aboutsummaryrefslogtreecommitdiff
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/command_classes/codecs/codec.cpp35
-rw-r--r--src/video_core/renderer_opengl/gl_device.cpp2
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.cpp12
-rw-r--r--src/video_core/renderer_vulkan/blit_image.cpp7
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.cpp6
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.cpp3
-rw-r--r--src/video_core/texture_cache/image_base.h3
-rw-r--r--src/video_core/texture_cache/texture_cache.h43
-rw-r--r--src/video_core/texture_cache/texture_cache_base.h11
-rw-r--r--src/video_core/vulkan_common/vulkan_device.cpp7
10 files changed, 37 insertions, 92 deletions
diff --git a/src/video_core/command_classes/codecs/codec.cpp b/src/video_core/command_classes/codecs/codec.cpp
index 81fac94bf..40f7755e8 100644
--- a/src/video_core/command_classes/codecs/codec.cpp
+++ b/src/video_core/command_classes/codecs/codec.cpp
@@ -56,6 +56,18 @@ 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) {
@@ -76,17 +88,6 @@ 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();
@@ -96,6 +97,8 @@ 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 {}",
@@ -127,15 +130,19 @@ bool Codec::CreateGpuAvDevice() {
av_codec->name, av_hwdevice_get_type_name(type));
break;
}
- 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) {
+ 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) {
// 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;
}
}
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index 656dd7eb0..597301eeb 100644
--- a/src/video_core/renderer_opengl/gl_device.cpp
+++ b/src/video_core/renderer_opengl/gl_device.cpp
@@ -282,7 +282,7 @@ void main() {
u64 Device::GetCurrentDedicatedVideoMemory() const {
GLint cur_avail_mem_kb = 0;
- glGetIntegerv(GL_GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX, &cur_avail_mem_kb);
+ glGetIntegerv(GL_GPU_MEMORY_INFO_TOTAL_AVAILABLE_MEMORY_NVX, &cur_avail_mem_kb);
return static_cast<u64>(cur_avail_mem_kb) * 1_KiB;
}
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 7e06d0069..e6f9ece8b 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -15,8 +15,9 @@
#include "common/logging/log.h"
#include "common/math_util.h"
#include "common/microprofile.h"
+#include "common/scope_exit.h"
#include "common/settings.h"
-#include "core/memory.h"
+
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
@@ -210,6 +211,7 @@ void RasterizerOpenGL::Clear() {
void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
MICROPROFILE_SCOPE(OpenGL_Drawing);
+ SCOPE_EXIT({ gpu.TickWork(); });
query_cache.UpdateCounters();
GraphicsPipeline* const pipeline{shader_cache.CurrentGraphicsPipeline()};
@@ -265,8 +267,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
++num_queued_commands;
has_written_global_memory |= pipeline->WritesGlobalMemory();
-
- gpu.TickWork();
}
void RasterizerOpenGL::DispatchCompute() {
@@ -352,7 +352,7 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
shader_cache.OnCPUWrite(addr, size);
{
std::scoped_lock lock{texture_cache.mutex};
- texture_cache.CachedWriteMemory(addr, size);
+ texture_cache.WriteMemory(addr, size);
}
{
std::scoped_lock lock{buffer_cache.mutex};
@@ -364,10 +364,6 @@ void RasterizerOpenGL::SyncGuestHost() {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
shader_cache.SyncGuestHost();
{
- std::scoped_lock lock{texture_cache.mutex};
- texture_cache.FlushCachedWrites();
- }
- {
std::scoped_lock lock{buffer_cache.mutex};
buffer_cache.FlushCachedWrites();
}
diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp
index ec03cca38..abda1c490 100644
--- a/src/video_core/renderer_vulkan/blit_image.cpp
+++ b/src/video_core/renderer_vulkan/blit_image.cpp
@@ -367,17 +367,14 @@ 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>)) {
- if (device.IsExtShaderStencilExportSupported()) {
- blit_depth_stencil_frag = BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_FRAG_SPV);
- }
-}
+ nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {}
BlitImageHelper::~BlitImageHelper() = default;
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index dd6e0027e..fa87d37f8 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -408,7 +408,7 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
pipeline_cache.OnCPUWrite(addr, size);
{
std::scoped_lock lock{texture_cache.mutex};
- texture_cache.CachedWriteMemory(addr, size);
+ texture_cache.WriteMemory(addr, size);
}
{
std::scoped_lock lock{buffer_cache.mutex};
@@ -419,10 +419,6 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
void RasterizerVulkan::SyncGuestHost() {
pipeline_cache.SyncGuestHost();
{
- std::scoped_lock lock{texture_cache.mutex};
- texture_cache.FlushCachedWrites();
- }
- {
std::scoped_lock lock{buffer_cache.mutex};
buffer_cache.FlushCachedWrites();
}
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index f2890d263..2c2ccc7c6 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -1451,8 +1451,7 @@ 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 (!runtime->device.IsBlitDepthStencilSupported() &&
- aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
+ } else if (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);
}
diff --git a/src/video_core/texture_cache/image_base.h b/src/video_core/texture_cache/image_base.h
index cc7999027..dd0106432 100644
--- a/src/video_core/texture_cache/image_base.h
+++ b/src/video_core/texture_cache/image_base.h
@@ -39,9 +39,6 @@ enum class ImageFlagBits : u32 {
Rescaled = 1 << 13,
CheckingRescalable = 1 << 14,
IsRescalable = 1 << 15,
-
- // Cached CPU
- CachedCpuModified = 1 << 16, ///< Contents have been modified from the CPU
};
DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits)
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 099b2ae1b..8fef74117 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -438,23 +438,6 @@ void TextureCache<P>::WriteMemory(VAddr cpu_addr, size_t size) {
}
template <class P>
-void TextureCache<P>::CachedWriteMemory(VAddr cpu_addr, size_t size) {
- const VAddr new_cpu_addr = Common::AlignDown(cpu_addr, CPU_PAGE_SIZE);
- const size_t new_size = Common::AlignUp(size + cpu_addr - new_cpu_addr, CPU_PAGE_SIZE);
- ForEachImageInRegion(new_cpu_addr, new_size, [this](ImageId image_id, Image& image) {
- if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
- return;
- }
- image.flags |= ImageFlagBits::CachedCpuModified;
- cached_cpu_invalidate.insert(image_id);
-
- if (True(image.flags & ImageFlagBits::Tracked)) {
- UntrackImage(image, image_id);
- }
- });
-}
-
-template <class P>
void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) {
std::vector<ImageId> images;
ForEachImageInRegion(cpu_addr, size, [this, &images](ImageId image_id, ImageBase& image) {
@@ -512,18 +495,6 @@ void TextureCache<P>::UnmapGPUMemory(GPUVAddr gpu_addr, size_t size) {
}
template <class P>
-void TextureCache<P>::FlushCachedWrites() {
- for (ImageId image_id : cached_cpu_invalidate) {
- Image& image = slot_images[image_id];
- if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
- image.flags &= ~ImageFlagBits::CachedCpuModified;
- image.flags |= ImageFlagBits::CpuModified;
- }
- }
- cached_cpu_invalidate.clear();
-}
-
-template <class P>
void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
const Tegra::Engines::Fermi2D::Surface& src,
const Tegra::Engines::Fermi2D::Config& copy) {
@@ -1109,8 +1080,6 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
Image& overlap = slot_images[overlap_id];
if (True(overlap.flags & ImageFlagBits::GpuModified)) {
new_image.flags |= ImageFlagBits::GpuModified;
- new_image.modification_tick =
- std::max(overlap.modification_tick, new_image.modification_tick);
}
if (overlap.info.num_samples != new_image.info.num_samples) {
LOG_WARNING(HW_GPU, "Copying between images with different samples is not implemented");
@@ -1589,9 +1558,6 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
template <class P>
void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) {
ASSERT(False(image.flags & ImageFlagBits::Tracked));
- if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
- return;
- }
image.flags |= ImageFlagBits::Tracked;
if (False(image.flags & ImageFlagBits::Sparse)) {
rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1);
@@ -1648,9 +1614,6 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
}
total_used_memory -= Common::AlignUp(tentative_size, 1024);
- if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
- cached_cpu_invalidate.erase(image_id);
- }
const GPUVAddr gpu_addr = image.gpu_addr;
const auto alloc_it = image_allocs_table.find(gpu_addr);
if (alloc_it == image_allocs_table.end()) {
@@ -1817,11 +1780,7 @@ template <class P>
void TextureCache<P>::PrepareImage(ImageId image_id, bool is_modification, bool invalidate) {
Image& image = slot_images[image_id];
if (invalidate) {
- if (True(image.flags & ImageFlagBits::CachedCpuModified)) {
- cached_cpu_invalidate.erase(image_id);
- }
- image.flags &= ~(ImageFlagBits::CpuModified | ImageFlagBits::GpuModified |
- ImageFlagBits::CachedCpuModified);
+ image.flags &= ~(ImageFlagBits::CpuModified | ImageFlagBits::GpuModified);
if (False(image.flags & ImageFlagBits::Tracked)) {
TrackImage(image, image_id);
}
diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h
index ad5978a33..b1324edf3 100644
--- a/src/video_core/texture_cache/texture_cache_base.h
+++ b/src/video_core/texture_cache/texture_cache_base.h
@@ -8,7 +8,6 @@
#include <span>
#include <type_traits>
#include <unordered_map>
-#include <unordered_set>
#include <vector>
#include <queue>
@@ -51,9 +50,6 @@ class TextureCache {
/// Address shift for caching images into a hash table
static constexpr u64 PAGE_BITS = 20;
- static constexpr u64 CPU_PAGE_BITS = 12;
- static constexpr u64 CPU_PAGE_SIZE = 1ULL << CPU_PAGE_BITS;
-
/// Enables debugging features to the texture cache
static constexpr bool ENABLE_VALIDATION = P::ENABLE_VALIDATION;
/// Implement blits as copies between framebuffers
@@ -140,9 +136,6 @@ public:
/// Mark images in a range as modified from the CPU
void WriteMemory(VAddr cpu_addr, size_t size);
- /// Mark images in a range as modified from the CPU
- void CachedWriteMemory(VAddr cpu_addr, size_t size);
-
/// Download contents of host images to guest memory in a region
void DownloadMemory(VAddr cpu_addr, size_t size);
@@ -152,8 +145,6 @@ public:
/// Remove images in a region
void UnmapGPUMemory(GPUVAddr gpu_addr, size_t size);
- void FlushCachedWrites();
-
/// Blit an image with the given parameters
void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
const Tegra::Engines::Fermi2D::Surface& src,
@@ -375,8 +366,6 @@ private:
std::unordered_map<ImageId, std::vector<ImageViewId>> sparse_views;
- std::unordered_set<ImageId> cached_cpu_invalidate;
-
VAddr virtual_invalid_space{};
bool has_deleted_images = false;
diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp
index e142bee35..f3a05ada9 100644
--- a/src/video_core/vulkan_common/vulkan_device.cpp
+++ b/src/video_core/vulkan_common/vulkan_device.cpp
@@ -621,6 +621,11 @@ 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) {
@@ -731,7 +736,7 @@ VkFormat Device::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags
}
void Device::ReportLoss() const {
- LOG_CRITICAL(Render_Vulkan, "Device loss occured!");
+ LOG_CRITICAL(Render_Vulkan, "Device loss occurred!");
// Wait for the log to flush and for Nsight Aftermath to dump the results
std::this_thread::sleep_for(std::chrono::seconds{15});
generated by cgit v1.2.3 (git 2.25.1) at 2025-12-29 15:08:37 +0000