 };
 
@@ -65,36 +66,36 @@ struct FormatTuple {
 }
 
 static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_format_tuples = {{
-    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, false},                    // ABGR8
-    {GL_RGB, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV, false},                       // B5G6R5
-    {GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, false},              // A2B10G10R10
-    {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV, false},                // A1B5G5R5
-    {GL_R8, GL_RED, GL_UNSIGNED_BYTE, false},                                   // R8
-    {GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT, false},                                // RGBA16F
-    {GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV, false},        // R11FG11FB10F
-    {GL_COMPRESSED_RGB_S3TC_DXT1_EXT, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, true},   // DXT1
-    {GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true}, // DXT23
-    {GL_COMPRESSED_RGBA_S3TC_DXT5_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true}, // DXT45
-    {GL_COMPRESSED_RED_RGTC1, GL_RED, GL_UNSIGNED_INT_8_8_8_8, true},           // DXN1
-    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, false},                               // ASTC_2D_4X4
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, ComponentType::UNorm, false}, // ABGR8
+    {GL_RGB, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV, ComponentType::UNorm, false},    // B5G6R5
+    {GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, ComponentType::UNorm,
+     false}, // A2B10G10R10
+    {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV, ComponentType::UNorm, false}, // A1B5G5R5
+    {GL_R8, GL_RED, GL_UNSIGNED_BYTE, ComponentType::UNorm, false},                    // R8
+    {GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT, ComponentType::Float, false},                 // RGBA16F
+    {GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV, ComponentType::Float,
+     false},                                                                     // R11FG11FB10F
+    {GL_RGBA32UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT, ComponentType::UInt, false}, // RGBA32UI
+    {GL_COMPRESSED_RGB_S3TC_DXT1_EXT, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm,
+     true}, // DXT1
+    {GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm,
+     true}, // DXT23
+    {GL_COMPRESSED_RGBA_S3TC_DXT5_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm,
+     true},                                                                                 // DXT45
+    {GL_COMPRESSED_RED_RGTC1, GL_RED, GL_UNSIGNED_INT_8_8_8_8, ComponentType::UNorm, true}, // DXN1
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_4X4
+
+    // DepthStencil formats
+    {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8, ComponentType::UNorm,
+     false}, // Z24S8
 }};
 
 static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType component_type) {
-    const SurfaceType type = SurfaceParams::GetFormatType(pixel_format);
-    if (type == SurfaceType::ColorTexture) {
-        ASSERT(static_cast<size_t>(pixel_format) < tex_format_tuples.size());
-        // For now only UNORM components are supported, or either R11FG11FB10F or RGBA16F which
-        // are type FLOAT
-        ASSERT(component_type == ComponentType::UNorm || pixel_format == PixelFormat::RGBA16F ||
-               pixel_format == PixelFormat::R11FG11FB10F);
-        return tex_format_tuples[static_cast<unsigned int>(pixel_format)];
-    } else if (type == SurfaceType::Depth || type == SurfaceType::DepthStencil) {
-        // TODO(Subv): Implement depth formats
-        ASSERT_MSG(false, "Unimplemented");
-    }
+    ASSERT(static_cast<size_t>(pixel_format) < tex_format_tuples.size());
+    auto& format = tex_format_tuples[static_cast<unsigned int>(pixel_format)];
+    ASSERT(component_type == format.component_type);
 
-    UNREACHABLE();
-    return {};
+    return format;
 }
 
 VAddr SurfaceParams::GetCpuAddr() const {
@@ -144,11 +145,17 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, Tegra::
     const auto& gpu = Core::System::GetInstance().GPU();
 
     if (morton_to_gl) {
-        auto data = Tegra::Texture::UnswizzleTexture(
-            *gpu.memory_manager->GpuToCpuAddress(addr),
-            SurfaceParams::TextureFormatFromPixelFormat(format), stride, height, block_height);
-
-        std::memcpy(gl_buffer, data.data(), data.size());
+        if (SurfaceParams::GetFormatType(format) == SurfaceType::ColorTexture) {
+            auto data = Tegra::Texture::UnswizzleTexture(
+                *gpu.memory_manager->GpuToCpuAddress(addr),
+                SurfaceParams::TextureFormatFromPixelFormat(format), stride, height, block_height);
+            std::memcpy(gl_buffer, data.data(), data.size());
+        } else {
+            auto data = Tegra::Texture::UnswizzleDepthTexture(
+                *gpu.memory_manager->GpuToCpuAddress(addr),
+                SurfaceParams::DepthFormatFromPixelFormat(format), stride, height, block_height);
+            std::memcpy(gl_buffer, data.data(), data.size());
+        }
     } else {
         // TODO(bunnei): Assumes the default rendering GOB size of 16 (128 lines). We should
         // check the configuration for this and perform more generic un/swizzle
@@ -166,9 +173,10 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, Tegra::GPUVAddr),
         MortonCopy<true, PixelFormat::ABGR8>,        MortonCopy<true, PixelFormat::B5G6R5>,
         MortonCopy<true, PixelFormat::A2B10G10R10>,  MortonCopy<true, PixelFormat::A1B5G5R5>,
         MortonCopy<true, PixelFormat::R8>,           MortonCopy<true, PixelFormat::RGBA16F>,
-        MortonCopy<true, PixelFormat::R11FG11FB10F>, MortonCopy<true, PixelFormat::DXT1>,
-        MortonCopy<true, PixelFormat::DXT23>,        MortonCopy<true, PixelFormat::DXT45>,
-        MortonCopy<true, PixelFormat::DXN1>,         MortonCopy<true, PixelFormat::ASTC_2D_4X4>,
+        MortonCopy<true, PixelFormat::R11FG11FB10F>, MortonCopy<true, PixelFormat::RGBA32UI>,
+        MortonCopy<true, PixelFormat::DXT1>,         MortonCopy<true, PixelFormat::DXT23>,
+        MortonCopy<true, PixelFormat::DXT45>,        MortonCopy<true, PixelFormat::DXN1>,
+        MortonCopy<true, PixelFormat::ASTC_2D_4X4>,  MortonCopy<true, PixelFormat::Z24S8>,
 };
 
 static constexpr std::array<void (*)(u32, u32, u32, u8*, Tegra::GPUVAddr),
@@ -181,12 +189,14 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, Tegra::GPUVAddr),
         MortonCopy<false, PixelFormat::R8>,
         MortonCopy<false, PixelFormat::RGBA16F>,
         MortonCopy<false, PixelFormat::R11FG11FB10F>,
+        MortonCopy<false, PixelFormat::RGBA32UI>,
         // TODO(Subv): Swizzling the DXT1/DXT23/DXT45/DXN1 formats is not yet supported
         nullptr,
         nullptr,
         nullptr,
         nullptr,
         MortonCopy<false, PixelFormat::ABGR8>,
+        MortonCopy<false, PixelFormat::Z24S8>,
 };
 
 // Allocate an uninitialized texture of appropriate size and format for the surface
@@ -390,9 +400,15 @@ SurfaceSurfaceRect_Tuple RasterizerCacheOpenGL::GetFramebufferSurfaces(
 
     // get color and depth surfaces
     const SurfaceParams color_params{SurfaceParams::CreateForFramebuffer(regs.rt[0])};
-    const SurfaceParams depth_params{color_params};
+    SurfaceParams depth_params{color_params};
 
-    ASSERT_MSG(!using_depth_fb, "depth buffer is unimplemented");
+    if (using_depth_fb) {
+        depth_params.addr = regs.zeta.Address();
+        depth_params.pixel_format = SurfaceParams::PixelFormatFromDepthFormat(regs.zeta.format);
+        depth_params.component_type = SurfaceParams::ComponentTypeFromDepthFormat(regs.zeta.format);
+        depth_params.type = SurfaceParams::GetFormatType(depth_params.pixel_format);
+        depth_params.size_in_bytes = depth_params.SizeInBytes();
+    }
 
     MathUtil::Rectangle<u32> color_rect{};
     Surface color_surface;

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -30,13 +30,21 @@ struct SurfaceParams {
         R8 = 4,
         RGBA16F = 5,
         R11FG11FB10F = 6,
-        DXT1 = 7,
-        DXT23 = 8,
-        DXT45 = 9,
-        DXN1 = 10, // This is also known as BC4
-        ASTC_2D_4X4 = 11,
+        RGBA32UI = 7,
+        DXT1 = 8,
+        DXT23 = 9,
+        DXT45 = 10,
+        DXN1 = 11, // This is also known as BC4
+        ASTC_2D_4X4 = 12,
 
-        Max,
+        MaxColorFormat,
+
+        // DepthStencil formats
+        Z24S8 = 13,
+
+        MaxDepthStencilFormat,
+
+        Max = MaxDepthStencilFormat,
         Invalid = 255,
     };
 
@@ -77,11 +85,13 @@ struct SurfaceParams {
             1, // R8
             1, // RGBA16F
             1, // R11FG11FB10F
+            1, // RGBA32UI
             4, // DXT1
             4, // DXT23
             4, // DXT45
             4, // DXN1
             4, // ASTC_2D_4X4
+            1, // Z24S8
         }};
 
         ASSERT(static_cast<size_t>(format) < compression_factor_table.size());
@@ -100,11 +110,13 @@ struct SurfaceParams {
             8,   // R8
             64,  // RGBA16F
             32,  // R11FG11FB10F
+            128, // RGBA32UI
             64,  // DXT1
             128, // DXT23
             128, // DXT45
             64,  // DXN1
             32,  // ASTC_2D_4X4
+            32,  // Z24S8
         }};
 
         ASSERT(static_cast<size_t>(format) < bpp_table.size());
@@ -114,6 +126,16 @@ struct SurfaceParams {
         return GetFormatBpp(pixel_format);
     }
 
+    static PixelFormat PixelFormatFromDepthFormat(Tegra::DepthFormat format) {
+        switch (format) {
+        case Tegra::DepthFormat::Z24_S8_UNORM:
+            return PixelFormat::Z24S8;
+        default:
+            NGLOG_CRITICAL(HW_GPU, "Unimplemented format={}", static_cast<u32>(format));
+            UNREACHABLE();
+        }
+    }
+
     static PixelFormat PixelFormatFromRenderTargetFormat(Tegra::RenderTargetFormat format) {
         switch (format) {
         case Tegra::RenderTargetFormat::RGBA8_UNORM:
@@ -125,6 +147,8 @@ struct SurfaceParams {
             return PixelFormat::RGBA16F;
         case Tegra::RenderTargetFormat::R11G11B10_FLOAT:
             return PixelFormat::R11FG11FB10F;
+        case Tegra::RenderTargetFormat::RGBA32_UINT:
+            return PixelFormat::RGBA32UI;
         default:
             NGLOG_CRITICAL(HW_GPU, "Unimplemented format={}", static_cast<u32>(format));
             UNREACHABLE();
@@ -148,6 +172,8 @@ struct SurfaceParams {
             return PixelFormat::RGBA16F;
         case Tegra::Texture::TextureFormat::BF10GF11RF11:
             return PixelFormat::R11FG11FB10F;
+        case Tegra::Texture::TextureFormat::R32_G32_B32_A32:
+            return PixelFormat::RGBA32UI;
         case Tegra::Texture::TextureFormat::DXT1:
             return PixelFormat::DXT1;
         case Tegra::Texture::TextureFormat::DXT23:
@@ -181,6 +207,8 @@ struct SurfaceParams {
             return Tegra::Texture::TextureFormat::R16_G16_B16_A16;
         case PixelFormat::R11FG11FB10F:
             return Tegra::Texture::TextureFormat::BF10GF11RF11;
+        case PixelFormat::RGBA32UI:
+            return Tegra::Texture::TextureFormat::R32_G32_B32_A32;
         case PixelFormat::DXT1:
             return Tegra::Texture::TextureFormat::DXT1;
         case PixelFormat::DXT23:
@@ -196,6 +224,15 @@ struct SurfaceParams {
         }
     }
 
+    static Tegra::DepthFormat DepthFormatFromPixelFormat(PixelFormat format) {
+        switch (format) {
+        case PixelFormat::Z24S8:
+            return Tegra::DepthFormat::Z24_S8_UNORM;
+        default:
+            UNREACHABLE();
+        }
+    }
+
     static ComponentType ComponentTypeFromTexture(Tegra::Texture::ComponentType type) {
         // TODO(Subv): Implement more component types
         switch (type) {
@@ -217,6 +254,8 @@ struct SurfaceParams {
         case Tegra::RenderTargetFormat::RGBA16_FLOAT:
         case Tegra::RenderTargetFormat::R11G11B10_FLOAT:
             return ComponentType::Float;
+        case Tegra::RenderTargetFormat::RGBA32_UINT:
+            return ComponentType::UInt;
         default:
             NGLOG_CRITICAL(HW_GPU, "Unimplemented format={}", static_cast<u32>(format));
             UNREACHABLE();
@@ -233,11 +272,26 @@ struct SurfaceParams {
         }
     }
 
+    static ComponentType ComponentTypeFromDepthFormat(Tegra::DepthFormat format) {
+        switch (format) {
+        case Tegra::DepthFormat::Z24_S8_UNORM:
+            return ComponentType::UNorm;
+        default:
+            NGLOG_CRITICAL(HW_GPU, "Unimplemented format={}", static_cast<u32>(format));
+            UNREACHABLE();
+        }
+    }
+
     static SurfaceType GetFormatType(PixelFormat pixel_format) {
-        if (static_cast<size_t>(pixel_format) < MaxPixelFormat) {
+        if (static_cast<size_t>(pixel_format) < static_cast<size_t>(PixelFormat::MaxColorFormat)) {
             return SurfaceType::ColorTexture;
         }
 
+        if (static_cast<size_t>(pixel_format) <
+            static_cast<size_t>(PixelFormat::MaxDepthStencilFormat)) {
+            return SurfaceType::DepthStencil;
+        }
+
         // TODO(Subv): Implement the other formats
         ASSERT(false);
 

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -719,21 +719,31 @@ private:
     /**
      * Returns the comparison string to use to compare two values in the 'set' family of
      * instructions.
-     * @params condition The condition used in the 'set'-family instruction.
+     * @param condition The condition used in the 'set'-family instruction.
+     * @param op_a First operand to use for the comparison.
+     * @param op_b Second operand to use for the comparison.
      * @returns String corresponding to the GLSL operator that matches the desired comparison.
      */
-    std::string GetPredicateComparison(Tegra::Shader::PredCondition condition) const {
+    std::string GetPredicateComparison(Tegra::Shader::PredCondition condition,
+                                       const std::string& op_a, const std::string& op_b) const {
         using Tegra::Shader::PredCondition;
         static const std::unordered_map<PredCondition, const char*> PredicateComparisonStrings = {
-            {PredCondition::LessThan, "<"},   {PredCondition::Equal, "=="},
-            {PredCondition::LessEqual, "<="}, {PredCondition::GreaterThan, ">"},
-            {PredCondition::NotEqual, "!="},  {PredCondition::GreaterEqual, ">="},
+            {PredCondition::LessThan, "<"},         {PredCondition::Equal, "=="},
+            {PredCondition::LessEqual, "<="},       {PredCondition::GreaterThan, ">"},
+            {PredCondition::NotEqual, "!="},        {PredCondition::GreaterEqual, ">="},
+            {PredCondition::NotEqualWithNan, "!="},
         };
 
-        auto comparison = PredicateComparisonStrings.find(condition);
+        const auto& comparison{PredicateComparisonStrings.find(condition)};
         ASSERT_MSG(comparison != PredicateComparisonStrings.end(),
                    "Unknown predicate comparison operation");
-        return comparison->second;
+
+        std::string predicate{'(' + op_a + ") " + comparison->second + " (" + op_b + ')'};
+        if (condition == PredCondition::NotEqualWithNan) {
+            predicate += " || isnan(" + op_a + ") || isnan(" + op_b + ')';
+        }
+
+        return predicate;
     }
 
     /**
@@ -907,8 +917,8 @@ private:
                     regs.SetRegisterToFloat(instr.gpr0, 0, "inversesqrt(" + op_a + ')', 1, 1,
                                             instr.alu.saturate_d);
                     break;
-                case SubOp::Min:
-                    regs.SetRegisterToFloat(instr.gpr0, 0, "min(" + op_a + "," + op_b + ')', 1, 1,
+                case SubOp::Sqrt:
+                    regs.SetRegisterToFloat(instr.gpr0, 0, "sqrt(" + op_a + ')', 1, 1,
                                             instr.alu.saturate_d);
                     break;
                 default:
@@ -1415,10 +1425,9 @@ private:
             std::string second_pred =
                 GetPredicateCondition(instr.fsetp.pred39, instr.fsetp.neg_pred != 0);
 
-            std::string comparator = GetPredicateComparison(instr.fsetp.cond);
             std::string combiner = GetPredicateCombiner(instr.fsetp.op);
 
-            std::string predicate = '(' + op_a + ") " + comparator + " (" + op_b + ')';
+            std::string predicate = GetPredicateComparison(instr.fsetp.cond, op_a, op_b);
             // Set the primary predicate to the result of Predicate OP SecondPredicate
             SetPredicate(instr.fsetp.pred3,
                          '(' + predicate + ") " + combiner + " (" + second_pred + ')');
@@ -1453,10 +1462,9 @@ private:
             std::string second_pred =
                 GetPredicateCondition(instr.isetp.pred39, instr.isetp.neg_pred != 0);
 
-            std::string comparator = GetPredicateComparison(instr.isetp.cond);
             std::string combiner = GetPredicateCombiner(instr.isetp.op);
 
-            std::string predicate = '(' + op_a + ") " + comparator + " (" + op_b + ')';
+            std::string predicate = GetPredicateComparison(instr.isetp.cond, op_a, op_b);
             // Set the primary predicate to the result of Predicate OP SecondPredicate
             SetPredicate(instr.isetp.pred3,
                          '(' + predicate + ") " + combiner + " (" + second_pred + ')');
@@ -1503,11 +1511,10 @@ private:
             std::string second_pred =
                 GetPredicateCondition(instr.fset.pred39, instr.fset.neg_pred != 0);
 
-            std::string comparator = GetPredicateComparison(instr.fset.cond);
             std::string combiner = GetPredicateCombiner(instr.fset.op);
 
-            std::string predicate = "(((" + op_a + ") " + comparator + " (" + op_b + ")) " +
-                                    combiner + " (" + second_pred + "))";
+            std::string predicate = "((" + GetPredicateComparison(instr.fset.cond, op_a, op_b) +
+                                    ") " + combiner + " (" + second_pred + "))";
 
             if (instr.fset.bf) {
                 regs.SetRegisterToFloat(instr.gpr0, 0, predicate + " ? 1.0 : 0.0", 1, 1);
@@ -1538,11 +1545,10 @@ private:
             std::string second_pred =
                 GetPredicateCondition(instr.iset.pred39, instr.iset.neg_pred != 0);
 
-            std::string comparator = GetPredicateComparison(instr.iset.cond);
             std::string combiner = GetPredicateCombiner(instr.iset.op);
 
-            std::string predicate = "(((" + op_a + ") " + comparator + " (" + op_b + ")) " +
-                                    combiner + " (" + second_pred + "))";
+            std::string predicate = "((" + GetPredicateComparison(instr.iset.cond, op_a, op_b) +
+                                    ") " + combiner + " (" + second_pred + "))";
 
             if (instr.iset.bf) {
                 regs.SetRegisterToFloat(instr.gpr0, 0, predicate + " ? 1.0 : 0.0", 1, 1);

src/video_core/renderer_opengl/maxwell_to_gl.h

@@ -201,4 +201,54 @@ inline GLenum SwizzleSource(Tegra::Texture::SwizzleSource source) {
     return {};
 }
 
+inline GLenum ComparisonOp(Maxwell::ComparisonOp comparison) {
+    switch (comparison) {
+    case Maxwell::ComparisonOp::Never:
+        return GL_NEVER;
+    case Maxwell::ComparisonOp::Less:
+        return GL_LESS;
+    case Maxwell::ComparisonOp::Equal:
+        return GL_EQUAL;
+    case Maxwell::ComparisonOp::LessEqual:
+        return GL_LEQUAL;
+    case Maxwell::ComparisonOp::Greater:
+        return GL_GREATER;
+    case Maxwell::ComparisonOp::NotEqual:
+        return GL_NOTEQUAL;
+    case Maxwell::ComparisonOp::GreaterEqual:
+        return GL_GEQUAL;
+    case Maxwell::ComparisonOp::Always:
+        return GL_ALWAYS;
+    }
+    NGLOG_CRITICAL(Render_OpenGL, "Unimplemented comparison op={}", static_cast<u32>(comparison));
+    UNREACHABLE();
+    return {};
+}
+
+inline GLenum FrontFace(Maxwell::Cull::FrontFace front_face) {
+    switch (front_face) {
+    case Maxwell::Cull::FrontFace::ClockWise:
+        return GL_CW;
+    case Maxwell::Cull::FrontFace::CounterClockWise:
+        return GL_CCW;
+    }
+    NGLOG_CRITICAL(Render_OpenGL, "Unimplemented front face cull={}", static_cast<u32>(front_face));
+    UNREACHABLE();
+    return {};
+}
+
+inline GLenum CullFace(Maxwell::Cull::CullFace cull_face) {
+    switch (cull_face) {
+    case Maxwell::Cull::CullFace::Front:
+        return GL_FRONT;
+    case Maxwell::Cull::CullFace::Back:
+        return GL_BACK;
+    case Maxwell::Cull::CullFace::FrontAndBack:
+        return GL_FRONT_AND_BACK;
+    }
+    NGLOG_CRITICAL(Render_OpenGL, "Unimplemented cull face={}", static_cast<u32>(cull_face));
+    UNREACHABLE();
+    return {};
+}
+
 } // namespace MaxwellToGL

src/video_core/textures/decoders.cpp

@@ -5,6 +5,7 @@
 #include <cstring>
 #include "common/assert.h"
 #include "core/memory.h"
+#include "video_core/gpu.h"
 #include "video_core/textures/decoders.h"
 #include "video_core/textures/texture.h"
 
@@ -65,6 +66,18 @@ u32 BytesPerPixel(TextureFormat format) {
         return 1;
     case TextureFormat::R16_G16_B16_A16:
         return 8;
+    case TextureFormat::R32_G32_B32_A32:
+        return 16;
+    default:
+        UNIMPLEMENTED_MSG("Format not implemented");
+        break;
+    }
+}
+
+static u32 DepthBytesPerPixel(DepthFormat format) {
+    switch (format) {
+    case DepthFormat::Z24_S8_UNORM:
+        return 4;
     default:
         UNIMPLEMENTED_MSG("Format not implemented");
         break;
@@ -94,6 +107,7 @@ std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width,
     case TextureFormat::B5G6R5:
     case TextureFormat::R8:
     case TextureFormat::R16_G16_B16_A16:
+    case TextureFormat::R32_G32_B32_A32:
     case TextureFormat::BF10GF11RF11:
     case TextureFormat::ASTC_2D_4X4:
         CopySwizzledData(width, height, bytes_per_pixel, bytes_per_pixel, data,
@@ -107,6 +121,26 @@ std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width,
     return unswizzled_data;
 }
 
+std::vector<u8> UnswizzleDepthTexture(VAddr address, DepthFormat format, u32 width, u32 height,
+                                      u32 block_height) {
+    u8* data = Memory::GetPointer(address);
+    u32 bytes_per_pixel = DepthBytesPerPixel(format);
+
+    std::vector<u8> unswizzled_data(width * height * bytes_per_pixel);
+
+    switch (format) {
+    case DepthFormat::Z24_S8_UNORM:
+        CopySwizzledData(width, height, bytes_per_pixel, bytes_per_pixel, data,
+                         unswizzled_data.data(), true, block_height);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Format not implemented");
+        break;
+    }
+
+    return unswizzled_data;
+}
+
 std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat format, u32 width,
                               u32 height) {
     std::vector<u8> rgba_data;
@@ -124,6 +158,7 @@ std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat
     case TextureFormat::B5G6R5:
     case TextureFormat::R8:
     case TextureFormat::BF10GF11RF11:
+    case TextureFormat::R32_G32_B32_A32:
         // TODO(Subv): For the time being just forward the same data without any decoding.
         rgba_data = texture_data;
         break;

src/video_core/textures/decoders.h

@@ -17,6 +17,12 @@ namespace Texture {
 std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width, u32 height,
                                  u32 block_height = TICEntry::DefaultBlockHeight);
 
+/**
+ * Unswizzles a swizzled depth texture without changing its format.
+ */
+std::vector<u8> UnswizzleDepthTexture(VAddr address, DepthFormat format, u32 width, u32 height,
+                                      u32 block_height = TICEntry::DefaultBlockHeight);
+
 /// Copies texture data from a buffer and performs swizzling/unswizzling as necessary.
 void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
                       u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height);