Emulator-Archive/duckstation
1
0
Fork 0
mirror of https://github.com/stenzek/duckstation.git synced 2024-11-22 21:39:40 +00:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

GPU/SW: Split out rasterizer, add dynamic selection

Browse Source
This commit is contained in:
Stenzek 2023-12-23 20:38:41 +10:00
parent 3fbdcdb4ff
commit aa955b8ae2
No known key found for this signature in database
13 changed files with 1493 additions and 881 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/core/CMakeLists.txt
View File

@ -57,6 +57,8 @@ add_library(core
gpu_sw.h
gpu_sw_backend.cpp
gpu_sw_backend.h
gpu_sw_rasterizer.cpp
gpu_sw_rasterizer.h
gpu_types.h
guncon.cpp
guncon.h

11
src/core/core.vcxproj
View File

@ -50,6 +50,13 @@
<ClCompile Include="gpu_shadergen.cpp" />
<ClCompile Include="gpu_sw.cpp" />
<ClCompile Include="gpu_sw_backend.cpp" />
<ClCompile Include="gpu_sw_rasterizer.cpp" />
<ClCompile Include="gpu_sw_rasterizer_avx2.cpp">
<EnableEnhancedInstructionSet>AdvancedVectorExtensions2</EnableEnhancedInstructionSet>
<AdditionalOptions Condition="$(Configuration.Contains(Clang))">%(AdditionalOptions) -mavx2</AdditionalOptions>
<ExcludedFromBuild Condition="'$(Platform)'!='x64'">true</ExcludedFromBuild>
<PrecompiledHeader>NotUsing</PrecompiledHeader>
</ClCompile>
<ClCompile Include="gte.cpp" />
<ClCompile Include="dma.cpp" />
<ClCompile Include="gpu.cpp" />
@ -127,6 +134,7 @@
<ClInclude Include="gpu_shadergen.h" />
<ClInclude Include="gpu_sw.h" />
<ClInclude Include="gpu_sw_backend.h" />
<ClInclude Include="gpu_sw_rasterizer.h" />
<ClInclude Include="gpu_types.h" />
<ClInclude Include="gte.h" />
<ClInclude Include="cpu_types.h" />
@ -195,6 +203,9 @@
<Project>{57f6206d-f264-4b07-baf8-11b9bbe1f455}</Project>
</ProjectReference>
</ItemGroup>
<ItemGroup>
<None Include="gpu_sw_rasterizer.inl" />
</ItemGroup>
<PropertyGroup Label="Globals">
<ProjectGuid>{868B98C8-65A1-494B-8346-250A73A48C0A}</ProjectGuid>
</PropertyGroup>

6
src/core/core.vcxproj.filters
View File

@ -67,6 +67,8 @@
<ClCompile Include="justifier.cpp" />
<ClCompile Include="pine_server.cpp" />
<ClCompile Include="gdb_server.cpp" />
<ClCompile Include="gpu_sw_rasterizer.cpp" />
<ClCompile Include="gpu_sw_rasterizer_avx2.cpp" />
</ItemGroup>
<ItemGroup>
<ClInclude Include="types.h" />
@ -140,5 +142,9 @@
<ClInclude Include="justifier.h" />
<ClInclude Include="pine_server.h" />
<ClInclude Include="gdb_server.h" />
<ClInclude Include="gpu_sw_rasterizer.h" />
</ItemGroup>
<ItemGroup>
<None Include="gpu_sw_rasterizer.inl" />
</ItemGroup>
</Project>

6
src/core/gpu_backend.cpp
View File

@ -26,7 +26,7 @@ bool GPUBackend::Initialize(bool force_thread)
void GPUBackend::Reset()
{
Sync(true);
m_drawing_area = {};
DrawingAreaChanged(GPUDrawingArea{0, 0, 0, 0}, GSVector4i::zero());
}
void GPUBackend::UpdateSettings()
@ -310,8 +310,8 @@ void GPUBackend::HandleCommand(const GPUBackendCommand* cmd)
case GPUBackendCommandType::SetDrawingArea:
{
FlushRender();
m_drawing_area = static_cast<const GPUBackendSetDrawingAreaCommand*>(cmd)->new_area;
DrawingAreaChanged();
const GPUBackendSetDrawingAreaCommand* ccmd = static_cast<const GPUBackendSetDrawingAreaCommand*>(cmd);
DrawingAreaChanged(ccmd->new_area, GSVector4i::load<false>(ccmd->new_clamped_area));
}
break;

4
src/core/gpu_backend.h
View File

@ -62,13 +62,11 @@ protected:
virtual void DrawRectangle(const GPUBackendDrawRectangleCommand* cmd) = 0;
virtual void DrawLine(const GPUBackendDrawLineCommand* cmd) = 0;
virtual void FlushRender() = 0;
virtual void DrawingAreaChanged() = 0;
virtual void DrawingAreaChanged(const GPUDrawingArea& new_drawing_area, const GSVector4i clamped_drawing_area) = 0;
virtual void UpdateCLUT(GPUTexturePaletteReg reg, bool clut_is_8bit) = 0;
void HandleCommand(const GPUBackendCommand* cmd);
GPUDrawingArea m_drawing_area = {};
Threading::KernelSemaphore m_sync_semaphore;
std::atomic_bool m_gpu_thread_sleeping{false};
std::atomic_bool m_gpu_loop_done{false};

1
src/core/gpu_sw.cpp
View File

@ -501,6 +501,7 @@ void GPU_SW::DispatchRenderCommand()
{
GPUBackendSetDrawingAreaCommand* cmd = m_backend.NewSetDrawingAreaCommand();
cmd->new_area = m_drawing_area;
GSVector4i::store<false>(cmd->new_clamped_area, m_clamped_drawing_area);
m_backend.PushCommand(cmd);
m_drawing_area_changed = false;
}

750
src/core/gpu_sw_backend.cpp
View File

@ -3,6 +3,7 @@
#include "gpu_sw_backend.h"
#include "gpu.h"
#include "gpu_sw_rasterizer.h"
#include "system.h"
#include "util/gpu_device.h"
@ -15,6 +16,8 @@ GPU_SW_Backend::~GPU_SW_Backend() = default;
bool GPU_SW_Backend::Initialize(bool force_thread)
{
GPU_SW_Rasterizer::SelectImplementation();
return GPUBackend::Initialize(force_thread);
}
@ -28,688 +31,31 @@ void GPU_SW_Backend::DrawPolygon(const GPUBackendDrawPolygonCommand* cmd)
const GPURenderCommand rc{cmd->rc.bits};
const bool dithering_enable = rc.IsDitheringEnabled() && cmd->draw_mode.dither_enable;
const DrawTriangleFunction DrawFunction = GetDrawTriangleFunction(
const GPU_SW_Rasterizer::DrawTriangleFunction DrawFunction = GPU_SW_Rasterizer::GetDrawTriangleFunction(
rc.shading_enable, rc.texture_enable, rc.raw_texture_enable, rc.transparency_enable, dithering_enable);
(this->*DrawFunction)(cmd, &cmd->vertices[0], &cmd->vertices[1], &cmd->vertices[2]);
DrawFunction(cmd, &cmd->vertices[0], &cmd->vertices[1], &cmd->vertices[2]);
if (rc.quad_polygon)
(this->*DrawFunction)(cmd, &cmd->vertices[2], &cmd->vertices[1], &cmd->vertices[3]);
DrawFunction(cmd, &cmd->vertices[2], &cmd->vertices[1], &cmd->vertices[3]);
}
void GPU_SW_Backend::DrawRectangle(const GPUBackendDrawRectangleCommand* cmd)
{
const GPURenderCommand rc{cmd->rc.bits};
const DrawRectangleFunction DrawFunction =
GetDrawRectangleFunction(rc.texture_enable, rc.raw_texture_enable, rc.transparency_enable);
const GPU_SW_Rasterizer::DrawRectangleFunction DrawFunction =
GPU_SW_Rasterizer::GetDrawRectangleFunction(rc.texture_enable, rc.raw_texture_enable, rc.transparency_enable);
(this->*DrawFunction)(cmd);
DrawFunction(cmd);
}
void GPU_SW_Backend::DrawLine(const GPUBackendDrawLineCommand* cmd)
{
const DrawLineFunction DrawFunction =
GetDrawLineFunction(cmd->rc.shading_enable, cmd->rc.transparency_enable, cmd->IsDitheringEnabled());
const GPU_SW_Rasterizer::DrawLineFunction DrawFunction = GPU_SW_Rasterizer::GetDrawLineFunction(
cmd->rc.shading_enable, cmd->rc.transparency_enable, cmd->IsDitheringEnabled());
for (u16 i = 1; i < cmd->num_vertices; i++)
(this->*DrawFunction)(cmd, &cmd->vertices[i - 1], &cmd->vertices[i]);
}
constexpr GPU_SW_Backend::DitherLUT GPU_SW_Backend::ComputeDitherLUT()
{
DitherLUT lut = {};
for (u32 i = 0; i < DITHER_MATRIX_SIZE; i++)
{
for (u32 j = 0; j < DITHER_MATRIX_SIZE; j++)
{
for (u32 value = 0; value < DITHER_LUT_SIZE; value++)
{
const s32 dithered_value = (static_cast<s32>(value) + DITHER_MATRIX[i][j]) >> 3;
lut[i][j][value] = static_cast<u8>((dithered_value < 0) ? 0 : ((dithered_value > 31) ? 31 : dithered_value));
}
}
}
return lut;
}
static constexpr GPU_SW_Backend::DitherLUT s_dither_lut = GPU_SW_Backend::ComputeDitherLUT();
template<bool texture_enable, bool raw_texture_enable, bool transparency_enable, bool dithering_enable>
void ALWAYS_INLINE_RELEASE GPU_SW_Backend::ShadePixel(const GPUBackendDrawCommand* cmd, u32 x, u32 y, u8 color_r,
u8 color_g, u8 color_b, u8 texcoord_x, u8 texcoord_y)
{
VRAMPixel color;
if constexpr (texture_enable)
{
// Apply texture window
texcoord_x = (texcoord_x & cmd->window.and_x) | cmd->window.or_x;
texcoord_y = (texcoord_y & cmd->window.and_y) | cmd->window.or_y;
VRAMPixel texture_color;
switch (cmd->draw_mode.texture_mode)
{
case GPUTextureMode::Palette4Bit:
{
const u16 palette_value =
GetPixel((cmd->draw_mode.GetTexturePageBaseX() + ZeroExtend32(texcoord_x / 4)) % VRAM_WIDTH,
(cmd->draw_mode.GetTexturePageBaseY() + ZeroExtend32(texcoord_y)) % VRAM_HEIGHT);
const size_t palette_index = (palette_value >> ((texcoord_x % 4) * 4)) & 0x0Fu;
texture_color.bits = g_gpu_clut[palette_index];
}
break;
case GPUTextureMode::Palette8Bit:
{
const u16 palette_value =
GetPixel((cmd->draw_mode.GetTexturePageBaseX() + ZeroExtend32(texcoord_x / 2)) % VRAM_WIDTH,
(cmd->draw_mode.GetTexturePageBaseY() + ZeroExtend32(texcoord_y)) % VRAM_HEIGHT);
const size_t palette_index = (palette_value >> ((texcoord_x % 2) * 8)) & 0xFFu;
texture_color.bits = g_gpu_clut[palette_index];
}
break;
default:
{
texture_color.bits = GetPixel((cmd->draw_mode.GetTexturePageBaseX() + ZeroExtend32(texcoord_x)) % VRAM_WIDTH,
(cmd->draw_mode.GetTexturePageBaseY() + ZeroExtend32(texcoord_y)) % VRAM_HEIGHT);
}
break;
}
if (texture_color.bits == 0)
return;
if constexpr (raw_texture_enable)
{
color.bits = texture_color.bits;
}
else
{
const u32 dither_y = (dithering_enable) ? (y & 3u) : 2u;
const u32 dither_x = (dithering_enable) ? (x & 3u) : 3u;
color.bits = (ZeroExtend16(s_dither_lut[dither_y][dither_x][(u16(texture_color.r) * u16(color_r)) >> 4]) << 0) |
(ZeroExtend16(s_dither_lut[dither_y][dither_x][(u16(texture_color.g) * u16(color_g)) >> 4]) << 5) |
(ZeroExtend16(s_dither_lut[dither_y][dither_x][(u16(texture_color.b) * u16(color_b)) >> 4]) << 10) |
(texture_color.bits & 0x8000u);
}
}
else
{
const u32 dither_y = (dithering_enable) ? (y & 3u) : 2u;
const u32 dither_x = (dithering_enable) ? (x & 3u) : 3u;
// Non-textured transparent polygons don't set bit 15, but are treated as transparent.
color.bits = (ZeroExtend16(s_dither_lut[dither_y][dither_x][color_r]) << 0) |
(ZeroExtend16(s_dither_lut[dither_y][dither_x][color_g]) << 5) |
(ZeroExtend16(s_dither_lut[dither_y][dither_x][color_b]) << 10) | (transparency_enable ? 0x8000u : 0);
}
const VRAMPixel bg_color{GetPixel(static_cast<u32>(x), static_cast<u32>(y))};
if constexpr (transparency_enable)
{
if (color.bits & 0x8000u || !texture_enable)
{
// Based on blargg's efficient 15bpp pixel math.
u32 bg_bits = ZeroExtend32(bg_color.bits);
u32 fg_bits = ZeroExtend32(color.bits);
switch (cmd->draw_mode.transparency_mode)
{
case GPUTransparencyMode::HalfBackgroundPlusHalfForeground:
{
bg_bits |= 0x8000u;
color.bits = Truncate16(((fg_bits + bg_bits) - ((fg_bits ^ bg_bits) & 0x0421u)) >> 1);
}
break;
case GPUTransparencyMode::BackgroundPlusForeground:
{
bg_bits &= ~0x8000u;
const u32 sum = fg_bits + bg_bits;
const u32 carry = (sum - ((fg_bits ^ bg_bits) & 0x8421u)) & 0x8420u;
color.bits = Truncate16((sum - carry) | (carry - (carry >> 5)));
}
break;
case GPUTransparencyMode::BackgroundMinusForeground:
{
bg_bits |= 0x8000u;
fg_bits &= ~0x8000u;
const u32 diff = bg_bits - fg_bits + 0x108420u;
const u32 borrow = (diff - ((bg_bits ^ fg_bits) & 0x108420u)) & 0x108420u;
color.bits = Truncate16((diff - borrow) & (borrow - (borrow >> 5)));
}
break;
case GPUTransparencyMode::BackgroundPlusQuarterForeground:
{
bg_bits &= ~0x8000u;
fg_bits = ((fg_bits >> 2) & 0x1CE7u) | 0x8000u;
const u32 sum = fg_bits + bg_bits;
const u32 carry = (sum - ((fg_bits ^ bg_bits) & 0x8421u)) & 0x8420u;
color.bits = Truncate16((sum - carry) | (carry - (carry >> 5)));
}
break;
default:
break;
}
// See above.
if constexpr (!texture_enable)
color.bits &= ~0x8000u;
}
}
const u16 mask_and = cmd->params.GetMaskAND();
if ((bg_color.bits & mask_and) != 0)
return;
DebugAssert(static_cast<u32>(x) < VRAM_WIDTH && static_cast<u32>(y) < VRAM_HEIGHT);
SetPixel(static_cast<u32>(x), static_cast<u32>(y), color.bits | cmd->params.GetMaskOR());
}
template<bool texture_enable, bool raw_texture_enable, bool transparency_enable>
void GPU_SW_Backend::DrawRectangle(const GPUBackendDrawRectangleCommand* cmd)
{
const s32 origin_x = cmd->x;
const s32 origin_y = cmd->y;
const auto [r, g, b] = UnpackColorRGB24(cmd->color);
const auto [origin_texcoord_x, origin_texcoord_y] = UnpackTexcoord(cmd->texcoord);
for (u32 offset_y = 0; offset_y < cmd->height; offset_y++)
{
const s32 y = origin_y + static_cast<s32>(offset_y);
if (y < static_cast<s32>(m_drawing_area.top) || y > static_cast<s32>(m_drawing_area.bottom) ||
(cmd->params.interlaced_rendering && cmd->params.active_line_lsb == (Truncate8(static_cast<u32>(y)) & 1u)))
{
continue;
}
const u32 draw_y = static_cast<u32>(y) & VRAM_HEIGHT_MASK;
const u8 texcoord_y = Truncate8(ZeroExtend32(origin_texcoord_y) + offset_y);
for (u32 offset_x = 0; offset_x < cmd->width; offset_x++)
{
const s32 x = origin_x + static_cast<s32>(offset_x);
if (x < static_cast<s32>(m_drawing_area.left) || x > static_cast<s32>(m_drawing_area.right))
continue;
const u8 texcoord_x = Truncate8(ZeroExtend32(origin_texcoord_x) + offset_x);
ShadePixel<texture_enable, raw_texture_enable, transparency_enable, false>(cmd, static_cast<u32>(x), draw_y, r, g,
b, texcoord_x, texcoord_y);
}
}
}
//////////////////////////////////////////////////////////////////////////
// Polygon and line rasterization ported from Mednafen
//////////////////////////////////////////////////////////////////////////
#define COORD_FBS 12
#define COORD_MF_INT(n) ((n) << COORD_FBS)
#define COORD_POST_PADDING 12
static ALWAYS_INLINE_RELEASE s64 MakePolyXFP(s32 x)
{
return ((u64)x << 32) + ((1ULL << 32) - (1 << 11));
}
static ALWAYS_INLINE_RELEASE s64 MakePolyXFPStep(s32 dx, s32 dy)
{
s64 ret;
s64 dx_ex = (u64)dx << 32;
if (dx_ex < 0)
dx_ex -= dy - 1;
if (dx_ex > 0)
dx_ex += dy - 1;
ret = dx_ex / dy;
return (ret);
}
static ALWAYS_INLINE_RELEASE s32 GetPolyXFP_Int(s64 xfp)
{
return (xfp >> 32);
}
template<bool shading_enable, bool texture_enable>
bool ALWAYS_INLINE_RELEASE GPU_SW_Backend::CalcIDeltas(i_deltas& idl, const GPUBackendDrawPolygonCommand::Vertex* A,
const GPUBackendDrawPolygonCommand::Vertex* B,
const GPUBackendDrawPolygonCommand::Vertex* C)
{
#define CALCIS(x, y) (((B->x - A->x) * (C->y - B->y)) - ((C->x - B->x) * (B->y - A->y)))
s32 denom = CALCIS(x, y);
if (!denom)
return false;
if constexpr (shading_enable)
{
idl.dr_dx = (u32)(CALCIS(r, y) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
idl.dr_dy = (u32)(CALCIS(x, r) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
idl.dg_dx = (u32)(CALCIS(g, y) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
idl.dg_dy = (u32)(CALCIS(x, g) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
idl.db_dx = (u32)(CALCIS(b, y) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
idl.db_dy = (u32)(CALCIS(x, b) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
}
if constexpr (texture_enable)
{
idl.du_dx = (u32)(CALCIS(u, y) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
idl.du_dy = (u32)(CALCIS(x, u) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
idl.dv_dx = (u32)(CALCIS(v, y) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
idl.dv_dy = (u32)(CALCIS(x, v) * (1 << COORD_FBS) / denom) << COORD_POST_PADDING;
}
return true;
#undef CALCIS
}
template<bool shading_enable, bool texture_enable>
void ALWAYS_INLINE_RELEASE GPU_SW_Backend::AddIDeltas_DX(i_group& ig, const i_deltas& idl, u32 count /*= 1*/)
{
if constexpr (shading_enable)
{
ig.r += idl.dr_dx * count;
ig.g += idl.dg_dx * count;
ig.b += idl.db_dx * count;
}
if constexpr (texture_enable)
{
ig.u += idl.du_dx * count;
ig.v += idl.dv_dx * count;
}
}
template<bool shading_enable, bool texture_enable>
void ALWAYS_INLINE_RELEASE GPU_SW_Backend::AddIDeltas_DY(i_group& ig, const i_deltas& idl, u32 count /*= 1*/)
{
if constexpr (shading_enable)
{
ig.r += idl.dr_dy * count;
ig.g += idl.dg_dy * count;
ig.b += idl.db_dy * count;
}
if constexpr (texture_enable)
{
ig.u += idl.du_dy * count;
ig.v += idl.dv_dy * count;
}
}
template<bool shading_enable, bool texture_enable, bool raw_texture_enable, bool transparency_enable,
bool dithering_enable>
void GPU_SW_Backend::DrawSpan(const GPUBackendDrawPolygonCommand* cmd, s32 y, s32 x_start, s32 x_bound, i_group ig,
const i_deltas& idl)
{
if (cmd->params.interlaced_rendering && cmd->params.active_line_lsb == (Truncate8(static_cast<u32>(y)) & 1u))
return;
s32 x_ig_adjust = x_start;
s32 w = x_bound - x_start;
s32 x = TruncateGPUVertexPosition(x_start);
if (x < static_cast<s32>(m_drawing_area.left))
{
s32 delta = static_cast<s32>(m_drawing_area.left) - x;
x_ig_adjust += delta;
x += delta;
w -= delta;
}
if ((x + w) > (static_cast<s32>(m_drawing_area.right) + 1))
w = static_cast<s32>(m_drawing_area.right) + 1 - x;
if (w <= 0)
return;
AddIDeltas_DX<shading_enable, texture_enable>(ig, idl, x_ig_adjust);
AddIDeltas_DY<shading_enable, texture_enable>(ig, idl, y);
do
{
const u32 r = ig.r >> (COORD_FBS + COORD_POST_PADDING);
const u32 g = ig.g >> (COORD_FBS + COORD_POST_PADDING);
const u32 b = ig.b >> (COORD_FBS + COORD_POST_PADDING);
const u32 u = ig.u >> (COORD_FBS + COORD_POST_PADDING);
const u32 v = ig.v >> (COORD_FBS + COORD_POST_PADDING);
ShadePixel<texture_enable, raw_texture_enable, transparency_enable, dithering_enable>(
cmd, static_cast<u32>(x), static_cast<u32>(y), Truncate8(r), Truncate8(g), Truncate8(b), Truncate8(u),
Truncate8(v));
x++;
AddIDeltas_DX<shading_enable, texture_enable>(ig, idl);
} while (--w > 0);
}
template<bool shading_enable, bool texture_enable, bool raw_texture_enable, bool transparency_enable,
bool dithering_enable>
void GPU_SW_Backend::DrawTriangle(const GPUBackendDrawPolygonCommand* cmd,
const GPUBackendDrawPolygonCommand::Vertex* v0,
const GPUBackendDrawPolygonCommand::Vertex* v1,
const GPUBackendDrawPolygonCommand::Vertex* v2)
{
u32 core_vertex;
{
u32 cvtemp = 0;
if (v1->x <= v0->x)
{
if (v2->x <= v1->x)
cvtemp = (1 << 2);
else
cvtemp = (1 << 1);
}
else if (v2->x < v0->x)
cvtemp = (1 << 2);
else
cvtemp = (1 << 0);
if (v2->y < v1->y)
{
std::swap(v2, v1);
cvtemp = ((cvtemp >> 1) & 0x2) | ((cvtemp << 1) & 0x4) | (cvtemp & 0x1);
}
if (v1->y < v0->y)
{
std::swap(v1, v0);
cvtemp = ((cvtemp >> 1) & 0x1) | ((cvtemp << 1) & 0x2) | (cvtemp & 0x4);
}
if (v2->y < v1->y)
{
std::swap(v2, v1);
cvtemp = ((cvtemp >> 1) & 0x2) | ((cvtemp << 1) & 0x4) | (cvtemp & 0x1);
}
core_vertex = cvtemp >> 1;
}
if (v0->y == v2->y)
return;
if (static_cast<u32>(std::abs(v2->x - v0->x)) >= MAX_PRIMITIVE_WIDTH ||
static_cast<u32>(std::abs(v2->x - v1->x)) >= MAX_PRIMITIVE_WIDTH ||
static_cast<u32>(std::abs(v1->x - v0->x)) >= MAX_PRIMITIVE_WIDTH ||
static_cast<u32>(v2->y - v0->y) >= MAX_PRIMITIVE_HEIGHT)
{
return;
}
s64 base_coord = MakePolyXFP(v0->x);
s64 base_step = MakePolyXFPStep((v2->x - v0->x), (v2->y - v0->y));
s64 bound_coord_us;
s64 bound_coord_ls;
bool right_facing;
if (v1->y == v0->y)
{
bound_coord_us = 0;
right_facing = (bool)(v1->x > v0->x);
}
else
{
bound_coord_us = MakePolyXFPStep((v1->x - v0->x), (v1->y - v0->y));
right_facing = (bool)(bound_coord_us > base_step);
}
if (v2->y == v1->y)
bound_coord_ls = 0;
else
bound_coord_ls = MakePolyXFPStep((v2->x - v1->x), (v2->y - v1->y));
i_deltas idl;
if (!CalcIDeltas<shading_enable, texture_enable>(idl, v0, v1, v2))
return;
const GPUBackendDrawPolygonCommand::Vertex* vertices[3] = {v0, v1, v2};
i_group ig;
if constexpr (texture_enable)
{
ig.u = (COORD_MF_INT(vertices[core_vertex]->u) + (1 << (COORD_FBS - 1))) << COORD_POST_PADDING;
ig.v = (COORD_MF_INT(vertices[core_vertex]->v) + (1 << (COORD_FBS - 1))) << COORD_POST_PADDING;
}
ig.r = (COORD_MF_INT(vertices[core_vertex]->r) + (1 << (COORD_FBS - 1))) << COORD_POST_PADDING;
ig.g = (COORD_MF_INT(vertices[core_vertex]->g) + (1 << (COORD_FBS - 1))) << COORD_POST_PADDING;
ig.b = (COORD_MF_INT(vertices[core_vertex]->b) + (1 << (COORD_FBS - 1))) << COORD_POST_PADDING;
AddIDeltas_DX<shading_enable, texture_enable>(ig, idl, -vertices[core_vertex]->x);
AddIDeltas_DY<shading_enable, texture_enable>(ig, idl, -vertices[core_vertex]->y);
struct TriangleHalf
{
u64 x_coord[2];
u64 x_step[2];
s32 y_coord;
s32 y_bound;
bool dec_mode;
} tripart[2];
u32 vo = 0;
u32 vp = 0;
if (core_vertex != 0)
vo = 1;
if (core_vertex == 2)
vp = 3;
{
TriangleHalf* tp = &tripart[vo];
tp->y_coord = vertices[0 ^ vo]->y;
tp->y_bound = vertices[1 ^ vo]->y;
tp->x_coord[right_facing] = MakePolyXFP(vertices[0 ^ vo]->x);
tp->x_step[right_facing] = bound_coord_us;
tp->x_coord[!right_facing] = base_coord + ((vertices[vo]->y - vertices[0]->y) * base_step);
tp->x_step[!right_facing] = base_step;
tp->dec_mode = vo;
}
{
TriangleHalf* tp = &tripart[vo ^ 1];
tp->y_coord = vertices[1 ^ vp]->y;
tp->y_bound = vertices[2 ^ vp]->y;
tp->x_coord[right_facing] = MakePolyXFP(vertices[1 ^ vp]->x);
tp->x_step[right_facing] = bound_coord_ls;
tp->x_coord[!right_facing] =
base_coord + ((vertices[1 ^ vp]->y - vertices[0]->y) *
base_step); // base_coord + ((vertices[1].y - vertices[0].y) * base_step);
tp->x_step[!right_facing] = base_step;
tp->dec_mode = vp;
}
for (u32 i = 0; i < 2; i++)
{
s32 yi = tripart[i].y_coord;
s32 yb = tripart[i].y_bound;
u64 lc = tripart[i].x_coord[0];
u64 ls = tripart[i].x_step[0];
u64 rc = tripart[i].x_coord[1];
u64 rs = tripart[i].x_step[1];
if (tripart[i].dec_mode)
{
while (yi > yb)
{
yi--;
lc -= ls;
rc -= rs;
s32 y = TruncateGPUVertexPosition(yi);
if (y < static_cast<s32>(m_drawing_area.top))
break;
if (y > static_cast<s32>(m_drawing_area.bottom))
continue;
DrawSpan<shading_enable, texture_enable, raw_texture_enable, transparency_enable, dithering_enable>(
cmd, y & VRAM_HEIGHT_MASK, GetPolyXFP_Int(lc), GetPolyXFP_Int(rc), ig, idl);
}
}
else
{
while (yi < yb)
{
s32 y = TruncateGPUVertexPosition(yi);
if (y > static_cast<s32>(m_drawing_area.bottom))
break;
if (y >= static_cast<s32>(m_drawing_area.top))
{
DrawSpan<shading_enable, texture_enable, raw_texture_enable, transparency_enable, dithering_enable>(
cmd, y & VRAM_HEIGHT_MASK, GetPolyXFP_Int(lc), GetPolyXFP_Int(rc), ig, idl);
}
yi++;
lc += ls;
rc += rs;
}
}
}
}
enum
{
Line_XY_FractBits = 32
};
enum
{
Line_RGB_FractBits = 12
};
struct line_fxp_coord
{
u64 x, y;
u32 r, g, b;
};
struct line_fxp_step
{
s64 dx_dk, dy_dk;
s32 dr_dk, dg_dk, db_dk;
};
static ALWAYS_INLINE_RELEASE s64 LineDivide(s64 delta, s32 dk)
{
delta = (u64)delta << Line_XY_FractBits;
if (delta < 0)
delta -= dk - 1;
if (delta > 0)
delta += dk - 1;
return (delta / dk);
}
template<bool shading_enable, bool transparency_enable, bool dithering_enable>
void GPU_SW_Backend::DrawLine(const GPUBackendDrawLineCommand* cmd, const GPUBackendDrawLineCommand::Vertex* p0,
const GPUBackendDrawLineCommand::Vertex* p1)
{
const s32 i_dx = std::abs(p1->x - p0->x);
const s32 i_dy = std::abs(p1->y - p0->y);
const s32 k = (i_dx > i_dy) ? i_dx : i_dy;
if (i_dx >= MAX_PRIMITIVE_WIDTH || i_dy >= MAX_PRIMITIVE_HEIGHT)
return;
if (p0->x >= p1->x && k > 0)
std::swap(p0, p1);
line_fxp_step step;
if (k == 0)
{
step.dx_dk = 0;
step.dy_dk = 0;
if constexpr (shading_enable)
{
step.dr_dk = 0;
step.dg_dk = 0;
step.db_dk = 0;
}
}
else
{
step.dx_dk = LineDivide(p1->x - p0->x, k);
step.dy_dk = LineDivide(p1->y - p0->y, k);
if constexpr (shading_enable)
{
step.dr_dk = (s32)((u32)(p1->r - p0->r) << Line_RGB_FractBits) / k;
step.dg_dk = (s32)((u32)(p1->g - p0->g) << Line_RGB_FractBits) / k;
step.db_dk = (s32)((u32)(p1->b - p0->b) << Line_RGB_FractBits) / k;
}
}
line_fxp_coord cur_point;
cur_point.x = ((u64)p0->x << Line_XY_FractBits) | (1ULL << (Line_XY_FractBits - 1));
cur_point.y = ((u64)p0->y << Line_XY_FractBits) | (1ULL << (Line_XY_FractBits - 1));
cur_point.x -= 1024;
if (step.dy_dk < 0)
cur_point.y -= 1024;
if constexpr (shading_enable)
{
cur_point.r = (p0->r << Line_RGB_FractBits) | (1 << (Line_RGB_FractBits - 1));
cur_point.g = (p0->g << Line_RGB_FractBits) | (1 << (Line_RGB_FractBits - 1));
cur_point.b = (p0->b << Line_RGB_FractBits) | (1 << (Line_RGB_FractBits - 1));
}
for (s32 i = 0; i <= k; i++)
{
// Sign extension is not necessary here for x and y, due to the maximum values that ClipX1 and ClipY1 can contain.
const s32 x = (cur_point.x >> Line_XY_FractBits) & 2047;
const s32 y = (cur_point.y >> Line_XY_FractBits) & 2047;
if ((!cmd->params.interlaced_rendering || cmd->params.active_line_lsb != (Truncate8(static_cast<u32>(y)) & 1u)) &&
x >= static_cast<s32>(m_drawing_area.left) && x <= static_cast<s32>(m_drawing_area.right) &&
y >= static_cast<s32>(m_drawing_area.top) && y <= static_cast<s32>(m_drawing_area.bottom))
{
const u8 r = shading_enable ? static_cast<u8>(cur_point.r >> Line_RGB_FractBits) : p0->r;
const u8 g = shading_enable ? static_cast<u8>(cur_point.g >> Line_RGB_FractBits) : p0->g;
const u8 b = shading_enable ? static_cast<u8>(cur_point.b >> Line_RGB_FractBits) : p0->b;
ShadePixel<false, false, transparency_enable, dithering_enable>(
cmd, static_cast<u32>(x), static_cast<u32>(y) & VRAM_HEIGHT_MASK, r, g, b, 0, 0);
}
cur_point.x += step.dx_dk;
cur_point.y += step.dy_dk;
if constexpr (shading_enable)
{
cur_point.r += step.dr_dk;
cur_point.g += step.dg_dk;
cur_point.b += step.db_dk;
}
}
DrawFunction(cmd, &cmd->vertices[i - 1], &cmd->vertices[i]);
}
void GPU_SW_Backend::FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color, GPUBackendCommandParameters params)
@ -896,82 +242,16 @@ void GPU_SW_Backend::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 wi
}
}
void GPU_SW_Backend::FlushRender()
{
}
void GPU_SW_Backend::DrawingAreaChanged()
{
}
void GPU_SW_Backend::UpdateCLUT(GPUTexturePaletteReg reg, bool clut_is_8bit)
{
GPU::ReadCLUT(g_gpu_clut, reg, clut_is_8bit);
}
GPU_SW_Backend::DrawLineFunction GPU_SW_Backend::GetDrawLineFunction(bool shading_enable, bool transparency_enable,
bool dithering_enable)
void GPU_SW_Backend::DrawingAreaChanged(const GPUDrawingArea& new_drawing_area, const GSVector4i clamped_drawing_area)
{
static constexpr DrawLineFunction funcs[2][2][2] = {
{{&GPU_SW_Backend::DrawLine<false, false, false>, &GPU_SW_Backend::DrawLine<false, false, true>},
{&GPU_SW_Backend::DrawLine<false, true, false>, &GPU_SW_Backend::DrawLine<false, true, true>}},
{{&GPU_SW_Backend::DrawLine<true, false, false>, &GPU_SW_Backend::DrawLine<true, false, true>},
{&GPU_SW_Backend::DrawLine<true, true, false>, &GPU_SW_Backend::DrawLine<true, true, true>}}};
return funcs[u8(shading_enable)][u8(transparency_enable)][u8(dithering_enable)];
GPU_SW_Rasterizer::g_drawing_area = new_drawing_area;
}
GPU_SW_Backend::DrawRectangleFunction
GPU_SW_Backend::GetDrawRectangleFunction(bool texture_enable, bool raw_texture_enable, bool transparency_enable)
void GPU_SW_Backend::FlushRender()
{
static constexpr DrawRectangleFunction funcs[2][2][2] = {
{{&GPU_SW_Backend::DrawRectangle<false, false, false>, &GPU_SW_Backend::DrawRectangle<false, false, true>},
{&GPU_SW_Backend::DrawRectangle<false, false, false>, &GPU_SW_Backend::DrawRectangle<false, false, true>}},
{{&GPU_SW_Backend::DrawRectangle<true, false, false>, &GPU_SW_Backend::DrawRectangle<true, false, true>},
{&GPU_SW_Backend::DrawRectangle<true, true, false>, &GPU_SW_Backend::DrawRectangle<true, true, true>}}};
return funcs[u8(texture_enable)][u8(raw_texture_enable)][u8(transparency_enable)];
}
GPU_SW_Backend::DrawTriangleFunction GPU_SW_Backend::GetDrawTriangleFunction(bool shading_enable, bool texture_enable,
bool raw_texture_enable,
bool transparency_enable,
bool dithering_enable)
{
static constexpr DrawTriangleFunction funcs[2][2][2][2][2] = {
{{{{&GPU_SW_Backend::DrawTriangle<false, false, false, false, false>,
&GPU_SW_Backend::DrawTriangle<false, false, false, false, true>},
{&GPU_SW_Backend::DrawTriangle<false, false, false, true, false>,
&GPU_SW_Backend::DrawTriangle<false, false, false, true, true>}},
{{&GPU_SW_Backend::DrawTriangle<false, false, false, false, false>,
&GPU_SW_Backend::DrawTriangle<false, false, false, false, false>},
{&GPU_SW_Backend::DrawTriangle<false, false, false, true, false>,
&GPU_SW_Backend::DrawTriangle<false, false, false, true, false>}}},
{{{&GPU_SW_Backend::DrawTriangle<false, true, false, false, false>,
&GPU_SW_Backend::DrawTriangle<false, true, false, false, true>},
{&GPU_SW_Backend::DrawTriangle<false, true, false, true, false>,
&GPU_SW_Backend::DrawTriangle<false, true, false, true, true>}},
{{&GPU_SW_Backend::DrawTriangle<false, true, true, false, false>,
&GPU_SW_Backend::DrawTriangle<false, true, true, false, false>},
{&GPU_SW_Backend::DrawTriangle<false, true, true, true, false>,
&GPU_SW_Backend::DrawTriangle<false, true, true, true, false>}}}},
{{{{&GPU_SW_Backend::DrawTriangle<true, false, false, false, false>,
&GPU_SW_Backend::DrawTriangle<true, false, false, false, true>},
{&GPU_SW_Backend::DrawTriangle<true, false, false, true, false>,
&GPU_SW_Backend::DrawTriangle<true, false, false, true, true>}},
{{&GPU_SW_Backend::DrawTriangle<true, false, false, false, false>,
&GPU_SW_Backend::DrawTriangle<true, false, false, false, false>},
{&GPU_SW_Backend::DrawTriangle<true, false, false, true, false>,
&GPU_SW_Backend::DrawTriangle<true, false, false, true, false>}}},
{{{&GPU_SW_Backend::DrawTriangle<true, true, false, false, false>,
&GPU_SW_Backend::DrawTriangle<true, true, false, false, true>},
{&GPU_SW_Backend::DrawTriangle<true, true, false, true, false>,
&GPU_SW_Backend::DrawTriangle<true, true, false, true, true>}},
{{&GPU_SW_Backend::DrawTriangle<true, true, true, false, false>,
&GPU_SW_Backend::DrawTriangle<true, true, true, false, false>},
{&GPU_SW_Backend::DrawTriangle<true, true, true, true, false>,
&GPU_SW_Backend::DrawTriangle<true, true, true, true, false>}}}}};
return funcs[u8(shading_enable)][u8(texture_enable)][u8(raw_texture_enable)][u8(transparency_enable)]
[u8(dithering_enable)];
}

142
src/core/gpu_sw_backend.h
View File

@ -17,77 +17,7 @@ public:
bool Initialize(bool force_thread) override;
void Reset() override;
ALWAYS_INLINE_RELEASE u16 GetPixel(const u32 x, const u32 y) const { return g_vram[VRAM_WIDTH * y + x]; }
ALWAYS_INLINE_RELEASE const u16* GetPixelPtr(const u32 x, const u32 y) const { return &g_vram[VRAM_WIDTH * y + x]; }
ALWAYS_INLINE_RELEASE u16* GetPixelPtr(const u32 x, const u32 y) { return &g_vram[VRAM_WIDTH * y + x]; }
ALWAYS_INLINE_RELEASE void SetPixel(const u32 x, const u32 y, const u16 value) { g_vram[VRAM_WIDTH * y + x] = value; }
// this is actually (31 * 255) >> 4) == 494, but to simplify addressing we use the next power of two (512)
static constexpr u32 DITHER_LUT_SIZE = 512;
using DitherLUT = std::array<std::array<std::array<u8, 512>, DITHER_MATRIX_SIZE>, DITHER_MATRIX_SIZE>;
static constexpr DitherLUT ComputeDitherLUT();
protected:
union VRAMPixel
{
u16 bits;
BitField<u16, u8, 0, 5> r;
BitField<u16, u8, 5, 5> g;
BitField<u16, u8, 10, 5> b;
BitField<u16, bool, 15, 1> c;
void Set(u8 r_, u8 g_, u8 b_, bool c_ = false)
{
bits = (ZeroExtend16(r_)) | (ZeroExtend16(g_) << 5) | (ZeroExtend16(b_) << 10) | (static_cast<u16>(c_) << 15);
}
void ClampAndSet(u8 r_, u8 g_, u8 b_, bool c_ = false)
{
Set(std::min<u8>(r_, 0x1F), std::min<u8>(g_, 0x1F), std::min<u8>(b_, 0x1F), c_);
}
void SetRGB24(u32 rgb24, bool c_ = false)
{
bits = Truncate16(((rgb24 >> 3) & 0x1F) | (((rgb24 >> 11) & 0x1F) << 5) | (((rgb24 >> 19) & 0x1F) << 10)) |
(static_cast<u16>(c_) << 15);
}
void SetRGB24(u8 r8, u8 g8, u8 b8, bool c_ = false)
{
bits = (ZeroExtend16(r8 >> 3)) | (ZeroExtend16(g8 >> 3) << 5) | (ZeroExtend16(b8 >> 3) << 10) |
(static_cast<u16>(c_) << 15);
}
void SetRGB24Dithered(u32 x, u32 y, u8 r8, u8 g8, u8 b8, bool c_ = false)
{
const s32 offset = DITHER_MATRIX[y & 3][x & 3];
r8 = static_cast<u8>(std::clamp<s32>(static_cast<s32>(ZeroExtend32(r8)) + offset, 0, 255));
g8 = static_cast<u8>(std::clamp<s32>(static_cast<s32>(ZeroExtend32(g8)) + offset, 0, 255));
b8 = static_cast<u8>(std::clamp<s32>(static_cast<s32>(ZeroExtend32(b8)) + offset, 0, 255));
SetRGB24(r8, g8, b8, c_);
}
u32 ToRGB24() const
{
const u32 r_ = ZeroExtend32(r.GetValue());
const u32 g_ = ZeroExtend32(g.GetValue());
const u32 b_ = ZeroExtend32(b.GetValue());
return ((r_ << 3) | (r_ & 7)) | (((g_ << 3) | (g_ & 7)) << 8) | (((b_ << 3) | (b_ & 7)) << 16);
}
};
static constexpr std::tuple<u8, u8> UnpackTexcoord(u16 texcoord)
{
return std::make_tuple(static_cast<u8>(texcoord), static_cast<u8>(texcoord >> 8));
}
static constexpr std::tuple<u8, u8, u8> UnpackColorRGB24(u32 rgb24)
{
return std::make_tuple(static_cast<u8>(rgb24), static_cast<u8>(rgb24 >> 8), static_cast<u8>(rgb24 >> 16));
}
void FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color, GPUBackendCommandParameters params) override;
void UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* data, GPUBackendCommandParameters params) override;
void CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height,
@ -96,75 +26,7 @@ protected:
void DrawPolygon(const GPUBackendDrawPolygonCommand* cmd) override;
void DrawLine(const GPUBackendDrawLineCommand* cmd) override;
void DrawRectangle(const GPUBackendDrawRectangleCommand* cmd) override;
void FlushRender() override;
void DrawingAreaChanged() override;
void DrawingAreaChanged(const GPUDrawingArea& new_drawing_area, const GSVector4i clamped_drawing_area) override;
void UpdateCLUT(GPUTexturePaletteReg reg, bool clut_is_8bit) override;
//////////////////////////////////////////////////////////////////////////
// Rasterization
//////////////////////////////////////////////////////////////////////////
template<bool texture_enable, bool raw_texture_enable, bool transparency_enable, bool dithering_enable>
void ShadePixel(const GPUBackendDrawCommand* cmd, u32 x, u32 y, u8 color_r, u8 color_g, u8 color_b, u8 texcoord_x,
u8 texcoord_y);
template<bool texture_enable, bool raw_texture_enable, bool transparency_enable>
void DrawRectangle(const GPUBackendDrawRectangleCommand* cmd);
using DrawRectangleFunction = void (GPU_SW_Backend::*)(const GPUBackendDrawRectangleCommand* cmd);
DrawRectangleFunction GetDrawRectangleFunction(bool texture_enable, bool raw_texture_enable,
bool transparency_enable);
//////////////////////////////////////////////////////////////////////////
// Polygon and line rasterization ported from Mednafen
//////////////////////////////////////////////////////////////////////////
struct i_deltas
{
u32 du_dx, dv_dx;
u32 dr_dx, dg_dx, db_dx;
u32 du_dy, dv_dy;
u32 dr_dy, dg_dy, db_dy;
};
struct i_group
{
u32 u, v;
u32 r, g, b;
};
template<bool shading_enable, bool texture_enable>
bool CalcIDeltas(i_deltas& idl, const GPUBackendDrawPolygonCommand::Vertex* A,
const GPUBackendDrawPolygonCommand::Vertex* B, const GPUBackendDrawPolygonCommand::Vertex* C);
template<bool shading_enable, bool texture_enable>
void AddIDeltas_DX(i_group& ig, const i_deltas& idl, u32 count = 1);
template<bool shading_enable, bool texture_enable>
void AddIDeltas_DY(i_group& ig, const i_deltas& idl, u32 count = 1);
template<bool shading_enable, bool texture_enable, bool raw_texture_enable, bool transparency_enable,
bool dithering_enable>
void DrawSpan(const GPUBackendDrawPolygonCommand* cmd, s32 y, s32 x_start, s32 x_bound, i_group ig,
const i_deltas& idl);
template<bool shading_enable, bool texture_enable, bool raw_texture_enable, bool transparency_enable,
bool dithering_enable>
void DrawTriangle(const GPUBackendDrawPolygonCommand* cmd, const GPUBackendDrawPolygonCommand::Vertex* v0,
const GPUBackendDrawPolygonCommand::Vertex* v1, const GPUBackendDrawPolygonCommand::Vertex* v2);
using DrawTriangleFunction = void (GPU_SW_Backend::*)(const GPUBackendDrawPolygonCommand* cmd,
const GPUBackendDrawPolygonCommand::Vertex* v0,
const GPUBackendDrawPolygonCommand::Vertex* v1,
const GPUBackendDrawPolygonCommand::Vertex* v2);
DrawTriangleFunction GetDrawTriangleFunction(bool shading_enable, bool texture_enable, bool raw_texture_enable,
bool transparency_enable, bool dithering_enable);
template<bool shading_enable, bool transparency_enable, bool dithering_enable>
void DrawLine(const GPUBackendDrawLineCommand* cmd, const GPUBackendDrawLineCommand::Vertex* p0,
const GPUBackendDrawLineCommand::Vertex* p1);
using DrawLineFunction = void (GPU_SW_Backend::*)(const GPUBackendDrawLineCommand* cmd,
const GPUBackendDrawLineCommand::Vertex* p0,
const GPUBackendDrawLineCommand::Vertex* p1);
DrawLineFunction GetDrawLineFunction(bool shading_enable, bool transparency_enable, bool dithering_enable);
void FlushRender() override;
};

100
src/core/gpu_sw_rasterizer.cpp Normal file
View File

@ -0,0 +1,100 @@
// SPDX-FileCopyrightText: 2019-2024 Connor McLaughlin <stenzek@gmail.com>
// SPDX-License-Identifier: CC-BY-NC-ND-4.0
#include "gpu_sw_rasterizer.h"
#include "gpu.h"
#include "cpuinfo.h"
#include "common/log.h"
#include "common/string_util.h"
Log_SetChannel(GPU_SW_Rasterizer);
namespace GPU_SW_Rasterizer {
// Default implementation, compatible with all ISAs.
extern const DrawRectangleFunctionTable DrawRectangleFunctions;
extern const DrawTriangleFunctionTable DrawTriangleFunctions;
extern const DrawLineFunctionTable DrawLineFunctions;
constinit const DitherLUT g_dither_lut = []() constexpr {
DitherLUT lut = {};
for (u32 i = 0; i < DITHER_MATRIX_SIZE; i++)
{
for (u32 j = 0; j < DITHER_MATRIX_SIZE; j++)
{
for (u32 value = 0; value < DITHER_LUT_SIZE; value++)
{
const s32 dithered_value = (static_cast<s32>(value) + DITHER_MATRIX[i][j]) >> 3;
lut[i][j][value] = static_cast<u8>((dithered_value < 0) ? 0 : ((dithered_value > 31) ? 31 : dithered_value));
}
}
}
return lut;
}();
GPUDrawingArea g_drawing_area = {};
} // namespace GPU_SW_Rasterizer
// Default implementation definitions.
namespace GPU_SW_Rasterizer {
#include "gpu_sw_rasterizer.inl"
}
// Default vector implementation definitions.
#if defined(CPU_ARCH_SSE) || defined(CPU_ARCH_NEON)
namespace GPU_SW_Rasterizer::SIMD {
#include "gpu_sw_rasterizer.inl"
}
#endif
// Initialize with default implementation.
namespace GPU_SW_Rasterizer {
const DrawRectangleFunctionTable* SelectedDrawRectangleFunctions = &DrawRectangleFunctions;
const DrawTriangleFunctionTable* SelectedDrawTriangleFunctions = &DrawTriangleFunctions;
const DrawLineFunctionTable* SelectedDrawLineFunctions = &DrawLineFunctions;
} // namespace GPU_SW_Rasterizer
// Declare alternative implementations.
void GPU_SW_Rasterizer::SelectImplementation()
{
static bool selected = false;
if (selected)
return;
selected = true;
#define SELECT_ALTERNATIVE_RASTERIZER(isa) \
do \
{ \
INFO_LOG("Using " #isa " software rasterizer implementation."); \
SelectedDrawRectangleFunctions = &isa::DrawRectangleFunctions; \
SelectedDrawTriangleFunctions = &isa::DrawTriangleFunctions; \
SelectedDrawLineFunctions = &isa::DrawLineFunctions; \
} while (0)
#if defined(CPU_ARCH_SSE) || defined(CPU_ARCH_NEON)
const char* use_isa = std::getenv("SW_USE_ISA");
// Default to scalar for now, until vector is finished.
use_isa = use_isa ? use_isa : "Scalar";
#if defined(CPU_ARCH_SSE) && defined(_MSC_VER)
if (cpuinfo_has_x86_avx2() && (!use_isa || StringUtil::Strcasecmp(use_isa, "AVX2") == 0))
{
SELECT_ALTERNATIVE_RASTERIZER(AVX2);
return;
}
#endif
if (!use_isa || StringUtil::Strcasecmp(use_isa, "SIMD") == 0)
{
SELECT_ALTERNATIVE_RASTERIZER(SIMD);
return;
}
#endif
INFO_LOG("Using scalar software rasterizer implementation.");
#undef SELECT_ALTERNATIVE_RASTERIZER
}

89
src/core/gpu_sw_rasterizer.h Normal file
View File

@ -0,0 +1,89 @@
// SPDX-FileCopyrightText: 2019-2024 Connor McLaughlin <stenzek@gmail.com>
// SPDX-License-Identifier: CC-BY-NC-ND-4.0
#pragma once
#include "gpu.h"
#include "gpu_types.h"
#include "common/intrin.h"
#include "common/types.h"
#include <algorithm>
#include <array>
namespace GPU_SW_Rasterizer {
// this is actually (31 * 255) >> 4) == 494, but to simplify addressing we use the next power of two (512)
static constexpr u32 DITHER_LUT_SIZE = 512;
using DitherLUT = std::array<std::array<std::array<u8, DITHER_LUT_SIZE>, DITHER_MATRIX_SIZE>, DITHER_MATRIX_SIZE>;
extern const DitherLUT g_dither_lut;
extern GPUDrawingArea g_drawing_area;
using DrawRectangleFunction = void (*)(const GPUBackendDrawRectangleCommand* cmd);
typedef const DrawRectangleFunction DrawRectangleFunctionTable[2][2][2];
using DrawTriangleFunction = void (*)(const GPUBackendDrawPolygonCommand* cmd,
const GPUBackendDrawPolygonCommand::Vertex* v0,
const GPUBackendDrawPolygonCommand::Vertex* v1,
const GPUBackendDrawPolygonCommand::Vertex* v2);
typedef const DrawTriangleFunction DrawTriangleFunctionTable[2][2][2][2][2];
using DrawLineFunction = void (*)(const GPUBackendDrawLineCommand* cmd, const GPUBackendDrawLineCommand::Vertex* p0,
const GPUBackendDrawLineCommand::Vertex* p1);
typedef const DrawLineFunction DrawLineFunctionTable[2][2][2];
// Default implementation, compatible with all ISAs.
extern const DrawRectangleFunctionTable DrawRectangleFunctions;
extern const DrawTriangleFunctionTable DrawTriangleFunctions;
extern const DrawLineFunctionTable DrawLineFunctions;
// Current implementation, selected at runtime.
extern const DrawRectangleFunctionTable* SelectedDrawRectangleFunctions;
extern const DrawTriangleFunctionTable* SelectedDrawTriangleFunctions;
extern const DrawLineFunctionTable* SelectedDrawLineFunctions;
extern void SelectImplementation();
ALWAYS_INLINE static DrawLineFunction GetDrawLineFunction(bool shading_enable, bool transparency_enable,
bool dithering_enable)
{
return (*SelectedDrawLineFunctions)[u8(shading_enable)][u8(transparency_enable)][u8(dithering_enable)];
}
ALWAYS_INLINE static DrawRectangleFunction GetDrawRectangleFunction(bool texture_enable, bool raw_texture_enable,
bool transparency_enable)
{
return (*SelectedDrawRectangleFunctions)[u8(texture_enable)][u8(raw_texture_enable)][u8(transparency_enable)];
}
ALWAYS_INLINE static DrawTriangleFunction GetDrawTriangleFunction(bool shading_enable, bool texture_enable,
bool raw_texture_enable, bool transparency_enable,
bool dithering_enable)
{
return (*SelectedDrawTriangleFunctions)[u8(shading_enable)][u8(texture_enable)][u8(raw_texture_enable)]
[u8(transparency_enable)][u8(dithering_enable)];
}
#define DECLARE_ALTERNATIVE_RASTERIZER(isa) \
namespace isa { \
extern const DrawRectangleFunctionTable DrawRectangleFunctions; \
extern const DrawTriangleFunctionTable DrawTriangleFunctions; \
extern const DrawLineFunctionTable DrawLineFunctions; \
}
// Have to define the symbols globally, because clang won't include them otherwise.
#if defined(CPU_ARCH_SSE) && defined(_MSC_VER)
#define ALTERNATIVE_RASTERIZER_LIST() DECLARE_ALTERNATIVE_RASTERIZER(AVX2)
#else
#define ALTERNATIVE_RASTERIZER_LIST()
#endif
ALTERNATIVE_RASTERIZER_LIST()
#undef DECLARE_ALTERNATIVE_RASTERIZER
} // namespace GPU_SW_Rasterizer
// static u32 s_bad_counter = 0;

1250
src/core/gpu_sw_rasterizer.inl Normal file
View File

File diff suppressed because it is too large Load Diff

12
src/core/gpu_sw_rasterizer_avx2.cpp Normal file
View File

@ -0,0 +1,12 @@
// SPDX-FileCopyrightText: 2019-2024 Connor McLaughlin <stenzek@gmail.com>
// SPDX-License-Identifier: CC-BY-NC-ND-4.0
#include "gpu_sw_rasterizer.h"
#include "common/assert.h"
#include "common/gsvector.h"
namespace GPU_SW_Rasterizer::AVX2 {
#define USE_VECTOR 1
#include "gpu_sw_rasterizer.inl"
}

1
src/core/gpu_types.h
View File

@ -333,6 +333,7 @@ struct GPUBackendCopyVRAMCommand : public GPUBackendCommand
struct GPUBackendSetDrawingAreaCommand : public GPUBackendCommand
{
GPUDrawingArea new_area;
s32 new_clamped_area[4];
};
struct GPUBackendUpdateCLUTCommand : public GPUBackendCommand