Merge branch 'stenzek:master' into master

yeager · Jan 4, 2025 · 61e179e · 61e179e
2 parents dce2d07 + a5d90c0
commit 61e179e
Show file tree

Hide file tree

Showing 49 changed files with 2,296 additions and 1,789 deletions.
diff --git a/scripts/deps/build-dependencies-linux-cross.sh b/scripts/deps/build-dependencies-linux-cross.sh
@@ -90,7 +90,7 @@ LIBJPEGTURBO=3.0.4
 LIBPNG=1.6.44
 LIBWEBP=1.4.0
 LIBZIP=1.11.2
-SDL2=2.30.9
+SDL2=2.30.11
 QT=6.8.1
 ZSTD=1.5.6
 
@@ -118,7 +118,7 @@ fi
 
 cat > SHASUMS <<EOF
 baf8aebd22002b762d803ba0e1e389b6b4415159334e9d34bba1a938f6de8ce6  $LIBBACKTRACE.tar.gz
-24b574f71c87a763f50704bbb630cbe38298d544a1f890f099a4696b1d6beba4  SDL2-$SDL2.tar.gz
+8b8d4aef2038533da814965220f88f77d60dfa0f32685f80ead65e501337da7f  SDL2-$SDL2.tar.gz
 e1351218d270db49c3dddcba04fb2153b09731ea3fa6830e423f5952f44585be  cpuinfo-$CPUINFO.tar.gz
 3eea5ccce6670c126282f1ba4d32c19d486db49a1a5cbfb8d6f48774784d310c  discord-rpc-$DISCORD_RPC.tar.gz
 3998b024b0d442614a9ee270e76e018bb37a17b8c6941212171731123cbbcac7  lunasvg-$LUNASVG.tar.gz

diff --git a/scripts/deps/build-dependencies-linux.sh b/scripts/deps/build-dependencies-linux.sh
@@ -72,7 +72,7 @@ LIBJPEGTURBO=3.0.4
 LIBPNG=1.6.44
 LIBWEBP=1.4.0
 LIBZIP=1.11.2
-SDL2=2.30.9
+SDL2=2.30.11
 QT=6.8.1
 ZSTD=1.5.6
 
@@ -99,7 +99,7 @@ fi
 
 cat > SHASUMS <<EOF
 baf8aebd22002b762d803ba0e1e389b6b4415159334e9d34bba1a938f6de8ce6  $LIBBACKTRACE.tar.gz
-24b574f71c87a763f50704bbb630cbe38298d544a1f890f099a4696b1d6beba4  SDL2-$SDL2.tar.gz
+8b8d4aef2038533da814965220f88f77d60dfa0f32685f80ead65e501337da7f  SDL2-$SDL2.tar.gz
 e1351218d270db49c3dddcba04fb2153b09731ea3fa6830e423f5952f44585be  cpuinfo-$CPUINFO.tar.gz
 3eea5ccce6670c126282f1ba4d32c19d486db49a1a5cbfb8d6f48774784d310c  discord-rpc-$DISCORD_RPC.tar.gz
 3998b024b0d442614a9ee270e76e018bb37a17b8c6941212171731123cbbcac7  lunasvg-$LUNASVG.tar.gz

diff --git a/scripts/deps/build-dependencies-mac.sh b/scripts/deps/build-dependencies-mac.sh
@@ -38,7 +38,7 @@ fi
 
 FREETYPE=2.13.3
 HARFBUZZ=10.1.0
-SDL2=2.30.9
+SDL2=2.30.11
 ZSTD=1.5.6
 LIBPNG=1.6.44
 LIBJPEGTURBO=3.0.4
@@ -86,7 +86,7 @@ c758fdce8587641b00403ee0df2cd5d30cbea7803d43c65fddd76224f7b49b88  harfbuzz-$HARF
 99130559e7d62e8d695f2c0eaeef912c5828d5b84a0537dcb24c9678c9d5b76b  libjpeg-turbo-$LIBJPEGTURBO.tar.gz
 61f873ec69e3be1b99535634340d5bde750b2e4447caa1db9f61be3fd49ab1e5  libwebp-$LIBWEBP.tar.gz
 5d471308cef4c4752bbcf973d9cd37ba4cb53739116c30349d4764ba1410dfc1  libzip-$LIBZIP.tar.xz
-24b574f71c87a763f50704bbb630cbe38298d544a1f890f099a4696b1d6beba4  SDL2-$SDL2.tar.gz
+8b8d4aef2038533da814965220f88f77d60dfa0f32685f80ead65e501337da7f  SDL2-$SDL2.tar.gz
 8c29e06cf42aacc1eafc4077ae2ec6c6fcb96a626157e0593d5e82a34fd403c1  zstd-$ZSTD.tar.gz
 40973d44970dbc83ef302b0609f2e74982be2d85916dd2ee7472d30678a7abe6  ffmpeg-$FFMPEG.tar.xz
 f415a09385030c6510a936155ce211f617c31506db5fbc563e804345f1ecf56e  v$MOLTENVK.tar.gz

diff --git a/scripts/deps/build-dependencies-windows-arm64.bat b/scripts/deps/build-dependencies-windows-arm64.bat
@@ -51,7 +51,7 @@ set LIBJPEGTURBO=3.0.4
 set LIBPNG=1644
 set QT=6.8.1
 set QTMINOR=6.8
-set SDL2=2.30.9
+set SDL2=2.30.11
 set WEBP=1.4.0
 set LIBZIP=1.11.2
 set ZLIB=1.3.1
@@ -71,7 +71,7 @@ call :downloadfile "freetype-%FREETYPE%.tar.gz" "https://download.savannah.gnu.o
 call :downloadfile "harfbuzz-%HARFBUZZ%.zip" "https://github.com/harfbuzz/harfbuzz/archive/refs/tags/%HARFBUZZ%.zip" f93ff7ec6f2fcb9242256976a7e6d1da2588b5e57a559fb71a025b74bd1f5539 || goto error
 call :downloadfile "lpng%LIBPNG%.zip" "https://download.sourceforge.net/libpng/lpng%LIBPNG%.zip" 7d7571a1faa1898b69888716dfdea0e4d466f1a5cf518e6aa626df2242bbadbe || goto error
 call :downloadfile "libjpeg-turbo-%LIBJPEGTURBO%.tar.gz" "https://github.com/libjpeg-turbo/libjpeg-turbo/releases/download/%LIBJPEGTURBO%/libjpeg-turbo-%LIBJPEGTURBO%.tar.gz" 99130559e7d62e8d695f2c0eaeef912c5828d5b84a0537dcb24c9678c9d5b76b || goto error
-call :downloadfile "SDL2-%SDL2%.zip" "https://github.com/libsdl-org/SDL/releases/download/release-%SDL2%/SDL2-%SDL2%.zip" ec855bcd815b4b63d0c958c42c2923311c656227d6e0c1ae1e721406d346444b || goto error
+call :downloadfile "SDL2-%SDL2%.zip" "https://github.com/libsdl-org/SDL/releases/download/release-%SDL2%/SDL2-%SDL2%.zip" a0b3e7ac5f708042683ff0f22e069bdf75563540c615f9854ecc9bc8913e2488 || goto error
 call :downloadfile "qtbase-everywhere-src-%QT%.zip" "https://download.qt.io/official_releases/qt/%QTMINOR%/%QT%/submodules/qtbase-everywhere-src-%QT%.zip" e22d997bd15b795a176c8da62c8c1da0a674eb534e02f7c01ca507bf11bce0c3 || goto error
 call :downloadfile "qtimageformats-everywhere-src-%QT%.zip" "https://download.qt.io/official_releases/qt/%QTMINOR%/%QT%/submodules/qtimageformats-everywhere-src-%QT%.zip" 247a0a58039275a5a4fb499a600a90f66dc6e00321bb6f86a9b8d8020344d853 || goto error
 call :downloadfile "qtsvg-everywhere-src-%QT%.zip" "https://download.qt.io/official_releases/qt/%QTMINOR%/%QT%/submodules/qtsvg-everywhere-src-%QT%.zip" 57bd332e5550ff70a852560c591b786b6ba587c5e41cb5ef91038d82db137ab9 || goto error

diff --git a/scripts/deps/build-dependencies-windows-x64.bat b/scripts/deps/build-dependencies-windows-x64.bat
@@ -49,7 +49,7 @@ set LIBJPEGTURBO=3.0.4
 set LIBPNG=1644
 set QT=6.8.1
 set QTMINOR=6.8
-set SDL2=2.30.9
+set SDL2=2.30.11
 set WEBP=1.4.0
 set LIBZIP=1.11.2
 set ZLIB=1.3.1
@@ -69,7 +69,7 @@ call :downloadfile "freetype-%FREETYPE%.tar.gz" "https://download.savannah.gnu.o
 call :downloadfile "harfbuzz-%HARFBUZZ%.zip" "https://github.com/harfbuzz/harfbuzz/archive/refs/tags/%HARFBUZZ%.zip" f93ff7ec6f2fcb9242256976a7e6d1da2588b5e57a559fb71a025b74bd1f5539 || goto error
 call :downloadfile "lpng%LIBPNG%.zip" "https://download.sourceforge.net/libpng/lpng%LIBPNG%.zip" 7d7571a1faa1898b69888716dfdea0e4d466f1a5cf518e6aa626df2242bbadbe || goto error
 call :downloadfile "libjpeg-turbo-%LIBJPEGTURBO%.tar.gz" "https://github.com/libjpeg-turbo/libjpeg-turbo/releases/download/%LIBJPEGTURBO%/libjpeg-turbo-%LIBJPEGTURBO%.tar.gz" 99130559e7d62e8d695f2c0eaeef912c5828d5b84a0537dcb24c9678c9d5b76b || goto error
-call :downloadfile "SDL2-%SDL2%.zip" "https://github.com/libsdl-org/SDL/releases/download/release-%SDL2%/SDL2-%SDL2%.zip" ec855bcd815b4b63d0c958c42c2923311c656227d6e0c1ae1e721406d346444b || goto error
+call :downloadfile "SDL2-%SDL2%.zip" "https://github.com/libsdl-org/SDL/releases/download/release-%SDL2%/SDL2-%SDL2%.zip" a0b3e7ac5f708042683ff0f22e069bdf75563540c615f9854ecc9bc8913e2488 || goto error
 call :downloadfile "qtbase-everywhere-src-%QT%.zip" "https://download.qt.io/official_releases/qt/%QTMINOR%/%QT%/submodules/qtbase-everywhere-src-%QT%.zip" e22d997bd15b795a176c8da62c8c1da0a674eb534e02f7c01ca507bf11bce0c3 || goto error
 call :downloadfile "qtimageformats-everywhere-src-%QT%.zip" "https://download.qt.io/official_releases/qt/%QTMINOR%/%QT%/submodules/qtimageformats-everywhere-src-%QT%.zip" 247a0a58039275a5a4fb499a600a90f66dc6e00321bb6f86a9b8d8020344d853 || goto error
 call :downloadfile "qtsvg-everywhere-src-%QT%.zip" "https://download.qt.io/official_releases/qt/%QTMINOR%/%QT%/submodules/qtsvg-everywhere-src-%QT%.zip" 57bd332e5550ff70a852560c591b786b6ba587c5e41cb5ef91038d82db137ab9 || goto error

diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
@@ -60,6 +60,8 @@ add_library(common
   thirdparty/SmallVector.h
   thirdparty/aes.cpp
   thirdparty/aes.h
+  task_queue.cpp
+  task_queue.h
   threading.cpp
   threading.h
   timer.cpp

diff --git a/src/common/common.vcxproj b/src/common/common.vcxproj
@@ -45,6 +45,7 @@
     <ClInclude Include="thirdparty\SmallVector.h" />
     <ClInclude Include="thirdparty\StackWalker.h" />
     <ClInclude Include="threading.h" />
+    <ClInclude Include="task_queue.h" />
     <ClInclude Include="timer.h" />
     <ClInclude Include="types.h" />
     <ClInclude Include="minizip_helpers.h" />
@@ -74,6 +75,7 @@
     <ClCompile Include="thirdparty\SmallVector.cpp" />
     <ClCompile Include="thirdparty\StackWalker.cpp" />
     <ClCompile Include="threading.cpp" />
+    <ClCompile Include="task_queue.cpp" />
     <ClCompile Include="timer.cpp" />
   </ItemGroup>
   <ItemGroup>

diff --git a/src/common/common.vcxproj.filters b/src/common/common.vcxproj.filters
@@ -52,6 +52,7 @@
     <ClInclude Include="log_channels.h" />
     <ClInclude Include="sha256_digest.h" />
     <ClInclude Include="thirdparty\aes.h" />
+    <ClInclude Include="task_queue.h" />
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="small_string.cpp" />
@@ -82,6 +83,7 @@
     <ClCompile Include="gsvector.cpp" />
     <ClCompile Include="sha256_digest.cpp" />
     <ClCompile Include="thirdparty\aes.cpp" />
+    <ClCompile Include="task_queue.cpp" />
   </ItemGroup>
   <ItemGroup>
     <Natvis Include="bitfield.natvis" />

diff --git a/src/common/gsvector.cpp b/src/common/gsvector.cpp
@@ -66,53 +66,11 @@ void GSMatrix2x2::store(void* m)
   std::memcpy(m, E, sizeof(E));
 }
 
-GSMatrix4x4::GSMatrix4x4(float e00, float e01, float e02, float e03, float e10, float e11, float e12, float e13,
-                         float e20, float e21, float e22, float e23, float e30, float e31, float e32, float e33)
-{
-  E[0][0] = e00;
-  E[0][1] = e01;
-  E[0][2] = e02;
-  E[0][3] = e03;
-  E[1][0] = e10;
-  E[1][1] = e11;
-  E[1][2] = e12;
-  E[1][3] = e13;
-  E[2][0] = e20;
-  E[2][1] = e21;
-  E[2][2] = e22;
-  E[2][3] = e23;
-  E[3][0] = e30;
-  E[3][1] = e31;
-  E[3][2] = e32;
-  E[3][3] = e33;
-}
-
-GSMatrix4x4::GSMatrix4x4(const GSMatrix2x2& m)
-{
-  E[0][0] = m.E[0][0];
-  E[0][1] = m.E[0][1];
-  E[0][2] = 0.0f;
-  E[0][3] = 0.0f;
-  E[1][0] = m.E[1][0];
-  E[1][1] = m.E[1][1];
-  E[1][2] = 0.0f;
-  E[1][3] = 0.0f;
-  E[2][0] = 0.0f;
-  E[2][1] = 0.0f;
-  E[2][2] = 1.0f;
-  E[2][3] = 0.0f;
-  E[3][0] = 0.0f;
-  E[3][1] = 0.0f;
-  E[3][2] = 0.0f;
-  E[3][3] = 1.0f;
-}
-
-GSMatrix4x4 GSMatrix4x4::operator*(const GSMatrix4x4& m) const
+ALWAYS_INLINE static void MatrixMult4x4(GSMatrix4x4& res, const GSMatrix4x4& lhs, const GSMatrix4x4& rhs)
 {
   // This isn't speedy by any means, but it's not hot code either.
-  GSMatrix4x4 res;
-
-#define MultRC(rw, cl) E[rw][0] * m.E[0][cl] + E[rw][1] * m.E[1][cl] + E[rw][2] * m.E[2][cl] + E[rw][3] * m.E[3][cl]
+#define MultRC(rw, cl)                                                                                                 \
+  lhs.E[rw][0] * rhs.E[0][cl] + lhs.E[rw][1] * rhs.E[1][cl] + lhs.E[rw][2] * rhs.E[2][cl] + lhs.E[rw][3] * rhs.E[3][cl]
 
   res.E[0][0] = MultRC(0, 0);
   res.E[0][1] = MultRC(0, 1);
@@ -132,75 +90,22 @@ GSMatrix4x4 GSMatrix4x4::operator*(const GSMatrix4x4& m) const
   res.E[3][3] = MultRC(3, 3);
 
 #undef MultRC
+}
 
+GSMatrix4x4 GSMatrix4x4::operator*(const GSMatrix4x4& m) const
+{
+  GSMatrix4x4 res;
+  MatrixMult4x4(res, *this, m);
   return res;
 }
 
 GSMatrix4x4& GSMatrix4x4::operator*=(const GSMatrix4x4& m)
 {
   const GSMatrix4x4 copy(*this);
-
-#define MultRC(rw, cl)                                                                                                 \
-  copy.E[rw][0] * m.E[0][cl] + copy.E[rw][1] * m.E[1][cl] + copy.E[rw][2] * m.E[2][cl] + copy.E[rw][3] * m.E[3][cl]
-
-  E[0][0] = MultRC(0, 0);
-  E[0][1] = MultRC(0, 1);
-  E[0][2] = MultRC(0, 2);
-  E[0][3] = MultRC(0, 3);
-  E[1][0] = MultRC(1, 0);
-  E[1][1] = MultRC(1, 1);
-  E[1][2] = MultRC(1, 2);
-  E[1][3] = MultRC(1, 3);
-  E[2][0] = MultRC(2, 0);
-  E[2][1] = MultRC(2, 1);
-  E[2][2] = MultRC(2, 2);
-  E[2][3] = MultRC(2, 3);
-  E[3][0] = MultRC(3, 0);
-  E[3][1] = MultRC(3, 1);
-  E[3][2] = MultRC(3, 2);
-  E[3][3] = MultRC(3, 3);
-
-#undef MultRC
-
+  MatrixMult4x4(*this, copy, m);
   return *this;
 }
 
-GSVector4 GSMatrix4x4::operator*(const GSVector4& v) const
-{
-  const GSVector4 r0 = row(0);
-  const GSVector4 r1 = row(1);
-  const GSVector4 r2 = row(2);
-  const GSVector4 r3 = row(3);
-
-  return GSVector4(r0.dot(v), r1.dot(v), r2.dot(v), r3.dot(v));
-}
-
-GSMatrix4x4 GSMatrix4x4::Identity()
-{
-  GSMatrix4x4 res;
-
-#define MultRC(rw, cl) E[rw][0] * m.E[0][cl] + E[rw][1] * m.E[1][cl] + E[rw][2] * m.E[2][cl] + E[rw][3] * m.E[3][cl]
-
-  res.E[0][0] = 1.0f;
-  res.E[0][1] = 0.0f;
-  res.E[0][2] = 0.0f;
-  res.E[0][3] = 0.0f;
-  res.E[1][0] = 0.0f;
-  res.E[1][1] = 1.0f;
-  res.E[1][2] = 0.0f;
-  res.E[1][3] = 0.0f;
-  res.E[2][0] = 0.0f;
-  res.E[2][1] = 0.0f;
-  res.E[2][2] = 1.0f;
-  res.E[2][3] = 0.0f;
-  res.E[3][0] = 0.0f;
-  res.E[3][1] = 0.0f;
-  res.E[3][2] = 0.0f;
-  res.E[3][3] = 1.0f;
-
-  return res;
-}
-
 GSMatrix4x4 GSMatrix4x4::RotationX(float angle_in_radians)
 {
   const float sin_angle = std::sin(angle_in_radians);
@@ -246,17 +151,7 @@ GSMatrix4x4 GSMatrix4x4::OffCenterOrthographicProjection(float width, float heig
   return OffCenterOrthographicProjection(0.0f, 0.0f, width, height, zNear, zFar);
 }
 
-GSVector4 GSMatrix4x4::row(size_t i) const
-{
-  return GSVector4::load<true>(&E[i][0]);
-}
-
-GSVector4 GSMatrix4x4::col(size_t i) const
-{
-  return GSVector4(E[0][i], E[1][i], E[2][i], E[3][i]);
-}
-
-GSMatrix4x4 GSMatrix4x4::Invert() const
+GSMatrix4x4 GSMatrix4x4::invert() const
 {
   GSMatrix4x4 ret;
 

diff --git a/src/common/gsvector.h b/src/common/gsvector.h
@@ -41,17 +41,60 @@ class GSMatrix2x2
 class alignas(VECTOR_ALIGNMENT) GSMatrix4x4
 {
 public:
-  GSMatrix4x4() = default;
-  GSMatrix4x4(float e00, float e01, float e02, float e03, float e10, float e11, float e12, float e13, float e20,
-              float e21, float e22, float e23, float e30, float e31, float e32, float e33);
-  GSMatrix4x4(const GSMatrix2x2& m);
+  constexpr GSMatrix4x4() = default;
+  constexpr GSMatrix4x4(float e00, float e01, float e02, float e03, float e10, float e11, float e12, float e13,
+                        float e20, float e21, float e22, float e23, float e30, float e31, float e32, float e33)
+  {
+    E[0][0] = e00;
+    E[0][1] = e01;
+    E[0][2] = e02;
+    E[0][3] = e03;
+    E[1][0] = e10;
+    E[1][1] = e11;
+    E[1][2] = e12;
+    E[1][3] = e13;
+    E[2][0] = e20;
+    E[2][1] = e21;
+    E[2][2] = e22;
+    E[2][3] = e23;
+    E[3][0] = e30;
+    E[3][1] = e31;
+    E[3][2] = e32;
+    E[3][3] = e33;
+  }
+
+  constexpr GSMatrix4x4(const GSMatrix2x2& m)
+  {
+    E[0][0] = m.E[0][0];
+    E[0][1] = m.E[0][1];
+    E[0][2] = 0.0f;
+    E[0][3] = 0.0f;
+    E[1][0] = m.E[1][0];
+    E[1][1] = m.E[1][1];
+    E[1][2] = 0.0f;
+    E[1][3] = 0.0f;
+    E[2][0] = 0.0f;
+    E[2][1] = 0.0f;
+    E[2][2] = 1.0f;
+    E[2][3] = 0.0f;
+    E[3][0] = 0.0f;
+    E[3][1] = 0.0f;
+    E[3][2] = 0.0f;
+    E[3][3] = 1.0f;
+  }
 
   GSMatrix4x4 operator*(const GSMatrix4x4& m) const;
   GSMatrix4x4& operator*=(const GSMatrix4x4& m);
 
-  GSVector4 operator*(const GSVector4& v) const;
+  GSVector4 operator*(const GSVector4& v) const
+  {
+    return GSVector4(row(0).dot(v), row(1).dot(v), row(2).dot(v), row(3).dot(v));
+  }
 
-  static GSMatrix4x4 Identity();
+  static constexpr GSMatrix4x4 Identity()
+  {
+    return GSMatrix4x4(1.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f, 0.0f, 0.0f, 0.0f, 0.0f, 1.0f);
+  }
 
   static GSMatrix4x4 RotationX(float angle_in_radians);
   static GSMatrix4x4 RotationY(float angle_in_radians);
@@ -62,10 +105,19 @@ class alignas(VECTOR_ALIGNMENT) GSMatrix4x4
                                                      float zFar);
   static GSMatrix4x4 OffCenterOrthographicProjection(float width, float height, float zNear, float zFar);
 
-  GSVector4 row(size_t i) const;
-  GSVector4 col(size_t i) const;
+  GSVector4 row(size_t i) const { return GSVector4::load<true>(&E[i][0]); }
+  GSVector4 col(size_t i) const { return GSVector4(E[0][i], E[1][i], E[2][i], E[3][i]); }
 
-  GSMatrix4x4 Invert() const;
+  void set_row(size_t i, GSVector4 row) { GSVector4::store<true>(&E[i][0], row); }
+  void set_col(size_t i, GSVector4 col)
+  {
+    E[0][i] = col.x;
+    E[1][i] = col.y;
+    E[2][i] = col.z;
+    E[3][i] = col.w;
+  }
+
+  GSMatrix4x4 invert() const;
 
   void store(void* m);