ShuriZma
/
duckstation
mirror of https://github.com/stenzek/duckstation
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

GPU: Implement PGXP for lines

This commit is contained in:
Stenzek 2024-12-22 14:06:13 +10:00
parent b81287efd2
commit 0b4e302c22
No known key found for this signature in database
13 changed files with 314 additions and 89 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/common/gsvector_neon.h
View File

@ -3131,6 +3131,11 @@ public:
ALWAYS_INLINE GSVector2 zw() const { return GSVector2(vget_high_s32(v4s)); }
ALWAYS_INLINE static GSVector4 xyxy(const GSVector2& l, const GSVector2& h)
{
return GSVector4(vcombine_f32(l.v2s, h.v2s));
}
#define VECTOR4_SHUFFLE_4(xs, xn, ys, yn, zs, zn, ws, wn) \
ALWAYS_INLINE GSVector4 xs##ys##zs##ws() const \
{ \

5
src/common/gsvector_nosimd.h
View File

@ -2316,6 +2316,11 @@ public:
ALWAYS_INLINE GSVector2 zw() const { return GSVector2(z, w); }
ALWAYS_INLINE static GSVector4 xyxy(const GSVector2& l, const GSVector2& h)
{
return GSVector4(l.x, l.y, h.x, h.y);
}
#define VECTOR4_SHUFFLE_4(xs, xn, ys, yn, zs, zn, ws, wn) \
ALWAYS_INLINE GSVector4 xs##ys##zs##ws() const { return GSVector4(F32[xn], F32[yn], F32[zn], F32[wn]); }

14
src/core/gpu.h
View File

@ -279,7 +279,7 @@ private:
/// Returns the number of vertices in the buffered poly-line.
ALWAYS_INLINE u32 GetPolyLineVertexCount() const
{
return (static_cast<u32>(m_blit_buffer.size()) + BoolToUInt32(m_render_command.shading_enable)) >>
return (static_cast<u32>(m_polyline_buffer.size()) + BoolToUInt32(m_render_command.shading_enable)) >>
BoolToUInt8(m_render_command.shading_enable);
}
@ -520,20 +520,20 @@ private:
u16 row;
} m_vram_transfer = {};
std::unique_ptr<GPUDump::Recorder> m_gpu_dump;
HeapFIFOQueue<u64, MAX_FIFO_SIZE> m_fifo;
std::vector<u32> m_blit_buffer;
TickCount m_max_run_ahead = 128;
u32 m_fifo_size = 128;
u32 m_blit_remaining_words;
GPURenderCommand m_render_command{};
std::unique_ptr<GPUDump::Recorder> m_gpu_dump;
std::vector<u32> m_blit_buffer;
std::vector<u64> m_polyline_buffer;
ALWAYS_INLINE u32 FifoPop() { return Truncate32(m_fifo.Pop()); }
ALWAYS_INLINE u32 FifoPeek() { return Truncate32(m_fifo.Peek()); }
ALWAYS_INLINE u32 FifoPeek(u32 i) { return Truncate32(m_fifo.Peek(i)); }
TickCount m_max_run_ahead = 128;
u32 m_fifo_size = 128;
private:
using GP0CommandHandler = bool (GPU::*)();
using GP0CommandHandlerTable = std::array<GP0CommandHandler, 256>;

19
src/core/gpu_backend.cpp
View File

@ -249,6 +249,16 @@ GPUBackendDrawLineCommand* GPUBackend::NewDrawLineCommand(u32 num_vertices)
return cmd;
}
GPUBackendDrawPreciseLineCommand* GPUBackend::NewDrawPreciseLineCommand(u32 num_vertices)
{
const u32 size =
sizeof(GPUBackendDrawPreciseLineCommand) + (num_vertices * sizeof(GPUBackendDrawPreciseLineCommand::Vertex));
GPUBackendDrawPreciseLineCommand* cmd = static_cast<GPUBackendDrawPreciseLineCommand*>(
GPUThread::AllocateCommand(GPUBackendCommandType::DrawPreciseLine, size));
cmd->num_vertices = Truncate16(num_vertices);
return cmd;
}
void GPUBackend::PushCommand(GPUThreadCommand* cmd)
{
GPUThread::PushCommand(cmd);
@ -489,6 +499,15 @@ void GPUBackend::HandleCommand(const GPUThreadCommand* cmd)
s_counters.num_primitives += ccmd->num_vertices / 2;
DrawLine(ccmd);
}
break;
case GPUBackendCommandType::DrawPreciseLine:
{
const GPUBackendDrawPreciseLineCommand* ccmd = static_cast<const GPUBackendDrawPreciseLineCommand*>(cmd);
s_counters.num_vertices += ccmd->num_vertices;
s_counters.num_primitives += ccmd->num_vertices / 2;
DrawPreciseLine(ccmd);
}
break;
DefaultCaseIsUnreachable();

2
src/core/gpu_backend.h
View File

@ -51,6 +51,7 @@ public:
static GPUBackendDrawPrecisePolygonCommand* NewDrawPrecisePolygonCommand(u32 num_vertices);
static GPUBackendDrawRectangleCommand* NewDrawRectangleCommand();
static GPUBackendDrawLineCommand* NewDrawLineCommand(u32 num_vertices);
static GPUBackendDrawPreciseLineCommand* NewDrawPreciseLineCommand(u32 num_vertices);
static void PushCommand(GPUThreadCommand* cmd);
static void PushCommandAndWakeThread(GPUThreadCommand* cmd);
static void PushCommandAndSync(GPUThreadCommand* cmd, bool spin);
@ -125,6 +126,7 @@ protected:
virtual void DrawPrecisePolygon(const GPUBackendDrawPrecisePolygonCommand* cmd) = 0;
virtual void DrawSprite(const GPUBackendDrawRectangleCommand* cmd) = 0;
virtual void DrawLine(const GPUBackendDrawLineCommand* cmd) = 0;
virtual void DrawPreciseLine(const GPUBackendDrawPreciseLineCommand* cmd) = 0;
virtual void DrawingAreaChanged() = 0;
virtual void UpdateCLUT(GPUTexturePaletteReg reg, bool clut_is_8bit) = 0;

135
src/core/gpu_commands.cpp
View File

@ -73,7 +73,7 @@ void GPU::TryExecuteCommands()
{
const u32 words_per_vertex = m_render_command.shading_enable ? 2 : 1;
u32 terminator_index =
m_render_command.shading_enable ? ((static_cast<u32>(m_blit_buffer.size()) & 1u) ^ 1u) : 0u;
m_render_command.shading_enable ? ((static_cast<u32>(m_polyline_buffer.size()) & 1u) ^ 1u) : 0u;
for (; terminator_index < m_fifo.GetSize(); terminator_index += words_per_vertex)
{
// polyline must have at least two vertices, and the terminator is (word & 0xf000f000) == 0x50005000.
@ -86,9 +86,9 @@ void GPU::TryExecuteCommands()
const u32 words_to_copy = std::min(terminator_index, m_fifo.GetSize());
if (words_to_copy > 0)
{
m_blit_buffer.reserve(m_blit_buffer.size() + words_to_copy);
m_polyline_buffer.reserve(m_polyline_buffer.size() + words_to_copy);
for (u32 i = 0; i < words_to_copy; i++)
m_blit_buffer.push_back(FifoPop());
m_polyline_buffer.push_back(m_fifo.Pop());
}
DEBUG_LOG("Added {} words to polyline", words_to_copy);
@ -98,7 +98,7 @@ void GPU::TryExecuteCommands()
m_fifo.RemoveOne();
DEBUG_LOG("Drawing poly-line with {} vertices", GetPolyLineVertexCount());
FinishPolyline();
m_blit_buffer.clear();
m_polyline_buffer.clear();
EndCommand();
continue;
}
@ -713,7 +713,53 @@ bool GPU::HandleRenderLineCommand()
m_fifo.RemoveOne();
PrepareForDraw();
GPUBackendDrawLineCommand* cmd = GPUBackend::NewDrawLineCommand(2);
if (g_settings.gpu_pgxp_enable)
{
GPUBackendDrawPreciseLineCommand* RESTRICT cmd = GPUBackend::NewDrawPreciseLineCommand(2);
FillDrawCommand(cmd, rc);
cmd->palette.bits = 0;
bool valid_w = g_settings.gpu_pgxp_texture_correction;
for (u32 i = 0; i < 2; i++)
{
const u32 color = ((i != 0 && rc.shading_enable) ? FifoPop() : rc.bits) & UINT32_C(0x00FFFFFF);
const u64 maddr_and_pos = m_fifo.Pop();
const GPUVertexPosition vp{Truncate32(maddr_and_pos)};
GPUBackendDrawPreciseLineCommand::Vertex* RESTRICT vert = &cmd->vertices[i];
vert->native_x = m_drawing_offset.x + vp.x;
vert->native_y = m_drawing_offset.y + vp.y;
vert->color = color;
valid_w &= CPU::PGXP::GetPreciseVertex(Truncate32(maddr_and_pos >> 32), vp.bits, vert->native_x, vert->native_y,
m_drawing_offset.x, m_drawing_offset.y, &vert->x, &vert->y, &vert->w);
}
if (!(cmd->valid_w = valid_w))
{
for (u32 i = 0; i < 2; i++)
cmd->vertices[i].w = 1.0f;
}
const GSVector2 v0f = GSVector2::load<false>(&cmd->vertices[0].x);
const GSVector2 v1f = GSVector2::load<false>(&cmd->vertices[1].x);
const GSVector4i rect =
GSVector4i(GSVector4(v0f.min(v1f)).upld(GSVector4(v0f.max(v1f)))).add32(GSVector4i::cxpr(0, 0, 1, 1));
const GSVector4i clamped_rect = rect.rintersect(m_clamped_drawing_area);
if (rect.width() > MAX_PRIMITIVE_WIDTH || rect.height() > MAX_PRIMITIVE_HEIGHT || clamped_rect.rempty())
{
DEBUG_LOG("Culling too-large/off-screen line: {},{} - {},{}", cmd->vertices[0].y, cmd->vertices[0].y,
cmd->vertices[1].x, cmd->vertices[1].y);
EndCommand();
return true;
}
AddDrawLineTicks(clamped_rect, rc.shading_enable);
GPUBackend::PushCommand(cmd);
}
else
{
GPUBackendDrawLineCommand* RESTRICT cmd = GPUBackend::NewDrawLineCommand(2);
FillDrawCommand(cmd, rc);
cmd->palette.bits = 0;
@ -758,6 +804,8 @@ bool GPU::HandleRenderLineCommand()
AddDrawLineTicks(clamped_rect, rc.shading_enable);
GPUBackend::PushCommand(cmd);
}
EndCommand();
return true;
}
@ -784,9 +832,9 @@ bool GPU::HandleRenderPolyLineCommand()
const u32 words_to_pop = min_words - 1;
// m_blit_buffer.resize(words_to_pop);
// FifoPopRange(m_blit_buffer.data(), words_to_pop);
m_blit_buffer.reserve(words_to_pop);
m_polyline_buffer.reserve(words_to_pop);
for (u32 i = 0; i < words_to_pop; i++)
m_blit_buffer.push_back(Truncate32(FifoPop()));
m_polyline_buffer.push_back(m_fifo.Pop());
// polyline goes via a different path through the blit buffer
m_blitter_state = BlitterState::DrawingPolyLine;
@ -801,11 +849,73 @@ void GPU::FinishPolyline()
const u32 num_vertices = GetPolyLineVertexCount();
DebugAssert(num_vertices >= 2);
GPUBackendDrawLineCommand* cmd = GPUBackend::NewDrawLineCommand((num_vertices - 1) * 2);
if (g_settings.gpu_pgxp_enable)
{
GPUBackendDrawPreciseLineCommand* RESTRICT cmd = GPUBackend::NewDrawPreciseLineCommand((num_vertices - 1) * 2);
FillDrawCommand(cmd, m_render_command);
cmd->palette.bits = 0;
u32 buffer_pos = 0;
const GPUVertexPosition start_vp{m_blit_buffer[buffer_pos++]};
u32 out_vertex_count = 0;
const bool shaded = m_render_command.shading_enable;
bool valid_w = g_settings.gpu_pgxp_texture_correction;
GPUBackendDrawPreciseLineCommand::Vertex start, end;
const auto read_vertex = [this, &buffer_pos, &valid_w](GPUBackendDrawPreciseLineCommand::Vertex& RESTRICT dest,
u32 color) {
const u64 maddr_and_pos = m_polyline_buffer[buffer_pos++];
const GPUVertexPosition vp{Truncate32(maddr_and_pos)};
dest.native_x = m_drawing_offset.x + vp.x;
dest.native_y = m_drawing_offset.y + vp.y;
dest.color = color;
valid_w &= CPU::PGXP::GetPreciseVertex(Truncate32(maddr_and_pos >> 32), vp.bits, dest.native_x, dest.native_y,
m_drawing_offset.x, m_drawing_offset.y, &dest.x, &dest.y, &dest.w);
};
read_vertex(start, m_render_command.color_for_first_vertex);
for (u32 i = 1; i < num_vertices; i++)
{
const u32 color =
(shaded ? Truncate32(m_polyline_buffer[buffer_pos++]) : m_render_command.bits) & UINT32_C(0x00FFFFFF);
read_vertex(end, color);
const GSVector2 start_pos = GSVector2::load<false>(&start.x);
const GSVector2 end_pos = GSVector2::load<false>(&end.x);
const GSVector4i rect =
GSVector4i(GSVector4::xyxy(start_pos.min(end_pos), start_pos.max(end_pos))).add32(GSVector4i::cxpr(0, 0, 1, 1));
const GSVector4i clamped_rect = rect.rintersect(m_clamped_drawing_area);
if (rect.width() > MAX_PRIMITIVE_WIDTH || rect.height() > MAX_PRIMITIVE_HEIGHT || clamped_rect.rempty())
{
DEBUG_LOG("Culling too-large/off-screen line: {},{} - {},{}", start_pos.x, start_pos.y, end_pos.x, end_pos.y);
}
else
{
AddDrawLineTicks(clamped_rect, m_render_command.shading_enable);
cmd->vertices[out_vertex_count++] = start;
cmd->vertices[out_vertex_count++] = end;
}
start = end;
}
if (out_vertex_count > 0)
{
DebugAssert(out_vertex_count <= cmd->num_vertices);
cmd->num_vertices = Truncate16(out_vertex_count);
GPUBackend::PushCommand(cmd);
}
}
else
{
GPUBackendDrawLineCommand* RESTRICT cmd = GPUBackend::NewDrawLineCommand((num_vertices - 1) * 2);
FillDrawCommand(cmd, m_render_command);
cmd->palette.bits = 0;
u32 buffer_pos = 0;
const GPUVertexPosition start_vp{Truncate32(m_polyline_buffer[buffer_pos++])};
const GSVector2i draw_offset = GSVector2i::load<false>(&m_drawing_offset.x);
GSVector2i start_pos = GSVector2i(start_vp.x, start_vp.y).add32(draw_offset);
u32 start_color = m_render_command.color_for_first_vertex;
@ -814,9 +924,9 @@ void GPU::FinishPolyline()
u32 out_vertex_count = 0;
for (u32 i = 1; i < num_vertices; i++)
{
const u32 end_color =
shaded ? (m_blit_buffer[buffer_pos++] & UINT32_C(0x00FFFFFF)) : m_render_command.color_for_first_vertex;
const GPUVertexPosition vp{m_blit_buffer[buffer_pos++]};
const u32 end_color = shaded ? (Truncate32(m_polyline_buffer[buffer_pos++] & UINT32_C(0x00FFFFFF))) :
m_render_command.color_for_first_vertex;
const GPUVertexPosition vp{Truncate32(m_polyline_buffer[buffer_pos++])};
const GSVector2i end_pos = GSVector2i(vp.x, vp.y).add32(draw_offset);
const GSVector4i rect =
@ -850,6 +960,7 @@ void GPU::FinishPolyline()
cmd->num_vertices = Truncate16(out_vertex_count);
GPUBackend::PushCommand(cmd);
}
}
}
bool GPU::HandleFillRectangleCommand()

49
src/core/gpu_hw.cpp
View File

@ -2500,6 +2500,54 @@ void GPU_HW::DrawLine(const GPUBackendDrawLineCommand* cmd)
}
}
void GPU_HW::DrawPreciseLine(const GPUBackendDrawPreciseLineCommand* cmd)
{
PrepareDraw(cmd);
const bool use_depth = m_pgxp_depth_buffer && cmd->valid_w;
SetBatchDepthBuffer(cmd, use_depth);
const u32 num_vertices = cmd->num_vertices;
DebugAssert(m_batch_vertex_space >= (num_vertices * 4) && m_batch_index_space >= (num_vertices * 6));
const float depth = GetCurrentNormalizedVertexDepth();
for (u32 i = 0; i < num_vertices; i += 2)
{
const GSVector2 start_pos = GSVector2::load<false>(&cmd->vertices[i].x);
const u32 start_color = cmd->vertices[i].color;
const GSVector2 end_pos = GSVector2::load<false>(&cmd->vertices[i + 1].x);
const u32 end_color = cmd->vertices[i + 1].color;
const GSVector4 bounds = GSVector4::xyxy(start_pos, end_pos);
const GSVector4i rect =
GSVector4i(GSVector4::xyxy(start_pos.min(end_pos), start_pos.max(end_pos))).add32(GSVector4i::cxpr(0, 0, 1, 1));
const GSVector4i clamped_rect = rect.rintersect(m_clamped_drawing_area);
DebugAssert(rect.width() <= MAX_PRIMITIVE_WIDTH && rect.height() <= MAX_PRIMITIVE_HEIGHT && !clamped_rect.rempty());
AddDrawnRectangle(clamped_rect);
DrawLine(bounds, start_color, end_color, depth);
}
if (ShouldDrawWithSoftwareRenderer())
{
const GPU_SW_Rasterizer::DrawLineFunction DrawFunction =
GPU_SW_Rasterizer::GetDrawLineFunction(cmd->shading_enable, cmd->transparency_enable);
for (u32 i = 0; i < cmd->num_vertices; i += 2)
{
const GPUBackendDrawPreciseLineCommand::Vertex& RESTRICT start = cmd->vertices[i];
const GPUBackendDrawPreciseLineCommand::Vertex& RESTRICT end = cmd->vertices[i + 1];
const GPUBackendDrawLineCommand::Vertex vertices[2] = {
{.x = start.native_x, .y = start.native_y, .color = start.color},
{.x = end.native_x, .y = end.native_y, .color = end.color},
};
DrawFunction(cmd, &vertices[0], &vertices[1]);
}
}
}
void GPU_HW::DrawLine(const GSVector4 bounds, u32 col0, u32 col1, float depth)
{
DebugAssert(m_batch_vertex_space >= 4 && m_batch_index_space >= 6);
@ -3003,6 +3051,7 @@ void GPU_HW::EnsureVertexBufferSpaceForCommand(const GPUBackendDrawCommand* cmd)
required_indices = MAX_VERTICES_FOR_RECTANGLE;
break;
case GPUBackendCommandType::DrawLine:
case GPUBackendCommandType::DrawPreciseLine:
{
// assume expansion
const GPUBackendDrawLineCommand* lcmd = static_cast<const GPUBackendDrawLineCommand*>(cmd);

1
src/core/gpu_hw.h
View File

@ -89,6 +89,7 @@ protected:
void DrawPrecisePolygon(const GPUBackendDrawPrecisePolygonCommand* cmd) override;
void DrawSprite(const GPUBackendDrawRectangleCommand* cmd) override;
void DrawLine(const GPUBackendDrawLineCommand* cmd) override;
void DrawPreciseLine(const GPUBackendDrawPreciseLineCommand* cmd) override;
void FlushRender() override;
void DrawingAreaChanged() override;

20
src/core/gpu_sw.cpp
View File

@ -150,6 +150,26 @@ void GPU_SW::DrawLine(const GPUBackendDrawLineCommand* cmd)
DrawFunction(cmd, &cmd->vertices[i], &cmd->vertices[i + 1]);
}
void GPU_SW::DrawPreciseLine(const GPUBackendDrawPreciseLineCommand* cmd)
{
const GPU_SW_Rasterizer::DrawLineFunction DrawFunction =
GPU_SW_Rasterizer::GetDrawLineFunction(cmd->shading_enable, cmd->transparency_enable);
// Need to cut out the irrelevant bits.
// TODO: In _theory_ we could use the fixed-point parts here.
for (u32 i = 0; i < cmd->num_vertices; i += 2)
{
const GPUBackendDrawPreciseLineCommand::Vertex& RESTRICT start = cmd->vertices[i];
const GPUBackendDrawPreciseLineCommand::Vertex& RESTRICT end = cmd->vertices[i + 1];
const GPUBackendDrawLineCommand::Vertex vertices[2] = {
{.x = start.native_x, .y = start.native_y, .color = start.color},
{.x = end.native_x, .y = end.native_y, .color = end.color},
};
DrawFunction(cmd, &vertices[0], &vertices[1]);
}
}
void GPU_SW::DrawingAreaChanged()
{
// GPU_SW_Rasterizer::g_drawing_area set by base class.

1
src/core/gpu_sw.h
View File

@ -36,6 +36,7 @@ protected:
void DrawPolygon(const GPUBackendDrawPolygonCommand* cmd) override;
void DrawPrecisePolygon(const GPUBackendDrawPrecisePolygonCommand* cmd) override;
void DrawLine(const GPUBackendDrawLineCommand* cmd) override;
void DrawPreciseLine(const GPUBackendDrawPreciseLineCommand* cmd) override;
void DrawSprite(const GPUBackendDrawRectangleCommand* cmd) override;
void DrawingAreaChanged() override;
void ClearCache() override;

2
src/core/gpu_sw_rasterizer.h
View File

@ -32,7 +32,7 @@ using DrawTriangleFunction = void (*)(const GPUBackendDrawCommand* cmd, const GP
const GPUBackendDrawPolygonCommand::Vertex* v2);
typedef const DrawTriangleFunction DrawTriangleFunctionTable[2][2][2][2];
using DrawLineFunction = void (*)(const GPUBackendDrawLineCommand* cmd, const GPUBackendDrawLineCommand::Vertex* p0,
using DrawLineFunction = void (*)(const GPUBackendDrawCommand* cmd, const GPUBackendDrawLineCommand::Vertex* p0,
const GPUBackendDrawLineCommand::Vertex* p1);
typedef const DrawLineFunction DrawLineFunctionTable[2][2];

3
src/core/gpu_sw_rasterizer.inl
View File

@ -766,8 +766,7 @@ static void DrawRectangle(const GPUBackendDrawRectangleCommand* RESTRICT cmd)
// TODO: Vectorize line draw.
template<bool shading_enable, bool transparency_enable>
static void DrawLine(const GPUBackendDrawLineCommand* RESTRICT cmd,
const GPUBackendDrawLineCommand::Vertex* RESTRICT p0,
static void DrawLine(const GPUBackendDrawCommand* RESTRICT cmd, const GPUBackendDrawLineCommand::Vertex* RESTRICT p0,
const GPUBackendDrawLineCommand::Vertex* RESTRICT p1)
{
static constexpr u32 XY_SHIFT = 32;

13
src/core/gpu_thread_commands.h
View File

@ -54,6 +54,7 @@ enum class GPUBackendCommandType : u8
DrawPrecisePolygon,
DrawRectangle,
DrawLine,
DrawPreciseLine,
};
struct GPUThreadCommand
@ -317,6 +318,18 @@ struct GPUBackendDrawLineCommand : public GPUBackendDrawCommand
Vertex vertices[0];
};
struct GPUBackendDrawPreciseLineCommand : public GPUBackendDrawCommand
{
struct Vertex
{
float x, y, w;
s32 native_x, native_y;
u32 color;
};
Vertex vertices[0];
};
#ifdef _MSC_VER
#pragma warning(pop)
#endif