swung0x48/MobileGL
1
0
Fork 0
mirror of https://github.com/MobileGL-Dev/MobileGL synced 2026-09-12 14:18:31 +09:00
Code Issues Packages Projects Releases Wiki Activity
1,669 Commits 28 Branches 0 Tags
d4922cb0fb88c6dadc44ce6417fac47d6474e656
Commit Graph
100 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
swung0x48 d4922cb0fb [Fix] (MG_Backend, MG_Impl/GLImpl, MG_Util): make anisotropic filtering actually reachable - advertise GL_EXT/ARB_texture_filter_anisotropic only where the host driver or the samplerAnisotropy device feature supports it, answer GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT from the backend limit, and honor the sampler state on DirectVulkan (feature enable, limit clamp, LINEAR-only gate, resolved value in the sampler cache key) 2026-07-16 22:59:11 -04:00
swung0x48 e2f873c95c [Fix] (MG_Util/ShaderTranspiler): retry a legacy shader at 460 when it fails to parse as normalized 330 core, so sources using 420-era syntax without the matching #extension line keep compiling as they did on real drivers 2026-07-16 22:33:30 -04:00
swung0x48 efd7b47388 [Chore] (MG_Test): assert exact GL error counts - name-lifecycle regression tests per object family, plus fixtures that drain on setup and fail the test that leaks an unconsumed error 2026-07-16 22:12:12 -04:00
swung0x48 a08669df72 [Fix] (MG_Impl/GLImpl): stop recording GL errors on the delete/query paths of every object family - glDeleteBuffers/VertexArrays/Renderbuffers/Framebuffers must silently ignore unknown names and glIsTexture must never raise, while glBindSampler now reports INVALID_OPERATION like the other bind entry points 2026-07-16 22:11:43 -04:00
swung0x48 b8db509581 [Fix] (MG_Util/ShaderTranspiler): normalize legacy desktop shaders to GLSL 330 2026-07-16 21:39:39 -04:00
swung0x48 5d6b544021 [Fix] (MG_Impl/Texture): support anisotropic sampler parameters 2026-07-16 21:39:38 -04:00
swung0x48 346cd417ca [Fix] (MG_Backend/DirectVulkan): fix ERROR-level vertex stream build 2026-07-16 21:36:22 -04:00
swung0x48 f61675e9ce [Fix] (MG_Impl/Texture): validate the bound texture before dereferencing it in TexSubImage2D, and stop recording an error when glDeleteTextures is handed unknown names 2026-07-16 21:34:40 -04:00
swung0x48 8026838563 [Fix] (MG_Backend/DirectVulkan): honor scissor in glClearBuffer*/glClearNamedFramebuffer* and clamp their depth clear values to [0,1] 2026-07-16 21:18:01 -04:00
swung0x48 5bd8fa8c4e [Perf] (MG_Backend/DirectVulkan): route full-coverage scissored glClear back to the deferred loadOp path, skip render-pass churn for no-op clears, and drop the per-clear heap Vector (extracted PrepareScissoredClear) 2026-07-16 21:18:00 -04:00
swung0x48 37ef2cb600 [Fix] (MG_Backend/DirectVulkan): materialize mid-pass pending color clears at the subpass color slot index, not the compacted description index 2026-07-16 21:18:00 -04:00
swung0x48 5184901a5b [Fix] (MG_Backend/DirectVulkan): clear every layer of layered framebuffers in vkCmdClearAttachments paths (rename the never-read RenderPassEntry::subpass to layers) 2026-07-16 21:17:59 -04:00
swung0x48 4172959e49 [Fix] (MG_Backend/DirectVulkan): honor the front stencil write mask in scissored glClear and drop a redundant compatibility re-check 2026-07-16 21:17:58 -04:00
swung0x48 eb5b4bca3e [Fix] (MG_Backend/DirectVulkan): ignore glClear/glClearBuffer* while GL_RASTERIZER_DISCARD is enabled, and early-out on an empty clear mask 2026-07-16 21:17:58 -04:00
swung0x48 56db115a1a [Fix] (MG_Impl/GLImpl): clamp glClearDepth to [0,1] per GL 3.3 (Vulkan clear values require it) 2026-07-16 21:17:57 -04:00
swung0x48 b14edd515a Merge branch 'worktree-agent-a1381c9e6b342e8d4' into dev 2026-07-16 19:47:52 -04:00
swung0x48 57030b5fb9 [Chore] (MG_Test/Texture): regression tests for 2D-array unpack subcuboid selection and TexStorage3D layer-count semantics, DirectGLES 2D-array target support, packed-format canonical transfer types, and the GL_RGB565 enum round-trip 2026-07-16 19:44:37 -04:00
swung0x48 72532de780 [Fix] (MG_Util/Texture, MG_Util/Converters): canonical packed transfer types for RGBA4/RGB565/RGB10_A2UI - NormalizePixelFormat's default handed backends GL_UNSIGNED_BYTE (and a non-integer GL_RGB transfer format for RGB10_A2UI), so uploads read 4 bytes per texel from 2-byte packed shadow rows (rgba4 layers shifted by 2x slice stride) or were rejected outright; also map GL_RGB565 <-> TextureInternalFormat::RGB5 (the enum had no GL_RGB565 mapping at all, glTexImage* with it failed as unknown) 2026-07-16 19:44:36 -04:00
swung0x48 25323bfb8e [Fix] (MG_Backend/DirectGLES, MG_Impl/GLImpl): sync GL_TEXTURE_2D_ARRAY textures to the ES backend - the target was skipped as unsupported so array textures never uploaded or bound (every KHR-GL33.pixelstoragemodes.teximage3d case failed); also keep array layer counts constant across mip levels in TexStorage3D and generated-mip storage allocation (only true 3D textures halve depth) 2026-07-16 19:44:20 -04:00
swung0x48 9ed5dbf483 [Perf] (MG_Backend/DirectVulkan): return VkBufferResource by raw pointer from GetOrCreateResource to drop per-draw SharedPtr refcounting 2026-07-16 18:34:55 -04:00
swung0x48 075471cdbd Merge branch 'worktree-agent-a7bc533bede191e68' into dev 2026-07-16 16:37:58 -04:00
swung0x48 8bf8f6f906 [Fix] (MG_Backend/DirectGLES, MG_Util/ShaderTranspiler): make the uniform-block cross-stage precision fix surgical - revert the global SPVC ES highp-default options (they changed emission for EVERY fragment shader: sampling code that used to inherit the effective highp default was suddenly printed as explicit mediump, regressing KHR-GL3x.texture_repeat_mode NPOT mip cases on device) and instead strip RelaxedPrecision member decorations from uniform-block-reachable structs in a DirectGLES-only SPIR-V pass, so matched blocks declare identical (highp) member precision in both stages and every other shader keeps its previous emission byte-for-byte 2026-07-16 16:36:28 -04:00
swung0x48 6a843b3088 [Fix] (MG_Backend/DirectVulkan): honor scissor in glClear 2026-07-16 12:39:55 -04:00
swung0x48 c59c15f66a Merge branch 'worktree-agent-a7bc533bede191e68' into dev 2026-07-16 12:12:47 -04:00
swung0x48 254cf1dc21 [Fix] (MG_Util/ShaderTranspiler, MG_State, MG_Impl/GLImpl, MG_Backend/DirectGLES): GL CTS uniform_block - coerce packed/shared block layouts to std140 at source preprocess (glslang rejects them when targeting SPIR-V; std140 is the only UBO layout the pipeline emits), GL-style block reflection (array "[0]" names, per-element struct-array expansion, unused members and declared-but-unread blocks stay active), vec4-padded GL_UNIFORM_BLOCK_DATA_SIZE, std140 array strides for struct-nested arrays (glslang reflects tight strides there), arrayed-block instances share the first instance member set, glDeleteShader-flagged names stay usable while attached, and backend ESSL emits against highp default precision so relaxed block members match across stages (KHR-GL33.shaders.uniform_block on llvmpipe: 659 Fail -> 828/828 Pass) 2026-07-16 12:09:11 -04:00
swung0x48 df0e9fca71 Merge branch 'worktree-agent-af155789372f0003b' into dev 2026-07-16 11:30:03 -04:00
swung0x48 0005a50517 [Feat] (MG_Backend/DirectGLES): packed-type readback encoding - repack wide RGBA reads into all GL 3.3 packed pixel types (3_3_2/2_3_3_REV, 5_6_5(_REV), 4_4_4_4(_REV), 5_5_5_1/1_5_5_5_REV, 8_8_8_8(_REV), 10_10_10_2/2_10_10_10_REV, packed-float 10F_11F_11F_REV and shared-exponent 5_9_9_9_REV) for ReadPixels/GetTexImage; conversion helpers extracted to context-free ReadbackImpl (Utils.cpp) with unit tests asserting exact packed words against the GL CTS pack_* oracle layouts 2026-07-16 11:08:02 -04:00
swung0x48 f896c7396f [Fix] (MG_Impl/GLImpl): TexImage3D - apply ConvertInternalFormatToSized like 2D/1D so unsized-internal 3D uploads get channel/type conversion, skip proxy shadow allocation, auto-generate mipmaps; TexSubImage3D - bound level and region against the target mip 2026-07-16 10:54:48 -04:00
swung0x48 1cefb9780b [Feat] (MG_State, MG_Util, MG_Impl/GLImpl, MG_Backend/DirectGLES): desktop-GL single-channel client formats GL_GREEN/GL_BLUE/GL_ALPHA and _INTEGER variants - validate and readback via wide-RGBA channel extraction (GL CTS packed_pixels rgba8_format_green/blue read with them), unpack GREEN/BLUE(_INTEGER) TexImage uploads into the named channel with 0/1 defaults per table 3.3; GL_ALPHA keeps the legacy Red upload mapping (R8 storage + 000R swizzle), its readback corrected at the backend to source channel 3 2026-07-16 06:47:33 -04:00
swung0x48 bae222227a [Fix] (MG_Backend/DirectGLES, MG_Impl/GLImpl): ReadPixels - fall back to wide-format conversion when the ES driver rejects a legacy native read combo (Adreno errors on e.g. GL_RED/GL_UNSIGNED_INT and leaves the buffer untouched), and enforce packed-type/format pairing at the state layer via shared ValidateClientFormatTypePairing (GL_RED + GL_UNSIGNED_SHORT_5_6_5 now raises GL_INVALID_OPERATION) 2026-07-16 06:11:53 -04:00
swung0x48 12a67f596c [Fix] (MG_Backend/DirectGLES): upload shadow mips with UNPACK_ALIGNMENT=1 - shadow rows are tightly packed but uploads ran with alignment 4, shifting every row of non-multiple-of-4-width textures by one pixel (R8 7-wide CTS gradients read back diagonally) 2026-07-16 05:52:01 -04:00
swung0x48 274c234aff [Fix] (MG_State/ErrorState): GL error flags are sticky per error code, not an unbounded queue - repeated same-code errors accumulated and leaked into later unrelated glGetError checks (GL CTS "Texture state reset failed" deinit noise and false "Error during glGetTexImage" failures) 2026-07-16 05:29:46 -04:00
swung0x48 20e1b417cc [Fix] (MG_Util/Texture): expand channels and convert component types on texture unpack to the internal shadow layout 2026-07-16 04:39:57 -04:00
swung0x48 4dbdbd3bcd Merge branch 'worktree-agent-acd555500daee6840' into dev 2026-07-16 04:13:25 -04:00
swung0x48 5eeba29579 [Fix] (CI): resume interrupted fixture downloads 2026-07-16 04:11:48 -04:00
swung0x48 dae8a40af1 Merge branch 'worktree-agent-a08726da20d78531a' into dev 
# Conflicts:
 2026-07-16 03:57:12 -04:00
swung0x48 8b78379f6f [Feat] (MG_Backend/DirectVulkan): implement combined depth-stencil texture upload via per-aspect de-interleaved staging copies 2026-07-16 03:54:33 -04:00
swung0x48 535e9f8b15 [Fix] (MG_Util/Metrics): correct Depth32FStencil8 shadow texel size (16->8) and FLOAT_32_UNSIGNED_INT_24_8_REV type size (4->8) to the GL client transfer layout 2026-07-16 03:54:33 -04:00
swung0x48 b9844ed7c1 [Fix] (MG_Util/Texture): infer RGBA8 for packed RGBA uploads 
(cherry picked from commit 1146188ed4)
 2026-07-16 03:41:29 -04:00
swung0x48 f0ed5c1b8e [Fix] (MG_Backend/DirectGLES): convert narrow client formats (RED/RG/RGB/BGR/BGRA + integer variants, byte/short/half/8888 types) in GetTexImage/ReadPixels via wide RGBA readback 2026-07-16 03:40:54 -04:00
swung0x48 b604188849 Merge branch 'worktree-agent-aae96e27d8b07e982' into dev 2026-07-16 03:32:39 -04:00
swung0x48 cf8f928db8 [Fix] (MG_Impl/GLImpl, MG_State): fallback UBO backing for optimizer-eliminated uniforms (null-MapUBO SIGSEGV in KHR-GL33 do_while loops) + per-element locations/offsets for array uniforms incl. nested struct arrays (size assert in KHR-GL33 struct.uniform); demote uniform write assert to log-and-clamp 2026-07-16 03:31:32 -04:00
swung0x48 176d130f09 [Fix] (MG_Backend): GetTexImage level-range check was off-by-one (max level is inclusive; single-level textures asserted on level 0), demoted to logged skip; advertise GL_ARB_texture_storage_multisample (entry points already implemented - unadvertised extension left null glw pointers and CTS framebuffer_blit jumped to address 0) 2026-07-16 03:06:03 -04:00
swung0x48 0b94e02de5 [Fix] (MG_Backend/DirectVulkan): skip combined depth-stencil texture data uploads instead of recording invalid single-copy with multi-bit aspect mask (VK_INCOMPLETE at vkEndCommandBuffer killed the process); proper per-aspect de-interleave tracked separately 2026-07-16 02:36:07 -04:00
swung0x48 fe7a5ee1b2 [Fix] (MG_Impl/GLImpl, MG_State, MG_Backend/DirectGLES): eliminate packed_pixels SIGTRAPs - complete TexImage format/type/internalformat validation matrix (depth-stencil family, integer-ness, packed-type pairing, 3D depth rejection), fix inverted UpdateSubData assert with clamped copy, demote unimplemented readback asserts to logged no-ops 2026-07-16 02:23:18 -04:00
swung0x48 5331150cb9 [Fix] (MG_Util/ShaderTranspiler): stop stripping precision qualifiers - the strip corrupted "precision highp float;" into invalid syntax; glslang accepts and ignores them natively in 460 core (unblocks ~2000 GL CTS cases per backend) 2026-07-16 01:50:33 -04:00
swung0x48 dc2f3a477b [Chore] (MG_Backend/DirectGLES): demote UBO ring creation log to debug level 2026-07-16 00:10:14 -04:00
swung0x48 21753b0e4c [Fix] (MG_Backend/DirectGLES): UBO ring - preserve generation across ES context recreation, retire frame marks at Present to bound growth 2026-07-16 00:10:14 -04:00
swung0x48 f509b19b1d [Fix] (MG_Backend/DirectGLES): UBO ring review fixes - invalidate array-buffer bind cache on failed ring creation, bump generation on emergency drain, division-based alignment rounding 2026-07-16 00:10:13 -04:00
swung0x48 c7ac5de28e [Chore] (MG_Backend/DirectGLES): log global-UBO ring creation 2026-07-16 00:10:13 -04:00
swung0x48 2e7073a890 [Perf] (MG_Backend/DirectGLES): replace per-draw global-UBO glBufferSubData with a persistent-mapped ring allocator (fence-watermark reclaimed, MOBILEGL_DISABLE_UBO_RING opt-out); scrub stale indexed-binding shadow on glDeleteBuffers 2026-07-16 00:10:12 -04:00
swung0x48 305701326c [Fix] (MG_Util/SelfTest): preserve POST checks before format probing 2026-07-16 00:10:12 -04:00
swung0x48 8c89b1618a [Feat] (MG_Backend, android-plugin): show format capability tables in POST 2026-07-16 00:10:11 -04:00
swung0x48 3e4ce5caa7 [Refactor] (MG_Impl/GLImpl): replace Flywheel dispatch sync hack with MOBILEGL_COHERENT_AS_FLUSH 2026-07-15 21:51:20 -04:00
swung0x48 f80f6f4a62 [Fix] (CI): retain fixtures for failed retraces 2026-07-15 04:36:15 -04:00
swung0x48 66caf907fd [Fix] (CI): remove intermediate trace artifacts 2026-07-15 02:52:03 -04:00
swung0x48 d3150399c7 [Fix] (CI): download direct trace APK artifact 2026-07-15 02:01:53 -04:00
swung0x48 f1c6a12c06 [Fix] (CI): use full commit SHAs 2026-07-15 01:33:56 -04:00
swung0x48 947442ec78 [Fix] (CI): resolve trace APK by SHA 2026-07-15 01:27:09 -04:00
swung0x48 165dd003d7 [Chore] (CI): clean intermediate artifacts 2026-07-15 00:53:50 -04:00
swung0x48 b852ced3c1 [Chore] (CI): upgrade GitHub Actions runtimes 2026-07-15 00:36:08 -04:00
swung0x48 4a03d62b91 [Fix] (CI): publish APK artifacts directly 2026-07-15 00:34:52 -04:00
swung0x48 056574eebe [Fix] (CI): name APK artifacts by commit 2026-07-15 00:22:12 -04:00
swung0x48 e61685547a [Fix] (CI): locate unified trace APK 2026-07-15 00:17:08 -04:00
swung0x48 7ebaf43282 [Refactor] (Android plugin): unify renderer APK with Plugin V2 DSL 2026-07-14 23:37:21 -04:00
swung0x48 15580ff6a6 [Fix] (Retrace): preserve bundled ANGLE binaries 2026-07-15 08:46:28 +08:00
swung0x48 fd6f5bca83 [Fix] (Retrace): rely on APK signing for bundled ANGLE 2026-07-15 07:50:40 +08:00
swung0x48 b6d311f20b [Fix] (Retrace): bundle and select signed ANGLE variants 2026-07-15 06:48:10 +08:00
swung0x48 9c0d5517bd [Refactor] (MG_Backend/DirectVulkan): replace null-renderer guards with MOBILEGL_ASSERT 
Drops the if (!pVulkanRenderer) { return; } / !MG_State::pGLContext early-return guards across DirectVulkan.cpp in favor of MOBILEGL_ASSERT, matching the pattern already used by the rest of the backend. Legitimate runtime conditions (index bounds, sync/query handle nullness, renderer-generation mismatch, timer-query support) are kept as real checks; only the null-pointer defenses are converted.
 2026-07-14 03:55:41 -04:00
swung0x48 3445ab9304 [Refactor] (MG_Backend/DirectVulkan, trace-replay): extract dump-image capture out of MobileGL backend 
Moves snapshot capture entirely into the apitrace retrace layer (glReadPixels + PNG encode). Drops the MOBILEGL_PRESENT_DUMP_PATH / MOBILEGL_PRESENT_STATS / MOBILEGL_PRESENT_DUMP_CALL / MOBILEGL_PRESENT_CURRENT_CALL / MOBILEGL_TRACE_CURRENT_CALL_OVERRIDE plumbing from Config, ConfigLoader, VulkanRenderer (GetPresentedDumpPixel/WritePresentedDumpPpm + present-stats readback), the EGL/GLX/Android ws shims, and the Android trace_replay_core PPM reader.

DirectVulkan ReadPixels on the default framebuffer now remaps raw swapchain pixels (top-left origin, preTransform-rotated) to GL orientation (bottom-left origin) so the retrace snapshot matches the golden; SwapchainObject also resizes the default-FBO stencil attachment to the swapchain extent to fix GL_INVALID_FRAMEBUFFER_OPERATION under the glReadPixels completeness check.
 2026-07-14 02:40:20 -04:00
swung0x48 533219ede7 [Fix] (Retrace): load ANGLE through LD_LIBRARY_PATH 2026-07-13 21:20:02 -04:00
swung0x48 b1f55026af [Chore] (MobileGL/ConfigLoader): rename/clean up more env var clutter 2026-07-13 20:45:40 -04:00
swung0x48 a26e9aaf25 [Refactor] (MG_Config, MG_Backend, trace-replay): remove unused stats instrumentation 2026-07-13 20:33:31 -04:00
swung0x48 a55a0645e2 [Refactor] (MG_Config, MG_Backend/DirectVulkan, trace-replay): centralize Magma env parsing and rename R11G11B10F fallback 2026-07-13 19:58:06 -04:00
swung0x48 e529e12d27 [Perf] (MG_Backend/DirectVulkan): bind GetBoundObject/GetSamplerObject const-ref returns by reference in BindProgramUniformBuffers resolve paths (UBO/SSBO/texel/sampler override) instead of copying the SharedPtr; ResolveUniformBufferPayload 2.2%->1.6% 2026-07-13 08:27:12 -04:00
swung0x48 e78eee972e [Perf] (MG_Backend/DirectVulkan): store the resolved TextureResource pointer in the per-draw sync memo so repeat SyncTextureAndGetDescriptor calls skip the resource-map lookup 2026-07-13 06:53:37 -04:00
swung0x48 37cd5b42de [Perf] (MG_Backend/DirectVulkan): drop redundant per-draw work in UploadAndBindVertexBuffers - pass programObj from SetupDraw, use the VAO attribute's buffer SharedPtr directly instead of re-resolving by external index; 8.5%->6.2% 2026-07-13 05:21:57 -04:00
swung0x48 7fe5247626 [Fix] (MG_Backend/DirectVulkan): key the sampled-set walk-skip on a program lifetime id, not the recyclable GL name, so a deleted+recreated program can't false-hit the cache 2026-07-13 04:58:05 -04:00
swung0x48 f098983c9f [Perf] (MG_Backend/DirectVulkan): skip the per-draw sampled-texture walk when the bound set is unchanged (texture-bind generation + program state version); CollectSampledTextures 5.0%->0.2%, fps 228->249 2026-07-13 04:41:14 -04:00
swung0x48 516d2a659e [Perf] (MG_Backend/DirectVulkan): cache resolved VkSampler per binding to skip the per-draw sampler key hash, keyed on a new sampler lifetime id 2026-07-13 02:04:57 -04:00
swung0x48 808c5dcc46 [Perf] (MG_Backend/DirectVulkan): re-land content-version texture early-out; bump content version on glGenerateMipmap so cached sampled views re-sync (fixes Iris shader retrace) 2026-07-13 00:50:09 -04:00
swung0x48 ecea8054a6 Revert content-version texture early-out (e8e1521): the SyncTexture cross-draw skip breaks Iris shader retrace validation (correctness regression) 2026-07-12 23:27:54 -04:00
swung0x48 b0076af9bd [Perf] (MG_Backend/DirectVulkan): raw-ptr in ResolveSamplerDescriptor drops the per-draw SharedPtr refcount on the sampler descriptor path 2026-07-12 22:48:05 -04:00
swung0x48 24cf1e3a7f [Perf] (MG_Backend/DirectVulkan): frames-in-flight from MOBILEGL_MAGMA_FRAMESINFLIGHT env (fallback 3), clamped to surface maxImageCount at init 2026-07-12 20:35:05 -04:00
swung0x48 e5ee4cde4f [Perf] (MG_Backend/DirectVulkan): deepen frame pipeline 2->3 to hide GPU-completion latency; cross-frame glClientWaitSync fence stalls -28% 2026-07-12 19:42:35 -04:00
swung0x48 e8e1521972 [Perf] (MG_Backend/DirectVulkan): skip cross-draw re-sync of unchanged textures via a content-version early-out; SyncTextureAndGetDescriptor 8.9%->2.4% 2026-07-12 18:40:48 -04:00
swung0x48 6f53b9a6bb [Perf] (MG_Backend/DirectVulkan): raw-ptr sampled-texture walk skips SharedPtr refcount churn per draw 2026-07-12 10:51:15 -04:00
swung0x48 acaa9f6dc7 [Perf] (MG_Backend/DirectVulkan): zero-copy UBO bind - point descriptor at the app's persistent VkBuffer instead of a per-draw transient copy; fps 127->166 2026-07-12 10:11:49 -04:00
swung0x48 375f2df694 [Perf] (MG_Backend/DirectVulkan): skip per-draw pipeline resolution when pipeline state unchanged; SetupDraw 54%->51%, fps 109->127 2026-07-12 09:28:30 -04:00
swung0x48 542e50be33 [Perf] (MG_Backend/DirectVulkan): skip per-draw render-pass hash when framebuffer state unchanged; SetupDraw 60%->54%, fps 96->109 2026-07-12 08:18:37 -04:00
swung0x48 ad9ee99521 [Perf] (MG_Backend/DirectGLES): dedup per-draw indexed UBO/SSBO binds with a shadow cache; BindCurrentProgramWithResources 5.1% -> 3.1% 2026-07-12 06:32:18 -04:00
swung0x48 25395a9f9a [Docs] (MG_Backend/DirectGLES): TODO for buffer-pool Phase 2 orphan-on-respecify 2026-07-12 05:37:03 -04:00
swung0x48 d7029952bb [Perf] (MG_Backend/DirectGLES): recycle idle GL buffers via a fence-gated size pool instead of glDeleteBuffers; pinned fps 184->220 2026-07-12 05:28:34 -04:00
swung0x48 340449b77e [Perf] (MG_State, MG_Backend/DirectGLES): skip never-touched buffer bind points via high-water mark; SyncNeccessaryBuffers 15.7% -> 4.2% 2026-07-12 04:31:34 -04:00
swung0x48 527e229ac8 [Perf] (MG_Backend/DirectGLES): drop redundant per-draw UBO binding-point sync; BindCurrentProgramWithResources already rebinds them 2026-07-12 04:11:23 -04:00
swung0x48 d5bc753764 [Perf] (MG_Backend/DirectGLES): skip scratch bind + upload for fully-synced mipmap textures 2026-07-12 04:11:22 -04:00
swung0x48 436f7f7e86 [Perf] (DirectGLES): shadow-track unpack state instead of glGetIntegerv 
ScopedDefaultUnpackState saved the backend GL unpack state with 6 glGetIntegerv
calls on every construction. glGetIntegerv forces a driver pipeline sync, and
because it ran per dirty texture per frame in the texture upload path, it
dominated the DirectGLES draw path - and stalling the pipeline serialized CPU-GPU
work far beyond its raw CPU cost.

The backend unpack state is set only by MobileGL's own save/restore helpers
(ScopedDefaultUnpackState, TempPixelStoreParameterSync, the R32F copy path), all
of which restore to the resting GL default, so it can be shadow-tracked: read the
previous state from a static shadow (no query), pin the backend to the known
default once up front, and set state with compare-and-set so the paired
glPixelStorei calls also usually no-op.

Device-verified on Adreno 830 (MC 26.3-snapshot3, Espryt, CPU pinned to 1.56/1.96
GHz for a thermally-comparable measurement): rendering correct; fps 105 -> 147
(+40%); render-thread profile: glGetIntegerv ~9% -> below noise, SyncNeccessary-
Textures 25% -> 12%, SyncMipmapsToBackend 23% -> 9%.
 2026-07-12 03:11:28 -04:00
swung0x48 009e37ec6f [Perf] (DirectVulkan): reuse descriptor set across draws with identical bindings 
BindProgramUniformBuffers rebuilt a fresh descriptor set and called
vkUpdateDescriptorSets on every draw, even when consecutive draws bound the exact
same textures/samplers/buffers (common in MC: many draws share a program + atlas).
Now, after resolving the bindings (still needed for the UBO dynamic offset),
compute a cheap word-wise signature of the resolved descriptor content + layout;
when it matches the previous draw, reuse that descriptor set and skip
AcquireDescriptorSet + vkUpdateDescriptorSets - only the bind-time dynamic offsets
differ.

Correct by construction: bindings are re-resolved every draw so the signature
always reflects current state and reuse only happens on an exact match; the reused
set is never re-acquired within a frame (the acquire cursor only advances); the
descriptor set layout is in the signature so reuse never crosses programs; the
cache resets each frame in BeginFrame when the frame's sets are recycled; sampler
overrides (blits) bypass and invalidate it. The signature hashes 64-bit words (the
Vk*Info payloads are 8-byte-multiple sized and value-initialized) so its own
per-draw cost stays small.

Device-verified on Adreno 830 (MC 26.3-snapshot3, optimized -O2 Magma): rendering
correct, no validation errors. Render-thread wall-clock profile:
BindProgramUniformBuffers 22.85% -> 19.82% (vkUpdateDescriptorSets ~5% dropped below
noise; word-wise signature adds ~0.6% self), SetupDraw 62% -> 60%.
 2026-07-12 01:55:31 -04:00
swung0x48andClaude Opus 4.8 b253df881d [Perf] (DirectVulkan): memoize per-draw texture sync in SetupDraw 
Each sampled texture was resolved ~3x per draw: SetupDraw's layout-probe
loop, its post-transition loop, and again inside ResolveSamplerDescriptor.
No GL texture mutation happens mid-SetupDraw, and layout is tracked on the
TextureResource independently of SyncTexture, so the repeat SyncTexture work
(mip-completeness / resource+view resync / dirty scan) is pure redundancy.

Add a per-draw memo in VkTextureManager (BeginDrawSyncScope/EndDrawSyncScope
+ RAII DrawSyncScope guard around SetupDraw): after the first successful sync
of a texture in a draw, repeat SyncTextureAndGetDescriptor calls short-circuit
to the already-synced resource.

Device-verified on Adreno 830 (MC 26.3-snapshot3, Magma): rendering correct,
no validation errors; wall-clock profile of the render thread shows
SyncTextureAndGetDescriptor dropping from 15.2% to ~5% and SetupDraw from
43.7% to 28.9%.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
 2026-07-11 22:06:28 -04:00
swung0x48 c1743aa42d [Refactor] (MG_State, MG_Backend): PipeResource storage layer + zero-copy coherent persistent maps 
Introduce a Mesa pipe_resource-style PipeResource that owns a GL buffer's bytes
and its backend GPU resource, abstracting WHERE the authoritative bytes live:
 - Shadow mode (non-persistent buffers): a CPU Vector; the backend keeps its own
   GPU copy in sync via BufferBackendOps, exactly as before.
 - Persistent mode (coherent GL_MAP_PERSISTENT maps): the backend's host-visible,
   COHERENT, persistently-mapped GPU memory is the single source of truth. The app
   writes into it directly, every reader resolves against it, and NO per-write
   backend transfer happens. The CPU shadow is released.

BufferObject no longer owns a raw shadow Vector; it holds a PipeResource and
exposes one accessor, MappedData(), that all readers go through. Every buffer-data
consumer (UBO payload, PBO texture upload, indirect draws, resident/streamed
uploads, both backends) was migrated from GetDataReadOnly()->data() to
MappedData(), so a persistent buffer's readers see GPU memory - not a stale
shadow. That stale-shadow inconsistency is what corrupted rendering (wrong UBOs ->
misplaced/"lost" vertices) in the first zero-copy attempt (625c8a6, reverted in
896cafc); routing every consumer through one accessor makes it structurally
impossible.

Backends provide the map via BufferBackendOps::AcquirePersistentMap:
 - DirectVulkan: a HOST_VISIBLE|HOST_COHERENT (required, not just requested),
   persistently mapped resident VkBuffer carrying every usage, seeded from the
   shadow, never recreated; AcquireResidentSlice binds it directly.
 - DirectGLES: EXT_buffer_storage immutable persistent+coherent glMapBufferRange,
   falling back to the shadow when the extension is absent.

Fixes the ~7GB GpuMemory OOM + 100%-CPU/ANR running modern Blaze3D Minecraft on
both Magma and Espryt (per-draw whole-buffer re-upload of the coherent persistent
ring buffer), without the coherency/stale-read hazards of the reverted attempt.

BufferTest: zero-copy stress guard (15,360 draws -> 0 per-draw transfers, and every
reader resolves to GPU memory) + a shadow-fallback test. Host suite: 203/203 pass.
Device verification pending.
 2026-07-11 20:18:38 -04:00