graphics: execute DB_RENDER_OVERRIDE depth and stencil copies

This commit is contained in:
nmzik
2026-09-09 15:19:37 +04:00
parent 57844a07e2
commit 0b4e78c0b2
6 changed files with 520 additions and 108 deletions
@@ -315,8 +315,17 @@ static void HwCtxSetDepthBoundsRegister(CommandProcessor& cp, uint32_t cmd_offse
}
}
static void HwCtxIgnoreDepthMetadataRegister([[maybe_unused]] uint32_t cmd_offset,
[[maybe_unused]] uint32_t value) {}
static void HwCtxSetDepthMetadataRegister(CommandProcessor& cp, uint32_t cmd_offset,
uint32_t value) {
if (cmd_offset == Pm4::DB_RENDER_OVERRIDE) {
HW::DepthRenderOverride control;
control.force_z_valid = (value & 0x20000000u) != 0;
control.force_z_dirty = (value & 0x08000000u) != 0;
control.force_stencil_valid = (value & 0x40000000u) != 0;
control.force_stencil_dirty = (value & 0x10000000u) != 0;
cp.GetCtx().SetDepthRenderOverride(control);
}
}
static void HwCtxIgnoreFovWindow([[maybe_unused]] uint32_t value) {}
@@ -389,7 +398,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetDepthMetadataRegisters) {
auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
for (uint32_t i = 0; i < num_values; i++) {
HwCtxIgnoreDepthMetadataRegister(cmd_offset + i, buffer[i]);
HwCtxSetDepthMetadataRegister(cp, cmd_offset + i, buffer[i]);
}
return num_values;
@@ -3003,28 +3012,28 @@ void GraphicsInitJmpTablesCxIndirect() {
HwCtxIgnoreCbDccControl(value);
};
g_hw_ctx_indirect_func[Pm4::DB_COUNT_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
};
for (auto cmd_offset = Pm4::DB_SRESULTS_COMPARE_STATE0;
cmd_offset <= Pm4::DB_SRESULTS_COMPARE_STATE1; cmd_offset++) {
g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
};
}
g_hw_ctx_indirect_func[Pm4::DB_RENDER_OVERRIDE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
};
g_hw_ctx_indirect_func[Pm4::DB_RENDER_OVERRIDE2] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
};
g_hw_ctx_indirect_func[Pm4::DB_DFSM_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
};
g_hw_ctx_indirect_func[Pm4::DB_RMI_L2_CACHE_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
};
g_hw_ctx_indirect_func[Pm4::CB_RMI_GL2_CACHE_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
};
g_hw_ctx_indirect_func[Pm4::TA_BC_BASE_ADDR] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
HwCtxIgnoreBorderColorTableAddr(cmd_offset, value);
+23 -9
View File
@@ -202,6 +202,13 @@ struct RenderControl {
uint8_t copy_sample = 0;
};
struct DepthRenderOverride {
bool force_z_valid = false;
bool force_z_dirty = false;
bool force_stencil_valid = false;
bool force_stencil_dirty = false;
};
struct GdsOaCounter {
uint32_t counter = 0;
uint32_t address = 0;
@@ -809,6 +816,12 @@ public:
void SetClipControl(const ClipControl& control) { m_clip_control = control; }
[[nodiscard]] const RenderControl& GetRenderControl() const { return m_render_control; }
void SetRenderControl(const RenderControl& control) { m_render_control = control; }
[[nodiscard]] const DepthRenderOverride& GetDepthRenderOverride() const {
return m_depth_render_override;
}
void SetDepthRenderOverride(const DepthRenderOverride& control) {
m_depth_render_override = control;
}
[[nodiscard]] const DepthControl& GetDepthControl() const { return m_depth_control; }
void SetDepthControl(const DepthControl& control) { m_depth_control = control; }
[[nodiscard]] const ModeControl& GetModeControl() const { return m_mode_control; }
@@ -904,15 +917,16 @@ private:
uint32_t m_shader_stages = 0;
DepthRenderTarget m_depth_render_target;
RenderControl m_render_control;
DepthControl m_depth_control;
StencilControl m_stencil_control;
StencilMask m_stencil_mask;
float m_depth_clear_value = 0.0f;
float m_depth_bounds_min = 0.0f;
float m_depth_bounds_max = 1.0f;
uint8_t m_stencil_clear_value = 0;
DepthRenderTarget m_depth_render_target;
RenderControl m_render_control;
DepthRenderOverride m_depth_render_override;
DepthControl m_depth_control;
StencilControl m_stencil_control;
StencilMask m_stencil_mask;
float m_depth_clear_value = 0.0f;
float m_depth_bounds_min = 0.0f;
float m_depth_bounds_max = 1.0f;
uint8_t m_stencil_clear_value = 0;
ModeControl m_mode_control;
PolyOffset m_poly_offset;
@@ -10,6 +10,7 @@
#include "graphics/guest_gpu/hardwareContext.h"
#include "graphics/guest_gpu/tile.h"
#include "graphics/host_gpu/graphicContext.h"
#include "graphics/host_gpu/renderer/commandScheduler.h"
#include "graphics/host_gpu/renderer/debug.h"
#include "graphics/host_gpu/renderer/image/imageView.h"
#include "graphics/host_gpu/renderer/image/textureCommon.h"
@@ -96,51 +97,17 @@ static bool UsesStencilOpValue(uint8_t fail, uint8_t pass, uint8_t depth_fail) {
return vk::Format::eUndefined;
}
// NOLINTNEXTLINE(readability-function-cognitive-complexity)
void RenderExecutor::ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepthInfo& r) {
KYTY_PROFILER_FUNCTION();
const auto& hw = buffer.GetRegisters();
const auto& z = hw.GetDepthRenderTarget();
const auto& rc = hw.GetRenderControl();
const auto& dc = hw.GetDepthControl();
const auto& sc = hw.GetStencilControl();
const auto& sm = hw.GetStencilMask();
const bool has_stencil = z.stencil_info.format != Prospero::StencilFormat::kInvalid;
const bool depth_active = dc.z_enable || dc.depth_bounds_enable ||
rc.depth_clear_enable || rc.copy_depth_to_color;
const bool stencil_active =
has_stencil && (dc.stencil_enable || rc.stencil_clear_enable || rc.copy_stencil_to_color);
if (!depth_active && !stencil_active) {
return;
}
static TextureCache::ImageDesc MakeDepthTargetDesc(const CommandBuffer& buffer,
const HW::DepthRenderTarget& z,
bool write_buffer = false) {
const bool has_stencil = z.stencil_info.format != Prospero::StencilFormat::kInvalid;
const auto depth_address = write_buffer ? z.z_write_base_addr : z.z_read_base_addr;
const auto stencil_address =
write_buffer ? z.stencil_write_base_addr : z.stencil_read_base_addr;
if (!z.z_info.HasValidTextureCompatibility() ||
!z.stencil_info.HasValidTextureCompatibility()) {
DepthFatal("invalid PS5 depth texture-compatibility encoding");
}
const bool attachment_unbound =
z.z_info.format == Prospero::DepthFormat::kInvalid &&
z.stencil_info.format == Prospero::StencilFormat::kInvalid && z.z_info.num_samples == 0 &&
z.z_info.texture_compatibility == Prospero::TextureCompatiblePlaneCompression::kDisable &&
!z.z_info.expclear_enabled && !z.z_info.partially_resident && z.z_info.max_mip_level == 0 &&
z.stencil_info.texture_compatibility == Prospero::TextureCompatibleStencil::kDisable &&
!z.stencil_info.expclear_enabled && !z.stencil_info.partially_resident &&
z.depth_view.slice_start == 0 && z.depth_view.slice_max == 0 &&
z.depth_view.current_mip_level == 0 && !z.depth_view.depth_write_disable &&
!z.depth_view.stencil_write_disable && z.z_read_base_addr == 0 && z.z_write_base_addr == 0 &&
z.stencil_read_base_addr == 0 && z.stencil_write_base_addr == 0 &&
z.htile_data_base_addr == 0 &&
// DB_DEPTH_SIZE_XY is independent state and may remain programmed after the attachment
// formats and addresses are unbound. A zero encoding is the valid 1x1 value, so its
// presence alone must not manufacture a depth attachment.
!z.z_info.htile_acceleration && z.shading_rate_encoding == 0 && z.size.x_max == 0 &&
z.size.y_max == 0;
if (attachment_unbound) {
static std::atomic_bool logged = false;
if (!logged.exchange(true, std::memory_order_relaxed)) {
LOGF("DepthTarget: ignoring enabled depth/stencil state without a bound attachment\n");
}
return;
}
const bool has_htile = z.z_info.htile_acceleration;
const bool unsupported_shading_rate_encoding =
z.shading_rate_encoding > 1 || (z.shading_rate_encoding != 0 && !has_htile);
@@ -158,23 +125,16 @@ void RenderExecutor::ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepth
DepthFatal("invalid depth view: base=%u last=%u", z.depth_view.slice_start,
z.depth_view.slice_max);
}
if (rc.copy_depth_to_color || rc.copy_stencil_to_color || rc.copy_centroid ||
rc.copy_sample != 0 || z.z_info.expclear_enabled ||
z.stencil_info.expclear_enabled || z.z_info.partially_resident ||
if (z.z_info.expclear_enabled || z.stencil_info.expclear_enabled ||
z.z_info.partially_resident ||
z.stencil_info.partially_resident || z.z_info.max_mip_level != 0 ||
z.depth_view.current_mip_level != 0 || unsupported_shading_rate_encoding ||
z.z_read_base_addr == 0 ||
(!z.depth_view.depth_write_disable && z.z_write_base_addr != z.z_read_base_addr) ||
(z.z_read_base_addr & 0xffffu) != 0 ||
dc.zfunc > static_cast<uint8_t>(vk::CompareOp::eAlways)) {
depth_address == 0 || (depth_address & 0xffffu) != 0) {
DepthFatal("unsupported depth register state");
}
if (has_stencil) {
if (z.stencil_info.format != Prospero::StencilFormat::k8UInt || !htile_stencil_compat ||
z.stencil_read_base_addr == 0 ||
(!z.depth_view.stencil_write_disable &&
z.stencil_write_base_addr != z.stencil_read_base_addr) ||
(z.stencil_read_base_addr & 0xffffu) != 0) {
stencil_address == 0 || (stencil_address & 0xffffu) != 0) {
DepthFatal("unsupported stencil attachment state");
}
} else if (z.stencil_read_base_addr != 0 || z.stencil_write_base_addr != 0) {
@@ -227,11 +187,92 @@ void RenderExecutor::ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepth
const auto depth_backing_size = depth_size.size * view.image_layers;
const auto stencil_backing_size = stencil_size.size * view.image_layers;
const auto htile_backing_size = htile_size.size * view.image_layers;
if (!GuestRange {z.z_read_base_addr, depth_backing_size}.Valid() ||
(has_stencil && !GuestRange {z.stencil_read_base_addr, stencil_backing_size}.Valid()) ||
if (!GuestRange {depth_address, depth_backing_size}.Valid() ||
(has_stencil && !GuestRange {stencil_address, stencil_backing_size}.Valid()) ||
(has_htile && !GuestRange {z.htile_data_base_addr, htile_backing_size}.Valid())) {
DepthFatal("layered depth backing range is invalid");
}
TextureCache::ImageDesc desc {};
desc.type = TextureCache::BindingType::DepthTarget;
desc.info.data = {depth_address, depth_backing_size};
desc.info.stencil =
has_stencil ? GuestRange {stencil_address, stencil_backing_size} : GuestRange {};
desc.info.pixel_format = format;
desc.info.guest_format = guest_format;
desc.info.type = Prospero::ImageType::kColor2D;
desc.info.extent = {width, height, 1};
desc.info.resources = {1, view.image_layers};
desc.info.pitch = pitch;
desc.info.bytes_per_block = bytes;
desc.info.samples = samples;
desc.info.tile_mode = Prospero::TileMode::kDepth;
desc.info.mip_layout[0] = {0, depth_backing_size, pitch, height};
desc.info.metadata.range =
has_htile ? GuestRange {z.htile_data_base_addr, htile_backing_size} : GuestRange {};
desc.info.metadata.kind = has_htile ? ImageMetadataKind::Htile : ImageMetadataKind::None;
desc.info.metadata.stencil_compressed =
has_stencil && has_htile && !z.stencil_info.htile_stencil_disabled;
desc.view_info.format = format;
desc.view_info.type =
view.layer_count == 1 ? vk::ImageViewType::e2D : vk::ImageViewType::e2DArray;
desc.view_info.aspect = ImageViewOps::DepthAspectMask(format);
desc.view_info.base_level = 0;
desc.view_info.level_count = 1;
desc.view_info.base_layer = view.base_layer;
desc.view_info.layer_count = view.layer_count;
desc.view_info.usage = vk::ImageUsageFlagBits::eDepthStencilAttachment;
return desc;
}
// NOLINTNEXTLINE(readability-function-cognitive-complexity)
void RenderExecutor::ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepthInfo& r) {
KYTY_PROFILER_FUNCTION();
const auto& hw = buffer.GetRegisters();
const auto& z = hw.GetDepthRenderTarget();
const auto& rc = hw.GetRenderControl();
const auto& dc = hw.GetDepthControl();
const auto& sc = hw.GetStencilControl();
const auto& sm = hw.GetStencilMask();
const bool has_stencil = z.stencil_info.format != Prospero::StencilFormat::kInvalid;
const bool depth_active = dc.z_enable || dc.depth_bounds_enable ||
rc.depth_clear_enable || rc.copy_depth_to_color;
const bool stencil_active =
has_stencil && (dc.stencil_enable || rc.stencil_clear_enable || rc.copy_stencil_to_color);
if (!depth_active && !stencil_active) {
return;
}
const bool attachment_unbound =
z.z_info.format == Prospero::DepthFormat::kInvalid &&
z.stencil_info.format == Prospero::StencilFormat::kInvalid && z.z_info.num_samples == 0 &&
z.z_info.texture_compatibility == Prospero::TextureCompatiblePlaneCompression::kDisable &&
!z.z_info.expclear_enabled && !z.z_info.partially_resident && z.z_info.max_mip_level == 0 &&
z.stencil_info.texture_compatibility == Prospero::TextureCompatibleStencil::kDisable &&
!z.stencil_info.expclear_enabled && !z.stencil_info.partially_resident &&
z.depth_view.slice_start == 0 && z.depth_view.slice_max == 0 &&
z.depth_view.current_mip_level == 0 && !z.depth_view.depth_write_disable &&
!z.depth_view.stencil_write_disable && z.z_read_base_addr == 0 && z.z_write_base_addr == 0 &&
z.stencil_read_base_addr == 0 && z.stencil_write_base_addr == 0 &&
z.htile_data_base_addr == 0 &&
// DB_DEPTH_SIZE_XY is independent state and may remain programmed after the attachment
// formats and addresses are unbound. A zero encoding is the valid 1x1 value, so its
// presence alone must not manufacture a depth attachment.
!z.z_info.htile_acceleration && z.shading_rate_encoding == 0 && z.size.x_max == 0 &&
z.size.y_max == 0;
if (attachment_unbound) {
static std::atomic_bool logged = false;
if (!logged.exchange(true, std::memory_order_relaxed)) {
LOGF("DepthTarget: ignoring enabled depth/stencil state without a bound attachment\n");
}
return;
}
if (rc.copy_depth_to_color || rc.copy_stencil_to_color || rc.copy_centroid ||
rc.copy_sample != 0 || dc.zfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
(!z.depth_view.depth_write_disable && z.z_write_base_addr != z.z_read_base_addr) ||
(has_stencil && !z.depth_view.stencil_write_disable &&
z.stencil_write_base_addr != z.stencil_read_base_addr)) {
DepthFatal("unsupported depth register state");
}
r.desc = MakeDepthTargetDesc(buffer, z);
r.depth_clear_enable = rc.depth_clear_enable;
r.depth_load_clear_enable = r.depth_clear_enable;
r.depth_clear_value = hw.GetDepthClearValue();
@@ -282,41 +323,73 @@ void RenderExecutor::ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepth
r.stencil_dynamic_back = r.stencil_dynamic_front;
}
}
TextureCache::ImageDesc desc {};
desc.type = TextureCache::BindingType::DepthTarget;
desc.info.data = {z.z_read_base_addr, depth_backing_size};
desc.info.stencil =
has_stencil ? GuestRange {z.stencil_read_base_addr, stencil_backing_size} : GuestRange {};
desc.info.pixel_format = format;
desc.info.guest_format = guest_format;
desc.info.type = Prospero::ImageType::kColor2D;
desc.info.extent = {width, height, 1};
desc.info.resources = {1, view.image_layers};
desc.info.pitch = pitch;
desc.info.bytes_per_block = bytes;
desc.info.samples = samples;
desc.info.tile_mode = Prospero::TileMode::kDepth;
desc.info.mip_layout[0] = {0, depth_backing_size, pitch, height};
desc.info.metadata.range =
has_htile ? GuestRange {z.htile_data_base_addr, htile_backing_size} : GuestRange {};
desc.info.metadata.kind = has_htile ? ImageMetadataKind::Htile : ImageMetadataKind::None;
desc.info.metadata.stencil_compressed =
has_stencil && has_htile && !z.stencil_info.htile_stencil_disabled;
desc.view_info.format = format;
desc.view_info.type =
view.layer_count == 1 ? vk::ImageViewType::e2D : vk::ImageViewType::e2DArray;
desc.view_info.aspect = ImageViewOps::DepthAspectMask(format);
desc.view_info.base_level = 0;
desc.view_info.level_count = 1;
desc.view_info.base_layer = view.base_layer;
desc.view_info.layer_count = view.layer_count;
desc.view_info.usage = vk::ImageUsageFlagBits::eDepthStencilAttachment;
r.desc = std::move(desc);
auto& cache = m_context.GetTextureCache();
r.image_id = cache.FindImage(r.desc);
auto& cache = m_context.GetTextureCache();
r.image_id = cache.FindImage(r.desc);
BindRenderTarget(r.image_id);
}
bool RenderExecutor::DepthStencilCopy(CommandBuffer& buffer) {
const auto& hw = buffer.GetRegisters();
const auto& z = hw.GetDepthRenderTarget();
const auto& override = hw.GetDepthRenderOverride();
if (hw.GetColorControl().mode != 0) {
return false;
}
const bool depth_copy = override.force_z_dirty && override.force_z_valid &&
z.z_info.format != Prospero::DepthFormat::kInvalid &&
z.z_read_base_addr != 0 && z.z_write_base_addr != 0 &&
z.z_read_base_addr != z.z_write_base_addr;
const bool stencil_copy = override.force_stencil_dirty && override.force_stencil_valid &&
z.stencil_info.format != Prospero::StencilFormat::kInvalid &&
z.stencil_read_base_addr != 0 && z.stencil_write_base_addr != 0 &&
z.stencil_read_base_addr != z.stencil_write_base_addr;
if (!depth_copy && !stencil_copy) {
return false;
}
auto read_desc = MakeDepthTargetDesc(buffer, z);
auto write_desc = MakeDepthTargetDesc(buffer, z, true);
auto& cache = m_context.GetTextureCache();
const auto read_id = cache.FindImage(read_desc);
BindRenderTarget(read_id);
const auto write_id = cache.FindImage(write_desc);
BindRenderTarget(write_id);
cache.UpdateImage(read_id);
cache.UpdateImage(write_id);
cache.MarkGpuWritten(write_id);
auto& source = cache.GetImage(read_id);
auto& destination = cache.GetImage(write_id);
EXIT_IF(read_id == write_id || source.backing.format != destination.backing.format);
auto& scheduler = m_context.GetCommandScheduler();
scheduler.EndRendering();
const auto command = scheduler.Current().Handle();
const ImageSubresourceRange range {read_desc.view_info.base_level, 1,
read_desc.view_info.base_layer,
read_desc.view_info.layer_count};
source.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead,
range, command);
destination.Transit(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite,
range, command);
std::array<vk::ImageCopy, 2> regions {};
uint32_t count = 0;
for (const auto aspect: {vk::ImageAspectFlagBits::eDepth, vk::ImageAspectFlagBits::eStencil}) {
if ((aspect == vk::ImageAspectFlagBits::eDepth && !depth_copy) ||
(aspect == vk::ImageAspectFlagBits::eStencil && !stencil_copy)) {
continue;
}
auto& region = regions[count++];
region.srcSubresource = {aspect, range.base_level, range.base_layer,
range.layer_count};
region.dstSubresource = region.srcSubresource;
region.extent = write_desc.info.extent;
}
command.copyImage(source.backing.image, vk::ImageLayout::eTransferSrcOptimal,
destination.backing.image, vk::ImageLayout::eTransferDstOptimal,
count, regions.data());
return true;
}
vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
const auto format = desc.view_info.format;
if (format == vk::Format::eUndefined) {
+1
View File
@@ -183,6 +183,7 @@ private:
uint32_t render_target_slice_offset, uint32_t render_target_slot,
bool ignore_target_mask = false, bool exact_format = false);
void ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepthInfo& target);
[[nodiscard]] bool DepthStencilCopy(CommandBuffer& buffer);
[[nodiscard]] bool PrepareDrawRenderState(CommandBuffer& buffer,
const DrawCallInfo& draw,
uint32_t render_target_slice_offset,
@@ -1238,7 +1238,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
return;
}
if (ConsumeMetadataColorOperation(buffer)) {
if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer)) {
ResetBindings();
return;
}
@@ -1354,7 +1354,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
return;
}
if (ConsumeMetadataColorOperation(buffer)) {
if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer)) {
ResetBindings();
return;
}