hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 3927

Browse Source
This commit is contained in:
pineappleEA
2023-10-12 19:22:22 +02:00
parent 91ade31ddd
commit 7fe7262395
29 changed files with 330 additions and 106 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/video_core/buffer_cache/buffer_cache_base.h
View File

@@ -62,7 +62,11 @@ using BufferId = SlotId;
using VideoCore::Surface::PixelFormat;
using namespace Common::Literals;
#ifdef __APPLE__
constexpr u32 NUM_VERTEX_BUFFERS = 16;
#else
constexpr u32 NUM_VERTEX_BUFFERS = 32;
#endif
constexpr u32 NUM_TRANSFORM_FEEDBACK_BUFFERS = 4;
constexpr u32 NUM_GRAPHICS_UNIFORM_BUFFERS = 18;
constexpr u32 NUM_COMPUTE_UNIFORM_BUFFERS = 8;

8
src/video_core/host_shaders/convert_d24s8_to_abgr8.frag
View File

@@ -3,16 +3,16 @@
#version 450
precision mediump int;
precision highp float;
layout(binding = 0) uniform sampler2D depth_tex;
layout(binding = 1) uniform isampler2D stencil_tex;
layout(binding = 1) uniform usampler2D stencil_tex;
layout(location = 0) out vec4 color;
void main() {
ivec2 coord = ivec2(gl_FragCoord.xy);
uint depth = uint(textureLod(depth_tex, coord, 0).r * (exp2(24.0) - 1.0f));
uint stencil = uint(textureLod(stencil_tex, coord, 0).r);
highp uint depth_val =
uint(textureLod(depth_tex, coord, 0).r * (exp2(32.0) - 1.0));
lowp uint stencil_val = textureLod(stencil_tex, coord, 0).r;

8
src/video_core/host_shaders/convert_s8d24_to_abgr8.frag
View File

@@ -3,16 +3,16 @@
#version 450
precision mediump int;
precision highp float;
layout(binding = 0) uniform sampler2D depth_tex;
layout(binding = 1) uniform isampler2D stencil_tex;
layout(binding = 1) uniform usampler2D stencil_tex;
layout(location = 0) out vec4 color;
void main() {
ivec2 coord = ivec2(gl_FragCoord.xy);
uint depth = uint(textureLod(depth_tex, coord, 0).r * (exp2(24.0) - 1.0f));
uint stencil = uint(textureLod(stencil_tex, coord, 0).r);
highp uint depth_val =
uint(textureLod(depth_tex, coord, 0).r * (exp2(32.0) - 1.0));
lowp uint stencil_val = textureLod(stencil_tex, coord, 0).r;

1
src/video_core/renderer_vulkan/vk_query_cache.cpp
View File

@@ -1436,6 +1436,7 @@ void QueryCacheRuntime::Barriers(bool is_prebarrier) {
.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
};
impl->scheduler.RequestOutsideRenderPassOperationContext();
if (is_prebarrier) {
impl->scheduler.Record([](vk::CommandBuffer cmdbuf) {
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,

13
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@@ -975,6 +975,19 @@ void RasterizerVulkan::UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs
if (!state_tracker.TouchScissors()) {
return;
}
if (!regs.viewport_scale_offset_enabled) {
const auto x = static_cast<float>(regs.surface_clip.x);
const auto y = static_cast<float>(regs.surface_clip.y);
const auto width = static_cast<float>(regs.surface_clip.width);
const auto height = static_cast<float>(regs.surface_clip.height);
VkRect2D scissor;
scissor.offset.x = static_cast<u32>(x);
scissor.offset.y = static_cast<u32>(y);
scissor.extent.width = static_cast<u32>(width != 0.0f ? width : 1.0f);
scissor.extent.height = static_cast<u32>(height != 0.0f ? height : 1.0f);
scheduler.Record([scissor](vk::CommandBuffer cmdbuf) { cmdbuf.SetScissor(0, scissor); });
return;
}
u32 up_scale = 1;
u32 down_shift = 0;
if (texture_cache.IsRescaling()) {

2
src/video_core/renderer_vulkan/vk_render_pass_cache.cpp
View File

@@ -19,7 +19,7 @@ VkAttachmentDescription AttachmentDescription(const Device& device, PixelFormat
VkSampleCountFlagBits samples) {
using MaxwellToVK::SurfaceFormat;
return {
.flags = VK_ATTACHMENT_DESCRIPTION_MAY_ALIAS_BIT,
.flags = {},
.format = SurfaceFormat(device, FormatType::Optimal, true, format).format,
.samples = samples,
.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,

8
src/video_core/texture_cache/formatter.cpp
View File

@@ -10,19 +10,23 @@
#include "video_core/texture_cache/image_info.h"
#include "video_core/texture_cache/image_view_base.h"
#include "video_core/texture_cache/render_targets.h"
#include "video_core/texture_cache/samples_helper.h"
namespace VideoCommon {
std::string Name(const ImageBase& image) {
const GPUVAddr gpu_addr = image.gpu_addr;
const ImageInfo& info = image.info;
const u32 width = info.size.width;
const u32 height = info.size.height;
u32 width = info.size.width;
u32 height = info.size.height;
const u32 depth = info.size.depth;
const u32 num_layers = image.info.resources.layers;
const u32 num_levels = image.info.resources.levels;
std::string resource;
if (image.info.num_samples > 1) {
const auto [samples_x, samples_y] = VideoCommon::SamplesLog2(image.info.num_samples);
width >>= samples_x;
height >>= samples_y;
resource += fmt::format(":{}xMSAA", image.info.num_samples);
}
if (num_layers > 1) {

2
src/video_core/texture_cache/samples_helper.h
View File

@@ -24,7 +24,7 @@ namespace VideoCommon {
return {2, 2};
}
ASSERT_MSG(false, "Invalid number of samples={}", num_samples);
return {1, 1};
return {0, 0};
}
[[nodiscard]] inline int NumSamples(Tegra::Texture::MsaaMode msaa_mode) {

41
src/video_core/texture_cache/util.cpp
View File

@@ -68,7 +68,6 @@ struct LevelInfo {
Extent2D tile_size;
u32 bpp_log2;
u32 tile_width_spacing;
u32 num_levels;
};
[[nodiscard]] constexpr u32 AdjustTileSize(u32 shift, u32 unit_factor, u32 dimension) {
@@ -119,11 +118,11 @@ template <u32 GOB_EXTENT>
}
[[nodiscard]] constexpr Extent3D AdjustMipBlockSize(Extent3D num_tiles, Extent3D block_size,
u32 level, u32 num_levels) {
u32 level) {
return {
.width = AdjustMipBlockSize<GOB_SIZE_X>(num_tiles.width, block_size.width, level),
.height = AdjustMipBlockSize<GOB_SIZE_Y>(num_tiles.height, block_size.height, level),
.depth = level == 0 && num_levels == 1
.depth = level == 0
? block_size.depth
: AdjustMipBlockSize<GOB_SIZE_Z>(num_tiles.depth, block_size.depth, level),
};
@@ -167,6 +166,13 @@ template <u32 GOB_EXTENT>
}
[[nodiscard]] constexpr Extent3D TileShift(const LevelInfo& info, u32 level) {
if (level == 0) {
return Extent3D{
.width = info.block.width,
.height = info.block.height,
.depth = info.block.depth,
};
}
const Extent3D blocks = NumLevelBlocks(info, level);
return Extent3D{
.width = AdjustTileSize(info.block.width, GOB_SIZE_X, blocks.width),
@@ -251,7 +257,7 @@ template <u32 GOB_EXTENT>
}
[[nodiscard]] constexpr LevelInfo MakeLevelInfo(PixelFormat format, Extent3D size, Extent3D block,
u32 tile_width_spacing, u32 num_levels) {
u32 tile_width_spacing) {
const u32 bytes_per_block = BytesPerBlock(format);
return {
.size =
@@ -264,18 +270,16 @@ template <u32 GOB_EXTENT>
.tile_size = DefaultBlockSize(format),
.bpp_log2 = BytesPerBlockLog2(bytes_per_block),
.tile_width_spacing = tile_width_spacing,
.num_levels = num_levels,
};
}
[[nodiscard]] constexpr LevelInfo MakeLevelInfo(const ImageInfo& info) {
return MakeLevelInfo(info.format, info.size, info.block, info.tile_width_spacing,
info.resources.levels);
return MakeLevelInfo(info.format, info.size, info.block, info.tile_width_spacing);
}
[[nodiscard]] constexpr u32 CalculateLevelOffset(PixelFormat format, Extent3D size, Extent3D block,
u32 tile_width_spacing, u32 level) {
const LevelInfo info = MakeLevelInfo(format, size, block, tile_width_spacing, 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);
@@ -462,7 +466,7 @@ template <u32 GOB_EXTENT>
};
const u32 bpp_log2 = BytesPerBlockLog2(info.format);
const u32 alignment = StrideAlignment(num_tiles, info.block, bpp_log2, info.tile_width_spacing);
const Extent3D mip_block = AdjustMipBlockSize(num_tiles, info.block, 0, info.resources.levels);
const Extent3D mip_block = AdjustMipBlockSize(num_tiles, info.block, 0);
return Extent3D{
.width = Common::AlignUpLog2(num_tiles.width, alignment),
.height = Common::AlignUpLog2(num_tiles.height, GOB_SIZE_Y_SHIFT + mip_block.height),
@@ -529,8 +533,7 @@ void SwizzleBlockLinearImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr
UNIMPLEMENTED_IF(copy.image_extent != level_size);
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
const Extent3D block =
AdjustMipBlockSize(num_tiles, level_info.block, level, level_info.num_levels);
const Extent3D block = AdjustMipBlockSize(num_tiles, level_info.block, level);
size_t host_offset = copy.buffer_offset;
@@ -695,7 +698,7 @@ u32 CalculateLevelStrideAlignment(const ImageInfo& info, u32 level) {
const Extent2D tile_size = DefaultBlockSize(info.format);
const Extent3D level_size = AdjustMipSize(info.size, level);
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
const Extent3D block = AdjustMipBlockSize(num_tiles, info.block, level, info.resources.levels);
const Extent3D block = AdjustMipBlockSize(num_tiles, info.block, level);
const u32 bpp_log2 = BytesPerBlockLog2(info.format);
return StrideAlignment(num_tiles, block, bpp_log2, info.tile_width_spacing);
}
@@ -884,8 +887,7 @@ boost::container::small_vector<BufferImageCopy, 16> UnswizzleImage(Tegra::Memory
.image_extent = level_size,
};
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
const Extent3D block =
AdjustMipBlockSize(num_tiles, level_info.block, level, level_info.num_levels);
const Extent3D block = AdjustMipBlockSize(num_tiles, level_info.block, level);
const u32 stride_alignment = StrideAlignment(num_tiles, info.block, gob, bpp_log2);
size_t guest_layer_offset = 0;
@@ -1039,7 +1041,7 @@ Extent3D MipBlockSize(const ImageInfo& info, u32 level) {
const Extent2D tile_size = DefaultBlockSize(info.format);
const Extent3D level_size = AdjustMipSize(info.size, level);
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
return AdjustMipBlockSize(num_tiles, level_info.block, level, level_info.num_levels);
return AdjustMipBlockSize(num_tiles, level_info.block, level);
}
boost::container::small_vector<SwizzleParameters, 16> FullUploadSwizzles(const ImageInfo& info) {
@@ -1061,8 +1063,7 @@ boost::container::small_vector<SwizzleParameters, 16> FullUploadSwizzles(const I
for (s32 level = 0; level < num_levels; ++level) {
const Extent3D level_size = AdjustMipSize(size, level);
const Extent3D num_tiles = AdjustTileSize(level_size, tile_size);
const Extent3D block =
AdjustMipBlockSize(num_tiles, level_info.block, level, level_info.num_levels);
const Extent3D block = AdjustMipBlockSize(num_tiles, level_info.block, level);
params[level] = SwizzleParameters{
.num_tiles = num_tiles,
.block = block,
@@ -1291,11 +1292,11 @@ u32 MapSizeBytes(const ImageBase& image) {
}
}
static_assert(CalculateLevelSize(LevelInfo{{1920, 1080, 1}, {0, 2, 0}, {1, 1}, 2, 0, 1}, 0) ==
static_assert(CalculateLevelSize(LevelInfo{{1920, 1080, 1}, {0, 2, 0}, {1, 1}, 2, 0}, 0) ==
0x7f8000);
static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0, 1}, 0) == 0x4000);
static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x40000);
static_assert(CalculateLevelSize(LevelInfo{{128, 8, 1}, {0, 4, 0}, {1, 1}, 4, 0, 1}, 0) == 0x4000);
static_assert(CalculateLevelSize(LevelInfo{{128, 8, 1}, {0, 4, 0}, {1, 1}, 4, 0}, 0) == 0x40000);
static_assert(CalculateLevelOffset(PixelFormat::R8_SINT, {1920, 1080, 1}, {0, 2, 0}, 0, 7) ==
0x2afc00);