Merge pull request #2042 from ReinUsesLisp/nouveau-tex

maxwell_3d: Allow texture handles with TIC id zero

src/video_core/dma_pusher.cpp

@@ -35,8 +35,10 @@ void DmaPusher::DispatchCalls() {
 bool DmaPusher::Step() {
     if (dma_get != dma_put) {
         // Push buffer non-empty, read a word
-        const CommandHeader command_header{
+        const auto address = gpu.MemoryManager().GpuToCpuAddress(dma_get);
-            Memory::Read32(*gpu.MemoryManager().GpuToCpuAddress(dma_get))};
+        ASSERT_MSG(address, "Invalid GPU address");
+
+        const CommandHeader command_header{Memory::Read32(*address)};
 
         dma_get += sizeof(u32);
 

src/video_core/engines/fermi_2d.cpp

@@ -42,8 +42,10 @@ void Fermi2D::HandleSurfaceCopy() {
     // TODO(Subv): Only raw copies are implemented.
     ASSERT(regs.operation == Regs::Operation::SrcCopy);
 
-    const VAddr source_cpu = *memory_manager.GpuToCpuAddress(source);
+    const auto source_cpu = memory_manager.GpuToCpuAddress(source);
-    const VAddr dest_cpu = *memory_manager.GpuToCpuAddress(dest);
+    const auto dest_cpu = memory_manager.GpuToCpuAddress(dest);
+    ASSERT_MSG(source_cpu, "Invalid source GPU address");
+    ASSERT_MSG(dest_cpu, "Invalid destination GPU address");
 
     u32 src_bytes_per_pixel = RenderTargetBytesPerPixel(regs.src.format);
     u32 dst_bytes_per_pixel = RenderTargetBytesPerPixel(regs.dst.format);
@@ -52,22 +54,22 @@ void Fermi2D::HandleSurfaceCopy() {
         // All copies here update the main memory, so mark all rasterizer states as invalid.
         Core::System::GetInstance().GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
 
-        rasterizer.FlushRegion(source_cpu, src_bytes_per_pixel * regs.src.width * regs.src.height);
+        rasterizer.FlushRegion(*source_cpu, src_bytes_per_pixel * regs.src.width * regs.src.height);
         // We have to invalidate the destination region to evict any outdated surfaces from the
         // cache. We do this before actually writing the new data because the destination address
         // might contain a dirty surface that will have to be written back to memory.
-        rasterizer.InvalidateRegion(dest_cpu,
+        rasterizer.InvalidateRegion(*dest_cpu,
                                     dst_bytes_per_pixel * regs.dst.width * regs.dst.height);
 
         if (regs.src.linear == regs.dst.linear) {
             // If the input layout and the output layout are the same, just perform a raw copy.
             ASSERT(regs.src.BlockHeight() == regs.dst.BlockHeight());
-            Memory::CopyBlock(dest_cpu, source_cpu,
+            Memory::CopyBlock(*dest_cpu, *source_cpu,
                               src_bytes_per_pixel * regs.dst.width * regs.dst.height);
             return;
         }
-        u8* src_buffer = Memory::GetPointer(source_cpu);
+        u8* src_buffer = Memory::GetPointer(*source_cpu);
-        u8* dst_buffer = Memory::GetPointer(dest_cpu);
+        u8* dst_buffer = Memory::GetPointer(*dest_cpu);
         if (!regs.src.linear && regs.dst.linear) {
             // If the input is tiled and the output is linear, deswizzle the input and copy it over.
             Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth,

src/video_core/engines/kepler_memory.cpp

@@ -39,16 +39,17 @@ void KeplerMemory::ProcessData(u32 data) {
     ASSERT_MSG(regs.exec.linear, "Non-linear uploads are not supported");
     ASSERT(regs.dest.x == 0 && regs.dest.y == 0 && regs.dest.z == 0);
 
-    GPUVAddr address = regs.dest.Address();
+    const GPUVAddr address = regs.dest.Address();
-    VAddr dest_address =
+    const auto dest_address =
-        *memory_manager.GpuToCpuAddress(address + state.write_offset * sizeof(u32));
+        memory_manager.GpuToCpuAddress(address + state.write_offset * sizeof(u32));
+    ASSERT_MSG(dest_address, "Invalid GPU address");
 
     // We have to invalidate the destination region to evict any outdated surfaces from the cache.
     // We do this before actually writing the new data because the destination address might contain
     // a dirty surface that will have to be written back to memory.
-    rasterizer.InvalidateRegion(dest_address, sizeof(u32));
+    rasterizer.InvalidateRegion(*dest_address, sizeof(u32));
 
-    Memory::Write32(dest_address, data);
+    Memory::Write32(*dest_address, data);
     Core::System::GetInstance().GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
 
     state.write_offset++;

src/video_core/engines/maxwell_3d.cpp

@@ -273,7 +273,8 @@ void Maxwell3D::ProcessQueryGet() {
     GPUVAddr sequence_address = regs.query.QueryAddress();
     // Since the sequence address is given as a GPU VAddr, we have to convert it to an application
     // VAddr before writing.
-    std::optional<VAddr> address = memory_manager.GpuToCpuAddress(sequence_address);
+    const auto address = memory_manager.GpuToCpuAddress(sequence_address);
+    ASSERT_MSG(address, "Invalid GPU address");
 
     // TODO(Subv): Support the other query units.
     ASSERT_MSG(regs.query.query_get.unit == Regs::QueryUnit::Crop,
@@ -386,14 +387,14 @@ void Maxwell3D::ProcessCBBind(Regs::ShaderStage stage) {
 
 void Maxwell3D::ProcessCBData(u32 value) {
     // Write the input value to the current const buffer at the current position.
-    GPUVAddr buffer_address = regs.const_buffer.BufferAddress();
+    const GPUVAddr buffer_address = regs.const_buffer.BufferAddress();
     ASSERT(buffer_address != 0);
 
     // Don't allow writing past the end of the buffer.
     ASSERT(regs.const_buffer.cb_pos + sizeof(u32) <= regs.const_buffer.cb_size);
 
-    std::optional<VAddr> address =
+    const auto address = memory_manager.GpuToCpuAddress(buffer_address + regs.const_buffer.cb_pos);
-        memory_manager.GpuToCpuAddress(buffer_address + regs.const_buffer.cb_pos);
+    ASSERT_MSG(address, "Invalid GPU address");
 
     Memory::Write32(*address, value);
     dirty_flags.OnMemoryWrite();
@@ -403,10 +404,11 @@ void Maxwell3D::ProcessCBData(u32 value) {
 }
 
 Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
-    GPUVAddr tic_base_address = regs.tic.TICAddress();
+    const GPUVAddr tic_base_address = regs.tic.TICAddress();
 
-    GPUVAddr tic_address_gpu = tic_base_address + tic_index * sizeof(Texture::TICEntry);
+    const GPUVAddr tic_address_gpu = tic_base_address + tic_index * sizeof(Texture::TICEntry);
-    std::optional<VAddr> tic_address_cpu = memory_manager.GpuToCpuAddress(tic_address_gpu);
+    const auto tic_address_cpu = memory_manager.GpuToCpuAddress(tic_address_gpu);
+    ASSERT_MSG(tic_address_cpu, "Invalid GPU address");
 
     Texture::TICEntry tic_entry;
     Memory::ReadBlock(*tic_address_cpu, &tic_entry, sizeof(Texture::TICEntry));
@@ -415,10 +417,10 @@ Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
                    tic_entry.header_version == Texture::TICHeaderVersion::Pitch,
                "TIC versions other than BlockLinear or Pitch are unimplemented");
 
-    auto r_type = tic_entry.r_type.Value();
+    const auto r_type = tic_entry.r_type.Value();
-    auto g_type = tic_entry.g_type.Value();
+    const auto g_type = tic_entry.g_type.Value();
-    auto b_type = tic_entry.b_type.Value();
+    const auto b_type = tic_entry.b_type.Value();
-    auto a_type = tic_entry.a_type.Value();
+    const auto a_type = tic_entry.a_type.Value();
 
     // TODO(Subv): Different data types for separate components are not supported
     ASSERT(r_type == g_type && r_type == b_type && r_type == a_type);
@@ -427,10 +429,11 @@ Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
 }
 
 Texture::TSCEntry Maxwell3D::GetTSCEntry(u32 tsc_index) const {
-    GPUVAddr tsc_base_address = regs.tsc.TSCAddress();
+    const GPUVAddr tsc_base_address = regs.tsc.TSCAddress();
 
-    GPUVAddr tsc_address_gpu = tsc_base_address + tsc_index * sizeof(Texture::TSCEntry);
+    const GPUVAddr tsc_address_gpu = tsc_base_address + tsc_index * sizeof(Texture::TSCEntry);
-    std::optional<VAddr> tsc_address_cpu = memory_manager.GpuToCpuAddress(tsc_address_gpu);
+    const auto tsc_address_cpu = memory_manager.GpuToCpuAddress(tsc_address_gpu);
+    ASSERT_MSG(tsc_address_cpu, "Invalid GPU address");
 
     Texture::TSCEntry tsc_entry;
     Memory::ReadBlock(*tsc_address_cpu, &tsc_entry, sizeof(Texture::TSCEntry));
@@ -452,8 +455,10 @@ std::vector<Texture::FullTextureInfo> Maxwell3D::GetStageTextures(Regs::ShaderSt
     for (GPUVAddr current_texture = tex_info_buffer.address + TextureInfoOffset;
          current_texture < tex_info_buffer_end; current_texture += sizeof(Texture::TextureHandle)) {
 
-        Texture::TextureHandle tex_handle{
+        const auto address = memory_manager.GpuToCpuAddress(current_texture);
-            Memory::Read32(*memory_manager.GpuToCpuAddress(current_texture))};
+        ASSERT_MSG(address, "Invalid GPU address");
+
+        const Texture::TextureHandle tex_handle{Memory::Read32(*address)};
 
         Texture::FullTextureInfo tex_info{};
         // TODO(Subv): Use the shader to determine which textures are actually accessed.
@@ -462,22 +467,15 @@ std::vector<Texture::FullTextureInfo> Maxwell3D::GetStageTextures(Regs::ShaderSt
             sizeof(Texture::TextureHandle);
 
         // Load the TIC data.
-        if (tex_handle.tic_id != 0) {
-            tex_info.enabled = true;
-
         auto tic_entry = GetTICEntry(tex_handle.tic_id);
         // TODO(Subv): Workaround for BitField's move constructor being deleted.
         std::memcpy(&tex_info.tic, &tic_entry, sizeof(tic_entry));
-        }
 
         // Load the TSC data
-        if (tex_handle.tsc_id != 0) {
         auto tsc_entry = GetTSCEntry(tex_handle.tsc_id);
         // TODO(Subv): Workaround for BitField's move constructor being deleted.
         std::memcpy(&tex_info.tsc, &tsc_entry, sizeof(tsc_entry));
-        }
 
-        if (tex_info.enabled)
         textures.push_back(tex_info);
     }
 
@@ -490,31 +488,28 @@ Texture::FullTextureInfo Maxwell3D::GetStageTexture(Regs::ShaderStage stage,
     auto& tex_info_buffer = shader.const_buffers[regs.tex_cb_index];
     ASSERT(tex_info_buffer.enabled && tex_info_buffer.address != 0);
 
-    GPUVAddr tex_info_address = tex_info_buffer.address + offset * sizeof(Texture::TextureHandle);
+    const GPUVAddr tex_info_address =
+        tex_info_buffer.address + offset * sizeof(Texture::TextureHandle);
 
     ASSERT(tex_info_address < tex_info_buffer.address + tex_info_buffer.size);
 
-    std::optional<VAddr> tex_address_cpu = memory_manager.GpuToCpuAddress(tex_info_address);
+    const auto tex_address_cpu = memory_manager.GpuToCpuAddress(tex_info_address);
-    Texture::TextureHandle tex_handle{Memory::Read32(*tex_address_cpu)};
+    ASSERT_MSG(tex_address_cpu, "Invalid GPU address");
+
+    const Texture::TextureHandle tex_handle{Memory::Read32(*tex_address_cpu)};
 
     Texture::FullTextureInfo tex_info{};
     tex_info.index = static_cast<u32>(offset);
 
     // Load the TIC data.
-    if (tex_handle.tic_id != 0) {
-        tex_info.enabled = true;
-
     auto tic_entry = GetTICEntry(tex_handle.tic_id);
     // TODO(Subv): Workaround for BitField's move constructor being deleted.
     std::memcpy(&tex_info.tic, &tic_entry, sizeof(tic_entry));
-    }
 
     // Load the TSC data
-    if (tex_handle.tsc_id != 0) {
     auto tsc_entry = GetTSCEntry(tex_handle.tsc_id);
     // TODO(Subv): Workaround for BitField's move constructor being deleted.
     std::memcpy(&tex_info.tsc, &tsc_entry, sizeof(tsc_entry));
-    }
 
     return tex_info;
 }

src/video_core/engines/maxwell_dma.cpp

@@ -39,8 +39,10 @@ void MaxwellDMA::HandleCopy() {
     const GPUVAddr source = regs.src_address.Address();
     const GPUVAddr dest = regs.dst_address.Address();
 
-    const VAddr source_cpu = *memory_manager.GpuToCpuAddress(source);
+    const auto source_cpu = memory_manager.GpuToCpuAddress(source);
-    const VAddr dest_cpu = *memory_manager.GpuToCpuAddress(dest);
+    const auto dest_cpu = memory_manager.GpuToCpuAddress(dest);
+    ASSERT_MSG(source_cpu, "Invalid source GPU address");
+    ASSERT_MSG(dest_cpu, "Invalid destination GPU address");
 
     // TODO(Subv): Perform more research and implement all features of this engine.
     ASSERT(regs.exec.enable_swizzle == 0);
@@ -64,7 +66,7 @@ void MaxwellDMA::HandleCopy() {
         // buffer of length `x_count`, otherwise we copy a 2D image of dimensions (x_count,
         // y_count).
         if (!regs.exec.enable_2d) {
-            Memory::CopyBlock(dest_cpu, source_cpu, regs.x_count);
+            Memory::CopyBlock(*dest_cpu, *source_cpu, regs.x_count);
             return;
         }
 
@@ -73,8 +75,8 @@ void MaxwellDMA::HandleCopy() {
         // rectangle. There is no need to manually flush/invalidate the regions because
         // CopyBlock does that for us.
         for (u32 line = 0; line < regs.y_count; ++line) {
-            const VAddr source_line = source_cpu + line * regs.src_pitch;
+            const VAddr source_line = *source_cpu + line * regs.src_pitch;
-            const VAddr dest_line = dest_cpu + line * regs.dst_pitch;
+            const VAddr dest_line = *dest_cpu + line * regs.dst_pitch;
             Memory::CopyBlock(dest_line, source_line, regs.x_count);
         }
         return;
@@ -87,12 +89,12 @@ void MaxwellDMA::HandleCopy() {
     const auto FlushAndInvalidate = [&](u32 src_size, u64 dst_size) {
         // TODO(Subv): For now, manually flush the regions until we implement GPU-accelerated
         // copying.
-        rasterizer.FlushRegion(source_cpu, src_size);
+        rasterizer.FlushRegion(*source_cpu, src_size);
 
         // We have to invalidate the destination region to evict any outdated surfaces from the
         // cache. We do this before actually writing the new data because the destination address
         // might contain a dirty surface that will have to be written back to memory.
-        rasterizer.InvalidateRegion(dest_cpu, dst_size);
+        rasterizer.InvalidateRegion(*dest_cpu, dst_size);
     };
 
     if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
@@ -105,8 +107,8 @@ void MaxwellDMA::HandleCopy() {
                            copy_size * src_bytes_per_pixel);
 
         Texture::UnswizzleSubrect(regs.x_count, regs.y_count, regs.dst_pitch,
-                                  regs.src_params.size_x, src_bytes_per_pixel, source_cpu, dest_cpu,
+                                  regs.src_params.size_x, src_bytes_per_pixel, *source_cpu,
-                                  regs.src_params.BlockHeight(), regs.src_params.pos_x,
+                                  *dest_cpu, regs.src_params.BlockHeight(), regs.src_params.pos_x,
                                   regs.src_params.pos_y);
     } else {
         ASSERT(regs.dst_params.size_z == 1);
@@ -119,7 +121,7 @@ void MaxwellDMA::HandleCopy() {
 
         // If the input is linear and the output is tiled, swizzle the input and copy it over.
         Texture::SwizzleSubrect(regs.x_count, regs.y_count, regs.src_pitch, regs.dst_params.size_x,
-                                src_bpp, dest_cpu, source_cpu, regs.dst_params.BlockHeight());
+                                src_bpp, *dest_cpu, *source_cpu, regs.dst_params.BlockHeight());
     }
 }
 

src/video_core/memory_manager.cpp

@@ -154,7 +154,8 @@ std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) {
     const VAddr base_addr{PageSlot(gpu_addr)};
 
     if (base_addr == static_cast<u64>(PageStatus::Allocated) ||
-        base_addr == static_cast<u64>(PageStatus::Unmapped)) {
+        base_addr == static_cast<u64>(PageStatus::Unmapped) ||
+        base_addr == static_cast<u64>(PageStatus::Reserved)) {
         return {};
     }
 

src/video_core/renderer_opengl/gl_buffer_cache.cpp

@@ -19,7 +19,8 @@ OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size)
 GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size,
                                       std::size_t alignment, bool cache) {
     auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
-    const std::optional<VAddr> cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
+    const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
+    ASSERT_MSG(cpu_addr, "Invalid GPU address");
 
     // Cache management is a big overhead, so only cache entries with a given size.
     // TODO: Figure out which size is the best for given games.

src/video_core/renderer_opengl/gl_primitive_assembler.cpp

@@ -46,7 +46,9 @@ GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size
     auto [dst_pointer, index_offset] = buffer_cache.ReserveMemory(map_size);
 
     auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
-    const std::optional<VAddr> cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
+    const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
+    ASSERT_MSG(cpu_addr, "Invalid GPU address");
+
     const u8* source{Memory::GetPointer(*cpu_addr)};
 
     for (u32 primitive = 0; primitive < count / 4; ++primitive) {

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -1008,10 +1008,6 @@ void RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& s
         auto& unit = state.texture_units[current_bindpoint];
 
         const auto texture = maxwell3d.GetStageTexture(entry.GetStage(), entry.GetOffset());
-        if (!texture.enabled) {
-            unit.texture = 0;
-            continue;
-        }
 
         texture_samplers[current_bindpoint].SyncWithConfig(texture.tsc);
 

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -23,8 +23,10 @@ using VideoCommon::Shader::ProgramCode;
 static VAddr GetShaderAddress(Maxwell::ShaderProgram program) {
     const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
     const auto& shader_config = gpu.regs.shader_config[static_cast<std::size_t>(program)];
-    return *gpu.memory_manager.GpuToCpuAddress(gpu.regs.code_address.CodeAddress() +
+    const auto address = gpu.memory_manager.GpuToCpuAddress(gpu.regs.code_address.CodeAddress() +
                                                             shader_config.offset);
+    ASSERT_MSG(address, "Invalid GPU address");
+    return *address;
 }
 
 /// Gets the shader program code from memory for the specified address

src/video_core/textures/texture.h

@@ -317,7 +317,6 @@ struct FullTextureInfo {
     u32 index;
     TICEntry tic;
     TSCEntry tsc;
-    bool enabled;
 };
 
 /// Returns the number of bytes per pixel of the input texture format.