mirror of
https://git.eden-emu.dev/eden-emu/eden
synced 2026-05-16 19:07:00 +02:00
[video_core/engines/maxwell3d] memory inline DrawState to reduce indirection on hot paths (#3758)
usual indirection removal helps very slightly to codegen the idea is basically to reduce the amount of pointer deference overall in the code, and use idiomatic std::variant<>-isms to not rely on vtables/unique_ptr overhead this should allow the compiler to emit better code of course it's a tiny optimisation and only CPU side, but allows us to reduce indirection which is almost always a good thing "but youre passing more parameters to the function!!!" its literally memoized into a register my friend Signed-off-by: lizzie <lizzie@eden-emu.dev> Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3758 Reviewed-by: crueter <crueter@eden-emu.dev> Reviewed-by: MaranBr <maranbr@eden-emu.dev>
This commit is contained in:
lizzie
crueter
parent
ed225f8a8b
commit
77decca678
15 changed files with 231 additions and 351 deletions
|
|
@ -1,23 +1,24 @@
|
|||
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
|
||||
// SPDX-License-Identifier: GPL-3.0-or-later
|
||||
|
||||
// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
|
||||
// SPDX-License-Identifier: GPL-2.0-or-later
|
||||
|
||||
#include "common/settings.h"
|
||||
#include "video_core/dirty_flags.h"
|
||||
#include "video_core/engines/draw_manager.h"
|
||||
#include "video_core/engines/maxwell_3d.h"
|
||||
#include "video_core/rasterizer_interface.h"
|
||||
|
||||
namespace Tegra::Engines {
|
||||
DrawManager::DrawManager(Maxwell3D* maxwell3d_) : maxwell3d(maxwell3d_) {}
|
||||
|
||||
void DrawManager::ProcessMethodCall(u32 method, u32 argument) {
|
||||
const auto& regs{maxwell3d->regs};
|
||||
void Maxwell3D::DrawManager::ProcessMethodCall(Maxwell3D& maxwell3d, u32 method, u32 argument) {
|
||||
switch (method) {
|
||||
case MAXWELL3D_REG_INDEX(clear_surface):
|
||||
return Clear(1);
|
||||
return Clear(maxwell3d, 1);
|
||||
case MAXWELL3D_REG_INDEX(draw.begin):
|
||||
return DrawBegin();
|
||||
return DrawBegin(maxwell3d);
|
||||
case MAXWELL3D_REG_INDEX(draw.end):
|
||||
return DrawEnd();
|
||||
return DrawEnd(maxwell3d);
|
||||
case MAXWELL3D_REG_INDEX(vertex_buffer.first):
|
||||
case MAXWELL3D_REG_INDEX(vertex_buffer.count):
|
||||
case MAXWELL3D_REG_INDEX(index_buffer.first):
|
||||
|
|
@ -33,33 +34,29 @@ void DrawManager::ProcessMethodCall(u32 method, u32 argument) {
|
|||
case MAXWELL3D_REG_INDEX(index_buffer32_first):
|
||||
case MAXWELL3D_REG_INDEX(index_buffer16_first):
|
||||
case MAXWELL3D_REG_INDEX(index_buffer8_first):
|
||||
return DrawIndexSmall(argument);
|
||||
return DrawIndexSmall(maxwell3d, argument);
|
||||
case MAXWELL3D_REG_INDEX(draw_inline_index):
|
||||
SetInlineIndexBuffer(argument);
|
||||
SetInlineIndexBuffer(maxwell3d, argument);
|
||||
break;
|
||||
case MAXWELL3D_REG_INDEX(inline_index_2x16.even):
|
||||
SetInlineIndexBuffer(regs.inline_index_2x16.even);
|
||||
SetInlineIndexBuffer(regs.inline_index_2x16.odd);
|
||||
SetInlineIndexBuffer(maxwell3d, maxwell3d.regs.inline_index_2x16.even);
|
||||
SetInlineIndexBuffer(maxwell3d, maxwell3d.regs.inline_index_2x16.odd);
|
||||
break;
|
||||
case MAXWELL3D_REG_INDEX(inline_index_4x8.index0):
|
||||
SetInlineIndexBuffer(regs.inline_index_4x8.index0);
|
||||
SetInlineIndexBuffer(regs.inline_index_4x8.index1);
|
||||
SetInlineIndexBuffer(regs.inline_index_4x8.index2);
|
||||
SetInlineIndexBuffer(regs.inline_index_4x8.index3);
|
||||
SetInlineIndexBuffer(maxwell3d, maxwell3d.regs.inline_index_4x8.index0);
|
||||
SetInlineIndexBuffer(maxwell3d, maxwell3d.regs.inline_index_4x8.index1);
|
||||
SetInlineIndexBuffer(maxwell3d, maxwell3d.regs.inline_index_4x8.index2);
|
||||
SetInlineIndexBuffer(maxwell3d, maxwell3d.regs.inline_index_4x8.index3);
|
||||
break;
|
||||
case MAXWELL3D_REG_INDEX(vertex_array_instance_first):
|
||||
DrawArrayInstanced(regs.vertex_array_instance_first.topology.Value(),
|
||||
regs.vertex_array_instance_first.start.Value(),
|
||||
regs.vertex_array_instance_first.count.Value(), false);
|
||||
DrawArrayInstanced(maxwell3d, maxwell3d.regs.vertex_array_instance_first.topology.Value(), maxwell3d.regs.vertex_array_instance_first.start.Value(), maxwell3d.regs.vertex_array_instance_first.count.Value(), false);
|
||||
break;
|
||||
case MAXWELL3D_REG_INDEX(vertex_array_instance_subsequent): {
|
||||
DrawArrayInstanced(regs.vertex_array_instance_subsequent.topology.Value(),
|
||||
regs.vertex_array_instance_subsequent.start.Value(),
|
||||
regs.vertex_array_instance_subsequent.count.Value(), true);
|
||||
DrawArrayInstanced(maxwell3d, maxwell3d.regs.vertex_array_instance_subsequent.topology.Value(), maxwell3d.regs.vertex_array_instance_subsequent.start.Value(), maxwell3d.regs.vertex_array_instance_subsequent.count.Value(), true);
|
||||
break;
|
||||
}
|
||||
case MAXWELL3D_REG_INDEX(draw_texture.src_y0): {
|
||||
DrawTexture();
|
||||
DrawTexture(maxwell3d);
|
||||
break;
|
||||
}
|
||||
default:
|
||||
|
|
@ -67,101 +64,87 @@ void DrawManager::ProcessMethodCall(u32 method, u32 argument) {
|
|||
}
|
||||
}
|
||||
|
||||
void DrawManager::Clear(u32 layer_count) {
|
||||
if (maxwell3d->ShouldExecute()) {
|
||||
maxwell3d->rasterizer->Clear(layer_count);
|
||||
void Maxwell3D::DrawManager::Clear(Maxwell3D& maxwell3d, u32 layer_count) {
|
||||
if (maxwell3d.ShouldExecute()) {
|
||||
maxwell3d.rasterizer->Clear(layer_count);
|
||||
}
|
||||
}
|
||||
|
||||
void DrawManager::DrawDeferred() {
|
||||
void Maxwell3D::DrawManager::DrawDeferred(Maxwell3D& maxwell3d) {
|
||||
if (draw_state.draw_mode != DrawMode::Instance || draw_state.instance_count == 0) {
|
||||
return;
|
||||
}
|
||||
DrawEnd(draw_state.instance_count + 1, true);
|
||||
DrawEnd(maxwell3d, draw_state.instance_count + 1, true);
|
||||
draw_state.instance_count = 0;
|
||||
}
|
||||
|
||||
void DrawManager::DrawArray(PrimitiveTopology topology, u32 vertex_first, u32 vertex_count,
|
||||
u32 base_instance, u32 num_instances) {
|
||||
void Maxwell3D::DrawManager::DrawArray(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 vertex_first, u32 vertex_count, u32 base_instance, u32 num_instances) {
|
||||
draw_state.topology = topology;
|
||||
draw_state.vertex_buffer.first = vertex_first;
|
||||
draw_state.vertex_buffer.count = vertex_count;
|
||||
draw_state.base_instance = base_instance;
|
||||
ProcessDraw(false, num_instances);
|
||||
ProcessDraw(maxwell3d, false, num_instances);
|
||||
}
|
||||
|
||||
void DrawManager::DrawArrayInstanced(PrimitiveTopology topology, u32 vertex_first, u32 vertex_count,
|
||||
bool subsequent) {
|
||||
void Maxwell3D::DrawManager::DrawArrayInstanced(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 vertex_first, u32 vertex_count, bool subsequent) {
|
||||
draw_state.topology = topology;
|
||||
draw_state.vertex_buffer.first = vertex_first;
|
||||
draw_state.vertex_buffer.count = vertex_count;
|
||||
|
||||
if (!subsequent) {
|
||||
draw_state.instance_count = 1;
|
||||
}
|
||||
|
||||
draw_state.base_instance = draw_state.instance_count - 1;
|
||||
draw_state.draw_mode = DrawMode::Instance;
|
||||
draw_state.instance_count++;
|
||||
ProcessDraw(false, 1);
|
||||
ProcessDraw(maxwell3d, false, 1);
|
||||
}
|
||||
|
||||
void DrawManager::DrawIndex(PrimitiveTopology topology, u32 index_first, u32 index_count,
|
||||
u32 base_index, u32 base_instance, u32 num_instances) {
|
||||
const auto& regs{maxwell3d->regs};
|
||||
void Maxwell3D::DrawManager::DrawIndex(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 index_first, u32 index_count, u32 base_index, u32 base_instance, u32 num_instances) {
|
||||
draw_state.topology = topology;
|
||||
draw_state.index_buffer = regs.index_buffer;
|
||||
draw_state.index_buffer = maxwell3d.regs.index_buffer;
|
||||
draw_state.index_buffer.first = index_first;
|
||||
draw_state.index_buffer.count = index_count;
|
||||
draw_state.base_index = base_index;
|
||||
draw_state.base_instance = base_instance;
|
||||
ProcessDraw(true, num_instances);
|
||||
ProcessDraw(maxwell3d, true, num_instances);
|
||||
}
|
||||
|
||||
void DrawManager::DrawArrayIndirect(PrimitiveTopology topology) {
|
||||
void Maxwell3D::DrawManager::DrawArrayIndirect(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology) {
|
||||
draw_state.topology = topology;
|
||||
|
||||
ProcessDrawIndirect();
|
||||
ProcessDrawIndirect(maxwell3d);
|
||||
}
|
||||
|
||||
void DrawManager::DrawIndexedIndirect(PrimitiveTopology topology, u32 index_first,
|
||||
u32 index_count) {
|
||||
const auto& regs{maxwell3d->regs};
|
||||
void Maxwell3D::DrawManager::DrawIndexedIndirect(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 index_first, u32 index_count) {
|
||||
draw_state.topology = topology;
|
||||
draw_state.index_buffer = regs.index_buffer;
|
||||
draw_state.index_buffer = maxwell3d.regs.index_buffer;
|
||||
draw_state.index_buffer.first = index_first;
|
||||
draw_state.index_buffer.count = index_count;
|
||||
|
||||
ProcessDrawIndirect();
|
||||
ProcessDrawIndirect(maxwell3d);
|
||||
}
|
||||
|
||||
void DrawManager::SetInlineIndexBuffer(u32 index) {
|
||||
draw_state.inline_index_draw_indexes.push_back(static_cast<u8>(index & 0x000000ff));
|
||||
draw_state.inline_index_draw_indexes.push_back(static_cast<u8>((index & 0x0000ff00) >> 8));
|
||||
draw_state.inline_index_draw_indexes.push_back(static_cast<u8>((index & 0x00ff0000) >> 16));
|
||||
draw_state.inline_index_draw_indexes.push_back(static_cast<u8>((index & 0xff000000) >> 24));
|
||||
void Maxwell3D::DrawManager::SetInlineIndexBuffer(Maxwell3D& maxwell3d, u32 index) {
|
||||
draw_state.inline_index_draw_indexes.push_back(u8(index & 0x000000ff));
|
||||
draw_state.inline_index_draw_indexes.push_back(u8((index & 0x0000ff00) >> 8));
|
||||
draw_state.inline_index_draw_indexes.push_back(u8((index & 0x00ff0000) >> 16));
|
||||
draw_state.inline_index_draw_indexes.push_back(u8((index & 0xff000000) >> 24));
|
||||
draw_state.draw_mode = DrawMode::InlineIndex;
|
||||
}
|
||||
|
||||
void DrawManager::DrawBegin() {
|
||||
const auto& regs{maxwell3d->regs};
|
||||
auto reset_instance_count = regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::First;
|
||||
auto increment_instance_count =
|
||||
regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent;
|
||||
void Maxwell3D::DrawManager::DrawBegin(Maxwell3D& maxwell3d) {
|
||||
auto reset_instance_count = maxwell3d.regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::First;
|
||||
auto increment_instance_count = maxwell3d.regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent;
|
||||
if (reset_instance_count) {
|
||||
DrawDeferred();
|
||||
DrawDeferred(maxwell3d);
|
||||
draw_state.instance_count = 0;
|
||||
draw_state.draw_mode = DrawMode::General;
|
||||
} else if (increment_instance_count) {
|
||||
draw_state.instance_count++;
|
||||
draw_state.draw_mode = DrawMode::Instance;
|
||||
}
|
||||
|
||||
draw_state.topology = regs.draw.topology;
|
||||
draw_state.topology = maxwell3d.regs.draw.topology;
|
||||
}
|
||||
|
||||
void DrawManager::DrawEnd(u32 instance_count, bool force_draw) {
|
||||
const auto& regs{maxwell3d->regs};
|
||||
void Maxwell3D::DrawManager::DrawEnd(Maxwell3D& maxwell3d, u32 instance_count, bool force_draw) {
|
||||
switch (draw_state.draw_mode) {
|
||||
case DrawMode::Instance:
|
||||
if (!force_draw) {
|
||||
|
|
@ -169,119 +152,100 @@ void DrawManager::DrawEnd(u32 instance_count, bool force_draw) {
|
|||
}
|
||||
[[fallthrough]];
|
||||
case DrawMode::General:
|
||||
draw_state.base_instance = regs.global_base_instance_index;
|
||||
draw_state.base_index = regs.global_base_vertex_index;
|
||||
draw_state.base_instance = maxwell3d.regs.global_base_instance_index;
|
||||
draw_state.base_index = maxwell3d.regs.global_base_vertex_index;
|
||||
if (draw_state.draw_indexed) {
|
||||
draw_state.index_buffer = regs.index_buffer;
|
||||
ProcessDraw(true, instance_count);
|
||||
draw_state.index_buffer = maxwell3d.regs.index_buffer;
|
||||
ProcessDraw(maxwell3d, true, instance_count);
|
||||
} else {
|
||||
draw_state.vertex_buffer = regs.vertex_buffer;
|
||||
ProcessDraw(false, instance_count);
|
||||
draw_state.vertex_buffer = maxwell3d.regs.vertex_buffer;
|
||||
ProcessDraw(maxwell3d, false, instance_count);
|
||||
}
|
||||
draw_state.draw_indexed = false;
|
||||
break;
|
||||
case DrawMode::InlineIndex:
|
||||
draw_state.base_instance = regs.global_base_instance_index;
|
||||
draw_state.base_index = regs.global_base_vertex_index;
|
||||
draw_state.index_buffer = regs.index_buffer;
|
||||
draw_state.index_buffer.count =
|
||||
static_cast<u32>(draw_state.inline_index_draw_indexes.size() / 4);
|
||||
draw_state.base_instance = maxwell3d.regs.global_base_instance_index;
|
||||
draw_state.base_index = maxwell3d.regs.global_base_vertex_index;
|
||||
draw_state.index_buffer = maxwell3d.regs.index_buffer;
|
||||
draw_state.index_buffer.count = u32(draw_state.inline_index_draw_indexes.size() / 4);
|
||||
draw_state.index_buffer.format = Maxwell3D::Regs::IndexFormat::UnsignedInt;
|
||||
maxwell3d->dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
|
||||
ProcessDraw(true, instance_count);
|
||||
maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
|
||||
ProcessDraw(maxwell3d, true, instance_count);
|
||||
draw_state.inline_index_draw_indexes.clear();
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
void DrawManager::DrawIndexSmall(u32 argument) {
|
||||
const auto& regs{maxwell3d->regs};
|
||||
IndexBufferSmall index_small_params{argument};
|
||||
draw_state.base_instance = regs.global_base_instance_index;
|
||||
draw_state.base_index = regs.global_base_vertex_index;
|
||||
draw_state.index_buffer = regs.index_buffer;
|
||||
void Maxwell3D::DrawManager::DrawIndexSmall(Maxwell3D& maxwell3d, u32 argument) {
|
||||
Maxwell3D::Regs::IndexBufferSmall index_small_params{argument};
|
||||
draw_state.base_instance = maxwell3d.regs.global_base_instance_index;
|
||||
draw_state.base_index = maxwell3d.regs.global_base_vertex_index;
|
||||
draw_state.index_buffer = maxwell3d.regs.index_buffer;
|
||||
draw_state.index_buffer.first = index_small_params.first;
|
||||
draw_state.index_buffer.count = index_small_params.count;
|
||||
draw_state.topology = index_small_params.topology;
|
||||
maxwell3d->dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
|
||||
ProcessDraw(true, 1);
|
||||
maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
|
||||
ProcessDraw(maxwell3d, true, 1);
|
||||
}
|
||||
|
||||
void DrawManager::DrawTexture() {
|
||||
const auto& regs{maxwell3d->regs};
|
||||
draw_texture_state.dst_x0 = static_cast<float>(regs.draw_texture.dst_x0) / 4096.f;
|
||||
draw_texture_state.dst_y0 = static_cast<float>(regs.draw_texture.dst_y0) / 4096.f;
|
||||
const auto dst_width = static_cast<float>(regs.draw_texture.dst_width) / 4096.f;
|
||||
const auto dst_height = static_cast<float>(regs.draw_texture.dst_height) / 4096.f;
|
||||
const bool lower_left{regs.window_origin.mode !=
|
||||
Maxwell3D::Regs::WindowOrigin::Mode::UpperLeft};
|
||||
void Maxwell3D::DrawManager::DrawTexture(Maxwell3D& maxwell3d) {
|
||||
draw_texture_state.dst_x0 = f32(maxwell3d.regs.draw_texture.dst_x0) / 4096.f;
|
||||
draw_texture_state.dst_y0 = f32(maxwell3d.regs.draw_texture.dst_y0) / 4096.f;
|
||||
const auto dst_width = f32(maxwell3d.regs.draw_texture.dst_width) / 4096.f;
|
||||
const auto dst_height = f32(maxwell3d.regs.draw_texture.dst_height) / 4096.f;
|
||||
const bool lower_left{maxwell3d.regs.window_origin.mode != Maxwell3D::Regs::WindowOrigin::Mode::UpperLeft};
|
||||
if (lower_left) {
|
||||
draw_texture_state.dst_y0 =
|
||||
static_cast<f32>(regs.surface_clip.height) - draw_texture_state.dst_y0;
|
||||
draw_texture_state.dst_y0 = f32(maxwell3d.regs.surface_clip.height) - draw_texture_state.dst_y0;
|
||||
}
|
||||
draw_texture_state.dst_x1 = draw_texture_state.dst_x0 + dst_width;
|
||||
draw_texture_state.dst_y1 = draw_texture_state.dst_y0 + dst_height;
|
||||
draw_texture_state.src_x0 = static_cast<float>(regs.draw_texture.src_x0) / 4096.f;
|
||||
draw_texture_state.src_y0 = static_cast<float>(regs.draw_texture.src_y0) / 4096.f;
|
||||
draw_texture_state.src_x1 =
|
||||
(static_cast<float>(regs.draw_texture.dx_du) / 4294967296.f) * dst_width +
|
||||
draw_texture_state.src_x0;
|
||||
draw_texture_state.src_y1 =
|
||||
(static_cast<float>(regs.draw_texture.dy_dv) / 4294967296.f) * dst_height +
|
||||
draw_texture_state.src_y0;
|
||||
draw_texture_state.src_sampler = regs.draw_texture.src_sampler;
|
||||
draw_texture_state.src_texture = regs.draw_texture.src_texture;
|
||||
maxwell3d->rasterizer->DrawTexture();
|
||||
draw_texture_state.src_x0 = f32(maxwell3d.regs.draw_texture.src_x0) / 4096.f;
|
||||
draw_texture_state.src_y0 = f32(maxwell3d.regs.draw_texture.src_y0) / 4096.f;
|
||||
draw_texture_state.src_x1 = (f32(maxwell3d.regs.draw_texture.dx_du) / 4294967296.f) * dst_width + draw_texture_state.src_x0;
|
||||
draw_texture_state.src_y1 = (f32(maxwell3d.regs.draw_texture.dy_dv) / 4294967296.f) * dst_height + draw_texture_state.src_y0;
|
||||
draw_texture_state.src_sampler = maxwell3d.regs.draw_texture.src_sampler;
|
||||
draw_texture_state.src_texture = maxwell3d.regs.draw_texture.src_texture;
|
||||
maxwell3d.rasterizer->DrawTexture();
|
||||
}
|
||||
|
||||
void DrawManager::UpdateTopology() {
|
||||
const auto& regs{maxwell3d->regs};
|
||||
switch (regs.primitive_topology_control) {
|
||||
case PrimitiveTopologyControl::UseInBeginMethods:
|
||||
void Maxwell3D::DrawManager::UpdateTopology(Maxwell3D& maxwell3d) {
|
||||
switch (maxwell3d.regs.primitive_topology_control) {
|
||||
case Maxwell3D::Regs::PrimitiveTopologyControl::UseInBeginMethods:
|
||||
break;
|
||||
case PrimitiveTopologyControl::UseSeparateState:
|
||||
switch (regs.topology_override) {
|
||||
case PrimitiveTopologyOverride::None:
|
||||
case Maxwell3D::Regs::PrimitiveTopologyControl::UseSeparateState:
|
||||
switch (maxwell3d.regs.topology_override) {
|
||||
case Maxwell3D::Regs::PrimitiveTopologyOverride::None:
|
||||
break;
|
||||
case PrimitiveTopologyOverride::Points:
|
||||
draw_state.topology = PrimitiveTopology::Points;
|
||||
case Maxwell3D::Regs::PrimitiveTopologyOverride::Points:
|
||||
draw_state.topology = Maxwell3D::Regs::PrimitiveTopology::Points;
|
||||
break;
|
||||
case PrimitiveTopologyOverride::Lines:
|
||||
draw_state.topology = PrimitiveTopology::Lines;
|
||||
case Maxwell3D::Regs::PrimitiveTopologyOverride::Lines:
|
||||
draw_state.topology = Maxwell3D::Regs::PrimitiveTopology::Lines;
|
||||
break;
|
||||
case PrimitiveTopologyOverride::LineStrip:
|
||||
draw_state.topology = PrimitiveTopology::LineStrip;
|
||||
case Maxwell3D::Regs::PrimitiveTopologyOverride::LineStrip:
|
||||
draw_state.topology = Maxwell3D::Regs::PrimitiveTopology::LineStrip;
|
||||
break;
|
||||
default:
|
||||
draw_state.topology = static_cast<PrimitiveTopology>(regs.topology_override);
|
||||
draw_state.topology = Maxwell3D::Regs::PrimitiveTopology(maxwell3d.regs.topology_override);
|
||||
break;
|
||||
}
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
void DrawManager::ProcessDraw(bool draw_indexed, u32 instance_count) {
|
||||
LOG_TRACE(HW_GPU, "called, topology={}, count={}", draw_state.topology,
|
||||
draw_indexed ? draw_state.index_buffer.count : draw_state.vertex_buffer.count);
|
||||
|
||||
UpdateTopology();
|
||||
|
||||
if (maxwell3d->ShouldExecute()) {
|
||||
maxwell3d->rasterizer->Draw(draw_indexed, instance_count);
|
||||
void Maxwell3D::DrawManager::ProcessDraw(Maxwell3D& maxwell3d, bool draw_indexed, u32 instance_count) {
|
||||
LOG_TRACE(HW_GPU, "called, topology={}, count={}", draw_state.topology, draw_indexed ? draw_state.index_buffer.count : draw_state.vertex_buffer.count);
|
||||
UpdateTopology(maxwell3d);
|
||||
if (maxwell3d.ShouldExecute()) {
|
||||
maxwell3d.rasterizer->Draw(draw_indexed, instance_count);
|
||||
}
|
||||
}
|
||||
|
||||
void DrawManager::ProcessDrawIndirect() {
|
||||
LOG_TRACE(
|
||||
HW_GPU,
|
||||
"called, topology={}, is_indexed={}, includes_count={}, buffer_size={}, max_draw_count={}",
|
||||
draw_state.topology, indirect_state.is_indexed, indirect_state.include_count,
|
||||
indirect_state.buffer_size, indirect_state.max_draw_counts);
|
||||
|
||||
UpdateTopology();
|
||||
|
||||
if (maxwell3d->ShouldExecute()) {
|
||||
maxwell3d->rasterizer->DrawIndirect();
|
||||
void Maxwell3D::DrawManager::ProcessDrawIndirect(Maxwell3D& maxwell3d) {
|
||||
LOG_TRACE(HW_GPU, "called, topology={}, is_indexed={}, includes_count={}, buffer_size={}, max_draw_count={}", draw_state.topology, indirect_state.is_indexed, indirect_state.include_count, indirect_state.buffer_size, indirect_state.max_draw_counts);
|
||||
UpdateTopology(maxwell3d);
|
||||
if (maxwell3d.ShouldExecute()) {
|
||||
maxwell3d.rasterizer->DrawIndirect();
|
||||
}
|
||||
}
|
||||
} // namespace Tegra::Engines
|
||||
|
|
|
|||
|
|
@ -1,117 +0,0 @@
|
|||
// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
|
||||
// SPDX-License-Identifier: GPL-2.0-or-later
|
||||
|
||||
#pragma once
|
||||
#include "common/common_types.h"
|
||||
#include "video_core/engines/maxwell_3d.h"
|
||||
|
||||
namespace VideoCore {
|
||||
class RasterizerInterface;
|
||||
}
|
||||
|
||||
namespace Tegra::Engines {
|
||||
using PrimitiveTopologyControl = Maxwell3D::Regs::PrimitiveTopologyControl;
|
||||
using PrimitiveTopology = Maxwell3D::Regs::PrimitiveTopology;
|
||||
using PrimitiveTopologyOverride = Maxwell3D::Regs::PrimitiveTopologyOverride;
|
||||
using IndexBuffer = Maxwell3D::Regs::IndexBuffer;
|
||||
using VertexBuffer = Maxwell3D::Regs::VertexBuffer;
|
||||
using IndexBufferSmall = Maxwell3D::Regs::IndexBufferSmall;
|
||||
|
||||
class DrawManager {
|
||||
public:
|
||||
enum class DrawMode : u32 { General = 0, Instance, InlineIndex };
|
||||
struct State {
|
||||
PrimitiveTopology topology{};
|
||||
DrawMode draw_mode{};
|
||||
bool draw_indexed{};
|
||||
u32 base_index{};
|
||||
VertexBuffer vertex_buffer;
|
||||
IndexBuffer index_buffer;
|
||||
u32 base_instance{};
|
||||
u32 instance_count{};
|
||||
std::vector<u8> inline_index_draw_indexes;
|
||||
};
|
||||
|
||||
struct DrawTextureState {
|
||||
f32 dst_x0;
|
||||
f32 dst_y0;
|
||||
f32 dst_x1;
|
||||
f32 dst_y1;
|
||||
f32 src_x0;
|
||||
f32 src_y0;
|
||||
f32 src_x1;
|
||||
f32 src_y1;
|
||||
u32 src_sampler;
|
||||
u32 src_texture;
|
||||
};
|
||||
|
||||
struct IndirectParams {
|
||||
bool is_byte_count;
|
||||
bool is_indexed;
|
||||
bool include_count;
|
||||
GPUVAddr count_start_address;
|
||||
GPUVAddr indirect_start_address;
|
||||
size_t buffer_size;
|
||||
size_t max_draw_counts;
|
||||
size_t stride;
|
||||
};
|
||||
|
||||
explicit DrawManager(Maxwell3D* maxwell_3d);
|
||||
|
||||
void ProcessMethodCall(u32 method, u32 argument);
|
||||
|
||||
void Clear(u32 layer_count);
|
||||
|
||||
void DrawDeferred();
|
||||
|
||||
void DrawArray(PrimitiveTopology topology, u32 vertex_first, u32 vertex_count,
|
||||
u32 base_instance, u32 num_instances);
|
||||
void DrawArrayInstanced(PrimitiveTopology topology, u32 vertex_first, u32 vertex_count,
|
||||
bool subsequent);
|
||||
|
||||
void DrawIndex(PrimitiveTopology topology, u32 index_first, u32 index_count, u32 base_index,
|
||||
u32 base_instance, u32 num_instances);
|
||||
|
||||
void DrawArrayIndirect(PrimitiveTopology topology);
|
||||
|
||||
void DrawIndexedIndirect(PrimitiveTopology topology, u32 index_first, u32 index_count);
|
||||
|
||||
const State& GetDrawState() const {
|
||||
return draw_state;
|
||||
}
|
||||
|
||||
const DrawTextureState& GetDrawTextureState() const {
|
||||
return draw_texture_state;
|
||||
}
|
||||
|
||||
IndirectParams& GetIndirectParams() {
|
||||
return indirect_state;
|
||||
}
|
||||
|
||||
const IndirectParams& GetIndirectParams() const {
|
||||
return indirect_state;
|
||||
}
|
||||
|
||||
private:
|
||||
void SetInlineIndexBuffer(u32 index);
|
||||
|
||||
void DrawBegin();
|
||||
|
||||
void DrawEnd(u32 instance_count = 1, bool force_draw = false);
|
||||
|
||||
void DrawIndexSmall(u32 argument);
|
||||
|
||||
void DrawTexture();
|
||||
|
||||
void UpdateTopology();
|
||||
|
||||
void ProcessDraw(bool draw_indexed, u32 instance_count);
|
||||
|
||||
void ProcessDrawIndirect();
|
||||
|
||||
Maxwell3D* maxwell3d{};
|
||||
State draw_state{};
|
||||
DrawTextureState draw_texture_state{};
|
||||
IndirectParams indirect_state{};
|
||||
};
|
||||
} // namespace Tegra::Engines
|
||||
|
|
@ -13,7 +13,7 @@
|
|||
#include "core/core.h"
|
||||
#include "core/core_timing.h"
|
||||
#include "video_core/dirty_flags.h"
|
||||
#include "video_core/engines/draw_manager.h"
|
||||
#include "video_core/engines/maxwell_3d.h"
|
||||
#include "video_core/engines/maxwell_3d.h"
|
||||
#include "video_core/gpu.h"
|
||||
#include "video_core/memory_manager.h"
|
||||
|
|
@ -26,7 +26,8 @@ namespace Tegra::Engines {
|
|||
constexpr u32 MacroRegistersStart = 0xE00;
|
||||
|
||||
Maxwell3D::Maxwell3D(Core::System& system_, MemoryManager& memory_manager_)
|
||||
: draw_manager{std::make_unique<DrawManager>(this)}, system{system_}
|
||||
: draw_manager()
|
||||
, system{system_}
|
||||
, memory_manager{memory_manager_}
|
||||
#ifdef ARCHITECTURE_x86_64
|
||||
, macro_engine(bool(Settings::values.disable_macro_jit))
|
||||
|
|
@ -373,8 +374,7 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
|
|||
case MAXWELL3D_REG_INDEX(sync_info):
|
||||
return ProcessSyncPoint();
|
||||
case MAXWELL3D_REG_INDEX(launch_dma):
|
||||
return upload_state.ProcessExec(regs.launch_dma.memory_layout.Value() ==
|
||||
Regs::LaunchDMA::Layout::Pitch);
|
||||
return upload_state.ProcessExec(regs.launch_dma.memory_layout.Value() == Regs::LaunchDMA::Layout::Pitch);
|
||||
case MAXWELL3D_REG_INDEX(inline_data):
|
||||
upload_state.ProcessData(argument, is_last_call);
|
||||
return;
|
||||
|
|
@ -386,7 +386,7 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
|
|||
case MAXWELL3D_REG_INDEX(tiled_cache_barrier):
|
||||
return rasterizer->TiledCacheBarrier();
|
||||
default:
|
||||
draw_manager->ProcessMethodCall(method, argument);
|
||||
draw_manager.ProcessMethodCall(*this, method, argument);
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
|
@ -401,8 +401,7 @@ void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters)
|
|||
|
||||
// Execute the current macro.
|
||||
macro_engine.Execute(*this, macro_positions[entry], parameters);
|
||||
|
||||
draw_manager->DrawDeferred();
|
||||
draw_manager.DrawDeferred(*this);
|
||||
}
|
||||
|
||||
void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
|
||||
|
|
|
|||
|
|
@ -25,6 +25,7 @@
|
|||
#include "video_core/gpu.h"
|
||||
#include "video_core/macro.h"
|
||||
#include "video_core/textures/texture.h"
|
||||
#include "video_core/engines/maxwell_3d.h"
|
||||
|
||||
namespace Core {
|
||||
class System;
|
||||
|
|
@ -40,8 +41,6 @@ class RasterizerInterface;
|
|||
|
||||
namespace Tegra::Engines {
|
||||
|
||||
class DrawManager;
|
||||
|
||||
/**
|
||||
* This Engine is known as GF100_3D. Documentation can be found in:
|
||||
* https://github.com/NVIDIA/open-gpu-doc/blob/master/classes/3d/clb197.h
|
||||
|
|
@ -543,7 +542,7 @@ public:
|
|||
}
|
||||
GPUVAddr StorageLimitAddress() const {
|
||||
return (GPUVAddr{storage_limit_address_high} << 32) |
|
||||
GPUVAddr{storage_limit_address_low};
|
||||
GPUVAddr{storage_limit_address_low};
|
||||
}
|
||||
};
|
||||
|
||||
|
|
@ -819,7 +818,7 @@ public:
|
|||
|
||||
u32 Map(std::size_t index) const {
|
||||
const std::array<u32, NumRenderTargets> maps{target0, target1, target2, target3,
|
||||
target4, target5, target6, target7};
|
||||
target4, target5, target6, target7};
|
||||
ASSERT(index < maps.size());
|
||||
return maps[index];
|
||||
}
|
||||
|
|
@ -1831,7 +1830,7 @@ public:
|
|||
|
||||
bool AnyEnabled() const {
|
||||
return output0_enable || output1_enable || output2_enable || output3_enable ||
|
||||
output4_enable || output5_enable || output6_enable || output7_enable;
|
||||
output4_enable || output5_enable || output6_enable || output7_enable;
|
||||
}
|
||||
};
|
||||
|
||||
|
|
@ -1870,7 +1869,7 @@ public:
|
|||
|
||||
bool AnyEnabled() const {
|
||||
return plane0 || plane1 || plane2 || plane3 || plane4 || plane5 || plane6 ||
|
||||
plane7;
|
||||
plane7;
|
||||
}
|
||||
};
|
||||
|
||||
|
|
@ -3023,8 +3022,7 @@ public:
|
|||
u32 bindless_texture_const_buffer_slot; ///< 0x2608
|
||||
u32 trap_handler; ///< 0x260C
|
||||
INSERT_PADDING_BYTES_NOINIT(0x1F0);
|
||||
std::array<std::array<StreamOutLayout, 32>, NumTransformFeedbackBuffers>
|
||||
stream_out_layout; ///< 0x2800
|
||||
std::array<std::array<StreamOutLayout, 32>, NumTransformFeedbackBuffers> stream_out_layout; ///< 0x2800
|
||||
INSERT_PADDING_BYTES_NOINIT(0x93C);
|
||||
ShaderPerformance shader_performance; ///< 0x333C
|
||||
INSERT_PADDING_BYTES_NOINIT(0x18);
|
||||
|
|
@ -3035,6 +3033,62 @@ public:
|
|||
};
|
||||
// clang-format on
|
||||
|
||||
struct DrawManager {
|
||||
enum class DrawMode : u32 { General = 0, Instance, InlineIndex };
|
||||
struct State {
|
||||
Maxwell3D::Regs::PrimitiveTopology topology{};
|
||||
DrawMode draw_mode{};
|
||||
bool draw_indexed{};
|
||||
u32 base_index{};
|
||||
Maxwell3D::Regs::VertexBuffer vertex_buffer;
|
||||
Maxwell3D::Regs::IndexBuffer index_buffer;
|
||||
u32 base_instance{};
|
||||
u32 instance_count{};
|
||||
std::vector<u8> inline_index_draw_indexes;
|
||||
};
|
||||
struct DrawTextureState {
|
||||
f32 dst_x0;
|
||||
f32 dst_y0;
|
||||
f32 dst_x1;
|
||||
f32 dst_y1;
|
||||
f32 src_x0;
|
||||
f32 src_y0;
|
||||
f32 src_x1;
|
||||
f32 src_y1;
|
||||
u32 src_sampler;
|
||||
u32 src_texture;
|
||||
};
|
||||
struct IndirectParams {
|
||||
bool is_byte_count;
|
||||
bool is_indexed;
|
||||
bool include_count;
|
||||
GPUVAddr count_start_address;
|
||||
GPUVAddr indirect_start_address;
|
||||
size_t buffer_size;
|
||||
size_t max_draw_counts;
|
||||
size_t stride;
|
||||
};
|
||||
void ProcessMethodCall(Maxwell3D& maxwell3d, u32 method, u32 argument);
|
||||
void Clear(Maxwell3D& maxwell3d, u32 layer_count);
|
||||
void DrawDeferred(Maxwell3D& maxwell3d);
|
||||
void DrawArray(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 vertex_first, u32 vertex_count, u32 base_instance, u32 num_instances);
|
||||
void DrawArrayInstanced(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 vertex_first, u32 vertex_count, bool subsequent);
|
||||
void DrawIndex(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 index_first, u32 index_count, u32 base_index, u32 base_instance, u32 num_instances);
|
||||
void DrawArrayIndirect(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology);
|
||||
void DrawIndexedIndirect(Maxwell3D& maxwell3d, Maxwell3D::Regs::PrimitiveTopology topology, u32 index_first, u32 index_count);
|
||||
void SetInlineIndexBuffer(Maxwell3D& maxwell3d, u32 index);
|
||||
void DrawBegin(Maxwell3D& maxwell3d);
|
||||
void DrawEnd(Maxwell3D& maxwell3d, u32 instance_count = 1, bool force_draw = false);
|
||||
void DrawIndexSmall(Maxwell3D& maxwell3d, u32 argument);
|
||||
void DrawTexture(Maxwell3D& maxwell3d);
|
||||
void UpdateTopology(Maxwell3D& maxwell3d);
|
||||
void ProcessDraw(Maxwell3D& maxwell3d, bool draw_indexed, u32 instance_count);
|
||||
void ProcessDrawIndirect(Maxwell3D& maxwell3d);
|
||||
State draw_state{};
|
||||
DrawTextureState draw_texture_state{};
|
||||
IndirectParams indirect_state{};
|
||||
};
|
||||
|
||||
Regs regs{};
|
||||
|
||||
/// Store temporary hw register values, used by some calls to restore state after a operation
|
||||
|
|
@ -3102,8 +3156,7 @@ public:
|
|||
Tables tables{};
|
||||
} dirty;
|
||||
|
||||
std::unique_ptr<DrawManager> draw_manager;
|
||||
friend class DrawManager;
|
||||
DrawManager draw_manager;
|
||||
|
||||
GPUVAddr GetMacroAddress(size_t index) const {
|
||||
return macro_addresses[index];
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue