early-access version 2618

This commit is contained in:
pineappleEA
2022-03-29 10:45:55 +02:00
parent 64e77a60bf
commit 95e5011b2b
35 changed files with 118 additions and 500 deletions

View File

@@ -56,18 +56,6 @@ AVPixelFormat GetGpuFormat(AVCodecContext* av_codec_ctx, const AVPixelFormat* pi
av_codec_ctx->pix_fmt = PREFERRED_CPU_FMT;
return PREFERRED_CPU_FMT;
}
// List all the currently available hwcontext in ffmpeg
std::vector<AVHWDeviceType> ListSupportedContexts() {
std::vector<AVHWDeviceType> contexts{};
AVHWDeviceType current_device_type = AV_HWDEVICE_TYPE_NONE;
do {
current_device_type = av_hwdevice_iterate_types(current_device_type);
contexts.push_back(current_device_type);
} while (current_device_type != AV_HWDEVICE_TYPE_NONE);
return contexts;
}
} // namespace
void AVFrameDeleter(AVFrame* ptr) {
@@ -88,6 +76,17 @@ Codec::~Codec() {
av_buffer_unref(&av_gpu_decoder);
}
// List all the currently available hwcontext in ffmpeg
static std::vector<AVHWDeviceType> ListSupportedContexts() {
std::vector<AVHWDeviceType> contexts{};
AVHWDeviceType current_device_type = AV_HWDEVICE_TYPE_NONE;
do {
current_device_type = av_hwdevice_iterate_types(current_device_type);
contexts.push_back(current_device_type);
} while (current_device_type != AV_HWDEVICE_TYPE_NONE);
return contexts;
}
bool Codec::CreateGpuAvDevice() {
static constexpr auto HW_CONFIG_METHOD = AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX;
static const auto supported_contexts = ListSupportedContexts();
@@ -97,8 +96,6 @@ bool Codec::CreateGpuAvDevice() {
LOG_DEBUG(Service_NVDRV, "{} explicitly unsupported", av_hwdevice_get_type_name(type));
continue;
}
// Avoid memory leak from not cleaning up after av_hwdevice_ctx_create
av_buffer_unref(&av_gpu_decoder);
const int hwdevice_res = av_hwdevice_ctx_create(&av_gpu_decoder, type, nullptr, nullptr, 0);
if (hwdevice_res < 0) {
LOG_DEBUG(Service_NVDRV, "{} av_hwdevice_ctx_create failed {}",
@@ -130,19 +127,15 @@ bool Codec::CreateGpuAvDevice() {
av_codec->name, av_hwdevice_get_type_name(type));
break;
}
if ((config->methods & HW_CONFIG_METHOD) != 0 && config->device_type == type) {
#if defined(__unix__)
// Some linux decoding backends are reported to crash with this config method
// TODO(ameerj): Properly support this method
if ((config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_FRAMES_CTX) != 0) {
if (config->methods & HW_CONFIG_METHOD && config->device_type == type) {
av_codec_ctx->pix_fmt = config->pix_fmt;
if (config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_FRAMES_CTX) {
// skip zero-copy decoders, we don't currently support them
LOG_DEBUG(Service_NVDRV, "Skipping decoder {} with unsupported capability {}.",
av_hwdevice_get_type_name(type), config->methods);
continue;
}
#endif
LOG_INFO(Service_NVDRV, "Using {} GPU decoder", av_hwdevice_get_type_name(type));
av_codec_ctx->pix_fmt = config->pix_fmt;
return true;
}
}

View File

@@ -90,10 +90,6 @@ void ThreadManager::FlushRegion(VAddr addr, u64 size) {
return;
}
if (!Settings::IsGPULevelExtreme()) {
// Push a command and block here before proceeding, addresses a synchronization
// bug causing an SVC break in Kirby and the Forgotten Land
// GPUTickCommand is essentially a no-op if we don't RequestFlush()
PushCommand(GPUTickCommand(), true);
return;
}
auto& gpu = system.GPU();

View File

@@ -39,22 +39,7 @@ GPUVAddr MemoryManager::UpdateRange(GPUVAddr gpu_addr, PageEntry page_entry, std
return gpu_addr;
}
void MemoryManager::UnmapSubmappedRanges(GPUVAddr gpu_addr, std::size_t size) {
const auto submapped_ranges = GetSubmappedRange(gpu_addr, size);
for (const auto& [map_addr, map_size] : submapped_ranges) {
// Flush and invalidate through the GPU interface, to be asynchronous if possible.
const std::optional<VAddr> cpu_vaddr = GpuToCpuAddress(map_addr);
if (!cpu_vaddr) {
continue;
}
rasterizer->UnmapMemory(*cpu_vaddr, map_size);
}
}
GPUVAddr MemoryManager::Map(VAddr cpu_addr, GPUVAddr gpu_addr, std::size_t size) {
// Unmap any pre-existing rasterizer memory in this range
UnmapSubmappedRanges(gpu_addr, size);
const auto it = std::ranges::lower_bound(map_ranges, gpu_addr, {}, &MapRange::first);
if (it != map_ranges.end() && it->first == gpu_addr) {
it->second = size;
@@ -85,8 +70,16 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
} else {
UNREACHABLE_MSG("Unmapping non-existent GPU address=0x{:x}", gpu_addr);
}
const auto submapped_ranges = GetSubmappedRange(gpu_addr, size);
for (const auto& [map_addr, map_size] : submapped_ranges) {
// Flush and invalidate through the GPU interface, to be asynchronous if possible.
const std::optional<VAddr> cpu_addr = GpuToCpuAddress(map_addr);
ASSERT(cpu_addr);
rasterizer->UnmapMemory(*cpu_addr, map_size);
}
UnmapSubmappedRanges(gpu_addr, size);
UpdateRange(gpu_addr, PageEntry::State::Unmapped, size);
}

View File

@@ -149,7 +149,6 @@ private:
[[nodiscard]] PageEntry GetPageEntry(GPUVAddr gpu_addr) const;
void SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size = page_size);
GPUVAddr UpdateRange(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size);
void UnmapSubmappedRanges(GPUVAddr gpu_addr, std::size_t size);
[[nodiscard]] std::optional<GPUVAddr> FindFreeRange(std::size_t size, std::size_t align,
bool start_32bit_address = false) const;

View File

@@ -367,14 +367,17 @@ BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_,
PipelineLayoutCreateInfo(two_textures_set_layout.address()))),
full_screen_vert(BuildShader(device, FULL_SCREEN_TRIANGLE_VERT_SPV)),
blit_color_to_color_frag(BuildShader(device, VULKAN_BLIT_COLOR_FLOAT_FRAG_SPV)),
blit_depth_stencil_frag(BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_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)),
convert_d24s8_to_abgr8_frag(BuildShader(device, CONVERT_D24S8_TO_ABGR8_FRAG_SPV)),
convert_s8d24_to_abgr8_frag(BuildShader(device, CONVERT_S8D24_TO_ABGR8_FRAG_SPV)),
linear_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_LINEAR>)),
nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {}
nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {
if (device.IsExtShaderStencilExportSupported()) {
blit_depth_stencil_frag = BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_FRAG_SPV);
}
}
BlitImageHelper::~BlitImageHelper() = default;

View File

@@ -1451,7 +1451,8 @@ bool Image::BlitScaleHelper(bool scale_up) {
runtime->blit_image_helper.BlitColor(blit_framebuffer.get(), color_view, dst_region,
src_region, operation, BLIT_OPERATION);
} else if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
} else if (!runtime->device.IsBlitDepthStencilSupported() &&
aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
if (!blit_framebuffer) {
blit_framebuffer = std::make_unique<Framebuffer>(*runtime, nullptr, view_ptr, extent);
}

View File

@@ -621,11 +621,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
khr_push_descriptor = false;
break;
}
const u32 nv_major_version = (properties.driverVersion >> 22) & 0x3ff;
if (nv_major_version >= 510) {
LOG_WARNING(Render_Vulkan, "NVIDIA Drivers >= 510 do not support MSAA image blits");
cant_blit_msaa = true;
}
}
const bool is_radv = driver_id == VK_DRIVER_ID_MESA_RADV;
if (ext_extended_dynamic_state && is_radv) {