PPSSPP Git (2026/10/09) is compiled. PPSSPP is a fast and portable PSP emulator for Android, Windows, Mac, and Linux, written in C++.

PPSSPP Git Changelog:
* Merge pull request #22467 from hrydgard/softgpu-sampling
* SoftGPU: Use larger tiles now that pixels are cheaper
* SoftGPU: Let the quad sampler skip the level check when all is level 0
* SoftGPU: Lerp two channels per lane in the four-pixel bilinear sampler
* SoftGPU: Skip the level of detail when nothing depends on it
* SoftGPU: Filter the four-pixel bilinear samples a channel per vector
* Merge pull request #22468 from mohammadsyuhada/fix-strip-psp-case
* DirectoryFileSystem: Skip the stripped PSP/ prefix when fixing path case
* Merge pull request #22464 from hrydgard/softgpu-spans
* CrossSIMD: Keep the C++ fallback's integer ops defined, like the SIMD ones
* SoftGPU: Find each edge's row bounds once
* SoftGPU: Pick the span function by blend kind, and decide the vector path once
* SoftGPU: Early depth test of staged spans in vector lanes
* SoftGPU: Leave out a zero secondary color, and fill constant span buffers once
* SoftGPU: Draw rectangles in stages like triangle rows
* SoftGPU: Walk the staged rows' planes four pixels at a time
* SoftGPU: Compute the staged rows' coverage in vector lanes
* SoftGPU: Pass a span's colors a channel at a time
* SoftGPU: Draw clear mode, stencil and color test spans in vector lanes
* SoftGPU: Specialize the staged rows' texturing for the triangle
* SoftGPU: Check a triangle's UV planes once instead of every span
* SoftGPU: Draw triangle rows in stages
* SoftGPU: Apply the texture function to a span's four pixels in lanes
* SoftGPU: Draw the pixels of a span together, in vector lanes
* SoftGPU: Step a triangle's planes along the row
* SoftGPU: Walk triangle rows between their exact ends
* SoftGPU: Compute a span's exact texture coordinates in vector lanes
* SoftGPU: Sample a span's four pixels in vector lanes
* SoftGPU: Draw triangles and sprites in spans of four pixels in a row
* SoftGPU: Fewer address and plane computations per textured pixel
* SoftGPU: Inline GERecip, and pick the mip level's q once per quad row
* SoftGPU: Specialize the bilinear sampler per texture format without a JIT
* SoftGPU: Scale UV plane values by a power of two instead of ldexp
* Merge pull request #22465 from hrydgard/hw-ge-blends
* Headless: Add --dump-compat, to apply compat settings to frame dumps
* Read the framebuffer for DOUBLESRCALPHA + DOUBLESRCALPHA blends in through mode
* Read the framebuffer for DSTCOLOR + DOUBLESRCALPHA blends in through mode
* Truncate the color in the GE-exact FIXA/FIXB blend, like the GE's filter
* Compute FIXA/FIXB blends like the GE
* Merge pull request #22463 from hrydgard/hw-rect-culling
* Drop 3D rectangles with a vertex out of range, like the software renderer
* Merge pull request #22462 from hrydgard/softgpu-transform
* SoftGPU: Add the diffuse light without branching on its factor
* SoftGPU: Skin a vertex without per-operation fallback checks
* SoftGPU: Keep the vertex position, normal and light vectors as Vec4F32
* GEMath: Add GEDot3 and GENormalize4 on Vec4F32
* SoftGPU: Skin the x, y and z components together in vector lanes
* GEMath: Build GERecip and GERsqrt results from exponent bits
* GEMath: GERowSum4 takes its left operands as one vector
* SoftGPU: Use GERowSum4 for the transform's row sums
* GEMath: Add GERowSum4, four row sums at once
* Merge pull request #22461 from hrydgard/softgpu-accuracy
* GPU: Copy overlapping block transfer rows as the GE does
* GPU: Count the cycles of commands the hardware backends consume by hand
* SoftGPU: Carry UVs and normals into formats without them as the GE does
* Headless: Don't apply compat flags to frame dumps
* Merge pull request #22460 from hrydgard/android-config-changes
* Android: Re-report the surface when the display density changes
* Android: Keep the native smallestScreenWidthDp current across config changes
* Android: Trim SizeManager down to the surface callback, rename to NativeSurfaceManager
* Android: Handle rotation and window resizing without recreating the activity
* Merge pull request #22458 from hrydgard/android-small-fixes
* Android: Remove LifeCycle.java, keep the logging
* Android: Don't log every mouse movement
* Android: Mark the last two JNI-called Java methods @Keep
* Android: Forget a key when it's released
* Android: Make renderer_inited atomic
* Android: Create the TextRenderer paint only once
* Android: Hand out the same font ID for the same font file
* Android: Survive long enough to show the missing native library message
* Android: Release the camera if starting it fails
* Don't save log levels set from the command line
* Android: Parse a launch parameter sent to the running app as a command line
* Android: Declare the main activity resizeable explicitly
* Android: Stop location updates while paused
* Android: Release all pointers on ACTION_CANCEL, not just the first
* Merge pull request #22456 from hrydgard/spline-simd
* Splines: Vectorize the GE-exact spline evaluation
* Splines: Vectorize the GE-exact Bezier evaluation
* CrossSIMD: Add an arithmetic Vec4S32::Shr and Vec4S32FromBits
* Splines: Skip the second vertex decode of tessellated curves
* Splines: Use MulAdd, and round colors with one instruction
* CrossSIMD: Add MulAdd and Vec4S32FromF32Round
* Splines: Tessellate four vertices at a time with CrossSIMD
* CrossSIMD: Add RecipSqrt, Select and Vec4F32FromBits
* Merge pull request #22454 from hrydgard/android-activity-lifecycle
* Docs: How to test an Android build on a real device
* Android: A replaced activity must not start or stop the render thread
* Android: Don't let a late onDestroy shut down a newer activity's native state
* Android: Don't create the static location and IR helpers from the activity
* Android: Unregister the power saving settings observer
* Android: Share one TextRenderer class ref between TextDrawers
* Android: Don't leak the activity on recreate, and hand it over safely
* Android: Stop leaking JNI local references on native threads
* Merge pull request #22455 from hrydgard/fix-createdir-chmod
* Don't chmod existing parent directories in CreateFullPath
* Merge pull request #22435 from hrydgard/softgpu-depth-swizzle
* SoftGPU: Keep a not-black color test when dithering, fog or a logic op is on
* GE dump playback: Let the GE finish the list before reusing its buffer
* GE dump playback: Restore old-layout contexts from old dumps
* GE dump playback: Give unaligned CLUTs and textures an aligned copy
* GE dump playback: Ignore a second INIT in a dump
* GE dump playback: Load the initial CLUT, and show the last framebuffer when the display is off
* frametests: Update IL-2's software reference
* GE dump playback: Write recorded memory into the depth mirrors as the PSP does
* Core: Let 16-bit CPU accesses through the depth mirrors see the GE's layout
* SoftGPU: Store depth in VRAM as the GE does
* GPU: Add the GE's depth buffer layout in VRAM
* Merge pull request #22453 from hrydgard/vulkan-suboptimal-transform
* Vulkan: Don't recreate the swapchain over a pre-transform we chose ourselves
* Merge pull request #22452 from hrydgard/hw-render-fixes
* iOS: Link StoreKit in IAP builds
* iOS toolchain: Pass IOS_APP_STORE to CMake's compiler checks
* iOS App Store: Update the static MoltenVK to 1.4.2, raise the minimum to iOS 15
* Mac: Update the bundled MoltenVK to 1.4.2, raise the macOS minimum to 12.0
* Self-texture copy: Include the texel past each edge of the UV bounds
* Pixel depth rounding: Floor with room for float error, store Z like clears
* Software transform: Check rectangle depth against the range like the GE
* SoftGPU: Move the GE clip and viewport helpers to GEMath.h
* Clip and cull in software when the GPU lacks cull distances
* Software transform clipping: split clipped quads like the GE, cull beyond far first
* Don't write fragment depth when the depth test is effectively off
* GLES: Enable the clip planes the vertex shader actually writes
* Merge pull request #22451 from hrydgard/android-egl-replace-javagl
* Android: Converge the GL and Vulkan lifecycles on a shared RunGraphicsLoop
* Android GL: Delete the JavaGL path, leaving only native EGL
* Android GL: Add a native EGL path, behind a dev toggle
* Merge pull request #22330 from hrydgard/android-render-thread-lifecycle
* Android: Don't send the launch parameter to the native side twice
* Android GL: Don't run stale GL commands when joining the emu thread
* Android: Join the Vulkan render thread when the surface is destroyed
* Android: Recover from a Vulkan render thread that exited by itself
* Android: Always join the Vulkan render thread on an exit request
* Merge pull request #22449 from hrydgard/stencil-write-mask
* Apply the pixel masks to clears in the hardware backends
* Add gpu/stencil/writemask to the test list
* Always emulate partial color write masks in the shader, drop ShaderColorBitmask
* Merge pull request #22448 from hrydgard/gedump-id-from-name
* Read a frame dump's game ID after a leading issue number in its name

Download: PPSSPP Git (2026/10/09) x64
Download: PPSSPP Git (2026/10/09) Android
Source: Here