Merge 4199c99cb6 into 10798c3464

* fix use-after-free of inContent
* don't try to free the DStream twice

CMakeLists.txt

@@ -8,6 +8,7 @@ endif()
 set(CMAKE_POLICY_DEFAULT_CMP0069 NEW)
 
 set(CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})
+set(CMAKE_USER_MAKE_RULES_OVERRIDE "${CMAKE_SOURCE_DIR}/cmake/DefaultBuildFlags.cmake")
 
 option(USE_VCPKG "Use vcpkg for dependency packages" OFF)
 if (USE_VCPKG)
@@ -78,14 +79,6 @@ if (ENABLE_LTO)
     set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
 endif()
 
-if (CMAKE_CXX_COMPILER_ID STREQUAL GNU)
-    set(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} -Og")
-    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -Og")
-endif()
-
-string(REPLACE "-O2" "-O3" CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE}")
-string(REPLACE "-O2" "-O3" CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE}")
-
 if (NOT APPLE)
     set(CMAKE_EXE_LINKER_FLAGS_RELEASE "${CMAKE_EXE_LINKER_FLAGS_RELEASE} -s")
 endif()

README.md

@@ -6,10 +6,9 @@
 <a href="https://www.gnu.org/licenses/gpl-3.0" alt="License: GPLv3"><img src="https://img.shields.io/badge/License-GPL%20v3-%23ff554d.svg"></a>
 <a href="https://kiwiirc.com/client/irc.badnik.net/?nick=IRC-Source_?#melonds" alt="IRC channel: #melonds"><img src="https://img.shields.io/badge/IRC%20chat-%23melonds-%23dd2e44.svg"></a>
 <br>
-<a href="https://github.com/melonDS-emu/melonDS/actions?query=workflow%3A%22CMake+Build+%28Windows+x86-64%29%22+event%3Apush"><img src="https://img.shields.io/github/actions/workflow/status/melonDS-emu/melonDS/build-windows.yml?label=Windows%20x86-64&logo=GitHub&branch=master"></img></a>
-<a href="https://github.com/melonDS-emu/melonDS/actions?query=workflow%3A%22CMake+Build+%28Ubuntu+x86-64%29%22+event%3Apush"><img src="https://img.shields.io/github/actions/workflow/status/melonDS-emu/melonDS/build-ubuntu.yml?label=Linux%20x86-64&logo=GitHub"></img></a>
-<a href="https://github.com/melonDS-emu/melonDS/actions?query=workflow%3A%22CMake+Build+%28Ubuntu+aarch64%29%22+event%3Apush"><img src="https://img.shields.io/github/actions/workflow/status/melonDS-emu/melonDS/build-ubuntu-aarch64.yml?label=Linux%20ARM64&logo=GitHub"></img></a>
-<a href="https://github.com/melonDS-emu/melonDS/actions/workflows/build-macos-universal.yml?query=event%3Apush"><img src="https://img.shields.io/github/actions/workflow/status/melonDS-emu/melonDS/build-macos.yml?label=macOS%20Universal&logo=GitHub"></img></a>
+<a href="https://github.com/melonDS-emu/melonDS/actions/workflows/build-windows.yml?query=event%3Apush"><img src="https://github.com/melonDS-emu/melonDS/actions/workflows/build-windows.yml/badge.svg" /></a>
+<a href="https://github.com/melonDS-emu/melonDS/actions/workflows/build-ubuntu.yml?query=event%3Apush"><img src="https://github.com/melonDS-emu/melonDS/actions/workflows/build-ubuntu.yml/badge.svg" /></a>
+<a href="https://github.com/melonDS-emu/melonDS/actions/workflows/build-macos.yml?query=event%3Apush"><img src="https://github.com/melonDS-emu/melonDS/actions/workflows/build-macos.yml/badge.svg" /></a>
 </p>
 DS emulator, sorta
 

cmake/DefaultBuildFlags.cmake

@@ -0,0 +1,9 @@
+if (CMAKE_C_COMPILER_ID STREQUAL GNU)
+	set(CMAKE_C_FLAGS_DEBUG_INIT "-g -Og")
+endif()
+if (CMAKE_CXX_COMPILER_ID STREQUAL GNU)
+	set(CMAKE_CXX_FLAGS_DEBUG_INIT "-g -Og")
+endif()
+
+string(REPLACE "-O2" "-O3" CMAKE_C_FLAGS_RELEASE_INIT "${CMAKE_C_FLAGS_RELEASE_INIT}")
+string(REPLACE "-O2" "-O3" CMAKE_CXX_FLAGS_RELEASE_INIT "${CMAKE_CXX_FLAGS_RELEASE_INIT}")

src/CP15.cpp

@@ -186,10 +186,14 @@ void ARMv5::UpdatePURegion(u32 n)
         return;
     }
 
-    u32 start = rgn >> 12;
-    u32 sz = 2 << ((rgn >> 1) & 0x1F);
-    u32 end = start + (sz >> 12);
-    // TODO: check alignment of start
+    // notes:
+    // * min size of a pu region is 4KiB (12 bits)
+    // * size is calculated as size + 1, but the 12 lsb of address space are ignored, therefore we need it as size + 1 - 12, or size - 11
+    // * pu regions are aligned based on their size
+    u32 size = std::max((int)((rgn>>1) & 0x1F) - 11, 0); // obtain the size, subtract 11 and clamp to a min of 0.
+    u32 start = ((rgn >> 12) >> size) << size; // determine the start offset, and use shifts to force alignment with a multiple of the size.
+    u32 end = start + (1<<size); // add 1 left shifted by size to start to determine end point
+    // dont need to bounds check the end point because the force alignment inherently prevents it from breaking
 
     u8 usermask = 0;
     u8 privmask = 0;
@@ -239,7 +243,7 @@ void ARMv5::UpdatePURegion(u32 n)
         "PU region %d: %08X-%08X, user=%02X priv=%02X, %08X/%08X\n",
         n,
         start << 12,
-        end << 12,
+        (end << 12) - 1,
         usermask,
         privmask,
         PU_DataRW,
@@ -579,12 +583,12 @@ void ARMv5::CP15Write(u32 id, u32 val)
 
         std::snprintf(log_output,
                  sizeof(log_output),
-                 "PU: region %d = %08X : %s, %08X-%08X\n",
+                 "PU: region %d = %08X : %s, start: %08X size: %02X\n",
                  (id >> 4) & 0xF,
                  val,
                  val & 1 ? "enabled" : "disabled",
                  val & 0xFFFFF000,
-                 (val & 0xFFFFF000) + (2 << ((val & 0x3E) >> 1))
+                 (val & 0x3E) >> 1
         );
         Log(LogLevel::Debug, "%s", log_output);
         // Some implementations of Log imply a newline, so we build up the line before printing it

src/GPU3D.cpp

@@ -299,6 +299,7 @@ void GPU3D::Reset() noexcept
     NumPolygons = 0;
     CurRAMBank = 0;
 
+    ShadowSent = true;
     FlushRequest = 0;
     FlushAttributes = 0;
 
@@ -358,6 +359,7 @@ void GPU3D::DoSavestate(Savestate* file) noexcept
     file->Var32(&RenderClearAttr2);
     
     file->Bool32(&RenderRasterRev);
+    file->Bool32(&ShadowSent);
 
     file->Var16(&RenderXPos);
 
@@ -463,6 +465,7 @@ void GPU3D::DoSavestate(Savestate* file) noexcept
         file->Bool32(&poly->FacingView);
         file->Bool32(&poly->Translucent);
 
+        file->Bool32(&poly->ClearStencil);
         file->Bool32(&poly->IsShadowMask);
         file->Bool32(&poly->IsShadow);
 
@@ -1225,6 +1228,18 @@ void GPU3D::SubmitPolygon() noexcept
 
     poly->IsShadowMask = ((CurPolygonAttr & 0x3F000030) == 0x00000030);
     poly->IsShadow = ((CurPolygonAttr & 0x30) == 0x30) && !poly->IsShadowMask;
+            
+    // yes we need specifically the rasterizer bit, not the gx bit
+    poly->ClearStencil = false;
+    if (NDS.GetSCFGRasterBit() && (FlushAttributes & 1) && poly->Translucent)
+    {
+        if (poly->IsShadowMask && ShadowSent)
+        {
+            ShadowSent = false;
+            poly->ClearStencil = true;
+        }
+    }
+    if (poly->IsShadow) ShadowSent = true;
 
     if (!poly->Translucent) NumOpaquePolygons++;
 
@@ -2056,6 +2071,7 @@ void GPU3D::ExecuteCommand() noexcept
 
         case 0x50: // flush
             VertexPipelineCmdDelayed4();
+            ShadowSent = true;
             FlushRequest = 1;
             FlushAttributes = entry.Param & 0x3;
             CycleCount = 325;

src/GPU3D.h

@@ -68,6 +68,7 @@ struct Polygon
     bool FacingView;
     bool Translucent;
 
+    bool ClearStencil; // used by the dsi's revised rasterizer mode
     bool IsShadowMask;
     bool IsShadow;
 
@@ -328,6 +329,8 @@ public:
     // used to fix stencil buffer's frame-to-frame persistence not working properly under extreme misuse of shadow masks/shadows
     bool ForceRerender = false;
     bool DontRerenderLoop = false;
+    
+    bool ShadowSent = false; // used to track whether the clear stencil flag should be set when the raster scfg bit is active
 
     u32 FlushRequest = 0;
     u32 FlushAttributes = 0;

src/GPU3D_Soft.cpp

@@ -125,6 +125,8 @@ void SoftRenderer::Reset(GPU& gpu)
 
     ShadowRendered[0] = false;
     ShadowRendered[1] = false;
+    ShadowRenderedi[0] = false;
+    ShadowRenderedi[1] = false;
     SetupRenderThread(gpu);
     EnableRenderThread();
 }
@@ -762,10 +764,9 @@ void SoftRenderer::RenderShadowMaskScanline(const GPU3D& gpu3d, RendererPolygon*
     bool (*fnDepthTest)(s32 dstz, s32 z, u32 attr, u32 dstattr, u8 flags);
 
     // stencil buffer is only cleared when beginning a shadow mask after a shadow polygon is rendered
-    // the "Revised" Rasterizer Circuit bugs out stencil buffer clearing
-    // TODO: under some circumstances that i am struggling to reproduce it is possible for this bit to actually *fix* bugs with misused shadow masks
-    // while this is in effect, toggling the scfg bit appears to glitch the stencil buffer for a frame?
-    if (ShadowRendered[y&0x1] && !gpu3d.RenderRasterRev)
+    // the state of whether a polygon was or wasn't rendered can persist between frames
+    // the Revised Rasterizer Circuit handles clearing the stencil buffer elsewhere
+    if (ShadowRendered[y&0x1] && (!gpu3d.RenderRasterRev || polygon->ClearStencil))
     {
         StencilCleared = true;
         memset(&StencilBuffer[256 * (y&0x1)], 0, 256);
@@ -961,7 +962,7 @@ void SoftRenderer::RenderPolygonScanline(GPU& gpu, RendererPolygon* rp, s32 y)
 
     if (polygon->IsShadow)
     {
-        ShadowRendered[y&0x1] = true;
+        if (!gpu.GPU3D.RenderRasterRev || !polygon->Translucent) ShadowRendered[y&0x1] = true;
         if (wireframe) return; // TODO: this probably still counts towards timings.
         if (!StencilCleared)
         {
@@ -1593,6 +1594,22 @@ void SoftRenderer::RenderScanline(GPU& gpu, s32 y, int npolys)
     {
         RendererPolygon* rp = &PolygonList[i];
         Polygon* polygon = rp->PolyData;
+        
+        //we actually handle clearing the stencil buffer here when the revision bit is set, this allows for a polygon to clear it on every scanline, even ones it isn't part of.
+        if (gpu.GPU3D.RenderRasterRev)
+        {
+            if (polygon->ClearStencil && polygon->Translucent && ShadowRenderedi[(y&0x1)])
+            {
+                StencilCleared = true;
+                memset(&StencilBuffer[256 * (y&0x1)], 0, 256);
+                ShadowRenderedi[(y&0x1)] = false;
+
+            }
+            else if (polygon->IsShadow && polygon->Translucent)
+            {
+                ShadowRenderedi[(y&0x1)] = true;
+            }
+        }
 
         if (y >= polygon->YTop && (y < polygon->YBottom || (y == polygon->YTop && polygon->YBottom == polygon->YTop)))
         {

src/GPU3D_Soft.h

@@ -510,6 +510,7 @@ private:
 
     u8 StencilBuffer[256*2];
     bool ShadowRendered[2];
+    bool ShadowRenderedi[2];
     bool StencilCleared;
 
     bool Enabled;

src/frontend/qt_sdl/ROMManager.cpp

@@ -948,8 +948,8 @@ u32 DecompressROM(const u8* inContent, const u32 inSize, unique_ptr<u8[]>& outCo
 
     if (realSize != ZSTD_CONTENTSIZE_UNKNOWN)
     {
-        outContent = make_unique<u8[]>(realSize);
-        u64 decompressed = ZSTD_decompress(outContent.get(), realSize, inContent, inSize);
+        auto newOutContent = make_unique<u8[]>(realSize);
+        u64 decompressed = ZSTD_decompress(newOutContent.get(), realSize, inContent, inSize);
 
         if (ZSTD_isError(decompressed))
         {
@@ -957,6 +957,7 @@ u32 DecompressROM(const u8* inContent, const u32 inSize, unique_ptr<u8[]>& outCo
             return 0;
         }
 
+        outContent = std::move(newOutContent);
         return realSize;
     }
     else
@@ -1011,7 +1012,6 @@ u32 DecompressROM(const u8* inContent, const u32 inSize, unique_ptr<u8[]>& outCo
             }
         } while (inBuf.pos < inBuf.size);
 
-        ZSTD_freeDStream(dStream);
         outContent = make_unique<u8[]>(outBuf.pos);
         memcpy(outContent.get(), outBuf.dst, outBuf.pos);