RPCS3/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2024-11-26 04:32:35 +01:00
Code Issues Releases Wiki Activity

rsx: Refactor and fix framebuffer layout checks

Browse Source 
- Refactors shared code back into rsx core
- Adds extra check to avoid contest confusion
This commit is contained in:
kd-11 2018-07-24 00:55:15 +03:00 committed by kd-11
parent 0267221586
commit d0165290b6
5 changed files with 347 additions and 437 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

241
rpcs3/Emu/RSX/GL/GLRenderTargets.cpp
View File

@ -173,6 +173,12 @@ namespace
void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool skip_reading) void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool skip_reading)
{ {
if (m_framebuffer_state_contested && (m_framebuffer_contest_type != context))
{
// Clear commands affect contested memory
m_rtts_dirty = true;
}
if (draw_fbo && !m_rtts_dirty) if (draw_fbo && !m_rtts_dirty)
{ {
set_viewport(); set_viewport();
@ -180,221 +186,41 @@ void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool sk
} }
m_rtts_dirty = false; m_rtts_dirty = false;
zcull_surface_active = false;
const u16 clip_horizontal = rsx::method_registers.surface_clip_width();
const u16 clip_vertical = rsx::method_registers.surface_clip_height();
const u16 clip_x = rsx::method_registers.surface_clip_origin_x();
const u16 clip_y = rsx::method_registers.surface_clip_origin_y();
framebuffer_status_valid = false; framebuffer_status_valid = false;
m_framebuffer_state_contested = false; m_framebuffer_state_contested = false;
if (clip_horizontal == 0 || clip_vertical == 0) const auto layout = get_framebuffer_layout(context);
if (!framebuffer_status_valid)
{ {
LOG_ERROR(RSX, "Invalid framebuffer setup, w=%d, h=%d", clip_horizontal, clip_vertical);
return; return;
} }
auto surface_addresses = get_color_surface_addresses(); if (draw_fbo && layout.ignore_change)
auto depth_address = get_zeta_surface_address();
const auto pitchs = get_pitchs();
const auto zeta_pitch = rsx::method_registers.surface_z_pitch();
const auto surface_format = rsx::method_registers.surface_color();
const auto depth_format = rsx::method_registers.surface_depth_fmt();
const auto target = rsx::method_registers.surface_color_target();
const auto aa_mode = rsx::method_registers.surface_antialias();
const u32 aa_factor_u = (aa_mode == rsx::surface_antialiasing::center_1_sample) ? 1 : 2;
const u32 aa_factor_v = (aa_mode == rsx::surface_antialiasing::center_1_sample || aa_mode == rsx::surface_antialiasing::diagonal_centered_2_samples) ? 1 : 2;
//NOTE: Its is possible that some renders are done on a swizzled context. Pitch is meaningless in that case
//Seen in Nier (color) and GT HD concept (z buffer)
//Restriction is that the dimensions are powers of 2. Also, dimensions are passed via log2w and log2h entries
const auto required_zeta_pitch = std::max<u32>((u32)(depth_format == rsx::surface_depth_format::z16 ? clip_horizontal * 2 : clip_horizontal * 4) * aa_factor_u, 64u);
const auto required_color_pitch = std::max<u32>((u32)rsx::utility::get_packed_pitch(surface_format, clip_horizontal) * aa_factor_u, 64u);
const bool stencil_test_enabled = depth_format == rsx::surface_depth_format::z24s8 && rsx::method_registers.stencil_test_enabled();
const auto lg2w = rsx::method_registers.surface_log2_width();
const auto lg2h = rsx::method_registers.surface_log2_height();
const auto clipw_log2 = (u32)floor(log2(clip_horizontal));
const auto cliph_log2 = (u32)floor(log2(clip_vertical));
if (depth_address)
{ {
if (!rsx::method_registers.depth_test_enabled() && // Nothing has changed, we're still using the same framebuffer
!stencil_test_enabled && // Update flags to match current
target != rsx::surface_target::none) for (u32 index = 0; index < 4; index++)
{ {
//Disable depth buffer if depth testing is not enabled, unless a clear command is targeting the depth buffer if (auto surface = std::get<1>(m_rtts.m_bound_render_targets[index]))
const bool is_depth_clear = !!(context & rsx::framebuffer_creation_context::context_clear_depth);
if (!is_depth_clear)
{ {
depth_address = 0; surface->write_aa_mode = layout.aa_mode;
m_framebuffer_state_contested = true;
} }
} }
if (depth_address && zeta_pitch < required_zeta_pitch) if (auto ds = std::get<1>(m_rtts.m_bound_depth_stencil))
{ {
if (lg2w < clipw_log2 || lg2h < cliph_log2) ds->write_aa_mode = layout.aa_mode;
{
//Cannot fit
depth_address = 0;
if (lg2w > 0 || lg2h > 0)
{
//Something was actually declared for the swizzle context dimensions
LOG_WARNING(RSX, "Invalid swizzled context depth surface dims, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, clip_horizontal, clip_vertical);
}
}
else
{
LOG_TRACE(RSX, "Swizzled context depth surface, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, clip_horizontal, clip_vertical);
}
}
}
for (const auto &index : rsx::utility::get_rtt_indexes(target))
{
if (pitchs[index] < required_color_pitch)
{
if (lg2w < clipw_log2 || lg2h < cliph_log2)
{
surface_addresses[index] = 0;
if (lg2w > 0 || lg2h > 0)
{
//Something was actually declared for the swizzle context dimensions
LOG_WARNING(RSX, "Invalid swizzled context color surface dims, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, clip_horizontal, clip_vertical);
}
}
else
{
LOG_TRACE(RSX, "Swizzled context color surface, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, clip_horizontal, clip_vertical);
}
} }
if (surface_addresses[index] == depth_address)
{
LOG_TRACE(RSX, "Framebuffer at 0x%X has aliasing color/depth targets, zeta_pitch = %d, color_pitch=%d", depth_address, zeta_pitch, pitchs[index]);
//TODO: Research clearing both depth AND color
//TODO: If context is creation_draw, deal with possibility of a lost buffer clear
if (context == rsx::framebuffer_creation_context::context_clear_depth ||
rsx::method_registers.depth_test_enabled() || stencil_test_enabled ||
(!rsx::method_registers.color_write_enabled() && rsx::method_registers.depth_write_enabled()))
{
// Use address for depth data
surface_addresses[index] = 0;
}
else
{
// Use address for color data
depth_address = 0;
m_framebuffer_state_contested = true;
break;
}
}
if (surface_addresses[index])
framebuffer_status_valid = true;
}
if (!framebuffer_status_valid && !depth_address)
{
LOG_WARNING(RSX, "Framebuffer setup failed. Draw calls may have been lost");
return; return;
} }
//Window (raster) offsets m_rtts.prepare_render_target(nullptr, layout.color_format, layout.depth_format, layout.width, layout.height,
const auto window_offset_x = rsx::method_registers.window_offset_x(); layout.target, layout.color_addresses, layout.zeta_address);
const auto window_offset_y = rsx::method_registers.window_offset_y();
const auto window_clip_width = rsx::method_registers.window_clip_horizontal();
const auto window_clip_height = rsx::method_registers.window_clip_vertical();
const auto bpp = get_format_block_size_in_bytes(surface_format);
if (window_offset_x || window_offset_y)
{
//Window offset is what affects the raster position!
//Tested with Turbo: Super stunt squad that only changes the window offset to declare new framebuffers
//Sampling behavior clearly indicates the addresses are expected to have changed
if (auto clip_type = rsx::method_registers.window_clip_type())
LOG_ERROR(RSX, "Unknown window clip type 0x%X" HERE, clip_type);
for (const auto &index : rsx::utility::get_rtt_indexes(target))
{
if (surface_addresses[index])
{
const u32 window_offset_bytes = (std::max<u32>(pitchs[index], required_color_pitch) * window_offset_y) + ((aa_factor_u * bpp) * window_offset_x);
surface_addresses[index] += window_offset_bytes;
}
}
if (depth_address)
{
const auto depth_bpp = depth_format == rsx::surface_depth_format::z16 ? 2 : 4;
depth_address += (std::max<u32>(zeta_pitch, required_zeta_pitch) * window_offset_y) + ((aa_factor_u * depth_bpp) * window_offset_x);
}
}
if ((window_clip_width && window_clip_width < clip_horizontal) ||
(window_clip_height && window_clip_height < clip_vertical))
{
LOG_ERROR(RSX, "Unexpected window clip dimensions: window_clip=%dx%d, surface_clip=%dx%d",
window_clip_width, window_clip_height, clip_horizontal, clip_vertical);
}
if (draw_fbo)
{
bool really_changed = false;
auto sz = draw_fbo.get_extents();
if (sz.width == clip_horizontal && sz.height == clip_vertical)
{
for (u8 i = 0; i < rsx::limits::color_buffers_count; ++i)
{
if (m_surface_info[i].address != surface_addresses[i])
{
really_changed = true;
break;
}
}
if (!really_changed)
{
if (depth_address == m_depth_surface_info.address)
{
// Nothing has changed, we're still using the same framebuffer
// Update flags to match current
const auto aa_mode = rsx::method_registers.surface_antialias();
for (u32 index = 0; index < 4; index++)
{
if (auto surface = std::get<1>(m_rtts.m_bound_render_targets[index]))
{
surface->write_aa_mode = aa_mode;
}
}
if (auto ds = std::get<1>(m_rtts.m_bound_depth_stencil))
{
ds->write_aa_mode = aa_mode;
}
return;
}
}
}
}
m_rtts.prepare_render_target(nullptr, surface_format, depth_format, clip_horizontal, clip_vertical,
target, surface_addresses, depth_address);
draw_fbo.recreate(); draw_fbo.recreate();
draw_fbo.bind(); draw_fbo.bind();
draw_fbo.set_extents({ (int)clip_horizontal, (int)clip_vertical }); draw_fbo.set_extents({ (int)layout.width, (int)layout.height });
bool old_format_found = false; bool old_format_found = false;
gl::texture::format old_format; gl::texture::format old_format;
@ -421,13 +247,13 @@ void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool sk
auto rtt = std::get<1>(m_rtts.m_bound_render_targets[i]); auto rtt = std::get<1>(m_rtts.m_bound_render_targets[i]);
draw_fbo.color[i] = *rtt; draw_fbo.color[i] = *rtt;
rtt->set_rsx_pitch(pitchs[i]); rtt->set_rsx_pitch(layout.color_pitch[i]);
m_surface_info[i] = { surface_addresses[i], std::max(pitchs[i], required_color_pitch), false, surface_format, depth_format, clip_horizontal, clip_vertical }; m_surface_info[i] = { layout.color_addresses[i], layout.actual_color_pitch[i], false, layout.color_format, layout.depth_format, layout.width, layout.height };
rtt->tile = find_tile(color_offsets[i], color_locations[i]); rtt->tile = find_tile(color_offsets[i], color_locations[i]);
rtt->write_aa_mode = aa_mode; rtt->write_aa_mode = layout.aa_mode;
m_gl_texture_cache.notify_surface_changed(surface_addresses[i]); m_gl_texture_cache.notify_surface_changed(m_surface_info[i].address);
m_gl_texture_cache.tag_framebuffer(surface_addresses[i]); m_gl_texture_cache.tag_framebuffer(m_surface_info[i].address);
} }
else else
m_surface_info[i] = {}; m_surface_info[i] = {};
@ -445,18 +271,17 @@ void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool sk
} }
auto ds = std::get<1>(m_rtts.m_bound_depth_stencil); auto ds = std::get<1>(m_rtts.m_bound_depth_stencil);
if (depth_format == rsx::surface_depth_format::z24s8) if (layout.depth_format == rsx::surface_depth_format::z24s8)
draw_fbo.depth_stencil = *ds; draw_fbo.depth_stencil = *ds;
else else
draw_fbo.depth = *ds; draw_fbo.depth = *ds;
std::get<1>(m_rtts.m_bound_depth_stencil)->set_rsx_pitch(rsx::method_registers.surface_z_pitch()); std::get<1>(m_rtts.m_bound_depth_stencil)->set_rsx_pitch(rsx::method_registers.surface_z_pitch());
m_depth_surface_info = { depth_address, std::max(zeta_pitch, required_zeta_pitch), true, surface_format, depth_format, clip_horizontal, clip_vertical }; m_depth_surface_info = { layout.zeta_address, layout.actual_zeta_pitch, true, layout.color_format, layout.depth_format, layout.width, layout.height };
ds->write_aa_mode = aa_mode; ds->write_aa_mode = layout.aa_mode;
m_gl_texture_cache.notify_surface_changed(depth_address); m_gl_texture_cache.notify_surface_changed(layout.zeta_address);
m_gl_texture_cache.tag_framebuffer(layout.zeta_address);
m_gl_texture_cache.tag_framebuffer(depth_address);
} }
else else
m_depth_surface_info = {}; m_depth_surface_info = {};
@ -502,13 +327,13 @@ void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool sk
//Mark buffer regions as NO_ACCESS on Cell visible side //Mark buffer regions as NO_ACCESS on Cell visible side
if (g_cfg.video.write_color_buffers) if (g_cfg.video.write_color_buffers)
{ {
auto color_format = rsx::internals::surface_color_format_to_gl(surface_format); auto color_format = rsx::internals::surface_color_format_to_gl(layout.color_format);
for (u8 i = 0; i < rsx::limits::color_buffers_count; ++i) for (u8 i = 0; i < rsx::limits::color_buffers_count; ++i)
{ {
if (!m_surface_info[i].address || !m_surface_info[i].pitch) continue; if (!m_surface_info[i].address || !m_surface_info[i].pitch) continue;
const u32 range = m_surface_info[i].pitch * m_surface_info[i].height * aa_factor_v; const u32 range = m_surface_info[i].pitch * m_surface_info[i].height * layout.aa_factors[1];
m_gl_texture_cache.lock_memory_region(std::get<1>(m_rtts.m_bound_render_targets[i]), m_surface_info[i].address, range, m_surface_info[i].width, m_surface_info[i].height, m_surface_info[i].pitch, m_gl_texture_cache.lock_memory_region(std::get<1>(m_rtts.m_bound_render_targets[i]), m_surface_info[i].address, range, m_surface_info[i].width, m_surface_info[i].height, m_surface_info[i].pitch,
color_format.format, color_format.type, color_format.swap_bytes); color_format.format, color_format.type, color_format.swap_bytes);
} }
@ -518,8 +343,8 @@ void GLGSRender::init_buffers(rsx::framebuffer_creation_context context, bool sk
{ {
if (m_depth_surface_info.address && m_depth_surface_info.pitch) if (m_depth_surface_info.address && m_depth_surface_info.pitch)
{ {
const auto depth_format_gl = rsx::internals::surface_depth_format_to_gl(depth_format); const auto depth_format_gl = rsx::internals::surface_depth_format_to_gl(layout.depth_format);
const u32 range = m_depth_surface_info.pitch * m_depth_surface_info.height * aa_factor_v; const u32 range = m_depth_surface_info.pitch * m_depth_surface_info.height * layout.aa_factors[1];
m_gl_texture_cache.lock_memory_region(std::get<1>(m_rtts.m_bound_depth_stencil), m_depth_surface_info.address, range, m_depth_surface_info.width, m_depth_surface_info.height, m_depth_surface_info.pitch, m_gl_texture_cache.lock_memory_region(std::get<1>(m_rtts.m_bound_depth_stencil), m_depth_surface_info.address, range, m_depth_surface_info.width, m_depth_surface_info.height, m_depth_surface_info.pitch,
depth_format_gl.format, depth_format_gl.type, true); depth_format_gl.format, depth_format_gl.type, true);
} }

235
rpcs3/Emu/RSX/RSXThread.cpp
View File

@ -8,6 +8,7 @@
#include "Common/BufferUtils.h" #include "Common/BufferUtils.h"
#include "Common/texture_cache.h" #include "Common/texture_cache.h"
#include "Common/surface_store.h"
#include "Capture/rsx_capture.h" #include "Capture/rsx_capture.h"
#include "rsx_methods.h" #include "rsx_methods.h"
#include "rsx_utils.h" #include "rsx_utils.h"
@ -1443,6 +1444,240 @@ namespace rsx
return rsx::get_address(offset_zeta, m_context_dma_z); return rsx::get_address(offset_zeta, m_context_dma_z);
} }
framebuffer_layout thread::get_framebuffer_layout(rsx::framebuffer_creation_context context)
{
framebuffer_layout layout;
memset(&layout, 0, sizeof(layout));
layout.ignore_change = true;
layout.width = rsx::method_registers.surface_clip_width();
layout.height = rsx::method_registers.surface_clip_height();
framebuffer_status_valid = false;
m_framebuffer_state_contested = false;
m_framebuffer_contest_type = context;
if (layout.width == 0 || layout.height == 0)
{
LOG_TRACE(RSX, "Invalid framebuffer setup, w=%d, h=%d", layout.width, layout.height);
return layout;
}
const u16 clip_x = rsx::method_registers.surface_clip_origin_x();
const u16 clip_y = rsx::method_registers.surface_clip_origin_y();
layout.color_addresses = get_color_surface_addresses();
layout.zeta_address = get_zeta_surface_address();
layout.zeta_pitch = rsx::method_registers.surface_z_pitch();
layout.color_pitch =
{
rsx::method_registers.surface_a_pitch(),
rsx::method_registers.surface_b_pitch(),
rsx::method_registers.surface_c_pitch(),
rsx::method_registers.surface_d_pitch(),
};
layout.color_format = rsx::method_registers.surface_color();
layout.depth_format = rsx::method_registers.surface_depth_fmt();
layout.target = rsx::method_registers.surface_color_target();
const auto aa_mode = rsx::method_registers.surface_antialias();
const u32 aa_factor_u = (aa_mode == rsx::surface_antialiasing::center_1_sample) ? 1 : 2;
const u32 aa_factor_v = (aa_mode == rsx::surface_antialiasing::center_1_sample || aa_mode == rsx::surface_antialiasing::diagonal_centered_2_samples) ? 1 : 2;
// NOTE: Its is possible that some renders are done on a swizzled context. Pitch is meaningless in that case
// Seen in Nier (color) and GT HD concept (z buffer)
// Restriction is that the dimensions are powers of 2. Also, dimensions are passed via log2w and log2h entries
const auto required_zeta_pitch = std::max<u32>((u32)(layout.depth_format == rsx::surface_depth_format::z16 ? layout.width * 2 : layout.width * 4) * aa_factor_u, 64u);
const auto required_color_pitch = std::max<u32>((u32)rsx::utility::get_packed_pitch(layout.color_format, layout.width) * aa_factor_u, 64u);
const bool color_write_enabled = (context & rsx::framebuffer_creation_context::context_clear_color) ? true : rsx::method_registers.color_write_enabled();
const bool depth_write_enabled = (context & rsx::framebuffer_creation_context::context_clear_depth) ? true : rsx::method_registers.depth_write_enabled();
const auto lg2w = rsx::method_registers.surface_log2_width();
const auto lg2h = rsx::method_registers.surface_log2_height();
const auto clipw_log2 = (u32)floor(log2(layout.width));
const auto cliph_log2 = (u32)floor(log2(layout.height));
const bool stencil_test_enabled = layout.depth_format == rsx::surface_depth_format::z24s8 && rsx::method_registers.stencil_test_enabled();
const bool depth_test_enabled = rsx::method_registers.depth_test_enabled();
const bool ignore_depth = (context == rsx::framebuffer_creation_context::context_clear_color);
const bool ignore_color = (context == rsx::framebuffer_creation_context::context_clear_depth);
if (layout.zeta_address)
{
if (!depth_test_enabled &&
!stencil_test_enabled &&
layout.target != rsx::surface_target::none)
{
// Disable depth buffer if depth testing is not enabled, unless a clear command is targeting the depth buffer
const bool is_depth_clear = !!(context & rsx::framebuffer_creation_context::context_clear_depth);
if (!is_depth_clear)
{
layout.zeta_address = 0;
m_framebuffer_state_contested = true;
}
}
if (layout.zeta_address && layout.zeta_pitch < required_zeta_pitch)
{
if (lg2w < clipw_log2 || lg2h < cliph_log2)
{
// Cannot fit
layout.zeta_address = 0;
if (lg2w > 0 || lg2h > 0)
{
// Something was actually declared for the swizzle context dimensions
LOG_WARNING(RSX, "Invalid swizzled context depth surface dims, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, layout.width, layout.height);
}
}
else
{
LOG_TRACE(RSX, "Swizzled context depth surface, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, layout.width, layout.height);
}
}
if (layout.zeta_address)
{
// Still exists? Unlikely to get discarded
layout.actual_zeta_pitch = std::max(layout.zeta_pitch, required_zeta_pitch);
}
}
for (const auto &index : rsx::utility::get_rtt_indexes(layout.target))
{
if (layout.color_pitch[index] < required_color_pitch)
{
if (lg2w < clipw_log2 || lg2h < cliph_log2)
{
layout.color_addresses[index] = 0;
if (lg2w > 0 || lg2h > 0)
{
// Something was actually declared for the swizzle context dimensions
LOG_WARNING(RSX, "Invalid swizzled context color surface dims, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, layout.width, layout.height);
}
}
else
{
LOG_TRACE(RSX, "Swizzled context color surface, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, layout.width, layout.height);
}
}
if (layout.zeta_address && (layout.color_addresses[index] == layout.zeta_address))
{
LOG_TRACE(RSX, "Framebuffer at 0x%X has aliasing color/depth targets, color_index=%d, zeta_pitch = %d, color_pitch=%d, context=%d",
layout.zeta_address, index, layout.zeta_pitch, layout.color_pitch[index], (u32)context);
// TODO: Research clearing both depth AND color
// TODO: If context is creation_draw, deal with possibility of a lost buffer clear
if (!ignore_depth &&
(ignore_color || depth_test_enabled || stencil_test_enabled ||
(!color_write_enabled && depth_write_enabled)))
{
// Use address for depth data
layout.color_addresses[index] = 0;
}
else
{
// Use address for color data
layout.zeta_address = 0;
m_framebuffer_state_contested = true;
}
}
if (layout.color_addresses[index])
{
layout.actual_color_pitch[index] = std::max(layout.color_pitch[index], required_color_pitch);
framebuffer_status_valid = true;
}
}
if (!framebuffer_status_valid && !layout.zeta_address)
{
LOG_WARNING(RSX, "Framebuffer setup failed. Draw calls may have been lost");
return layout;
}
// At least one attachment exists
framebuffer_status_valid = true;
// Window (raster) offsets
const auto window_offset_x = rsx::method_registers.window_offset_x();
const auto window_offset_y = rsx::method_registers.window_offset_y();
const auto window_clip_width = rsx::method_registers.window_clip_horizontal();
const auto window_clip_height = rsx::method_registers.window_clip_vertical();
const auto bpp = get_format_block_size_in_bytes(layout.color_format);
if (window_offset_x || window_offset_y)
{
// Window offset is what affects the raster position!
// Tested with Turbo: Super stunt squad that only changes the window offset to declare new framebuffers
// Sampling behavior clearly indicates the addresses are expected to have changed
if (auto clip_type = rsx::method_registers.window_clip_type())
LOG_ERROR(RSX, "Unknown window clip type 0x%X" HERE, clip_type);
for (const auto &index : rsx::utility::get_rtt_indexes(layout.target))
{
if (layout.color_addresses[index])
{
const u32 window_offset_bytes = (std::max<u32>(layout.color_pitch[index], required_color_pitch) * window_offset_y) + ((aa_factor_u * bpp) * window_offset_x);
layout.color_addresses[index] += window_offset_bytes;
}
}
if (layout.zeta_address)
{
const auto depth_bpp = (layout.depth_format == rsx::surface_depth_format::z16) ? 2 : 4;
layout.zeta_address += (std::max<u32>(layout.zeta_pitch, required_zeta_pitch) * window_offset_y) + ((aa_factor_u * depth_bpp) * window_offset_x);
}
}
if ((window_clip_width && window_clip_width < layout.width) ||
(window_clip_height && window_clip_height < layout.height))
{
LOG_ERROR(RSX, "Unexpected window clip dimensions: window_clip=%dx%d, surface_clip=%dx%d",
window_clip_width, window_clip_height, layout.width, layout.height);
}
layout.aa_mode = aa_mode;
layout.aa_factors[0] = aa_factor_u;
layout.aa_factors[1] = aa_factor_v;
bool really_changed = false;
for (u8 i = 0; i < rsx::limits::color_buffers_count; ++i)
{
if (m_surface_info[i].address != layout.color_addresses[i])
{
really_changed = true;
break;
}
if (layout.color_addresses[i])
{
if (m_surface_info[i].width != layout.width ||
m_surface_info[i].height != layout.height)
{
really_changed = true;
break;
}
}
}
if (!really_changed)
{
if (layout.zeta_address == m_depth_surface_info.address)
{
// Same target is reused
return layout;
}
}
layout.ignore_change = false;
return layout;
}
void thread::get_current_vertex_program(const std::array<std::unique_ptr<rsx::sampled_image_descriptor_base>, rsx::limits::vertex_textures_count>& sampler_descriptors, bool skip_textures, bool skip_vertex_inputs) void thread::get_current_vertex_program(const std::array<std::unique_ptr<rsx::sampled_image_descriptor_base>, rsx::limits::vertex_textures_count>& sampler_descriptors, bool skip_textures, bool skip_vertex_inputs)
{ {
if (!(m_graphics_state & rsx::pipeline_state::vertex_program_dirty)) if (!(m_graphics_state & rsx::pipeline_state::vertex_program_dirty))

22
rpcs3/Emu/RSX/RSXThread.h
View File

@ -213,6 +213,24 @@ namespace rsx
std::array<attribute_buffer_placement, 16> attribute_placement; std::array<attribute_buffer_placement, 16> attribute_placement;
}; };
struct framebuffer_layout
{
u16 width;
u16 height;
std::array<u32, 4> color_addresses;
std::array<u32, 4> color_pitch;
std::array<u32, 4> actual_color_pitch;
u32 zeta_address;
u32 zeta_pitch;
u32 actual_zeta_pitch;
rsx::surface_target target;
rsx::surface_color_format color_format;
rsx::surface_depth_format depth_format;
rsx::surface_antialiasing aa_mode;
u32 aa_factors[2];
bool ignore_change;
};
namespace reports namespace reports
{ {
struct occlusion_query_info struct occlusion_query_info
@ -391,6 +409,8 @@ namespace rsx
bool m_textures_dirty[16]; bool m_textures_dirty[16];
bool m_vertex_textures_dirty[4]; bool m_vertex_textures_dirty[4];
bool m_framebuffer_state_contested = false; bool m_framebuffer_state_contested = false;
rsx::framebuffer_creation_context m_framebuffer_contest_type = rsx::framebuffer_creation_context::context_draw;
u32 m_graphics_state = 0; u32 m_graphics_state = 0;
u64 ROP_sync_timestamp = 0; u64 ROP_sync_timestamp = 0;
@ -401,6 +421,8 @@ namespace rsx
std::array<u32, 4> get_color_surface_addresses() const; std::array<u32, 4> get_color_surface_addresses() const;
u32 get_zeta_surface_address() const; u32 get_zeta_surface_address() const;
framebuffer_layout get_framebuffer_layout(rsx::framebuffer_creation_context context);
/** /**
* Analyze vertex inputs and group all interleaved blocks * Analyze vertex inputs and group all interleaved blocks
*/ */

278
rpcs3/Emu/RSX/VK/VKGSRender.cpp
View File

@ -2516,231 +2516,55 @@ void VKGSRender::open_command_buffer()
m_current_command_buffer->begin(); m_current_command_buffer->begin();
} }
void VKGSRender::prepare_rtts(rsx::framebuffer_creation_context context) void VKGSRender::prepare_rtts(rsx::framebuffer_creation_context context)
{ {
if (m_framebuffer_state_contested && (m_framebuffer_contest_type != context))
{
// Clear commands affect contested memory
m_rtts_dirty = true;
}
if (m_draw_fbo && !m_rtts_dirty) if (m_draw_fbo && !m_rtts_dirty)
return; return;
m_rtts_dirty = false; m_rtts_dirty = false;
u32 clip_width = rsx::method_registers.surface_clip_width();
u32 clip_height = rsx::method_registers.surface_clip_height();
u32 clip_x = rsx::method_registers.surface_clip_origin_x();
u32 clip_y = rsx::method_registers.surface_clip_origin_y();
framebuffer_status_valid = false; framebuffer_status_valid = false;
m_framebuffer_state_contested = false; m_framebuffer_state_contested = false;
if (clip_width == 0 || clip_height == 0) const auto layout = get_framebuffer_layout(context);
if (!framebuffer_status_valid)
{ {
LOG_ERROR(RSX, "Invalid framebuffer setup, w=%d, h=%d", clip_width, clip_height);
return; return;
} }
auto surface_addresses = get_color_surface_addresses(); if (m_draw_fbo && layout.ignore_change)
auto zeta_address = get_zeta_surface_address();
const auto zeta_pitch = rsx::method_registers.surface_z_pitch();
const u32 surface_pitchs[] = { rsx::method_registers.surface_a_pitch(), rsx::method_registers.surface_b_pitch(),
rsx::method_registers.surface_c_pitch(), rsx::method_registers.surface_d_pitch() };
const auto color_fmt = rsx::method_registers.surface_color();
const auto depth_fmt = rsx::method_registers.surface_depth_fmt();
const auto target = rsx::method_registers.surface_color_target();
const auto aa_mode = rsx::method_registers.surface_antialias();
const u32 aa_factor_u = (aa_mode == rsx::surface_antialiasing::center_1_sample) ? 1 : 2;
const u32 aa_factor_v = (aa_mode == rsx::surface_antialiasing::center_1_sample || aa_mode == rsx::surface_antialiasing::diagonal_centered_2_samples) ? 1 : 2;
//NOTE: Its is possible that some renders are done on a swizzled context. Pitch is meaningless in that case
//Seen in Nier (color) and GT HD concept (z buffer)
//Restriction is that the dimensions are powers of 2. Also, dimensions are passed via log2w and log2h entries
const auto required_zeta_pitch = std::max<u32>((u32)(depth_fmt == rsx::surface_depth_format::z16 ? clip_width * 2 : clip_width * 4) * aa_factor_u, 64u);
const auto required_color_pitch = std::max<u32>((u32)rsx::utility::get_packed_pitch(color_fmt, clip_width) * aa_factor_v, 64u);
const bool stencil_test_enabled = depth_fmt == rsx::surface_depth_format::z24s8 && rsx::method_registers.stencil_test_enabled();
const auto lg2w = rsx::method_registers.surface_log2_width();
const auto lg2h = rsx::method_registers.surface_log2_height();
const auto clipw_log2 = (u32)floor(log2(clip_width));
const auto cliph_log2 = (u32)floor(log2(clip_height));
if (zeta_address)
{ {
if (!rsx::method_registers.depth_test_enabled() && // Nothing has changed, we're still using the same framebuffer
!stencil_test_enabled && // Update flags to match current
target != rsx::surface_target::none)
const auto aa_mode = rsx::method_registers.surface_antialias();
for (u32 index = 0; index < 4; index++)
{ {
//Disable depth buffer if depth testing is not enabled, unless a clear command is targeting the depth buffer if (auto surface = std::get<1>(m_rtts.m_bound_render_targets[index]))
const bool is_depth_clear = !!(context & rsx::framebuffer_creation_context::context_clear_depth);
if (!is_depth_clear)
{ {
zeta_address = 0; surface->write_aa_mode = layout.aa_mode;
m_framebuffer_state_contested = true;
} }
} }
if (zeta_address && zeta_pitch < required_zeta_pitch) if (auto ds = std::get<1>(m_rtts.m_bound_depth_stencil))
{ {
if (lg2w < clipw_log2 || lg2h < cliph_log2) ds->write_aa_mode = layout.aa_mode;
{
//Cannot fit
zeta_address = 0;
if (lg2w > 0 || lg2h > 0)
{
//Something was actually declared for the swizzle context dimensions
LOG_WARNING(RSX, "Invalid swizzled context depth surface dims, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, clip_width, clip_height);
}
}
else
{
LOG_TRACE(RSX, "Swizzled context depth surface, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, clip_width, clip_height);
}
}
}
for (const auto &index : rsx::utility::get_rtt_indexes(target))
{
if (surface_pitchs[index] < required_color_pitch)
{
if (lg2w < clipw_log2 || lg2h < cliph_log2)
{
surface_addresses[index] = 0;
if (lg2w > 0 || lg2h > 0)
{
//Something was actually declared for the swizzle context dimensions
LOG_WARNING(RSX, "Invalid swizzled context color surface dims, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, clip_width, clip_height);
}
}
else
{
LOG_TRACE(RSX, "Swizzled context color surface, LG2W=%d, LG2H=%d, clip_w=%d, clip_h=%d", lg2w, lg2h, clip_width, clip_height);
}
} }
if (surface_addresses[index] == zeta_address)
{
LOG_TRACE(RSX, "Framebuffer at 0x%X has aliasing color/depth targets, zeta_pitch = %d, color_pitch=%d", zeta_address, zeta_pitch, surface_pitchs[index]);
if (context == rsx::framebuffer_creation_context::context_clear_depth ||
rsx::method_registers.depth_test_enabled() || stencil_test_enabled ||
(!rsx::method_registers.color_write_enabled() && rsx::method_registers.depth_write_enabled()))
{
// Use address for depth data
// TODO: create a temporary render buffer for this to keep MRT outputs aligned
surface_addresses[index] = 0;
}
else
{
// Use address for color data
zeta_address = 0;
m_framebuffer_state_contested = true;
break;
}
}
if (surface_addresses[index])
framebuffer_status_valid = true;
}
if (!framebuffer_status_valid && !zeta_address)
{
LOG_WARNING(RSX, "Framebuffer setup failed. Draw calls may have been lost");
return; return;
} }
//At least one attachment exists
framebuffer_status_valid = true;
const auto fbo_width = rsx::apply_resolution_scale(clip_width, true);
const auto fbo_height = rsx::apply_resolution_scale(clip_height, true);
const auto bpp = get_format_block_size_in_bytes(color_fmt);
//Window (raster) offsets
const auto window_offset_x = rsx::method_registers.window_offset_x();
const auto window_offset_y = rsx::method_registers.window_offset_y();
const auto window_clip_width = rsx::method_registers.window_clip_horizontal();
const auto window_clip_height = rsx::method_registers.window_clip_vertical();
if (window_offset_x || window_offset_y)
{
//Window offset is what affects the raster position!
//Tested with Turbo: Super stunt squad that only changes the window offset to declare new framebuffers
//Sampling behavior clearly indicates the addresses are expected to have changed
if (auto clip_type = rsx::method_registers.window_clip_type())
LOG_ERROR(RSX, "Unknown window clip type 0x%X" HERE, clip_type);
for (const auto &index : rsx::utility::get_rtt_indexes(target))
{
if (surface_addresses[index])
{
const u32 window_offset_bytes = (std::max<u32>(surface_pitchs[index], required_color_pitch) * window_offset_y) + ((aa_factor_u * bpp) * window_offset_x);
surface_addresses[index] += window_offset_bytes;
}
}
if (zeta_address)
{
const auto depth_bpp = (depth_fmt == rsx::surface_depth_format::z16 ? 2 : 4);
zeta_address += (std::max<u32>(zeta_pitch, required_zeta_pitch) * window_offset_y) + ((aa_factor_u * depth_bpp) * window_offset_x);
}
}
if ((window_clip_width && window_clip_width < clip_width) ||
(window_clip_height && window_clip_height < clip_height))
{
LOG_ERROR(RSX, "Unexpected window clip dimensions: window_clip=%dx%d, surface_clip=%dx%d",
window_clip_width, window_clip_height, clip_width, clip_height);
}
if (m_draw_fbo)
{
bool really_changed = false;
if (m_draw_fbo->width() == fbo_width && m_draw_fbo->height() == fbo_height)
{
for (u8 i = 0; i < rsx::limits::color_buffers_count; ++i)
{
if (m_surface_info[i].address != surface_addresses[i])
{
really_changed = true;
break;
}
}
if (!really_changed)
{
if (zeta_address == m_depth_surface_info.address)
{
// Nothing has changed, we're still using the same framebuffer
// Update flags to match current
const auto aa_mode = rsx::method_registers.surface_antialias();
for (u32 index = 0; index < 4; index++)
{
if (auto surface = std::get<1>(m_rtts.m_bound_render_targets[index]))
{
surface->write_aa_mode = aa_mode;
}
}
if (auto ds = std::get<1>(m_rtts.m_bound_depth_stencil))
{
ds->write_aa_mode = aa_mode;
}
return;
}
}
}
}
m_rtts.prepare_render_target(&*m_current_command_buffer, m_rtts.prepare_render_target(&*m_current_command_buffer,
color_fmt, depth_fmt, layout.color_format, layout.depth_format,
clip_width, clip_height, layout.width, layout.height,
target, layout.target,
surface_addresses, zeta_address, layout.color_addresses, layout.zeta_address,
(*m_device), &*m_current_command_buffer); (*m_device), &*m_current_command_buffer);
//Reset framebuffer information //Reset framebuffer information
@ -2759,9 +2583,9 @@ void VKGSRender::prepare_rtts(rsx::framebuffer_creation_context context)
} }
m_surface_info[i].address = m_surface_info[i].pitch = 0; m_surface_info[i].address = m_surface_info[i].pitch = 0;
m_surface_info[i].width = clip_width; m_surface_info[i].width = layout.width;
m_surface_info[i].height = clip_height; m_surface_info[i].height = layout.height;
m_surface_info[i].color_format = color_fmt; m_surface_info[i].color_format = layout.color_format;
} }
//Process depth surface as well //Process depth surface as well
@ -2775,13 +2599,13 @@ void VKGSRender::prepare_rtts(rsx::framebuffer_creation_context context)
} }
m_depth_surface_info.address = m_depth_surface_info.pitch = 0; m_depth_surface_info.address = m_depth_surface_info.pitch = 0;
m_depth_surface_info.width = clip_width; m_depth_surface_info.width = layout.width;
m_depth_surface_info.height = clip_height; m_depth_surface_info.height = layout.height;
m_depth_surface_info.depth_format = depth_fmt; m_depth_surface_info.depth_format = layout.depth_format;
} }
//Bind created rtts as current fbo... //Bind created rtts as current fbo...
std::vector<u8> draw_buffers = rsx::utility::get_rtt_indexes(target); const auto draw_buffers = rsx::utility::get_rtt_indexes(layout.target);
m_draw_buffers_count = 0; m_draw_buffers_count = 0;
std::vector<vk::image*> bound_images; std::vector<vk::image*> bound_images;
@ -2793,14 +2617,13 @@ void VKGSRender::prepare_rtts(rsx::framebuffer_creation_context context)
{ {
bound_images.push_back(surface); bound_images.push_back(surface);
m_surface_info[index].address = surface_addresses[index]; m_surface_info[index].address = layout.color_addresses[index];
m_surface_info[index].pitch = std::max(surface_pitchs[index], required_color_pitch); m_surface_info[index].pitch = layout.color_pitch[index];
surface->rsx_pitch = surface_pitchs[index]; surface->rsx_pitch = layout.color_pitch[index];
surface->write_aa_mode = aa_mode; surface->write_aa_mode = layout.aa_mode;
m_texture_cache.notify_surface_changed(surface_addresses[index]); m_texture_cache.notify_surface_changed(layout.color_addresses[index]);
m_texture_cache.tag_framebuffer(layout.color_addresses[index]);
m_texture_cache.tag_framebuffer(surface_addresses[index]);
m_draw_buffers_count++; m_draw_buffers_count++;
} }
} }
@ -2810,26 +2633,25 @@ void VKGSRender::prepare_rtts(rsx::framebuffer_creation_context context)
auto ds = std::get<1>(m_rtts.m_bound_depth_stencil); auto ds = std::get<1>(m_rtts.m_bound_depth_stencil);
bound_images.push_back(ds); bound_images.push_back(ds);
m_depth_surface_info.address = zeta_address; m_depth_surface_info.address = layout.zeta_address;
m_depth_surface_info.pitch = std::max(zeta_pitch, required_zeta_pitch); m_depth_surface_info.pitch = layout.zeta_pitch;
ds->rsx_pitch = m_depth_surface_info.pitch; ds->rsx_pitch = layout.zeta_pitch;
ds->write_aa_mode = aa_mode; ds->write_aa_mode = layout.aa_mode;
m_texture_cache.notify_surface_changed(zeta_address); m_texture_cache.notify_surface_changed(layout.zeta_address);
m_texture_cache.tag_framebuffer(layout.zeta_address);
m_texture_cache.tag_framebuffer(zeta_address);
} }
if (g_cfg.video.write_color_buffers) if (g_cfg.video.write_color_buffers)
{ {
const auto color_fmt_info = vk::get_compatible_gcm_format(color_fmt); const auto color_fmt_info = vk::get_compatible_gcm_format(layout.color_format);
for (u8 index : draw_buffers) for (u8 index : draw_buffers)
{ {
if (!m_surface_info[index].address || !m_surface_info[index].pitch) continue; if (!m_surface_info[index].address || !m_surface_info[index].pitch) continue;
const u32 range = m_surface_info[index].pitch * m_surface_info[index].height * aa_factor_v; const u32 range = m_surface_info[index].pitch * m_surface_info[index].height * layout.aa_factors[1];
m_texture_cache.lock_memory_region(std::get<1>(m_rtts.m_bound_render_targets[index]), m_surface_info[index].address, range, m_texture_cache.lock_memory_region(std::get<1>(m_rtts.m_bound_render_targets[index]), m_surface_info[index].address, range,
m_surface_info[index].width, m_surface_info[index].height, m_surface_info[index].pitch, color_fmt_info.first, color_fmt_info.second); m_surface_info[index].width, m_surface_info[index].height, layout.actual_color_pitch[index], color_fmt_info.first, color_fmt_info.second);
} }
} }
@ -2838,17 +2660,19 @@ void VKGSRender::prepare_rtts(rsx::framebuffer_creation_context context)
if (m_depth_surface_info.address && m_depth_surface_info.pitch) if (m_depth_surface_info.address && m_depth_surface_info.pitch)
{ {
const u32 gcm_format = (m_depth_surface_info.depth_format != rsx::surface_depth_format::z16)? CELL_GCM_TEXTURE_DEPTH16 : CELL_GCM_TEXTURE_DEPTH24_D8; const u32 gcm_format = (m_depth_surface_info.depth_format != rsx::surface_depth_format::z16)? CELL_GCM_TEXTURE_DEPTH16 : CELL_GCM_TEXTURE_DEPTH24_D8;
const u32 range = m_depth_surface_info.pitch * m_depth_surface_info.height * aa_factor_v; const u32 range = m_depth_surface_info.pitch * m_depth_surface_info.height * layout.aa_factors[1];
m_texture_cache.lock_memory_region(std::get<1>(m_rtts.m_bound_depth_stencil), m_depth_surface_info.address, range, m_texture_cache.lock_memory_region(std::get<1>(m_rtts.m_bound_depth_stencil), m_depth_surface_info.address, range,
m_depth_surface_info.width, m_depth_surface_info.height, m_depth_surface_info.pitch, gcm_format, false); m_depth_surface_info.width, m_depth_surface_info.height, layout.actual_zeta_pitch, gcm_format, false);
} }
} }
auto vk_depth_format = (zeta_address == 0) ? VK_FORMAT_UNDEFINED : vk::get_compatible_depth_surface_format(m_device->get_formats_support(), depth_fmt); auto vk_depth_format = (layout.zeta_address == 0) ? VK_FORMAT_UNDEFINED : vk::get_compatible_depth_surface_format(m_device->get_formats_support(), layout.depth_format);
m_current_renderpass_id = vk::get_render_pass_location(vk::get_compatible_surface_format(color_fmt).first, vk_depth_format, m_draw_buffers_count); m_current_renderpass_id = vk::get_render_pass_location(vk::get_compatible_surface_format(layout.color_format).first, vk_depth_format, m_draw_buffers_count);
//Search old framebuffers for this same configuration //Search old framebuffers for this same configuration
bool framebuffer_found = false; bool framebuffer_found = false;
const auto fbo_width = rsx::apply_resolution_scale(layout.width, true);
const auto fbo_height = rsx::apply_resolution_scale(layout.height, true);
for (auto &fbo : m_framebuffers_to_clean) for (auto &fbo : m_framebuffers_to_clean)
{ {

8
rpcs3/Emu/RSX/VK/VKHelpers.h
View File

@ -2688,8 +2688,12 @@ public:
void unmap() void unmap()
{ {
//mapped = false; if (g_cfg.video.disable_vulkan_mem_allocator)
//heap->unmap(); {
heap->unmap();
mapped = false;
_ptr = nullptr;
}
} }
}; };
} }