summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/buffer_cache/buffer_cache.h5
-rw-r--r--src/video_core/gpu.cpp5
-rw-r--r--src/video_core/gpu.h2
-rw-r--r--src/video_core/renderer_opengl/gl_texture_cache.cpp9
-rw-r--r--src/video_core/renderer_opengl/gl_texture_cache.h8
-rw-r--r--src/video_core/renderer_opengl/renderer_opengl.cpp1
-rw-r--r--src/video_core/renderer_vulkan/blit_image.cpp26
-rw-r--r--src/video_core/renderer_vulkan/blit_image.h10
-rw-r--r--src/video_core/renderer_vulkan/renderer_vulkan.cpp1
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.cpp38
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.h5
-rw-r--r--src/video_core/texture_cache/texture_cache.h57
-rw-r--r--src/video_core/texture_cache/types.h7
13 files changed, 112 insertions, 62 deletions
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 32dcbd693..de971041f 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -690,7 +690,10 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
const VAddr cpu_addr = binding.cpu_addr;
const u32 size = binding.size;
Buffer& buffer = slot_buffers[binding.buffer_id];
- if (size <= uniform_buffer_skip_cache_size && !buffer.IsRegionGpuModified(cpu_addr, size)) {
+ const bool use_fast_buffer = binding.buffer_id != NULL_BUFFER_ID &&
+ size <= uniform_buffer_skip_cache_size &&
+ !buffer.IsRegionGpuModified(cpu_addr, size);
+ if (use_fast_buffer) {
if constexpr (IS_OPENGL) {
if (runtime.HasFastBufferSubData()) {
// Fast path for Nvidia
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index a38024242..37f7b24e1 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -13,6 +13,7 @@
#include "core/frontend/emu_window.h"
#include "core/hardware_interrupt_manager.h"
#include "core/memory.h"
+#include "core/perf_stats.h"
#include "video_core/engines/fermi_2d.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/kepler_memory.h"
@@ -191,6 +192,10 @@ u64 GPU::GetTicks() const {
return nanoseconds_num * gpu_ticks_num + (nanoseconds_rem * gpu_ticks_num) / gpu_ticks_den;
}
+void GPU::RendererFrameEndNotify() {
+ system.GetPerfStats().EndGameFrame();
+}
+
void GPU::FlushCommands() {
rasterizer->FlushCommands();
}
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index 8669e9940..29a867863 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -247,6 +247,8 @@ public:
return use_nvdec;
}
+ void RendererFrameEndNotify();
+
enum class FenceOperation : u32 {
Acquire = 0,
Increment = 1,
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 623b43d8a..ffe9edc1b 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -543,8 +543,7 @@ void TextureCacheRuntime::EmulateCopyImage(Image& dst, Image& src,
}
void TextureCacheRuntime::BlitFramebuffer(Framebuffer* dst, Framebuffer* src,
- const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
+ const Region2D& dst_region, const Region2D& src_region,
Tegra::Engines::Fermi2D::Filter filter,
Tegra::Engines::Fermi2D::Operation operation) {
state_tracker.NotifyScissor0();
@@ -560,9 +559,9 @@ void TextureCacheRuntime::BlitFramebuffer(Framebuffer* dst, Framebuffer* src,
const GLbitfield buffer_bits = dst->BufferBits();
const bool has_depth = (buffer_bits & ~GL_COLOR_BUFFER_BIT) != 0;
const bool is_linear = !has_depth && filter == Tegra::Engines::Fermi2D::Filter::Bilinear;
- glBlitNamedFramebuffer(src->Handle(), dst->Handle(), src_region[0].x, src_region[0].y,
- src_region[1].x, src_region[1].y, dst_region[0].x, dst_region[0].y,
- dst_region[1].x, dst_region[1].y, buffer_bits,
+ glBlitNamedFramebuffer(src->Handle(), dst->Handle(), src_region.start.x, src_region.start.y,
+ src_region.end.x, src_region.end.y, dst_region.start.x,
+ dst_region.start.y, dst_region.end.x, dst_region.end.y, buffer_bits,
is_linear ? GL_LINEAR : GL_NEAREST);
}
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h
index 3c871541b..df8be12ff 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.h
+++ b/src/video_core/renderer_opengl/gl_texture_cache.h
@@ -28,7 +28,7 @@ using VideoCommon::ImageId;
using VideoCommon::ImageViewId;
using VideoCommon::ImageViewType;
using VideoCommon::NUM_RT;
-using VideoCommon::Offset2D;
+using VideoCommon::Region2D;
using VideoCommon::RenderTargets;
struct ImageBufferMap {
@@ -73,10 +73,8 @@ public:
void EmulateCopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
- void BlitFramebuffer(Framebuffer* dst, Framebuffer* src,
- const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
- Tegra::Engines::Fermi2D::Filter filter,
+ void BlitFramebuffer(Framebuffer* dst, Framebuffer* src, const Region2D& dst_region,
+ const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter,
Tegra::Engines::Fermi2D::Operation operation);
void AccelerateImageUpload(Image& image, const ImageBufferMap& map,
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index cc2e499f9..a718bff7a 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -155,6 +155,7 @@ void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
++m_current_frame;
+ gpu.RendererFrameEndNotify();
rasterizer.TickFrame();
context->SwapBuffers();
diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp
index 1f6a169ae..b7f5b8bc2 100644
--- a/src/video_core/renderer_vulkan/blit_image.cpp
+++ b/src/video_core/renderer_vulkan/blit_image.cpp
@@ -289,16 +289,15 @@ void UpdateTwoTexturesDescriptorSet(const Device& device, VkDescriptorSet descri
device.GetLogical().UpdateDescriptorSets(write_descriptor_sets, nullptr);
}
-void BindBlitState(vk::CommandBuffer cmdbuf, VkPipelineLayout layout,
- const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region) {
+void BindBlitState(vk::CommandBuffer cmdbuf, VkPipelineLayout layout, const Region2D& dst_region,
+ const Region2D& src_region) {
const VkOffset2D offset{
- .x = std::min(dst_region[0].x, dst_region[1].x),
- .y = std::min(dst_region[0].y, dst_region[1].y),
+ .x = std::min(dst_region.start.x, dst_region.end.x),
+ .y = std::min(dst_region.start.y, dst_region.end.y),
};
const VkExtent2D extent{
- .width = static_cast<u32>(std::abs(dst_region[1].x - dst_region[0].x)),
- .height = static_cast<u32>(std::abs(dst_region[1].y - dst_region[0].y)),
+ .width = static_cast<u32>(std::abs(dst_region.end.x - dst_region.start.x)),
+ .height = static_cast<u32>(std::abs(dst_region.end.y - dst_region.start.y)),
};
const VkViewport viewport{
.x = static_cast<float>(offset.x),
@@ -313,11 +312,12 @@ void BindBlitState(vk::CommandBuffer cmdbuf, VkPipelineLayout layout,
.offset = offset,
.extent = extent,
};
- const float scale_x = static_cast<float>(src_region[1].x - src_region[0].x);
- const float scale_y = static_cast<float>(src_region[1].y - src_region[0].y);
+ const float scale_x = static_cast<float>(src_region.end.x - src_region.start.x);
+ const float scale_y = static_cast<float>(src_region.end.y - src_region.start.y);
const PushConstants push_constants{
.tex_scale = {scale_x, scale_y},
- .tex_offset = {static_cast<float>(src_region[0].x), static_cast<float>(src_region[0].y)},
+ .tex_offset = {static_cast<float>(src_region.start.x),
+ static_cast<float>(src_region.start.y)},
};
cmdbuf.SetViewport(0, viewport);
cmdbuf.SetScissor(0, scissor);
@@ -353,8 +353,7 @@ BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_,
BlitImageHelper::~BlitImageHelper() = default;
void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
- const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
+ const Region2D& dst_region, const Region2D& src_region,
Tegra::Engines::Fermi2D::Filter filter,
Tegra::Engines::Fermi2D::Operation operation) {
const bool is_linear = filter == Tegra::Engines::Fermi2D::Filter::Bilinear;
@@ -383,8 +382,7 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageV
void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer,
VkImageView src_depth_view, VkImageView src_stencil_view,
- const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
+ const Region2D& dst_region, const Region2D& src_region,
Tegra::Engines::Fermi2D::Filter filter,
Tegra::Engines::Fermi2D::Operation operation) {
ASSERT(filter == Tegra::Engines::Fermi2D::Filter::Point);
diff --git a/src/video_core/renderer_vulkan/blit_image.h b/src/video_core/renderer_vulkan/blit_image.h
index 43fd3d737..0d81a06ed 100644
--- a/src/video_core/renderer_vulkan/blit_image.h
+++ b/src/video_core/renderer_vulkan/blit_image.h
@@ -13,7 +13,7 @@
namespace Vulkan {
-using VideoCommon::Offset2D;
+using VideoCommon::Region2D;
class Device;
class Framebuffer;
@@ -35,15 +35,13 @@ public:
~BlitImageHelper();
void BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
- const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
+ const Region2D& dst_region, const Region2D& src_region,
Tegra::Engines::Fermi2D::Filter filter,
Tegra::Engines::Fermi2D::Operation operation);
void BlitDepthStencil(const Framebuffer* dst_framebuffer, VkImageView src_depth_view,
- VkImageView src_stencil_view, const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
- Tegra::Engines::Fermi2D::Filter filter,
+ VkImageView src_stencil_view, const Region2D& dst_region,
+ const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter,
Tegra::Engines::Fermi2D::Operation operation);
void ConvertD32ToR32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view);
diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
index 2e0cf4232..3986eb172 100644
--- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp
+++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
@@ -154,6 +154,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
if (swapchain.Present(render_semaphore)) {
blit_screen.Recreate();
}
+ gpu.RendererFrameEndNotify();
rasterizer.TickFrame();
}
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index 017348e05..bdd0ce8bc 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -490,8 +490,7 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im
write_barrier);
}
-[[nodiscard]] VkImageBlit MakeImageBlit(const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
+[[nodiscard]] VkImageBlit MakeImageBlit(const Region2D& dst_region, const Region2D& src_region,
const VkImageSubresourceLayers& dst_layers,
const VkImageSubresourceLayers& src_layers) {
return VkImageBlit{
@@ -499,13 +498,13 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im
.srcOffsets =
{
{
- .x = src_region[0].x,
- .y = src_region[0].y,
+ .x = src_region.start.x,
+ .y = src_region.start.y,
.z = 0,
},
{
- .x = src_region[1].x,
- .y = src_region[1].y,
+ .x = src_region.end.x,
+ .y = src_region.end.y,
.z = 1,
},
},
@@ -513,42 +512,42 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im
.dstOffsets =
{
{
- .x = dst_region[0].x,
- .y = dst_region[0].y,
+ .x = dst_region.start.x,
+ .y = dst_region.start.y,
.z = 0,
},
{
- .x = dst_region[1].x,
- .y = dst_region[1].y,
+ .x = dst_region.end.x,
+ .y = dst_region.end.y,
.z = 1,
},
},
};
}
-[[nodiscard]] VkImageResolve MakeImageResolve(const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
+[[nodiscard]] VkImageResolve MakeImageResolve(const Region2D& dst_region,
+ const Region2D& src_region,
const VkImageSubresourceLayers& dst_layers,
const VkImageSubresourceLayers& src_layers) {
return VkImageResolve{
.srcSubresource = src_layers,
.srcOffset =
{
- .x = src_region[0].x,
- .y = src_region[0].y,
+ .x = src_region.start.x,
+ .y = src_region.start.y,
.z = 0,
},
.dstSubresource = dst_layers,
.dstOffset =
{
- .x = dst_region[0].x,
- .y = dst_region[0].y,
+ .x = dst_region.start.x,
+ .y = dst_region.start.y,
.z = 0,
},
.extent =
{
- .width = static_cast<u32>(dst_region[1].x - dst_region[0].x),
- .height = static_cast<u32>(dst_region[1].y - dst_region[0].y),
+ .width = static_cast<u32>(dst_region.end.x - dst_region.start.x),
+ .height = static_cast<u32>(dst_region.end.y - dst_region.start.y),
.depth = 1,
},
};
@@ -602,8 +601,7 @@ StagingBufferRef TextureCacheRuntime::DownloadStagingBuffer(size_t size) {
}
void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src,
- const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
+ const Region2D& dst_region, const Region2D& src_region,
Tegra::Engines::Fermi2D::Filter filter,
Tegra::Engines::Fermi2D::Operation operation) {
const VkImageAspectFlags aspect_mask = ImageAspectMask(src.format);
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h
index 628785d5e..4a57d378b 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.h
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.h
@@ -16,7 +16,7 @@ namespace Vulkan {
using VideoCommon::ImageId;
using VideoCommon::NUM_RT;
-using VideoCommon::Offset2D;
+using VideoCommon::Region2D;
using VideoCommon::RenderTargets;
using VideoCore::Surface::PixelFormat;
@@ -71,8 +71,7 @@ struct TextureCacheRuntime {
[[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size);
void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src,
- const std::array<Offset2D, 2>& dst_region,
- const std::array<Offset2D, 2>& src_region,
+ const Region2D& dst_region, const Region2D& src_region,
Tegra::Engines::Fermi2D::Filter filter,
Tegra::Engines::Fermi2D::Operation operation);
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 98e33c3a0..59b7c678b 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -148,7 +148,9 @@ public:
/// Blit an image with the given parameters
void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
const Tegra::Engines::Fermi2D::Surface& src,
- const Tegra::Engines::Fermi2D::Config& copy);
+ const Tegra::Engines::Fermi2D::Config& copy,
+ std::optional<Region2D> src_region_override = {},
+ std::optional<Region2D> dst_region_override = {});
/// Invalidate the contents of the color buffer index
/// These contents become unspecified, the cache can assume aggressive optimizations.
@@ -615,7 +617,9 @@ void TextureCache<P>::UnmapMemory(VAddr cpu_addr, size_t size) {
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) {
+ const Tegra::Engines::Fermi2D::Config& copy,
+ std::optional<Region2D> src_override,
+ std::optional<Region2D> dst_override) {
const BlitImages images = GetBlitImages(dst, src);
const ImageId dst_id = images.dst_id;
const ImageId src_id = images.src_id;
@@ -631,20 +635,42 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
const ImageViewInfo dst_view_info(ImageViewType::e2D, images.dst_format, dst_range);
const auto [dst_framebuffer_id, dst_view_id] = RenderTargetFromImage(dst_id, dst_view_info);
const auto [src_samples_x, src_samples_y] = SamplesLog2(src_image.info.num_samples);
- const std::array src_region{
- Offset2D{.x = copy.src_x0 >> src_samples_x, .y = copy.src_y0 >> src_samples_y},
- Offset2D{.x = copy.src_x1 >> src_samples_x, .y = copy.src_y1 >> src_samples_y},
+
+ // out of bounds texture blit checking
+ const bool use_override = src_override.has_value();
+ const s32 src_x0 = copy.src_x0 >> src_samples_x;
+ s32 src_x1 = use_override ? src_override->end.x : copy.src_x1 >> src_samples_x;
+ const s32 src_y0 = copy.src_y0 >> src_samples_y;
+ const s32 src_y1 = copy.src_y1 >> src_samples_y;
+
+ const auto src_width = static_cast<s32>(src_image.info.size.width);
+ const bool width_oob = src_x1 > src_width;
+ const auto width_diff = width_oob ? src_x1 - src_width : 0;
+ if (width_oob) {
+ src_x1 = src_width;
+ }
+
+ const Region2D src_dimensions{
+ Offset2D{.x = src_x0, .y = src_y0},
+ Offset2D{.x = src_x1, .y = src_y1},
};
+ const auto src_region = use_override ? *src_override : src_dimensions;
const std::optional src_base = src_image.TryFindBase(src.Address());
const SubresourceRange src_range{.base = src_base.value(), .extent = {1, 1}};
const ImageViewInfo src_view_info(ImageViewType::e2D, images.src_format, src_range);
const auto [src_framebuffer_id, src_view_id] = RenderTargetFromImage(src_id, src_view_info);
const auto [dst_samples_x, dst_samples_y] = SamplesLog2(dst_image.info.num_samples);
- const std::array dst_region{
- Offset2D{.x = copy.dst_x0 >> dst_samples_x, .y = copy.dst_y0 >> dst_samples_y},
- Offset2D{.x = copy.dst_x1 >> dst_samples_x, .y = copy.dst_y1 >> dst_samples_y},
+
+ const s32 dst_x0 = copy.dst_x0 >> dst_samples_x;
+ const s32 dst_x1 = copy.dst_x1 >> dst_samples_x;
+ const s32 dst_y0 = copy.dst_y0 >> dst_samples_y;
+ const s32 dst_y1 = copy.dst_y1 >> dst_samples_y;
+ const Region2D dst_dimensions{
+ Offset2D{.x = dst_x0, .y = dst_y0},
+ Offset2D{.x = dst_x1 - width_diff, .y = dst_y1},
};
+ const auto dst_region = use_override ? *dst_override : dst_dimensions;
// Always call this after src_framebuffer_id was queried, as the address might be invalidated.
Framebuffer* const dst_framebuffer = &slot_framebuffers[dst_framebuffer_id];
@@ -661,6 +687,21 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
runtime.BlitImage(dst_framebuffer, dst_view, src_view, dst_region, src_region, copy.filter,
copy.operation);
}
+
+ if (width_oob) {
+ // Continue copy of the oob region of the texture on the next row
+ auto oob_src = src;
+ oob_src.height++;
+ const Region2D src_region_override{
+ Offset2D{.x = 0, .y = src_y0 + 1},
+ Offset2D{.x = width_diff, .y = src_y1 + 1},
+ };
+ const Region2D dst_region_override{
+ Offset2D{.x = dst_x1 - width_diff, .y = dst_y0},
+ Offset2D{.x = dst_x1, .y = dst_y1},
+ };
+ BlitImage(dst, oob_src, copy, src_region_override, dst_region_override);
+ }
}
template <class P>
diff --git a/src/video_core/texture_cache/types.h b/src/video_core/texture_cache/types.h
index 2ad2d72a6..c9571f7e4 100644
--- a/src/video_core/texture_cache/types.h
+++ b/src/video_core/texture_cache/types.h
@@ -64,6 +64,13 @@ struct Offset3D {
s32 z;
};
+struct Region2D {
+ constexpr auto operator<=>(const Region2D&) const noexcept = default;
+
+ Offset2D start;
+ Offset2D end;
+};
+
struct Extent2D {
constexpr auto operator<=>(const Extent2D&) const noexcept = default;
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-26 07:40:29 +0100