renderer_vulkan: Add depth buffer support
This commit is contained in:
parent
12f1743d5f
commit
17cb4e0132
|
@ -10,10 +10,8 @@
|
||||||
#include "video_core/amdgpu/pixel_format.h"
|
#include "video_core/amdgpu/pixel_format.h"
|
||||||
|
|
||||||
#include <array>
|
#include <array>
|
||||||
#include <condition_variable>
|
|
||||||
#include <coroutine>
|
#include <coroutine>
|
||||||
#include <functional>
|
#include <mutex>
|
||||||
#include <future>
|
|
||||||
#include <span>
|
#include <span>
|
||||||
#include <thread>
|
#include <thread>
|
||||||
#include <queue>
|
#include <queue>
|
||||||
|
@ -333,6 +331,14 @@ struct Liverpool {
|
||||||
u32 Height() const {
|
u32 Height() const {
|
||||||
return (depth_size.height_tile_max + 1) << 3;
|
return (depth_size.height_tile_max + 1) << 3;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
u64 Address() const {
|
||||||
|
return u64(z_read_base) << 8;
|
||||||
|
}
|
||||||
|
|
||||||
|
[[nodiscard]] size_t GetSizeAligned() const {
|
||||||
|
return depth_slice.tile_max * 8;
|
||||||
|
}
|
||||||
};
|
};
|
||||||
|
|
||||||
enum class ClipSpace : u32 {
|
enum class ClipSpace : u32 {
|
||||||
|
@ -717,6 +723,14 @@ struct Liverpool {
|
||||||
CbColor7Base = 0xA381,
|
CbColor7Base = 0xA381,
|
||||||
};
|
};
|
||||||
|
|
||||||
|
struct PolygonOffset {
|
||||||
|
float depth_bias;
|
||||||
|
float front_scale;
|
||||||
|
float front_offset;
|
||||||
|
float back_scale;
|
||||||
|
float back_offset;
|
||||||
|
};
|
||||||
|
|
||||||
union Regs {
|
union Regs {
|
||||||
struct {
|
struct {
|
||||||
INSERT_PADDING_WORDS(0x2C08);
|
INSERT_PADDING_WORDS(0x2C08);
|
||||||
|
@ -726,8 +740,8 @@ struct Liverpool {
|
||||||
INSERT_PADDING_WORDS(0x2E00 - 0x2C4C - 16);
|
INSERT_PADDING_WORDS(0x2E00 - 0x2C4C - 16);
|
||||||
ComputeProgram cs_program;
|
ComputeProgram cs_program;
|
||||||
INSERT_PADDING_WORDS(0xA008 - 0x2E00 - 80);
|
INSERT_PADDING_WORDS(0xA008 - 0x2E00 - 80);
|
||||||
u32 depth_bounds_min;
|
float depth_bounds_min;
|
||||||
u32 depth_bounds_max;
|
float depth_bounds_max;
|
||||||
u32 stencil_clear;
|
u32 stencil_clear;
|
||||||
u32 depth_clear;
|
u32 depth_clear;
|
||||||
Scissor screen_scissor;
|
Scissor screen_scissor;
|
||||||
|
@ -776,7 +790,9 @@ struct Liverpool {
|
||||||
IndexBufferType index_buffer_type;
|
IndexBufferType index_buffer_type;
|
||||||
INSERT_PADDING_WORDS(0xA2A1 - 0xA29E - 2);
|
INSERT_PADDING_WORDS(0xA2A1 - 0xA29E - 2);
|
||||||
u32 enable_primitive_id;
|
u32 enable_primitive_id;
|
||||||
INSERT_PADDING_WORDS(0xA318 - 0xA2A1 - 1);
|
INSERT_PADDING_WORDS(0xA2DF - 0xA2A1 - 1);
|
||||||
|
PolygonOffset poly_offset;
|
||||||
|
INSERT_PADDING_WORDS(0xA318 - 0xA2DF - 5);
|
||||||
ColorBuffer color_buffers[NumColorBuffers];
|
ColorBuffer color_buffers[NumColorBuffers];
|
||||||
INSERT_PADDING_WORDS(0xC242 - 0xA390);
|
INSERT_PADDING_WORDS(0xC242 - 0xA390);
|
||||||
PrimitiveType primitive_type;
|
PrimitiveType primitive_type;
|
||||||
|
@ -930,6 +946,7 @@ static_assert(GFX6_3D_REG_INDEX(viewport_control) == 0xA206);
|
||||||
static_assert(GFX6_3D_REG_INDEX(vs_output_control) == 0xA207);
|
static_assert(GFX6_3D_REG_INDEX(vs_output_control) == 0xA207);
|
||||||
static_assert(GFX6_3D_REG_INDEX(index_buffer_type) == 0xA29F);
|
static_assert(GFX6_3D_REG_INDEX(index_buffer_type) == 0xA29F);
|
||||||
static_assert(GFX6_3D_REG_INDEX(enable_primitive_id) == 0xA2A1);
|
static_assert(GFX6_3D_REG_INDEX(enable_primitive_id) == 0xA2A1);
|
||||||
|
static_assert(GFX6_3D_REG_INDEX(poly_offset) == 0xA2DF);
|
||||||
static_assert(GFX6_3D_REG_INDEX(color_buffers[0].base_address) == 0xA318);
|
static_assert(GFX6_3D_REG_INDEX(color_buffers[0].base_address) == 0xA318);
|
||||||
static_assert(GFX6_3D_REG_INDEX(color_buffers[0].pitch) == 0xA319);
|
static_assert(GFX6_3D_REG_INDEX(color_buffers[0].pitch) == 0xA319);
|
||||||
static_assert(GFX6_3D_REG_INDEX(color_buffers[0].slice) == 0xA31A);
|
static_assert(GFX6_3D_REG_INDEX(color_buffers[0].slice) == 0xA31A);
|
||||||
|
|
|
@ -64,8 +64,10 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
|
||||||
.pVertexAttributeDescriptions = attributes.data(),
|
.pVertexAttributeDescriptions = attributes.data(),
|
||||||
};
|
};
|
||||||
|
|
||||||
ASSERT_MSG(key.prim_type != Liverpool::PrimitiveType::RectList || IsEmbeddedVs(),
|
if (key.prim_type == Liverpool::PrimitiveType::RectList && !IsEmbeddedVs()) {
|
||||||
|
LOG_WARNING(Render_Vulkan,
|
||||||
"Rectangle List primitive type is only supported for embedded VS");
|
"Rectangle List primitive type is only supported for embedded VS");
|
||||||
|
}
|
||||||
|
|
||||||
const vk::PipelineInputAssemblyStateCreateInfo input_assembly = {
|
const vk::PipelineInputAssemblyStateCreateInfo input_assembly = {
|
||||||
.topology = LiverpoolToVK::PrimitiveType(key.prim_type),
|
.topology = LiverpoolToVK::PrimitiveType(key.prim_type),
|
||||||
|
@ -76,11 +78,14 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
|
||||||
.depthClampEnable = false,
|
.depthClampEnable = false,
|
||||||
.rasterizerDiscardEnable = false,
|
.rasterizerDiscardEnable = false,
|
||||||
.polygonMode = LiverpoolToVK::PolygonMode(key.polygon_mode),
|
.polygonMode = LiverpoolToVK::PolygonMode(key.polygon_mode),
|
||||||
.cullMode = vk::CullModeFlagBits::eNone, /*LiverpoolToVK::CullMode(key.cull_mode),*/
|
.cullMode = LiverpoolToVK::CullMode(key.cull_mode),
|
||||||
.frontFace = key.front_face == Liverpool::FrontFace::Clockwise
|
.frontFace = key.front_face == Liverpool::FrontFace::Clockwise
|
||||||
? vk::FrontFace::eClockwise
|
? vk::FrontFace::eClockwise
|
||||||
: vk::FrontFace::eCounterClockwise,
|
: vk::FrontFace::eCounterClockwise,
|
||||||
.depthBiasEnable = false,
|
.depthBiasEnable = bool(key.depth_bias_enable),
|
||||||
|
.depthBiasConstantFactor = key.depth_bias_const_factor,
|
||||||
|
.depthBiasClamp = key.depth_bias_clamp,
|
||||||
|
.depthBiasSlopeFactor = key.depth_bias_slope_factor,
|
||||||
.lineWidth = 1.0f,
|
.lineWidth = 1.0f,
|
||||||
};
|
};
|
||||||
|
|
||||||
|
@ -103,7 +108,12 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
|
||||||
.extent = {1, 1},
|
.extent = {1, 1},
|
||||||
};
|
};
|
||||||
|
|
||||||
|
const vk::PipelineViewportDepthClipControlCreateInfoEXT clip_control = {
|
||||||
|
.negativeOneToOne = key.clip_space == Liverpool::ClipSpace::MinusWToW,
|
||||||
|
};
|
||||||
|
|
||||||
const vk::PipelineViewportStateCreateInfo viewport_info = {
|
const vk::PipelineViewportStateCreateInfo viewport_info = {
|
||||||
|
.pNext = &clip_control,
|
||||||
.viewportCount = 1,
|
.viewportCount = 1,
|
||||||
.pViewports = &viewport,
|
.pViewports = &viewport,
|
||||||
.scissorCount = 1,
|
.scissorCount = 1,
|
||||||
|
@ -150,6 +160,8 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
|
||||||
.writeMask = key.stencil_ref_back.stencil_write_mask,
|
.writeMask = key.stencil_ref_back.stencil_write_mask,
|
||||||
.reference = key.stencil_ref_back.stencil_test_val,
|
.reference = key.stencil_ref_back.stencil_test_val,
|
||||||
},
|
},
|
||||||
|
.minDepthBounds = key.depth_bounds_min,
|
||||||
|
.maxDepthBounds = key.depth_bounds_max,
|
||||||
};
|
};
|
||||||
|
|
||||||
u32 shader_count = 1;
|
u32 shader_count = 1;
|
||||||
|
|
|
@ -32,6 +32,12 @@ struct GraphicsPipelineKey {
|
||||||
vk::Format depth_format;
|
vk::Format depth_format;
|
||||||
|
|
||||||
Liverpool::DepthControl depth;
|
Liverpool::DepthControl depth;
|
||||||
|
float depth_bounds_min;
|
||||||
|
float depth_bounds_max;
|
||||||
|
float depth_bias_const_factor;
|
||||||
|
float depth_bias_slope_factor;
|
||||||
|
float depth_bias_clamp;
|
||||||
|
u32 depth_bias_enable;
|
||||||
Liverpool::StencilControl stencil;
|
Liverpool::StencilControl stencil;
|
||||||
Liverpool::StencilRefMask stencil_ref_front;
|
Liverpool::StencilRefMask stencil_ref_front;
|
||||||
Liverpool::StencilRefMask stencil_ref_back;
|
Liverpool::StencilRefMask stencil_ref_back;
|
||||||
|
@ -39,7 +45,7 @@ struct GraphicsPipelineKey {
|
||||||
Liverpool::PolygonMode polygon_mode;
|
Liverpool::PolygonMode polygon_mode;
|
||||||
Liverpool::CullMode cull_mode;
|
Liverpool::CullMode cull_mode;
|
||||||
Liverpool::FrontFace front_face;
|
Liverpool::FrontFace front_face;
|
||||||
u32 pad{};
|
Liverpool::ClipSpace clip_space;
|
||||||
std::array<Liverpool::BlendControl, Liverpool::NumColorBuffers> blend_controls;
|
std::array<Liverpool::BlendControl, Liverpool::NumColorBuffers> blend_controls;
|
||||||
std::array<vk::ColorComponentFlags, Liverpool::NumColorBuffers> write_masks;
|
std::array<vk::ColorComponentFlags, Liverpool::NumColorBuffers> write_masks;
|
||||||
|
|
||||||
|
@ -47,7 +53,6 @@ struct GraphicsPipelineKey {
|
||||||
return std::memcmp(this, &key, sizeof(key)) == 0;
|
return std::memcmp(this, &key, sizeof(key)) == 0;
|
||||||
}
|
}
|
||||||
};
|
};
|
||||||
static_assert(std::has_unique_object_representations_v<GraphicsPipelineKey>);
|
|
||||||
|
|
||||||
class GraphicsPipeline {
|
class GraphicsPipeline {
|
||||||
public:
|
public:
|
||||||
|
|
|
@ -155,6 +155,8 @@ bool Instance::CreateDevice() {
|
||||||
custom_border_color = add_extension(VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
|
custom_border_color = add_extension(VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
|
||||||
add_extension(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
|
add_extension(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
|
||||||
add_extension(VK_KHR_MAINTENANCE_4_EXTENSION_NAME);
|
add_extension(VK_KHR_MAINTENANCE_4_EXTENSION_NAME);
|
||||||
|
add_extension(VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME);
|
||||||
|
add_extension(VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME);
|
||||||
// The next two extensions are required to be available together in order to support write masks
|
// The next two extensions are required to be available together in order to support write masks
|
||||||
color_write_en = add_extension(VK_EXT_COLOR_WRITE_ENABLE_EXTENSION_NAME);
|
color_write_en = add_extension(VK_EXT_COLOR_WRITE_ENABLE_EXTENSION_NAME);
|
||||||
color_write_en &= add_extension(VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME);
|
color_write_en &= add_extension(VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME);
|
||||||
|
@ -227,6 +229,9 @@ bool Instance::CreateDevice() {
|
||||||
vk::PhysicalDeviceExtendedDynamicState3FeaturesEXT{
|
vk::PhysicalDeviceExtendedDynamicState3FeaturesEXT{
|
||||||
.extendedDynamicState3ColorWriteMask = true,
|
.extendedDynamicState3ColorWriteMask = true,
|
||||||
},
|
},
|
||||||
|
vk::PhysicalDeviceDepthClipControlFeaturesEXT{
|
||||||
|
.depthClipControl = true,
|
||||||
|
},
|
||||||
};
|
};
|
||||||
|
|
||||||
if (!color_write_en) {
|
if (!color_write_en) {
|
||||||
|
|
|
@ -88,12 +88,26 @@ void PipelineCache::RefreshGraphicsKey() {
|
||||||
auto& key = graphics_key;
|
auto& key = graphics_key;
|
||||||
|
|
||||||
key.depth = regs.depth_control;
|
key.depth = regs.depth_control;
|
||||||
|
key.depth_bounds_min = regs.depth_bounds_min;
|
||||||
|
key.depth_bounds_max = regs.depth_bounds_max;
|
||||||
|
key.depth_bias_enable = regs.polygon_control.enable_polygon_offset_back ||
|
||||||
|
regs.polygon_control.enable_polygon_offset_front ||
|
||||||
|
regs.polygon_control.enable_polygon_offset_para;
|
||||||
|
if (regs.polygon_control.enable_polygon_offset_front) {
|
||||||
|
key.depth_bias_const_factor = regs.poly_offset.front_offset;
|
||||||
|
key.depth_bias_slope_factor = regs.poly_offset.front_scale;
|
||||||
|
} else {
|
||||||
|
key.depth_bias_const_factor = regs.poly_offset.back_offset;
|
||||||
|
key.depth_bias_slope_factor = regs.poly_offset.back_scale;
|
||||||
|
}
|
||||||
|
key.depth_bias_clamp = regs.poly_offset.depth_bias;
|
||||||
key.stencil = regs.stencil_control;
|
key.stencil = regs.stencil_control;
|
||||||
key.stencil_ref_front = regs.stencil_ref_front;
|
key.stencil_ref_front = regs.stencil_ref_front;
|
||||||
key.stencil_ref_back = regs.stencil_ref_back;
|
key.stencil_ref_back = regs.stencil_ref_back;
|
||||||
key.prim_type = regs.primitive_type;
|
key.prim_type = regs.primitive_type;
|
||||||
key.polygon_mode = regs.polygon_control.PolyMode();
|
key.polygon_mode = regs.polygon_control.PolyMode();
|
||||||
key.cull_mode = regs.polygon_control.CullingMode();
|
key.cull_mode = regs.polygon_control.CullingMode();
|
||||||
|
key.clip_space = regs.clipper_control.clip_space;
|
||||||
key.front_face = regs.polygon_control.front_face;
|
key.front_face = regs.polygon_control.front_face;
|
||||||
|
|
||||||
const auto& db = regs.depth_buffer;
|
const auto& db = regs.depth_buffer;
|
||||||
|
@ -103,6 +117,9 @@ void PipelineCache::RefreshGraphicsKey() {
|
||||||
// `RenderingInfo` is assumed to be initialized with a contiguous array of valid color
|
// `RenderingInfo` is assumed to be initialized with a contiguous array of valid color
|
||||||
// attachments. This might be not a case as HW color buffers can be bound in an arbitrary order.
|
// attachments. This might be not a case as HW color buffers can be bound in an arbitrary order.
|
||||||
// We need to do some arrays compaction at this stage
|
// We need to do some arrays compaction at this stage
|
||||||
|
key.color_formats.fill(vk::Format::eUndefined);
|
||||||
|
key.blend_controls.fill({});
|
||||||
|
key.write_masks.fill({});
|
||||||
int remapped_cb{};
|
int remapped_cb{};
|
||||||
for (auto cb = 0u; cb < Liverpool::NumColorBuffers; ++cb) {
|
for (auto cb = 0u; cb < Liverpool::NumColorBuffers; ++cb) {
|
||||||
auto const& col_buf = regs.color_buffers[cb];
|
auto const& col_buf = regs.color_buffers[cb];
|
||||||
|
@ -112,6 +129,8 @@ void PipelineCache::RefreshGraphicsKey() {
|
||||||
key.color_formats[remapped_cb] =
|
key.color_formats[remapped_cb] =
|
||||||
LiverpoolToVK::SurfaceFormat(col_buf.info.format, col_buf.NumFormat());
|
LiverpoolToVK::SurfaceFormat(col_buf.info.format, col_buf.NumFormat());
|
||||||
key.blend_controls[remapped_cb] = regs.blend_control[cb];
|
key.blend_controls[remapped_cb] = regs.blend_control[cb];
|
||||||
|
key.blend_controls[remapped_cb].enable.Assign(key.blend_controls[remapped_cb].enable &&
|
||||||
|
!col_buf.info.blend_bypass);
|
||||||
key.write_masks[remapped_cb] = vk::ColorComponentFlags{regs.color_target_mask.GetMask(cb)};
|
key.write_masks[remapped_cb] = vk::ColorComponentFlags{regs.color_target_mask.GetMask(cb)};
|
||||||
|
|
||||||
++remapped_cb;
|
++remapped_cb;
|
||||||
|
|
|
@ -41,6 +41,8 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
|
||||||
|
|
||||||
boost::container::static_vector<vk::RenderingAttachmentInfo, Liverpool::NumColorBuffers>
|
boost::container::static_vector<vk::RenderingAttachmentInfo, Liverpool::NumColorBuffers>
|
||||||
color_attachments{};
|
color_attachments{};
|
||||||
|
vk::RenderingAttachmentInfo depth_attachment{};
|
||||||
|
u32 num_depth_attachments{};
|
||||||
for (auto col_buf_id = 0u; col_buf_id < Liverpool::NumColorBuffers; ++col_buf_id) {
|
for (auto col_buf_id = 0u; col_buf_id < Liverpool::NumColorBuffers; ++col_buf_id) {
|
||||||
const auto& col_buf = regs.color_buffers[col_buf_id];
|
const auto& col_buf = regs.color_buffers[col_buf_id];
|
||||||
if (!col_buf) {
|
if (!col_buf) {
|
||||||
|
@ -57,6 +59,17 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
|
||||||
.storeOp = vk::AttachmentStoreOp::eStore,
|
.storeOp = vk::AttachmentStoreOp::eStore,
|
||||||
});
|
});
|
||||||
}
|
}
|
||||||
|
if (regs.depth_control.depth_enable && regs.depth_buffer.Address() != 0) {
|
||||||
|
const auto& image_view =
|
||||||
|
texture_cache.DepthTarget(regs.depth_buffer, liverpool->last_db_extent);
|
||||||
|
depth_attachment = {
|
||||||
|
.imageView = *image_view.image_view,
|
||||||
|
.imageLayout = vk::ImageLayout::eGeneral,
|
||||||
|
.loadOp = vk::AttachmentLoadOp::eLoad,
|
||||||
|
.storeOp = vk::AttachmentStoreOp::eStore,
|
||||||
|
};
|
||||||
|
num_depth_attachments++;
|
||||||
|
}
|
||||||
|
|
||||||
// TODO: Don't restart renderpass every draw
|
// TODO: Don't restart renderpass every draw
|
||||||
const auto& scissor = regs.screen_scissor;
|
const auto& scissor = regs.screen_scissor;
|
||||||
|
@ -69,6 +82,7 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
|
||||||
.layerCount = 1,
|
.layerCount = 1,
|
||||||
.colorAttachmentCount = static_cast<u32>(color_attachments.size()),
|
.colorAttachmentCount = static_cast<u32>(color_attachments.size()),
|
||||||
.pColorAttachments = color_attachments.data(),
|
.pColorAttachments = color_attachments.data(),
|
||||||
|
.pDepthAttachment = num_depth_attachments ? &depth_attachment : nullptr,
|
||||||
};
|
};
|
||||||
|
|
||||||
UpdateDynamicState(*pipeline);
|
UpdateDynamicState(*pipeline);
|
||||||
|
@ -78,7 +92,9 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
|
||||||
if (is_indexed) {
|
if (is_indexed) {
|
||||||
cmdbuf.drawIndexed(num_indices, regs.num_instances.NumInstances(), 0, 0, 0);
|
cmdbuf.drawIndexed(num_indices, regs.num_instances.NumInstances(), 0, 0, 0);
|
||||||
} else {
|
} else {
|
||||||
const u32 num_vertices = pipeline->IsEmbeddedVs() ? 4 : regs.num_indices;
|
const u32 num_vertices = regs.primitive_type == AmdGpu::Liverpool::PrimitiveType::RectList
|
||||||
|
? 4
|
||||||
|
: regs.num_indices;
|
||||||
cmdbuf.draw(num_vertices, regs.num_instances.NumInstances(), 0, 0);
|
cmdbuf.draw(num_vertices, regs.num_instances.NumInstances(), 0, 0);
|
||||||
}
|
}
|
||||||
cmdbuf.endRendering();
|
cmdbuf.endRendering();
|
||||||
|
@ -156,13 +172,15 @@ void Rasterizer::UpdateDynamicState(const GraphicsPipeline& pipeline) {
|
||||||
void Rasterizer::UpdateViewportScissorState() {
|
void Rasterizer::UpdateViewportScissorState() {
|
||||||
auto& regs = liverpool->regs;
|
auto& regs = liverpool->regs;
|
||||||
|
|
||||||
|
const float reduce_z =
|
||||||
|
regs.clipper_control.clip_space == AmdGpu::Liverpool::ClipSpace::MinusWToW ? 1.0f : 0.0f;
|
||||||
const auto cmdbuf = scheduler.CommandBuffer();
|
const auto cmdbuf = scheduler.CommandBuffer();
|
||||||
const vk::Viewport viewport{
|
const vk::Viewport viewport{
|
||||||
.x = regs.viewports[0].xoffset - regs.viewports[0].xscale,
|
.x = regs.viewports[0].xoffset - regs.viewports[0].xscale,
|
||||||
.y = regs.viewports[0].yoffset - regs.viewports[0].yscale,
|
.y = regs.viewports[0].yoffset - regs.viewports[0].yscale,
|
||||||
.width = regs.viewports[0].xscale * 2.0f,
|
.width = regs.viewports[0].xscale * 2.0f,
|
||||||
.height = regs.viewports[0].yscale * 2.0f,
|
.height = regs.viewports[0].yscale * 2.0f,
|
||||||
.minDepth = regs.viewports[0].zoffset - regs.viewports[0].zscale,
|
.minDepth = regs.viewports[0].zoffset - regs.viewports[0].zscale * reduce_z,
|
||||||
.maxDepth = regs.viewports[0].zscale + regs.viewports[0].zoffset,
|
.maxDepth = regs.viewports[0].zscale + regs.viewports[0].zoffset,
|
||||||
};
|
};
|
||||||
const vk::Rect2D scissor{
|
const vk::Rect2D scissor{
|
||||||
|
|
|
@ -33,11 +33,23 @@ static vk::Format ConvertPixelFormat(const VideoOutFormat format) {
|
||||||
return {};
|
return {};
|
||||||
}
|
}
|
||||||
|
|
||||||
|
static bool IsDepthStencilFormat(vk::Format format) {
|
||||||
|
switch (format) {
|
||||||
|
case vk::Format::eD16Unorm:
|
||||||
|
case vk::Format::eD16UnormS8Uint:
|
||||||
|
case vk::Format::eD32Sfloat:
|
||||||
|
case vk::Format::eD32SfloatS8Uint:
|
||||||
|
return true;
|
||||||
|
default:
|
||||||
|
return false;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
static vk::ImageUsageFlags ImageUsageFlags(const vk::Format format) {
|
static vk::ImageUsageFlags ImageUsageFlags(const vk::Format format) {
|
||||||
vk::ImageUsageFlags usage = vk::ImageUsageFlagBits::eTransferSrc |
|
vk::ImageUsageFlags usage = vk::ImageUsageFlagBits::eTransferSrc |
|
||||||
vk::ImageUsageFlagBits::eTransferDst |
|
vk::ImageUsageFlagBits::eTransferDst |
|
||||||
vk::ImageUsageFlagBits::eSampled;
|
vk::ImageUsageFlagBits::eSampled;
|
||||||
if (false /*&& IsDepthStencilFormat(format)*/) {
|
if (IsDepthStencilFormat(format)) {
|
||||||
usage |= vk::ImageUsageFlagBits::eDepthStencilAttachment;
|
usage |= vk::ImageUsageFlagBits::eDepthStencilAttachment;
|
||||||
} else {
|
} else {
|
||||||
if (format != vk::Format::eBc3SrgbBlock) {
|
if (format != vk::Format::eBc3SrgbBlock) {
|
||||||
|
@ -54,9 +66,9 @@ static vk::ImageType ConvertImageType(AmdGpu::ImageType type) noexcept {
|
||||||
case AmdGpu::ImageType::Color2D:
|
case AmdGpu::ImageType::Color2D:
|
||||||
case AmdGpu::ImageType::Color1DArray:
|
case AmdGpu::ImageType::Color1DArray:
|
||||||
case AmdGpu::ImageType::Cube:
|
case AmdGpu::ImageType::Cube:
|
||||||
|
case AmdGpu::ImageType::Color2DArray:
|
||||||
return vk::ImageType::e2D;
|
return vk::ImageType::e2D;
|
||||||
case AmdGpu::ImageType::Color3D:
|
case AmdGpu::ImageType::Color3D:
|
||||||
case AmdGpu::ImageType::Color2DArray:
|
|
||||||
return vk::ImageType::e3D;
|
return vk::ImageType::e3D;
|
||||||
default:
|
default:
|
||||||
UNREACHABLE();
|
UNREACHABLE();
|
||||||
|
@ -98,6 +110,18 @@ ImageInfo::ImageInfo(const AmdGpu::Liverpool::ColorBuffer& buffer,
|
||||||
guest_size_bytes = buffer.GetSizeAligned();
|
guest_size_bytes = buffer.GetSizeAligned();
|
||||||
}
|
}
|
||||||
|
|
||||||
|
ImageInfo::ImageInfo(const AmdGpu::Liverpool::DepthBuffer& buffer,
|
||||||
|
const AmdGpu::Liverpool::CbDbExtent& hint) noexcept {
|
||||||
|
is_tiled = false;
|
||||||
|
pixel_format = LiverpoolToVK::DepthFormat(buffer.z_info.format, buffer.stencil_info.format);
|
||||||
|
type = vk::ImageType::e2D;
|
||||||
|
size.width = hint.Valid() ? hint.width : buffer.Pitch();
|
||||||
|
size.height = hint.Valid() ? hint.height : buffer.Height();
|
||||||
|
size.depth = 1;
|
||||||
|
pitch = size.width;
|
||||||
|
guest_size_bytes = buffer.GetSizeAligned();
|
||||||
|
}
|
||||||
|
|
||||||
ImageInfo::ImageInfo(const AmdGpu::Image& image) noexcept {
|
ImageInfo::ImageInfo(const AmdGpu::Image& image) noexcept {
|
||||||
is_tiled = image.IsTiled();
|
is_tiled = image.IsTiled();
|
||||||
tiling_mode = image.GetTilingMode();
|
tiling_mode = image.GetTilingMode();
|
||||||
|
@ -165,6 +189,13 @@ Image::Image(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_,
|
||||||
info.usage |= vk::ImageUsageFlagBits::eStorage;
|
info.usage |= vk::ImageUsageFlagBits::eStorage;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
if (info.pixel_format == vk::Format::eD32Sfloat) {
|
||||||
|
aspect_mask = vk::ImageAspectFlagBits::eDepth;
|
||||||
|
}
|
||||||
|
if (info.pixel_format == vk::Format::eD32SfloatS8Uint) {
|
||||||
|
aspect_mask = vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil;
|
||||||
|
}
|
||||||
|
|
||||||
const vk::ImageCreateInfo image_ci = {
|
const vk::ImageCreateInfo image_ci = {
|
||||||
.flags = flags,
|
.flags = flags,
|
||||||
.imageType = info.type,
|
.imageType = info.type,
|
||||||
|
@ -187,7 +218,7 @@ Image::Image(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_,
|
||||||
if (info.is_tiled) {
|
if (info.is_tiled) {
|
||||||
ImageViewInfo view_info;
|
ImageViewInfo view_info;
|
||||||
view_info.format = DemoteImageFormatForDetiling(info.pixel_format);
|
view_info.format = DemoteImageFormatForDetiling(info.pixel_format);
|
||||||
view_for_detiler.emplace(*instance, view_info, image);
|
view_for_detiler.emplace(*instance, view_info, *this);
|
||||||
}
|
}
|
||||||
|
|
||||||
Transit(vk::ImageLayout::eGeneral, vk::AccessFlagBits::eNone);
|
Transit(vk::ImageLayout::eGeneral, vk::AccessFlagBits::eNone);
|
||||||
|
@ -198,7 +229,8 @@ void Image::Transit(vk::ImageLayout dst_layout, vk::Flags<vk::AccessFlagBits> ds
|
||||||
return;
|
return;
|
||||||
}
|
}
|
||||||
|
|
||||||
const vk::ImageMemoryBarrier barrier = {.srcAccessMask = access_mask,
|
const vk::ImageMemoryBarrier barrier = {
|
||||||
|
.srcAccessMask = access_mask,
|
||||||
.dstAccessMask = dst_mask,
|
.dstAccessMask = dst_mask,
|
||||||
.oldLayout = layout,
|
.oldLayout = layout,
|
||||||
.newLayout = dst_layout,
|
.newLayout = dst_layout,
|
||||||
|
@ -211,10 +243,11 @@ void Image::Transit(vk::ImageLayout dst_layout, vk::Flags<vk::AccessFlagBits> ds
|
||||||
.levelCount = VK_REMAINING_MIP_LEVELS,
|
.levelCount = VK_REMAINING_MIP_LEVELS,
|
||||||
.baseArrayLayer = 0,
|
.baseArrayLayer = 0,
|
||||||
.layerCount = VK_REMAINING_ARRAY_LAYERS,
|
.layerCount = VK_REMAINING_ARRAY_LAYERS,
|
||||||
}};
|
},
|
||||||
|
};
|
||||||
|
|
||||||
// Adjust pipieline stage
|
// Adjust pipieline stage
|
||||||
vk::PipelineStageFlags dst_pl_stage =
|
const vk::PipelineStageFlags dst_pl_stage =
|
||||||
(dst_mask == vk::AccessFlagBits::eTransferRead ||
|
(dst_mask == vk::AccessFlagBits::eTransferRead ||
|
||||||
dst_mask == vk::AccessFlagBits::eTransferWrite)
|
dst_mask == vk::AccessFlagBits::eTransferWrite)
|
||||||
? vk::PipelineStageFlagBits::eTransfer
|
? vk::PipelineStageFlagBits::eTransfer
|
||||||
|
|
|
@ -38,6 +38,8 @@ struct ImageInfo {
|
||||||
explicit ImageInfo(const Libraries::VideoOut::BufferAttributeGroup& group) noexcept;
|
explicit ImageInfo(const Libraries::VideoOut::BufferAttributeGroup& group) noexcept;
|
||||||
explicit ImageInfo(const AmdGpu::Liverpool::ColorBuffer& buffer,
|
explicit ImageInfo(const AmdGpu::Liverpool::ColorBuffer& buffer,
|
||||||
const AmdGpu::Liverpool::CbDbExtent& hint = {}) noexcept;
|
const AmdGpu::Liverpool::CbDbExtent& hint = {}) noexcept;
|
||||||
|
explicit ImageInfo(const AmdGpu::Liverpool::DepthBuffer& buffer,
|
||||||
|
const AmdGpu::Liverpool::CbDbExtent& hint = {}) noexcept;
|
||||||
explicit ImageInfo(const AmdGpu::Image& image) noexcept;
|
explicit ImageInfo(const AmdGpu::Image& image) noexcept;
|
||||||
|
|
||||||
bool is_tiled = false;
|
bool is_tiled = false;
|
||||||
|
|
|
@ -3,6 +3,7 @@
|
||||||
|
|
||||||
#include "video_core/renderer_vulkan/liverpool_to_vk.h"
|
#include "video_core/renderer_vulkan/liverpool_to_vk.h"
|
||||||
#include "video_core/renderer_vulkan/vk_instance.h"
|
#include "video_core/renderer_vulkan/vk_instance.h"
|
||||||
|
#include "video_core/texture_cache/image.h"
|
||||||
#include "video_core/texture_cache/image_view.h"
|
#include "video_core/texture_cache/image_view.h"
|
||||||
|
|
||||||
namespace VideoCore {
|
namespace VideoCore {
|
||||||
|
@ -58,7 +59,7 @@ ImageViewInfo::ImageViewInfo(const AmdGpu::Image& image) noexcept {
|
||||||
mapping.a = ConvertComponentSwizzle(image.dst_sel_w);
|
mapping.a = ConvertComponentSwizzle(image.dst_sel_w);
|
||||||
}
|
}
|
||||||
|
|
||||||
ImageView::ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info_, vk::Image image,
|
ImageView::ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info_, Image& image,
|
||||||
std::optional<vk::ImageUsageFlags> usage_override /*= {}*/)
|
std::optional<vk::ImageUsageFlags> usage_override /*= {}*/)
|
||||||
: info{info_} {
|
: info{info_} {
|
||||||
vk::ImageViewUsageCreateInfo usage_ci{};
|
vk::ImageViewUsageCreateInfo usage_ci{};
|
||||||
|
@ -66,14 +67,20 @@ ImageView::ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info
|
||||||
usage_ci.usage = usage_override.value();
|
usage_ci.usage = usage_override.value();
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// When sampling D32 texture from shader, the T# specifies R32 Float format so adjust it.
|
||||||
|
vk::Format format = info.format;
|
||||||
|
if (image.aspect_mask & vk::ImageAspectFlagBits::eDepth && format == vk::Format::eR32Sfloat) {
|
||||||
|
format = vk::Format::eD32Sfloat;
|
||||||
|
}
|
||||||
|
|
||||||
const vk::ImageViewCreateInfo image_view_ci = {
|
const vk::ImageViewCreateInfo image_view_ci = {
|
||||||
.pNext = usage_override.has_value() ? &usage_ci : nullptr,
|
.pNext = usage_override.has_value() ? &usage_ci : nullptr,
|
||||||
.image = image,
|
.image = image.image,
|
||||||
.viewType = info.type,
|
.viewType = info.type,
|
||||||
.format = info.format,
|
.format = format,
|
||||||
.components = info.mapping,
|
.components = info.mapping,
|
||||||
.subresourceRange{
|
.subresourceRange{
|
||||||
.aspectMask = vk::ImageAspectFlagBits::eColor,
|
.aspectMask = image.aspect_mask,
|
||||||
.baseMipLevel = 0U,
|
.baseMipLevel = 0U,
|
||||||
.levelCount = 1,
|
.levelCount = 1,
|
||||||
.baseArrayLayer = 0,
|
.baseArrayLayer = 0,
|
||||||
|
|
|
@ -28,8 +28,10 @@ struct ImageViewInfo {
|
||||||
auto operator<=>(const ImageViewInfo&) const = default;
|
auto operator<=>(const ImageViewInfo&) const = default;
|
||||||
};
|
};
|
||||||
|
|
||||||
|
struct Image;
|
||||||
|
|
||||||
struct ImageView {
|
struct ImageView {
|
||||||
explicit ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info, vk::Image image,
|
explicit ImageView(const Vulkan::Instance& instance, const ImageViewInfo& info, Image& image,
|
||||||
std::optional<vk::ImageUsageFlags> usage_override = {});
|
std::optional<vk::ImageUsageFlags> usage_override = {});
|
||||||
~ImageView();
|
~ImageView();
|
||||||
|
|
||||||
|
|
|
@ -93,7 +93,7 @@ TextureCache::TextureCache(const Vulkan::Instance& instance_, Vulkan::Scheduler&
|
||||||
ASSERT(null_id.index == 0);
|
ASSERT(null_id.index == 0);
|
||||||
|
|
||||||
ImageViewInfo view_info;
|
ImageViewInfo view_info;
|
||||||
void(slot_image_views.insert(instance, view_info, slot_images[null_id].image));
|
void(slot_image_views.insert(instance, view_info, slot_images[null_id]));
|
||||||
}
|
}
|
||||||
|
|
||||||
TextureCache::~TextureCache() {
|
TextureCache::~TextureCache() {
|
||||||
|
@ -112,7 +112,7 @@ void TextureCache::OnCpuWrite(VAddr address) {
|
||||||
});
|
});
|
||||||
}
|
}
|
||||||
|
|
||||||
Image& TextureCache::FindImage(const ImageInfo& info, VAddr cpu_address) {
|
Image& TextureCache::FindImage(const ImageInfo& info, VAddr cpu_address, bool refresh_on_create) {
|
||||||
std::unique_lock lock{m_page_table};
|
std::unique_lock lock{m_page_table};
|
||||||
boost::container::small_vector<ImageId, 2> image_ids;
|
boost::container::small_vector<ImageId, 2> image_ids;
|
||||||
ForEachImageInRegion(cpu_address, info.guest_size_bytes, [&](ImageId image_id, Image& image) {
|
ForEachImageInRegion(cpu_address, info.guest_size_bytes, [&](ImageId image_id, Image& image) {
|
||||||
|
@ -132,7 +132,8 @@ Image& TextureCache::FindImage(const ImageInfo& info, VAddr cpu_address) {
|
||||||
}
|
}
|
||||||
|
|
||||||
Image& image = slot_images[image_id];
|
Image& image = slot_images[image_id];
|
||||||
if (True(image.flags & ImageFlagBits::CpuModified)) {
|
if (True(image.flags & ImageFlagBits::CpuModified) &&
|
||||||
|
(!image_ids.empty() || refresh_on_create)) {
|
||||||
RefreshImage(image);
|
RefreshImage(image);
|
||||||
TrackImage(image, image_id);
|
TrackImage(image, image_id);
|
||||||
}
|
}
|
||||||
|
@ -153,8 +154,7 @@ ImageView& TextureCache::RegisterImageView(Image& image, const ImageViewInfo& vi
|
||||||
usage_override = image.info.usage & ~vk::ImageUsageFlagBits::eStorage;
|
usage_override = image.info.usage & ~vk::ImageUsageFlagBits::eStorage;
|
||||||
}
|
}
|
||||||
|
|
||||||
const ImageViewId view_id =
|
const ImageViewId view_id = slot_image_views.insert(instance, view_info, image, usage_override);
|
||||||
slot_image_views.insert(instance, view_info, image.image, usage_override);
|
|
||||||
image.image_view_infos.emplace_back(view_info);
|
image.image_view_infos.emplace_back(view_info);
|
||||||
image.image_view_ids.emplace_back(view_id);
|
image.image_view_ids.emplace_back(view_id);
|
||||||
return slot_image_views[view_id];
|
return slot_image_views[view_id];
|
||||||
|
@ -177,6 +177,7 @@ ImageView& TextureCache::RenderTarget(const AmdGpu::Liverpool::ColorBuffer& buff
|
||||||
const AmdGpu::Liverpool::CbDbExtent& hint) {
|
const AmdGpu::Liverpool::CbDbExtent& hint) {
|
||||||
const ImageInfo info{buffer, hint};
|
const ImageInfo info{buffer, hint};
|
||||||
auto& image = FindImage(info, buffer.Address());
|
auto& image = FindImage(info, buffer.Address());
|
||||||
|
image.flags &= ~ImageFlagBits::CpuModified;
|
||||||
|
|
||||||
image.Transit(vk::ImageLayout::eColorAttachmentOptimal,
|
image.Transit(vk::ImageLayout::eColorAttachmentOptimal,
|
||||||
vk::AccessFlagBits::eColorAttachmentWrite |
|
vk::AccessFlagBits::eColorAttachmentWrite |
|
||||||
|
@ -187,6 +188,17 @@ ImageView& TextureCache::RenderTarget(const AmdGpu::Liverpool::ColorBuffer& buff
|
||||||
return RegisterImageView(image, view_info);
|
return RegisterImageView(image, view_info);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
ImageView& TextureCache::DepthTarget(const AmdGpu::Liverpool::DepthBuffer& buffer,
|
||||||
|
const AmdGpu::Liverpool::CbDbExtent& hint) {
|
||||||
|
const ImageInfo info{buffer, hint};
|
||||||
|
auto& image = FindImage(info, buffer.Address(), false);
|
||||||
|
image.flags &= ~ImageFlagBits::CpuModified;
|
||||||
|
|
||||||
|
ImageViewInfo view_info;
|
||||||
|
view_info.format = info.pixel_format;
|
||||||
|
return RegisterImageView(image, view_info);
|
||||||
|
}
|
||||||
|
|
||||||
void TextureCache::RefreshImage(Image& image) {
|
void TextureCache::RefreshImage(Image& image) {
|
||||||
// Mark image as validated.
|
// Mark image as validated.
|
||||||
image.flags &= ~ImageFlagBits::CpuModified;
|
image.flags &= ~ImageFlagBits::CpuModified;
|
||||||
|
|
|
@ -37,7 +37,8 @@ public:
|
||||||
void OnCpuWrite(VAddr address);
|
void OnCpuWrite(VAddr address);
|
||||||
|
|
||||||
/// Retrieves the image handle of the image with the provided attributes and address.
|
/// Retrieves the image handle of the image with the provided attributes and address.
|
||||||
[[nodiscard]] Image& FindImage(const ImageInfo& info, VAddr cpu_address);
|
[[nodiscard]] Image& FindImage(const ImageInfo& info, VAddr cpu_address,
|
||||||
|
bool refresh_on_create = true);
|
||||||
|
|
||||||
/// Retrieves an image view with the properties of the specified image descriptor.
|
/// Retrieves an image view with the properties of the specified image descriptor.
|
||||||
[[nodiscard]] ImageView& FindImageView(const AmdGpu::Image& image);
|
[[nodiscard]] ImageView& FindImageView(const AmdGpu::Image& image);
|
||||||
|
@ -45,6 +46,8 @@ public:
|
||||||
/// Retrieves the render target with specified properties
|
/// Retrieves the render target with specified properties
|
||||||
[[nodiscard]] ImageView& RenderTarget(const AmdGpu::Liverpool::ColorBuffer& buffer,
|
[[nodiscard]] ImageView& RenderTarget(const AmdGpu::Liverpool::ColorBuffer& buffer,
|
||||||
const AmdGpu::Liverpool::CbDbExtent& hint);
|
const AmdGpu::Liverpool::CbDbExtent& hint);
|
||||||
|
[[nodiscard]] ImageView& DepthTarget(const AmdGpu::Liverpool::DepthBuffer& buffer,
|
||||||
|
const AmdGpu::Liverpool::CbDbExtent& hint);
|
||||||
|
|
||||||
/// Reuploads image contents.
|
/// Reuploads image contents.
|
||||||
void RefreshImage(Image& image);
|
void RefreshImage(Image& image);
|
||||||
|
|
Loading…
Reference in New Issue