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.h78
-rw-r--r--src/video_core/engines/maxwell_dma.cpp20
-rw-r--r--src/video_core/engines/maxwell_dma.h2
-rw-r--r--src/video_core/renderer_opengl/gl_buffer_cache.cpp6
-rw-r--r--src/video_core/renderer_opengl/gl_buffer_cache.h2
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.cpp5
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.h2
-rw-r--r--src/video_core/renderer_opengl/gl_texture_cache.h2
-rw-r--r--src/video_core/renderer_vulkan/vk_buffer_cache.cpp55
-rw-r--r--src/video_core/renderer_vulkan/vk_buffer_cache.h8
-rw-r--r--src/video_core/renderer_vulkan/vk_compute_pass.cpp33
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.cpp5
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.h2
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.cpp6
-rw-r--r--src/video_core/texture_cache/texture_cache.h6
-rw-r--r--src/video_core/texture_cache/util.cpp86
16 files changed, 224 insertions, 94 deletions
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 2871682f6..7373cb62d 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -164,11 +164,16 @@ public:
/// Pop asynchronous downloads
void PopAsyncFlushes();
- [[nodiscard]] bool DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount);
+ bool DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount);
+
+ bool DMAClear(GPUVAddr src_address, u64 amount, u32 value);
/// Return true when a CPU region is modified from the GPU
[[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
+ /// Return true when a region is registered on the cache
+ [[nodiscard]] bool IsRegionRegistered(VAddr addr, size_t size);
+
/// Return true when a CPU region is modified from the CPU
[[nodiscard]] bool IsRegionCpuModified(VAddr addr, size_t size);
@@ -324,6 +329,8 @@ private:
[[nodiscard]] bool HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept;
+ void ClearDownload(IntervalType subtract_interval);
+
VideoCore::RasterizerInterface& rasterizer;
Tegra::Engines::Maxwell3D& maxwell3d;
Tegra::Engines::KeplerCompute& kepler_compute;
@@ -463,23 +470,28 @@ void BufferCache<P>::DownloadMemory(VAddr cpu_addr, u64 size) {
}
template <class P>
+void BufferCache<P>::ClearDownload(IntervalType subtract_interval) {
+ uncommitted_ranges.subtract(subtract_interval);
+ for (auto& interval_set : committed_ranges) {
+ interval_set.subtract(subtract_interval);
+ }
+}
+
+template <class P>
bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) {
const std::optional<VAddr> cpu_src_address = gpu_memory.GpuToCpuAddress(src_address);
const std::optional<VAddr> cpu_dest_address = gpu_memory.GpuToCpuAddress(dest_address);
if (!cpu_src_address || !cpu_dest_address) {
return false;
}
- const bool source_dirty = IsRegionGpuModified(*cpu_src_address, amount);
- const bool dest_dirty = IsRegionGpuModified(*cpu_dest_address, amount);
+ const bool source_dirty = IsRegionRegistered(*cpu_src_address, amount);
+ const bool dest_dirty = IsRegionRegistered(*cpu_dest_address, amount);
if (!source_dirty && !dest_dirty) {
return false;
}
const IntervalType subtract_interval{*cpu_dest_address, *cpu_dest_address + amount};
- uncommitted_ranges.subtract(subtract_interval);
- for (auto& interval_set : committed_ranges) {
- interval_set.subtract(subtract_interval);
- }
+ ClearDownload(subtract_interval);
BufferId buffer_a;
BufferId buffer_b;
@@ -510,12 +522,13 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
ForEachWrittenRange(*cpu_src_address, amount, mirror);
// This subtraction in this order is important for overlapping copies.
common_ranges.subtract(subtract_interval);
+ bool atleast_1_download = tmp_intervals.size() != 0;
for (const IntervalType add_interval : tmp_intervals) {
common_ranges.add(add_interval);
}
runtime.CopyBuffer(dest_buffer, src_buffer, copies);
- if (source_dirty) {
+ if (atleast_1_download) {
dest_buffer.MarkRegionAsGpuModified(*cpu_dest_address, amount);
}
std::vector<u8> tmp_buffer(amount);
@@ -525,6 +538,33 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
}
template <class P>
+bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) {
+ const std::optional<VAddr> cpu_dst_address = gpu_memory.GpuToCpuAddress(dst_address);
+ if (!cpu_dst_address) {
+ return false;
+ }
+ const bool dest_dirty = IsRegionRegistered(*cpu_dst_address, amount);
+ if (!dest_dirty) {
+ return false;
+ }
+
+ const size_t size = amount * sizeof(u32);
+ const IntervalType subtract_interval{*cpu_dst_address, *cpu_dst_address + size};
+ ClearDownload(subtract_interval);
+ common_ranges.subtract(subtract_interval);
+
+ BufferId buffer;
+ do {
+ has_deleted_buffers = false;
+ buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size));
+ } while (has_deleted_buffers);
+ auto& dest_buffer = slot_buffers[buffer];
+ const u32 offset = static_cast<u32>(*cpu_dst_address - dest_buffer.CpuAddr());
+ runtime.ClearBuffer(dest_buffer, offset, size, value);
+ return true;
+}
+
+template <class P>
void BufferCache<P>::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr,
u32 size) {
const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
@@ -782,6 +822,27 @@ bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
}
template <class P>
+bool BufferCache<P>::IsRegionRegistered(VAddr addr, size_t size) {
+ const VAddr end_addr = addr + size;
+ const u64 page_end = Common::DivCeil(end_addr, PAGE_SIZE);
+ for (u64 page = addr >> PAGE_BITS; page < page_end;) {
+ const BufferId buffer_id = page_table[page];
+ if (!buffer_id) {
+ ++page;
+ continue;
+ }
+ Buffer& buffer = slot_buffers[buffer_id];
+ const VAddr buf_start_addr = buffer.CpuAddr();
+ const VAddr buf_end_addr = buf_start_addr + buffer.SizeBytes();
+ if (buf_start_addr < end_addr && addr < buf_end_addr) {
+ return true;
+ }
+ page = Common::DivCeil(end_addr, PAGE_SIZE);
+ }
+ return false;
+}
+
+template <class P>
bool BufferCache<P>::IsRegionCpuModified(VAddr addr, size_t size) {
const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE);
for (u64 page = addr >> PAGE_BITS; page < page_end;) {
@@ -1425,6 +1486,7 @@ void BufferCache<P>::DownloadBufferMemory(Buffer& buffer, VAddr cpu_addr, u64 si
const VAddr end_address = start_address + range_size;
ForEachWrittenRange(start_address, range_size, add_download);
const IntervalType subtract_interval{start_address, end_address};
+ ClearDownload(subtract_interval);
common_ranges.subtract(subtract_interval);
});
if (total_size_bytes == 0) {
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index 24481952b..c51776466 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -4,6 +4,7 @@
#include "common/assert.h"
#include "common/logging/log.h"
+#include "common/microprofile.h"
#include "common/settings.h"
#include "core/core.h"
#include "video_core/engines/maxwell_3d.h"
@@ -12,6 +13,9 @@
#include "video_core/renderer_base.h"
#include "video_core/textures/decoders.h"
+MICROPROFILE_DECLARE(GPU_DMAEngine);
+MICROPROFILE_DEFINE(GPU_DMAEngine, "GPU", "DMA Engine", MP_RGB(224, 224, 128));
+
namespace Tegra::Engines {
using namespace Texture;
@@ -43,6 +47,7 @@ void MaxwellDMA::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
}
void MaxwellDMA::Launch() {
+ MICROPROFILE_SCOPE(GPU_DMAEngine);
LOG_TRACE(Render_OpenGL, "DMA copy 0x{:x} -> 0x{:x}", static_cast<GPUVAddr>(regs.offset_in),
static_cast<GPUVAddr>(regs.offset_out));
@@ -87,9 +92,11 @@ void MaxwellDMA::CopyPitchToPitch() {
// TODO: allow multisized components.
if (is_buffer_clear) {
ASSERT(regs.remap_const.component_size_minus_one == 3);
+ accelerate.BufferClear(regs.offset_out, regs.line_length_in, regs.remap_consta_value);
std::vector<u32> tmp_buffer(regs.line_length_in, regs.remap_consta_value);
- memory_manager.WriteBlock(regs.offset_out, reinterpret_cast<u8*>(tmp_buffer.data()),
- regs.line_length_in * sizeof(u32));
+ memory_manager.WriteBlockUnsafe(regs.offset_out,
+ reinterpret_cast<u8*>(tmp_buffer.data()),
+ regs.line_length_in * sizeof(u32));
return;
}
UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0);
@@ -179,8 +186,13 @@ void MaxwellDMA::CopyPitchToBlockLinear() {
write_buffer.resize(dst_size);
}
- memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
- memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
+ if (Settings::IsGPULevelExtreme()) {
+ memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
+ memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
+ } else {
+ memory_manager.ReadBlockUnsafe(regs.offset_in, read_buffer.data(), src_size);
+ memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size);
+ }
// If the input is linear and the output is tiled, swizzle the input and copy it over.
if (regs.dst_params.block_size.depth > 0) {
diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h
index 4ed0d0996..d3329b0f8 100644
--- a/src/video_core/engines/maxwell_dma.h
+++ b/src/video_core/engines/maxwell_dma.h
@@ -31,6 +31,8 @@ class AccelerateDMAInterface {
public:
/// Write the value to the register identified by method.
virtual bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) = 0;
+
+ virtual bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) = 0;
};
/**
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
index c225d1fc9..c4189fb60 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
@@ -98,6 +98,12 @@ void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer,
}
}
+void BufferCacheRuntime::ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value) {
+ glClearNamedBufferSubData(dest_buffer.Handle(), GL_R32UI, static_cast<GLintptr>(offset),
+ static_cast<GLsizeiptr>(size / sizeof(u32)), GL_RGBA, GL_UNSIGNED_INT,
+ &value);
+}
+
void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) {
if (has_unified_vertex_buffers) {
buffer.MakeResident(GL_READ_ONLY);
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h
index d8b20a9af..fe91aa452 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.h
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.h
@@ -57,6 +57,8 @@ public:
void CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer,
std::span<const VideoCommon::BufferCopy> copies);
+ void ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value);
+
void BindIndexBuffer(Buffer& buffer, u32 offset, u32 size);
void BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride);
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 82c84127a..ceb3abcb2 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -1407,4 +1407,9 @@ bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64
return buffer_cache.DMACopy(src_address, dest_address, amount);
}
+bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) {
+ std::scoped_lock lock{buffer_cache.mutex};
+ return buffer_cache.DMAClear(src_address, amount, value);
+}
+
} // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index ccee9ba33..d30ad698f 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -65,6 +65,8 @@ public:
bool BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) override;
+ bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override;
+
private:
BufferCache& buffer_cache;
};
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h
index 25fe61566..cf3b789e3 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.h
+++ b/src/video_core/renderer_opengl/gl_texture_cache.h
@@ -122,7 +122,7 @@ private:
bool has_broken_texture_view_formats = false;
StagingBuffers upload_buffers{GL_MAP_WRITE_BIT, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT};
- StagingBuffers download_buffers{GL_MAP_READ_BIT, GL_MAP_READ_BIT};
+ StagingBuffers download_buffers{GL_MAP_READ_BIT | GL_CLIENT_STORAGE_BIT, GL_MAP_READ_BIT};
OGLTexture null_image_1d_array;
OGLTexture null_image_cube_array;
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
index 0df4e1a1c..0def1e769 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
@@ -136,6 +136,30 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
});
}
+void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value) {
+ static constexpr VkMemoryBarrier READ_BARRIER{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
+ };
+ static constexpr VkMemoryBarrier WRITE_BARRIER{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+ };
+
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([dest_buffer, offset, size, value](vk::CommandBuffer cmdbuf) {
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+ 0, READ_BARRIER);
+ cmdbuf.FillBuffer(dest_buffer, offset, size, value);
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+ 0, WRITE_BARRIER);
+ });
+}
+
void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format,
u32 base_vertex, u32 num_indices, VkBuffer buffer,
u32 offset, [[maybe_unused]] u32 size) {
@@ -152,8 +176,8 @@ void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat
}
if (vk_buffer == VK_NULL_HANDLE) {
// Vulkan doesn't support null index buffers. Replace it with our own null buffer.
- ReserveNullIndexBuffer();
- vk_buffer = *null_index_buffer;
+ ReserveNullBuffer();
+ vk_buffer = *null_buffer;
}
scheduler.Record([vk_buffer, vk_offset, vk_index_type](vk::CommandBuffer cmdbuf) {
cmdbuf.BindIndexBuffer(vk_buffer, vk_offset, vk_index_type);
@@ -161,6 +185,13 @@ void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat
}
void BufferCacheRuntime::BindQuadArrayIndexBuffer(u32 first, u32 count) {
+ if (count == 0) {
+ ReserveNullBuffer();
+ scheduler.Record([this](vk::CommandBuffer cmdbuf) {
+ cmdbuf.BindIndexBuffer(*null_buffer, 0, VK_INDEX_TYPE_UINT32);
+ });
+ return;
+ }
ReserveQuadArrayLUT(first + count, true);
// The LUT has the indices 0, 1, 2, and 3 copied as an array
@@ -195,6 +226,14 @@ void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, VkBuffer buffer,
// Already logged in the rasterizer
return;
}
+ if (buffer == VK_NULL_HANDLE) {
+ // Vulkan doesn't support null transform feedback buffers.
+ // Replace it with our own null buffer.
+ ReserveNullBuffer();
+ buffer = *null_buffer;
+ offset = 0;
+ size = 0;
+ }
scheduler.Record([index, buffer, offset, size](vk::CommandBuffer cmdbuf) {
const VkDeviceSize vk_offset = offset;
const VkDeviceSize vk_size = size;
@@ -279,11 +318,11 @@ void BufferCacheRuntime::ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle
});
}
-void BufferCacheRuntime::ReserveNullIndexBuffer() {
- if (null_index_buffer) {
+void BufferCacheRuntime::ReserveNullBuffer() {
+ if (null_buffer) {
return;
}
- null_index_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
+ null_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
@@ -294,12 +333,12 @@ void BufferCacheRuntime::ReserveNullIndexBuffer() {
.pQueueFamilyIndices = nullptr,
});
if (device.HasDebuggingToolAttached()) {
- null_index_buffer.SetObjectNameEXT("Null index buffer");
+ null_buffer.SetObjectNameEXT("Null index buffer");
}
- null_index_buffer_commit = memory_allocator.Commit(null_index_buffer, MemoryUsage::DeviceLocal);
+ null_buffer_commit = memory_allocator.Commit(null_buffer, MemoryUsage::DeviceLocal);
scheduler.RequestOutsideRenderPassOperationContext();
- scheduler.Record([buffer = *null_index_buffer](vk::CommandBuffer cmdbuf) {
+ scheduler.Record([buffer = *null_buffer](vk::CommandBuffer cmdbuf) {
cmdbuf.FillBuffer(buffer, 0, VK_WHOLE_SIZE, 0);
});
}
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h
index 982e92191..3bb81d5b3 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.h
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h
@@ -60,6 +60,8 @@ public:
void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer,
std::span<const VideoCommon::BufferCopy> copies);
+ void ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value);
+
void BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 num_indices,
u32 base_vertex, VkBuffer buffer, u32 offset, u32 size);
@@ -92,7 +94,7 @@ private:
void ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle);
- void ReserveNullIndexBuffer();
+ void ReserveNullBuffer();
const Device& device;
MemoryAllocator& memory_allocator;
@@ -105,8 +107,8 @@ private:
VkIndexType quad_array_lut_index_type{};
u32 current_num_indices = 0;
- vk::Buffer null_index_buffer;
- MemoryCommit null_index_buffer_commit;
+ vk::Buffer null_buffer;
+ MemoryCommit null_buffer_commit;
Uint8Pass uint8_pass;
QuadIndexedPass quad_index_pass;
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
index 205cd3b05..4181d83ee 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@@ -374,20 +374,20 @@ void ASTCDecoderPass::MakeDataBuffer() {
scheduler.Record([src = staging_ref.buffer, offset = staging_ref.offset, dst = *data_buffer,
TOTAL_BUFFER_SIZE](vk::CommandBuffer cmdbuf) {
- cmdbuf.CopyBuffer(src, dst,
- VkBufferCopy{
- .srcOffset = offset,
- .dstOffset = 0,
- .size = TOTAL_BUFFER_SIZE,
- });
- cmdbuf.PipelineBarrier(
- VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
- VkMemoryBarrier{
- .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
- .pNext = nullptr,
- .srcAccessMask = 0,
- .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
- });
+ static constexpr VkMemoryBarrier write_barrier{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+ };
+ const VkBufferCopy copy{
+ .srcOffset = offset,
+ .dstOffset = 0,
+ .size = TOTAL_BUFFER_SIZE,
+ };
+ cmdbuf.CopyBuffer(src, dst, copy);
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+ 0, write_barrier);
});
}
@@ -411,7 +411,7 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
const VkImageMemoryBarrier image_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
- .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+ .srcAccessMask = is_initialized ? VK_ACCESS_SHADER_WRITE_BIT : VkAccessFlags{},
.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
.oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
@@ -426,7 +426,8 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
};
- cmdbuf.PipelineBarrier(is_initialized ? VK_PIPELINE_STAGE_ALL_COMMANDS_BIT : 0,
+ cmdbuf.PipelineBarrier(is_initialized ? VK_PIPELINE_STAGE_ALL_COMMANDS_BIT
+ : VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline);
});
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index a8ffbe6ba..f57c15b37 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -706,6 +706,11 @@ void RasterizerVulkan::FlushWork() {
AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {}
+bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) {
+ std::scoped_lock lock{buffer_cache.mutex};
+ return buffer_cache.DMAClear(src_address, amount, value);
+}
+
bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) {
std::scoped_lock lock{buffer_cache.mutex};
return buffer_cache.DMACopy(src_address, dest_address, amount);
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index 3a78de258..2065209be 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -56,6 +56,8 @@ public:
bool BufferCopy(GPUVAddr start_address, GPUVAddr end_address, u64 amount) override;
+ bool BufferClear(GPUVAddr src_address, u64 amount, u32 value) override;
+
private:
BufferCache& buffer_cache;
};
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index a2ab4d1ee..fd01c902c 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -608,7 +608,10 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
const VkImageAspectFlags aspect_mask = ImageAspectMask(src.format);
const bool is_dst_msaa = dst.Samples() != VK_SAMPLE_COUNT_1_BIT;
const bool is_src_msaa = src.Samples() != VK_SAMPLE_COUNT_1_BIT;
- ASSERT(aspect_mask == ImageAspectMask(dst.format));
+ if (aspect_mask != ImageAspectMask(dst.format)) {
+ UNIMPLEMENTED_MSG("Incompatible blit from format {} to {}", src.format, dst.format);
+ return;
+ }
if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT && !is_src_msaa && !is_dst_msaa) {
blit_image_helper.BlitColor(dst_framebuffer, src, dst_region, src_region, filter,
operation);
@@ -911,6 +914,7 @@ void Image::UploadMemory(const StagingBufferRef& map,
void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask);
+ scheduler->RequestOutsideRenderPassOperationContext();
scheduler->Record([buffer = map.buffer, image = *image, aspect_mask = aspect_mask,
vk_copies](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier read_barrier{
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 01de2d498..85ce06d56 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -599,6 +599,12 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) {
using namespace VideoCommon::Dirty;
auto& flags = maxwell3d.dirty.flags;
if (!flags[Dirty::RenderTargets]) {
+ for (size_t index = 0; index < NUM_RT; ++index) {
+ ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index];
+ PrepareImageView(color_buffer_id, true, is_clear && IsFullClear(color_buffer_id));
+ }
+ const ImageViewId depth_buffer_id = render_targets.depth_buffer_id;
+ PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id));
return;
}
flags[Dirty::RenderTargets] = false;
diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp
index c872517b8..59cf2f561 100644
--- a/src/video_core/texture_cache/util.cpp
+++ b/src/video_core/texture_cache/util.cpp
@@ -169,23 +169,6 @@ template <u32 GOB_EXTENT>
return Common::DivCeil(AdjustMipSize(size, level), block_size);
}
-[[nodiscard]] constexpr std::pair<int, int> Samples(int num_samples) {
- switch (num_samples) {
- case 1:
- return {1, 1};
- case 2:
- return {2, 1};
- case 4:
- return {2, 2};
- case 8:
- return {4, 2};
- case 16:
- return {4, 4};
- }
- UNREACHABLE_MSG("Invalid number of samples={}", num_samples);
- return {1, 1};
-}
-
[[nodiscard]] constexpr Extent2D DefaultBlockSize(PixelFormat format) {
return {DefaultBlockWidth(format), DefaultBlockHeight(format)};
}
@@ -283,14 +266,13 @@ template <u32 GOB_EXTENT>
}
[[nodiscard]] constexpr LevelInfo MakeLevelInfo(PixelFormat format, Extent3D size, Extent3D block,
- u32 num_samples, u32 tile_width_spacing) {
- const auto [samples_x, samples_y] = Samples(num_samples);
+ u32 tile_width_spacing) {
const u32 bytes_per_block = BytesPerBlock(format);
return {
.size =
{
- .width = size.width * samples_x,
- .height = size.height * samples_y,
+ .width = size.width,
+ .height = size.height,
.depth = size.depth,
},
.block = block,
@@ -301,14 +283,12 @@ template <u32 GOB_EXTENT>
}
[[nodiscard]] constexpr LevelInfo MakeLevelInfo(const ImageInfo& info) {
- return MakeLevelInfo(info.format, info.size, info.block, info.num_samples,
- info.tile_width_spacing);
+ return MakeLevelInfo(info.format, info.size, info.block, info.tile_width_spacing);
}
[[nodiscard]] constexpr u32 CalculateLevelOffset(PixelFormat format, Extent3D size, Extent3D block,
- u32 num_samples, u32 tile_width_spacing,
- u32 level) {
- const LevelInfo info = MakeLevelInfo(format, size, block, num_samples, tile_width_spacing);
+ u32 tile_width_spacing, u32 level) {
+ const LevelInfo info = MakeLevelInfo(format, size, block, tile_width_spacing);
u32 offset = 0;
for (u32 current_level = 0; current_level < level; ++current_level) {
offset += CalculateLevelSize(info, current_level);
@@ -645,8 +625,8 @@ u32 CalculateLayerStride(const ImageInfo& info) noexcept {
u32 CalculateLayerSize(const ImageInfo& info) noexcept {
ASSERT(info.type != ImageType::Linear);
- return CalculateLevelOffset(info.format, info.size, info.block, info.num_samples,
- info.tile_width_spacing, info.resources.levels);
+ return CalculateLevelOffset(info.format, info.size, info.block, info.tile_width_spacing,
+ info.resources.levels);
}
LevelArray CalculateMipLevelOffsets(const ImageInfo& info) noexcept {
@@ -1195,37 +1175,37 @@ static_assert(CalculateLevelSize(LevelInfo{{1920, 1080, 1}, {0, 2, 0}, {1, 1}, 2
0x7f8000);
static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x4000);
-static_assert(CalculateLevelOffset(PixelFormat::R8_SINT, {1920, 1080, 1}, {0, 2, 0}, 1, 0, 7) ==
+static_assert(CalculateLevelOffset(PixelFormat::R8_SINT, {1920, 1080, 1}, {0, 2, 0}, 0, 7) ==
0x2afc00);
-static_assert(CalculateLevelOffset(PixelFormat::ASTC_2D_12X12_UNORM, {8192, 4096, 1}, {0, 2, 0}, 1,
- 0, 12) == 0x50d200);
-
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 0) == 0);
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 1) == 0x400000);
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 2) == 0x500000);
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 3) == 0x540000);
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 4) == 0x550000);
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 5) == 0x554000);
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 6) == 0x555000);
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 7) == 0x555400);
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 8) == 0x555600);
-static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0,
- 9) == 0x555800);
+static_assert(CalculateLevelOffset(PixelFormat::ASTC_2D_12X12_UNORM, {8192, 4096, 1}, {0, 2, 0}, 0,
+ 12) == 0x50d200);
+
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 0) ==
+ 0);
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 1) ==
+ 0x400000);
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 2) ==
+ 0x500000);
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 3) ==
+ 0x540000);
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 4) ==
+ 0x550000);
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 5) ==
+ 0x554000);
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 6) ==
+ 0x555000);
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 7) ==
+ 0x555400);
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 8) ==
+ 0x555600);
+static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 0, 9) ==
+ 0x555800);
constexpr u32 ValidateLayerSize(PixelFormat format, u32 width, u32 height, u32 block_height,
u32 tile_width_spacing, u32 level) {
const Extent3D size{width, height, 1};
const Extent3D block{0, block_height, 0};
- const u32 offset = CalculateLevelOffset(format, size, block, 1, tile_width_spacing, level);
+ const u32 offset = CalculateLevelOffset(format, size, block, tile_width_spacing, level);
return AlignLayerSize(offset, size, block, DefaultBlockHeight(format), tile_width_spacing);
}
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-10 01:53:20 +0100