[Merge] (GLImpl, DirectGLES, DirectVulkan): complete direct state access and the compressed-texture family

MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.cpp

@@ -753,7 +753,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                         .TargetGLSLVersion = {4, 6, 0}, // Target Shading Language Version
                         // Baseline advertisement (no runtime capabilities yet); reconciled once
                         // the ES capabilities exist, see UpdateAdvertisedCapabilityExtensions.
-                        .Extensions = BuildAdvertisedExtensions(false, false, false, false, false),
+                        .Extensions = BuildAdvertisedExtensions(false, false, false, false, false, false),
                         .IsCompatibilityProfile = false // Is Compatibility Profile
                     },
                 .StaticBackendCapability = {.AllowVSOnlyPrograms = false} // Backend Capability
@@ -778,7 +778,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 AreTimerQueriesSupported(), capabilities.SupportsTextureFilterAnisotropy,
                 capabilities.SupportsDrawIndirect,
                 capabilities.SupportsDrawIndirect && capabilities.SupportsBaseInstance,
-                capabilities.SupportsTextureView);
+                capabilities.SupportsTextureView, capabilities.SupportsTextureCubeMapArray);
         }
     } // namespace
 
@@ -992,7 +992,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
     Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported,
                                                   Bool drawIndirectSupported,
                                                   Bool nonZeroIndirectBaseInstanceSupported,
-                                                  Bool textureViewSupported) {
+                                                  Bool textureViewSupported, Bool cubeMapArraySupported) {
         Vector<GLExtension> extensions = {
             // The version tokens have to reach the version the backend actually claims:
             // TargetGLVersion is {4,3,0}, and a list that stopped at OpenGL40 told an
@@ -1030,6 +1030,22 @@ namespace MobileGL::MG_Backend::DirectGLES {
             // was simply never emitted, which left KHR-GL4*.draw_elements_base_vertex_tests
             // NotSupported on a feature that works.
             E_GL_ARB_draw_elements_base_vertex,
+            // The whole sync-object family is real and core since 3.2: glFenceSync, glIsSync,
+            // glDeleteSync, glClientWaitSync, glWaitSync and glGetSynciv all live in GLImpl over a
+            // backend fence (a host GLsync here, a VkFence on DirectVulkan), and glGetInteger64v
+            // answers GL_MAX_SERVER_WAIT_TIMEOUT. The string matters for the same reason
+            // ARB_uniform_buffer_object's does: LWJGL builds GLCapabilities from the extension
+            // list, and a caller that finds GL_ARB_sync missing never resolves the entry points -
+            // then calls through null if it uses fences anyway. Nothing in the CTS gates on this
+            // string, so it is advertised on the strength of the implementation, not a test unlock.
+            E_GL_ARB_sync,
+            // Atomic counters, core since 4.2. glGetActiveAtomicCounterBufferiv and the whole
+            // GL_ATOMIC_COUNTER_BUFFER_* query family are real in GLImpl, and SyncAtomicCounterBuffers
+            // re-issues the counter buffer as an SSBO binding in the range reserved at the top of
+            // the ES driver's shader-storage points, so a counter dispatch reads and writes the
+            // buffer the application bound. DirectVulkan reaches the same place through its own
+            // descriptor resolution, so the string is symmetric.
+            E_GL_ARB_shader_atomic_counters,
             // glVertexAttribDivisor, core since 3.3 and real on both backends. Applications
             // (Better Clouds' GLCompat among them) accept the extension string as an
             // ALTERNATIVE to a 3.3 context when deciding whether instanced rendering is
@@ -1039,6 +1055,46 @@ namespace MobileGL::MG_Backend::DirectGLES {
             // object-label table are MobileGL's own state, not the host driver's - so it is as
             // available here as it is on DirectVulkan, which has advertised it all along.
             E_GL_KHR_debug,
+            // Core GL 3.0-4.3 plumbing that has been real here for as long as the backend has
+            // existed, and that was simply never named. None of these unlocks a single CTS case -
+            // the conformance suite reaches all of them through the version - so they are
+            // advertised for the OTHER consumer of this list: LWJGL builds GLCapabilities from the
+            // string set, and an application that gates its ENTRY POINTS on the string rather than
+            // on the version never resolves them and then calls through null. Each is backed by
+            // the entry points named beside it.
+            //
+            // glBindVertexArray / glGenVertexArrays / glDeleteVertexArrays / glIsVertexArray.
+            E_GL_ARB_vertex_array_object,
+            // The 14 glSamplerParameter* / glGetSamplerParameter* entry points, including the
+            // integer-valued Iiv/Iuiv forms.
+            E_GL_ARB_sampler_objects,
+            // glMapBufferRange + glFlushMappedBufferRange, which ARB_buffer_storage's persistent
+            // maps are already built on top of.
+            E_GL_ARB_map_buffer_range,
+            // glCopyBufferSubData plus the GL_COPY_READ_BUFFER / GL_COPY_WRITE_BUFFER targets.
+            E_GL_ARB_copy_buffer,
+            // glCopyImageSubData, wired to a real backend hook on both backends.
+            E_GL_ARB_copy_image,
+            // GL_TEXTURE_SWIZZLE_{R,G,B,A,RGBA}, which this backend syncs through to the ES
+            // driver's identical parameters.
+            E_GL_ARB_texture_swizzle,
+            // GL_INT_2_10_10_10_REV / GL_UNSIGNED_INT_2_10_10_10_REV on glVertexAttribPointer plus
+            // the eight glVertexAttribP* entry points.
+            E_GL_ARB_vertex_type_2_10_10_10_rev,
+            // The R/RG internal formats. Named separately from the float ones because an
+            // application may check either.
+            E_GL_ARB_texture_rg,
+            // GL_DEPTH_COMPONENT32F and GL_DEPTH32F_STENCIL8.
+            E_GL_ARB_depth_buffer_float,
+            // The floating-point colour formats. Unlike the rest of this block this string DOES
+            // gate CTS cases - KHR-GL4*.internalformat.texture2d.*{16f,32f} is keyed on it with no
+            // core-version fallback, so eight cases per version list were NotSupported on formats
+            // the backend has always had.
+            E_GL_ARB_texture_float,
+            // glViewportArrayv / glViewportIndexedf{,v} / glScissorArrayv / glScissorIndexed{,v} /
+            // glDepthRangeArrayv / glDepthRangeIndexed / glGetFloati_v / glGetDoublei_v, over the
+            // 16 viewports GL_MAX_VIEWPORTS reports and the per-viewport routing emulation.
+            E_GL_ARB_viewport_array,
             // Advertised with GL_NUM_PROGRAM_BINARY_FORMATS = 0, which the
             // extension explicitly permits. It is also the only thing that
             // exposes glProgramParameteri before GL 4.1.
@@ -1087,6 +1143,19 @@ namespace MobileGL::MG_Backend::DirectGLES {
         if (timerQueriesSupported && !MG_Config::Features.DisableTimerQuery) {
             extensions.push_back(E_GL_ARB_timer_query);
         }
+        // Cube map arrays are core from GL 4.0 and from ES 3.2, but on a pre-ES-3.2 driver without
+        // EXT/OES_texture_cube_map_array there is nothing underneath: the texture gets no storage
+        // and a samplerCubeArray shader does not even compile, which is exactly what the POST
+        // reports. So the string follows the host capability rather than the version.
+        //
+        // Named for the application's benefit rather than the suite's: measured on Adreno 830,
+        // KHR-GL43.texture_gather.plain-gather-*-cube-array already passed without the string, so
+        // this unlocks no conformance case. It is advertised because the feature is real and
+        // because an application that feature-detects cube map arrays off the string (rather than
+        // off the 4.0 version) would otherwise decline a path this backend serves.
+        if (cubeMapArraySupported) {
+            extensions.push_back(E_GL_ARB_texture_cube_map_array);
+        }
         // Only advertised when the host ES driver has EXT/OES_texture_view. ES has no core
         // texture views at any version and no honest emulation exists: a view is a SECOND NAME
         // over the SAME storage, so that writes through either are visible through the other and

MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.h

@@ -83,7 +83,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
     Vector<GLExtension> BuildAdvertisedExtensions(Bool timerQueriesSupported, Bool anisotropicFilteringSupported,
                                                   Bool drawIndirectSupported,
                                                   Bool nonZeroIndirectBaseInstanceSupported,
-                                                  Bool textureViewSupported);
+                                                  Bool textureViewSupported, Bool cubeMapArraySupported);
 
     // Format: <OpenGL ES Renderer>, OpenGL ES <Major>.<Minor> — the exact string an
     // initialized backend returns from GetBackendAPIVersionString (and that ends up

MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.cpp

@@ -504,7 +504,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                .TargetGLSLVersion = {4, 6, 0},
                                // Baseline advertisement (no runtime-gated capabilities); a live
                                // backend reconciles its copy in UpdateAdvertisedExtensions.
-                               .Extensions = BuildAdvertisedExtensions(false, false, false, false),
+                               .Extensions = BuildAdvertisedExtensions(false, false, false, false, false),
                                .IsCompatibilityProfile = false},
             .StaticBackendCapability = {.AllowVSOnlyPrograms = false}};
         return rendererInfo;
@@ -512,7 +512,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
     Vector<GLExtension> BuildAdvertisedExtensions(Bool shaderSubgroupSupported, Bool timerQueriesSupported,
                                                   Bool anisotropicFilteringSupported,
-                                                  Bool nonZeroIndirectBaseInstanceSupported) {
+                                                  Bool nonZeroIndirectBaseInstanceSupported,
+                                                  Bool cubeMapArraySupported) {
         Vector<GLExtension> extensions = {
             // The version tokens have to reach the version the backend actually claims:
             // TargetGLVersion is {4,3,0}, and a list that stopped at OpenGL40 told an
@@ -550,11 +551,67 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // was simply never emitted, which left KHR-GL4*.draw_elements_base_vertex_tests
             // NotSupported on a feature that works.
             E_GL_ARB_draw_elements_base_vertex,
+            // The whole sync-object family is real and core since 3.2: glFenceSync, glIsSync,
+            // glDeleteSync, glClientWaitSync, glWaitSync and glGetSynciv all live in GLImpl over a
+            // backend fence (a VkFence here, an EGLSync/GLsync on DirectGLES), and glGetInteger64v
+            // answers GL_MAX_SERVER_WAIT_TIMEOUT. The string matters for the same reason
+            // ARB_uniform_buffer_object's does: LWJGL builds GLCapabilities from the extension
+            // list, and a caller that finds GL_ARB_sync missing never resolves the entry points -
+            // then calls through null if it uses fences anyway. Nothing in the CTS gates on this
+            // string, so it is advertised on the strength of the implementation, not a test unlock.
+            E_GL_ARB_sync,
+            // Atomic counters, core since 4.2. glGetActiveAtomicCounterBufferiv and the whole
+            // GL_ATOMIC_COUNTER_BUFFER_* query family are real in GLImpl, and the counter buffer
+            // now reaches the shader on BOTH backends - Magma resolves the lowered
+            // gl_AtomicCounterBlock_<N> from the atomic-counter binding points rather than the
+            // shader-storage ones (see ResolveStorageBufferDescriptor). Withheld here until that
+            // landed, because the counter silently read whatever was bound as SSBO N instead.
+            E_GL_ARB_shader_atomic_counters,
             // glVertexAttribDivisor, core since 3.3 and real on both backends. Applications
             // (Better Clouds' GLCompat among them) accept the extension string as an
             // ALTERNATIVE to a 3.3 context when deciding whether instanced rendering is
             // available, so withholding it makes MobileGL look less capable than it is.
             E_GL_ARB_instanced_arrays,
+            // Core GL 3.0-4.3 plumbing that has been real here for as long as the backend has
+            // existed, and that was simply never named. None of these unlocks a single CTS case -
+            // the conformance suite reaches all of them through the version - so they are
+            // advertised for the OTHER consumer of this list: LWJGL builds GLCapabilities from the
+            // string set, and an application that gates its ENTRY POINTS on the string rather than
+            // on the version never resolves them and then calls through null. Each is backed by
+            // the entry points named beside it. Kept identical to the DirectGLES block so the two
+            // backends do not disagree about what MobileGL is.
+            //
+            // glBindVertexArray / glGenVertexArrays / glDeleteVertexArrays / glIsVertexArray.
+            E_GL_ARB_vertex_array_object,
+            // The 14 glSamplerParameter* / glGetSamplerParameter* entry points, including the
+            // integer-valued Iiv/Iuiv forms.
+            E_GL_ARB_sampler_objects,
+            // glMapBufferRange + glFlushMappedBufferRange, which ARB_buffer_storage's persistent
+            // maps are already built on top of.
+            E_GL_ARB_map_buffer_range,
+            // glCopyBufferSubData plus the GL_COPY_READ_BUFFER / GL_COPY_WRITE_BUFFER targets.
+            E_GL_ARB_copy_buffer,
+            // glCopyImageSubData, wired to a real backend hook on both backends.
+            E_GL_ARB_copy_image,
+            // GL_TEXTURE_SWIZZLE_{R,G,B,A,RGBA}, which map onto a VkImageView's component swizzle.
+            E_GL_ARB_texture_swizzle,
+            // GL_INT_2_10_10_10_REV / GL_UNSIGNED_INT_2_10_10_10_REV on glVertexAttribPointer plus
+            // the eight glVertexAttribP* entry points.
+            E_GL_ARB_vertex_type_2_10_10_10_rev,
+            // The R/RG internal formats. Named separately from the float ones because an
+            // application may check either.
+            E_GL_ARB_texture_rg,
+            // GL_DEPTH_COMPONENT32F and GL_DEPTH32F_STENCIL8.
+            E_GL_ARB_depth_buffer_float,
+            // The floating-point colour formats. Unlike the rest of this block this string DOES
+            // gate CTS cases - KHR-GL4*.internalformat.texture2d.*{16f,32f} is keyed on it with no
+            // core-version fallback, so eight cases per version list were NotSupported on formats
+            // the backend has always had.
+            E_GL_ARB_texture_float,
+            // glViewportArrayv / glViewportIndexedf{,v} / glScissorArrayv / glScissorIndexed{,v} /
+            // glDepthRangeArrayv / glDepthRangeIndexed / glGetFloati_v / glGetDoublei_v, over the
+            // 16 viewports GL_MAX_VIEWPORTS reports.
+            E_GL_ARB_viewport_array,
             // Advertised with GL_NUM_PROGRAM_BINARY_FORMATS = 0, which the
             // extension explicitly permits. It is also the only thing that
             // exposes glProgramParameteri before GL 4.1.
@@ -607,6 +664,18 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             extensions.push_back(E_GL_EXT_texture_filter_anisotropic);
             extensions.push_back(E_GL_ARB_texture_filter_anisotropic);
         }
+        // A cube map array is a 6n-layer VkImage viewed as VK_IMAGE_VIEW_TYPE_CUBE_ARRAY, and that
+        // view type cannot be created without the imageCubeArray device feature - so the string
+        // follows the feature, not the version, exactly as the per-layer attachment bit does.
+        //
+        // Named for the application's benefit rather than the suite's: measured on Adreno 830,
+        // KHR-GL43.texture_gather.plain-gather-*-cube-array already passed without the string, so
+        // this unlocks no conformance case. It is advertised because the feature is real and
+        // because an application that feature-detects cube map arrays off the string (rather than
+        // off the 4.0 version) would otherwise decline a path this backend serves.
+        if (cubeMapArraySupported) {
+            extensions.push_back(E_GL_ARB_texture_cube_map_array);
+        }
         return extensions;
     }
 
@@ -737,7 +806,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_rendererInfo.RendererGLInfo.Extensions = BuildAdvertisedExtensions(
             subgroupSupportAdvertised, pVulkanRenderer && pVulkanRenderer->IsTimerQuerySupported(),
             pVulkanRenderer && pVulkanRenderer->IsSamplerAnisotropySupported(),
-            pVulkanRenderer && pVulkanRenderer->IsNonZeroIndirectBaseInstanceSupported());
+            pVulkanRenderer && pVulkanRenderer->IsNonZeroIndirectBaseInstanceSupported(),
+            m_vulkanCaps.SupportsImageCubeArray);
     }
 
     void BackendObject_DirectVulkan::UpdateDynamicBackendParameters() {

MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.h

@@ -75,7 +75,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     // the detected device support (passing an already-gated value is harmless).
     Vector<GLExtension> BuildAdvertisedExtensions(Bool shaderSubgroupSupported, Bool timerQueriesSupported,
                                                   Bool anisotropicFilteringSupported,
-                                                  Bool nonZeroIndirectBaseInstanceSupported);
+                                                  Bool nonZeroIndirectBaseInstanceSupported,
+                                                  Bool cubeMapArraySupported);
 
     // Format: <GPU Name>, Vulkan <Vulkan Version>, Driver <Driver Version> — the exact
     // string an initialized backend returns from GetBackendAPIVersionString (and that

MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.cpp

@@ -17,6 +17,7 @@
 #include "MG_Util/Converters/MGToStr/FramebufferEnumConverter.h"
 #include "MG_Util/Converters/MGToVk/TextureEnumConverter.h"
 #include "MG_Util/Metrics/TextureMetrics.h"
+#include "MG_Util/ShaderTranspiler/Types.h"
 #include <Config.h>
 #include <algorithm>
 #include <cstdio>
@@ -923,22 +924,46 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         const Int blockIndex = programObj.storageBlockIndexByBinding[binding];
         MOBILEGL_ASSERT(blockIndex >= 0, "ResolveStorageBufferDescriptor: no SSBO block mapped to binding %u",
                         binding);
+        // An atomic counter is not an SSBO the application ever declared: glslang lowers every
+        // atomic_uint onto a synthesized gl_AtomicCounterBlock_<N> storage block, where N is the
+        // GL ATOMIC-COUNTER binding. That block arrives here auto-mapped to an arbitrary
+        // storage-block slot, so resolving it the SSBO way looked up GL_SHADER_STORAGE_BUFFER
+        // point N' - which is never where glBindBufferBase(GL_ATOMIC_COUNTER_BUFFER, N, ...) put
+        // the buffer. The counter therefore never reached the shader (KHR-GL43
+        // shader_atomic_counters.advanced-usage-*), and when the application also bound an SSBO at
+        // the colliding slot the descriptor silently aliased it, so the dispatch wrote over the
+        // application's own buffer. DirectGLES has always taken this branch explicitly
+        // (SyncAtomicCounterBuffers); this is the same rule in Magma's descriptor resolution.
+        //
+        // Only the SOURCE of the handle differs. The per-counter layout(offset=) is already folded
+        // into the block's SPIR-V member offsets on this path (FlattenAtomicCounterBlockPass is
+        // DirectGLES-only), so everything below - residency, the glBindBufferRange window, the
+        // descriptor fill - is target-agnostic and stays exactly as it was.
+        const String& blockName = programObj.storageBlockNameByBinding[binding];
+        const Int atomicCounterBinding = MG_Util::ShaderTranspiler::AtomicCounterBlockGlBinding(blockName);
+        const Bool isAtomicCounterBlock = atomicCounterBinding >= 0;
+        const BufferTarget bufferTarget =
+            isAtomicCounterBlock ? BufferTarget::AtomicCounter : BufferTarget::ShaderStorage;
         // A block instance array declares one block whose elements take consecutive GL binding
         // points from the declared one (GL 4.6 core 7.8), and the reflection collapses the whole
-        // array to that one block - so the element index IS the offset from its binding.
+        // array to that one block - so the element index IS the offset from its binding. glslang
+        // synthesizes one counter block per GL binding, so a counter block is never an instance
+        // array and `element` is always 0 there; the +element rule stays with the SSBO case.
         const GLuint frontendBinding =
-            GetShaderStorageBlockBinding(program, static_cast<GLuint>(blockIndex)) + element;
+            isAtomicCounterBlock
+                ? static_cast<GLuint>(atomicCounterBinding)
+                : GetShaderStorageBlockBinding(program, static_cast<GLuint>(blockIndex)) + element;
         const Uint32 bindingPointCount =
-            static_cast<Uint32>(MG_State::pGLContext->GetBufferBindingPointCount(BufferTarget::ShaderStorage));
+            static_cast<Uint32>(MG_State::pGLContext->GetBufferBindingPointCount(bufferTarget));
         MOBILEGL_ASSERT(frontendBinding < bindingPointCount,
-                        "ResolveStorageBufferDescriptor: frontend SSBO binding %u out of range for block '%s'",
+                        "ResolveStorageBufferDescriptor: frontend binding %u out of range for block '%s'",
-                        frontendBinding, programObj.storageBlockNameByBinding[binding].c_str());
+                        frontendBinding, blockName.c_str());
 
-        auto& bindingPoint = MG_State::pGLContext->GetBufferBindingPoint(BufferTarget::ShaderStorage, frontendBinding);
+        auto& bindingPoint = MG_State::pGLContext->GetBufferBindingPoint(bufferTarget, frontendBinding);
         const auto& bufferObject = bindingPoint.GetBoundObject();
         if (bufferObject == nullptr) {
-            MGLOG_E_ONCE("ResolveStorageBufferDescriptor: no SSBO bound at frontend binding %u for block '%s'",
+            MGLOG_E_ONCE("ResolveStorageBufferDescriptor: no buffer bound at frontend binding %u for block '%s'",
-                    frontendBinding, programObj.storageBlockNameByBinding[binding].c_str());
+                    frontendBinding, blockName.c_str());
             return false;
         }
 

MobileGL/MG_Impl/GLImpl/Drawing/GL_Drawing.cpp

@@ -713,7 +713,34 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
     }
 
+    namespace {
+        // GL 4.6 core 7.11.2 (and ARB_shader_image_load_store, which introduced the call): the
+        // barrier bitfield is INVALID_VALUE unless every bit is one of the defined ones, with
+        // GL_ALL_BARRIER_BITS - which is 0xFFFFFFFF, not the union of the list - accepted whole.
+        // Forwarding an undefined bit to the host driver let a caller that had computed its mask
+        // wrongly (or reused an ES-only bit) get silence instead of the error the spec promises.
+        constexpr GLbitfield kAllDefinedBarrierBits =
+            GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT | GL_ELEMENT_ARRAY_BARRIER_BIT | GL_UNIFORM_BARRIER_BIT |
+            GL_TEXTURE_FETCH_BARRIER_BIT | GL_SHADER_IMAGE_ACCESS_BARRIER_BIT | GL_COMMAND_BARRIER_BIT |
+            GL_PIXEL_BUFFER_BARRIER_BIT | GL_TEXTURE_UPDATE_BARRIER_BIT | GL_BUFFER_UPDATE_BARRIER_BIT |
+            GL_FRAMEBUFFER_BARRIER_BIT | GL_TRANSFORM_FEEDBACK_BARRIER_BIT | GL_ATOMIC_COUNTER_BARRIER_BIT |
+            GL_SHADER_STORAGE_BARRIER_BIT | GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT | GL_QUERY_BUFFER_BARRIER_BIT;
+
+        Bool ValidateMemoryBarrierBits(const char* function, GLbitfield barriers) {
+            if (barriers == GL_ALL_BARRIER_BITS) return true;
+            if ((barriers & ~kAllDefinedBarrierBits) != 0) {
+                MG_State::pGLContext->RecordError(
+                    ErrorCode::InvalidValue,
+                    MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", function,
+                                                 "barriers contains bits that are not defined barrier bits."));
+                return false;
+            }
+            return true;
+        }
+    } // namespace
+
     void MemoryBarrier(GLbitfield barriers) {
+        if (!ValidateMemoryBarrierBits(__func__, barriers)) return;
         auto memoryBarrier = MG_Backend::gBackendFunctionsTable.GL.MemoryBarrier;
         if (!memoryBarrier) {
             MG_State::pGLContext->RecordError(
@@ -725,6 +752,7 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     void MemoryBarrierByRegion(GLbitfield barriers) {
+        if (!ValidateMemoryBarrierBits(__func__, barriers)) return;
         auto memoryBarrierByRegion = MG_Backend::gBackendFunctionsTable.GL.MemoryBarrierByRegion;
         if (!memoryBarrierByRegion) {
             MG_State::pGLContext->RecordError(

MobileGL/MG_Impl/GLImpl/Exporting/Definitions.cpp

@@ -1047,9 +1047,9 @@ DECLARE_GL_FUNCTION_HEAD(void, TextureStorage3DMultisample, GLuint texture, GLsi
 DECLARE_GL_FUNCTION_HEAD(void, TextureSubImage1D, GLuint texture, GLint level, GLint xoffset, GLsizei width, GLenum format, GLenum type, const void* pixels) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureSubImage1D, texture, level, xoffset, width, format, type, pixels)
 DECLARE_GL_FUNCTION_HEAD(void, TextureSubImage2D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLsizei width, GLsizei height, GLenum format, GLenum type, const void* pixels) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureSubImage2D, texture, level, xoffset, yoffset, width, height, format, type, pixels)
 DECLARE_GL_FUNCTION_HEAD(void, TextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLenum type, const void* pixels) DECLARE_GL_FUNCTION_END_NO_RETURN(void, TextureSubImage3D, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, type, pixels)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage1D, GLuint texture, GLint level, GLint xoffset, GLsizei width, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage1D, texture, level, xoffset, width, format, imageSize, data)
+DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage1D, GLuint texture, GLint level, GLint xoffset, GLsizei width, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage1D, texture, level, xoffset, width, format, imageSize, data)
 DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage2D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLsizei width, GLsizei height, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage2D, texture, level, xoffset, yoffset, width, height, format, imageSize, data)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, imageSize, data)
+DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize, const void* data) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, imageSize, data)
 DECLARE_GL_FUNCTION_HEAD(void, CopyTextureSubImage1D, GLuint texture, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CopyTextureSubImage1D, texture, level, xoffset, x, y, width)
 DECLARE_GL_FUNCTION_HEAD(void, CopyTextureSubImage2D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width, GLsizei height) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CopyTextureSubImage2D, texture, level, xoffset, yoffset, x, y, width, height)
 DECLARE_GL_FUNCTION_HEAD(void, CopyTextureSubImage3D, GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLint x, GLint y, GLsizei width, GLsizei height) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CopyTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, x, y, width, height)
@@ -1835,9 +1835,9 @@ DECLARE_GL_FUNCTION_STUB_HEAD(void, GetBooleanIndexedvEXT, GLenum target, GLuint
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureImage3DEXT, GLuint texture, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureImage3DEXT, texture, target, level, internalformat, width, height, depth, border, imageSize, bits)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureImage2DEXT, GLuint texture, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureImage2DEXT, texture, target, level, internalformat, width, height, border, imageSize, bits)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureImage1DEXT, GLuint texture, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureImage1DEXT, texture, target, level, internalformat, width, border, imageSize, bits)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage3DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage3DEXT, texture, target, level, xoffset, yoffset, zoffset, width, height, depth, format, imageSize, bits)
+DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage3DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage3D, texture, level, xoffset, yoffset, zoffset, width, height, depth, format, imageSize, bits)
 DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage2DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLsizei width, GLsizei height, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage2D, texture, level, xoffset, yoffset, width, height, format, imageSize, bits)
-DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedTextureSubImage1DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLsizei width, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedTextureSubImage1DEXT, texture, target, level, xoffset, width, format, imageSize, bits)
+DECLARE_GL_FUNCTION_HEAD(void, CompressedTextureSubImage1DEXT, GLuint texture, GLenum target, GLint level, GLint xoffset, GLsizei width, GLenum format, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_END_NO_RETURN(void, CompressedTextureSubImage1D, texture, level, xoffset, width, format, imageSize, bits)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, GetCompressedTextureImageEXT, GLuint texture, GLenum target, GLint lod, void* img) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, GetCompressedTextureImageEXT, texture, target, lod, img)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedMultiTexImage3DEXT, GLenum texunit, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height, GLsizei depth, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedMultiTexImage3DEXT, texunit, target, level, internalformat, width, height, depth, border, imageSize, bits)
 DECLARE_GL_FUNCTION_STUB_HEAD(void, CompressedMultiTexImage2DEXT, GLenum texunit, GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height, GLint border, GLsizei imageSize, const void* bits) DECLARE_GL_FUNCTION_STUB_END_NO_RETURN(void, CompressedMultiTexImage2DEXT, texunit, target, level, internalformat, width, height, border, imageSize, bits)

MobileGL/MG_Impl/GLImpl/Sync/GL_Sync.cpp

@@ -69,8 +69,25 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     GLenum ClientWaitSync(GLsync sync, GLbitfield flags, GLuint64 timeout) {
+        // GL 4.6 core 4.1.1: GL_SYNC_FLUSH_COMMANDS_BIT is the only bit this call accepts, and
+        // any other bit is INVALID_VALUE. Silently ignoring the stray bits used to make a caller
+        // that passed, say, GL_SYNC_GPU_COMMANDS_COMPLETE by mistake think it had asked for a
+        // flush it never got.
+        if ((flags & ~static_cast<GLbitfield>(GL_SYNC_FLUSH_COMMANDS_BIT)) != 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "flags must be zero or GL_SYNC_FLUSH_COMMANDS_BIT."));
+            return GL_WAIT_FAILED;
+        }
         const auto* syncObject = FindSyncObject(sync);
         if (!syncObject) {
+            // The spec pairs the GL_WAIT_FAILED return with a recorded INVALID_VALUE; returning
+            // the enum alone left glGetError() clean and the failure indistinguishable from a
+            // genuine wait failure on a live sync.
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "sync is not the name of a sync object."));
             return GL_WAIT_FAILED;
         }
         const auto backendClientWaitSync = MG_Backend::gBackendFunctionsTable.GL.ClientWaitSync;
@@ -95,6 +112,9 @@ namespace MobileGL::MG_Impl::GLImpl {
         }
         const auto* syncObject = FindSyncObject(sync);
         if (!syncObject) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "sync is not the name of a sync object."));
             return;
         }
         const auto backendWaitSync = MG_Backend::gBackendFunctionsTable.GL.WaitSync;
@@ -125,8 +145,22 @@ namespace MobileGL::MG_Impl::GLImpl {
     }
 
     void GetSynciv(GLsync sync, GLenum pname, GLsizei bufSize, GLsizei* length, GLint* values) {
+        // GL 4.6 core 4.1: a negative bufSize is INVALID_VALUE, an unnamed sync is INVALID_VALUE
+        // and an unrecognised pname is INVALID_ENUM. All three used to leave glGetError() clean
+        // and write a plausible-looking zero, which is the one failure mode a caller cannot tell
+        // apart from a real answer - GL_SYNC_STATUS legitimately answers GL_UNSIGNALED (0x9118),
+        // but a mistyped pname answered a bare 0 that no query ever returns.
+        if (bufSize < 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "bufSize must not be negative."));
+            return;
+        }
         const auto* syncObject = FindSyncObject(sync);
         if (!syncObject) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "sync is not the name of a sync object."));
             if (length) {
                 *length = 0;
             }
@@ -152,7 +186,15 @@ namespace MobileGL::MG_Impl::GLImpl {
             value = static_cast<GLint>(syncObject->flags);
             break;
         default:
-            break;
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "pname must be GL_OBJECT_TYPE, GL_SYNC_STATUS, GL_SYNC_CONDITION or "
+                                             "GL_SYNC_FLAGS."));
+            if (length) {
+                *length = 0;
+            }
+            return;
         }
 
         if (length) {

MobileGL/MG_Impl/GLImpl/Texture/GL_Texture.cpp

@@ -4099,11 +4099,171 @@ namespace MobileGL::MG_Impl::GLImpl {
                                          "1D textures are not supported by this implementation"));
     }
 
+    // The three-dimensional twin of CompressedTexSubImage2D_State: a block-aligned box of the
+    // compressed image the level shadows is replaced, slice by slice. Same deviation as the 2D form
+    // - the uncompressed texel shadow beside it is NOT touched, so what changes is the image
+    // glGetCompressedTexImage hands back, not what the level samples as.
     void CompressedTexSubImage3D_State(GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint zoffset,
                                        GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize,
                                        const void* data) {
-        // TODO: implement compressed upload - see CompressedTexImage2D_State.
+        // ======================= Converting ================================
-        RecordUnsupportedCompressedFormat(__func__);
+        const auto textureUploadTarget = MG_Util::ConvertGLEnumToTextureUploadTarget(target);
+        const auto textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
+        // Zero block width doubles as "format is not a specific compressed format", the
+        // INVALID_ENUM case - one lookup answers both questions.
+        const auto compressedInfo = MG_Util::GetCompressedFormatInfo(format);
+
+        // ===================== Error Checking ==============================
+        if (!TextureImpl::ValidateTextureUploadTarget(textureUploadTarget)) return;
+        // A proxy holds no image to modify; only the glTexImage*/glCompressedTexImage* pair
+        // accepts one.
+        if (TextureImpl::IsProxyTextureTarget(textureUploadTarget)) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidEnum,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "A proxy target has no texture image to modify."));
+            return;
+        }
+        if (!TextureImpl::ValidateTextureLevelNumber(level)) return;
+        if (!TextureImpl::ValidateTextureLevelWithUploadTarget(textureUploadTarget, level)) return;
+        if (width < 0 || height < 0 || depth < 0) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "width, height and depth must be non-negative."));
+            return;
+        }
+        if (compressedInfo.blockWidth == 0) {
+            RecordUnsupportedCompressedFormat(__func__);
+            return;
+        }
+
+        auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+        if (!TextureImpl::ValidateTextureObject(textureObject)) return;
+        auto* textureMipmapObject = MG_State::GLState::AsMipmapTexture(textureObject.get());
+        if (textureMipmapObject == nullptr) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__, "Texture storage is not mipmap-backed."));
+            return;
+        }
+        // GL 4.6 core 8.7: INVALID_OPERATION unless the image being modified is stored in
+        // exactly this compressed format. That is also what makes the block arithmetic below
+        // sound - the level's grid is measured with THIS format's block size.
+        const GLenum levelFormat =
+            textureMipmapObject->GetMipmapCompressedFormat(textureUploadTarget, static_cast<Uint>(level));
+        if (levelFormat != format) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "format does not match the internal format of the texture image."));
+            return;
+        }
+
+        const IntVec3 levelSize = textureMipmapObject->GetMipmapTexelSize(textureUploadTarget, static_cast<Uint>(level));
+        const Int levelDepth = std::max(levelSize.z(), 1);
+        // Subtractions rather than sums for the reason CompressedTexSubImage2D_State spells out:
+        // offset + extent are both application-supplied GLints and a signed overflow is undefined.
+        if (xoffset < 0 || yoffset < 0 || zoffset < 0 || width > levelSize.x() - xoffset ||
+            height > levelSize.y() - yoffset || depth > levelDepth - zoffset) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "The replaced region does not lie within the texture image."));
+            return;
+        }
+        // GL 4.6 core 8.7 for block-based formats: the region must start on a block boundary
+        // and must either be a whole number of blocks wide/high or run to the image's edge. Every
+        // format that reaches here is 4x4x1, so the depth axis carries no block alignment rule -
+        // each slice is its own block grid.
+        const Int blockWidth = static_cast<Int>(compressedInfo.blockWidth);
+        const Int blockHeight = static_cast<Int>(compressedInfo.blockHeight);
+        const Bool alignedX = (xoffset % blockWidth == 0) &&
+                              (width % blockWidth == 0 || xoffset + width == levelSize.x());
+        const Bool alignedY = (yoffset % blockHeight == 0) &&
+                              (height % blockHeight == 0 || yoffset + height == levelSize.y());
+        if (!alignedX || !alignedY) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "The replaced region is not aligned to the format's compressed blocks."));
+            return;
+        }
+        // Exactly the size the format and dimensions imply, which is also what keeps the copy
+        // below in bounds.
+        const SizeT expectedImageSize =
+            MG_Util::CalculateCompressedTextureImageSize(compressedInfo, {width, height, depth});
+        if (imageSize < 0 || static_cast<SizeT>(imageSize) != expectedImageSize) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "imageSize does not match the compressed image size."));
+            return;
+        }
+
+        // ======================= Processing ================================
+        if (!ValidateCompressedUnpackBufferSource(data, expectedImageSize, __func__)) return;
+        const void* compressedBytes = CompressedUnpackSource(data);
+        if (expectedImageSize == 0) return; // a zero-sized region is a legal no-op
+        if (compressedBytes == nullptr) {
+            // No unpack buffer and a null client pointer: there is nothing to read. GL leaves
+            // this undefined rather than erroring, and dereferencing it is the one answer that
+            // is never acceptable.
+            MGLOG_D("%s: null data with no pixel unpack buffer bound, nothing to replace", __func__);
+            return;
+        }
+
+        static std::atomic<Bool> announcedNoCodec3D{false};
+        if (!announcedNoCodec3D.exchange(true)) {
+            MGLOG_W("%s: the compressed blocks are stored verbatim and returned by "
+                    "glGetCompressedTexImage, but there is no BC/ETC decoder here, so they do not "
+                    "reach the texels this level SAMPLES as. Upload through glTexSubImage3D for "
+                    "that.",
+                    __func__);
+        }
+
+        const SizeT blobSize =
+            textureMipmapObject->GetMipmapCompressedByteSize(textureUploadTarget, static_cast<Uint>(level));
+        const void* existing =
+            textureMipmapObject->MapMipmapCompressedImage(textureUploadTarget, static_cast<Uint>(level));
+        if (blobSize == 0 || existing == nullptr) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidOperation,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "The texture level holds no compressed image to modify."));
+            return;
+        }
+        Vector<Uint8> blob(blobSize);
+        Memcpy(blob.data(), existing, blobSize);
+
+        const SizeT blockByteSize = compressedInfo.blockByteSize;
+        const SizeT levelBlocksX = (static_cast<SizeT>(levelSize.x()) + compressedInfo.blockWidth - 1) /
+                                   compressedInfo.blockWidth;
+        const SizeT levelBlocksY = (static_cast<SizeT>(levelSize.y()) + compressedInfo.blockHeight - 1) /
+                                   compressedInfo.blockHeight;
+        const SizeT levelRowBytes = levelBlocksX * blockByteSize;
+        const SizeT levelSliceBytes = levelRowBytes * levelBlocksY;
+        const SizeT regionBlocksX = (static_cast<SizeT>(width) + compressedInfo.blockWidth - 1) /
+                                    compressedInfo.blockWidth;
+        const SizeT regionBlocksY = (static_cast<SizeT>(height) + compressedInfo.blockHeight - 1) /
+                                    compressedInfo.blockHeight;
+        const SizeT firstBlockX = static_cast<SizeT>(xoffset) / compressedInfo.blockWidth;
+        const SizeT firstBlockY = static_cast<SizeT>(yoffset) / compressedInfo.blockHeight;
+        const SizeT regionRowBytes = regionBlocksX * blockByteSize;
+        const SizeT regionSliceBytes = regionRowBytes * regionBlocksY;
+        const auto* source = static_cast<const Uint8*>(compressedBytes);
+        for (SizeT slice = 0; slice < static_cast<SizeT>(depth); ++slice) {
+            const SizeT destSliceBase = (static_cast<SizeT>(zoffset) + slice) * levelSliceBytes;
+            for (SizeT row = 0; row < regionBlocksY; ++row) {
+                const SizeT destOffset =
+                    destSliceBase + (firstBlockY + row) * levelRowBytes + firstBlockX * blockByteSize;
+                if (destOffset + regionRowBytes > blobSize) break; // a level whose blob predates its size
+                Memcpy(blob.data() + destOffset, source + slice * regionSliceBytes + row * regionRowBytes,
+                       regionRowBytes);
+            }
+        }
+        textureMipmapObject->SetMipmapCompressedImage(textureUploadTarget, static_cast<Uint>(level), format,
+                                                      blob.data(), blobSize);
     }
 
     // Replaces a block-aligned rectangle of the compressed image glCompressedTexImage2D (or a
@@ -4278,15 +4438,83 @@ namespace MobileGL::MG_Impl::GLImpl {
         RecordUnsupportedCompressedFormat(__func__);
     }
 
+    // The three-dimensional twin of CompressedTexImage2D_State, and the same deviation applies: the
+    // blocks are shadowed verbatim for glGetCompressedTexImage while the texels this level SAMPLES
+    // as stay zero, because there is no BC/ETC decoder here. A 3D compressed image is a stack of
+    // `depth` two-dimensional block grids - every format that reaches here has a 4x4x1 block - so
+    // the blob layout is slice-major and CalculateCompressedTextureImageSize already multiplies by
+    // depth.
     void CompressedTexImage3D_State(GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height,
                                     GLsizei depth, GLint border, GLsizei imageSize, const void* data) {
+        // ======================= Converting ================================
         const auto textureUploadTarget = MG_Util::ConvertGLEnumToTextureUploadTarget(target);
         const auto textureTarget = MG_Util::ConvertGLEnumToTextureTarget(target);
-        auto& textureObject = GetTextureObjectByTarget(textureUploadTarget, textureTarget);
+        // Zero block width doubles as "internalformat is not a specific compressed format", which is
+        // the INVALID_ENUM case - one lookup answers both questions.
+        const auto compressedInfo = MG_Util::GetCompressedFormatInfo(internalformat);
+
+        // ===================== Error Checking ==============================
+        if (!TextureImpl::ValidateTextureUploadTarget(textureUploadTarget)) return;
+        if (!TextureImpl::ValidateTextureLevelNumber(level)) return;
+        if (!TextureImpl::ValidateTextureSizeWithTextureUploadTarget(textureUploadTarget, width, height)) return;
+        if (!TextureImpl::ValidateTextureSizeRange(width, height, depth)) return;
+        if (!TextureImpl::ValidateTextureBorderNumber(border)) return;
+        if (!TextureImpl::ValidateTextureLevelWithUploadTarget(textureUploadTarget, level)) return;
+        if (compressedInfo.blockWidth == 0) {
+            RecordUnsupportedCompressedFormat(__func__);
+            return;
+        }
+        // GL 4.6 core 8.7: imageSize must be exactly the size the format and dimensions imply,
+        // otherwise INVALID_VALUE. This is also the guard that keeps the copy below in bounds.
+        const SizeT expectedImageSize =
+            MG_Util::CalculateCompressedTextureImageSize(compressedInfo, {width, height, depth});
+        if (imageSize < 0 || static_cast<SizeT>(imageSize) != expectedImageSize) {
+            MG_State::pGLContext->RecordError(
+                ErrorCode::InvalidValue,
+                MakeUnique<GenericErrorInfo>("MG_Impl/GLImpl", __func__,
+                                             "imageSize does not match the compressed image size."));
+            return;
+        }
+
+        // Object resolution copied from TexImage3D_State rather than routed through
+        // GetTextureObjectByTarget, for the reason CompressedTexImage2D_State gives: a proxy target
+        // is legal here and only CreateOrReplaceProxyTextureObject gives it an object to answer the
+        // level queries from.
+        auto& activeUnit = MG_State::pGLContext->GetTextureUnitObject(MG_State::pGLContext->GetActiveTextureUnit());
+        auto& bindingSlot = activeUnit.GetBindingSlot(textureTarget);
+        const Bool isProxy = TextureImpl::IsProxyTextureTarget(textureUploadTarget);
+        auto& textureObject =
+            isProxy ? TextureImpl::pProxyTextureManager->CreateOrReplaceProxyTextureObject(textureUploadTarget)
+                    : bindingSlot.GetBoundObject();
+        if (!TextureImpl::ValidateTextureObject(textureObject)) return;
         if (!ValidateTextureMutable(textureObject, __func__)) return;
 
-        // TODO: implement compressed upload - see CompressedTexImage2D_State.
+        // ======================= Processing ================================
-        RecordUnsupportedCompressedFormat(__func__);
+        const TextureInternalFormat textureInternalFormat =
+            MG_Util::ConvertGLEnumToTextureInternalFormat(internalformat);
+        textureObject->SetInternalFormat(textureInternalFormat);
+
+        // A proxy records the format and nothing else - it must never take storage, and it must never
+        // be tagged compressed, or GL_TEXTURE_COMPRESSED_IMAGE_SIZE on a proxy would stop being
+        // INVALID_OPERATION.
+        if (isProxy) return;
+
+        const SizeT internalBpp =
+            MG_Util::GetInternalBytesPerPixel(textureInternalFormat, TexturePixelDataType::UnsignedByte);
+        const SizeT internalBytes =
+            static_cast<SizeT>(width) * static_cast<SizeT>(height) * static_cast<SizeT>(depth) * internalBpp;
+
+        auto* textureMipmapObject = static_cast<MG_State::GLState::TextureObjectMipmap*>(textureObject.get());
+        DiscardMipmapChainOnBaseRespecification(textureMipmapObject, textureUploadTarget, level);
+        // AllocateStorage clears any compressed image the level used to hold, so this must run before
+        // SetMipmapCompressedImage re-arms it.
+        textureMipmapObject->AllocateStorage(textureUploadTarget, level, {{width, height, depth}, internalBytes});
+
+        if (!ValidateCompressedUnpackBufferSource(data, expectedImageSize, __func__)) return;
+        const void* compressedBytes = CompressedUnpackSource(data);
+        textureMipmapObject->SetMipmapCompressedImage(textureUploadTarget, level, internalformat, compressedBytes,
+                                                      expectedImageSize);
+        textureMipmapObject->MarkStorageDirty(textureUploadTarget, level, true);
     }
 
     void CompressedTexImage2D_State(GLenum target, GLint level, GLenum internalformat, GLsizei width, GLsizei height,
@@ -5521,6 +5749,14 @@ namespace MobileGL::MG_Impl::GLImpl {
         free(processedPixels);
     }
 
+    void CompressedTextureSubImage1D(GLuint texture, GLint level, GLint xoffset, GLsizei width, GLenum format,
+                                     GLsizei imageSize, const void* data) {
+        auto textureObject = GetTextureObjectByName(texture, __func__);
+        WithTemporarilyBoundNamedTexture(textureObject, [&](GLenum target) {
+            CompressedTexSubImage1D_State(target, level, xoffset, width, format, imageSize, data);
+        });
+    }
+
     void CompressedTextureSubImage2D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLsizei width,
                                      GLsizei height, GLenum format, GLsizei imageSize, const void* data) {
         auto textureObject = GetTextureObjectByName(texture, __func__);
@@ -5529,6 +5765,16 @@ namespace MobileGL::MG_Impl::GLImpl {
         });
     }
 
+    void CompressedTextureSubImage3D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset,
+                                     GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize,
+                                     const void* data) {
+        auto textureObject = GetTextureObjectByName(texture, __func__);
+        WithTemporarilyBoundNamedTexture(textureObject, [&](GLenum target) {
+            CompressedTexSubImage3D_State(target, level, xoffset, yoffset, zoffset, width, height, depth, format,
+                                          imageSize, data);
+        });
+    }
+
     void TextureSubImage3D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width,
                            GLsizei height, GLsizei depth, GLenum format, GLenum type, const void* pixels) {
         auto textureObject = GetTextureObjectByName(texture, __func__);

MobileGL/MG_Impl/GLImpl/Texture/GL_Texture.h

@@ -37,8 +37,13 @@ namespace MobileGL::MG_Impl::GLImpl {
                            GLenum format, GLenum type, const void* pixels);
     void TextureSubImage3D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset, GLsizei width,
                            GLsizei height, GLsizei depth, GLenum format, GLenum type, const void* pixels);
+    void CompressedTextureSubImage1D(GLuint texture, GLint level, GLint xoffset, GLsizei width, GLenum format,
+                                     GLsizei imageSize, const void* data);
     void CompressedTextureSubImage2D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLsizei width,
                                      GLsizei height, GLenum format, GLsizei imageSize, const void* data);
+    void CompressedTextureSubImage3D(GLuint texture, GLint level, GLint xoffset, GLint yoffset, GLint zoffset,
+                                     GLsizei width, GLsizei height, GLsizei depth, GLenum format, GLsizei imageSize,
+                                     const void* data);
     void TextureParameterf(GLuint texture, GLenum pname, GLfloat param);
     void TextureParameterfv(GLuint texture, GLenum pname, const GLfloat* params);
     void TextureParameteri(GLuint texture, GLenum pname, GLint param);

MobileGL/MG_IntegrationTest/Scenarios/AtomicCounterScenario.cpp

@@ -87,11 +87,6 @@ void main() {
                                  << " and GL_MAX_COMPUTE_ATOMIC_COUNTER_BUFFERS is " << buffers
                                  << "; this needs 3 and 2";
                 }
-                if (!AtomicCountersAreWired()) {
-                    GTEST_SKIP() << "atomic counter buffers are not wired up on " << Gl().BackendName()
-                                 << " yet: glslang lowers them onto a storage block and that block's descriptor "
-                                 << "is still resolved from the shader-storage binding points";
-                }
                 m_program = CompileComputeProgram(kCounterComputeSource);
                 ASSERT_NE(m_program, 0u) << m_buildLog;
             }
@@ -105,12 +100,6 @@ void main() {
                 m_program = 0;
             }
 
-            // Magma binds the lowered block as an ordinary storage-buffer descriptor resolved
-            // from GL_SHADER_STORAGE_BUFFER point N, so the counter buffer never reaches it. The
-            // frontend half (limits, reflection queries, the link-time offset rules) is
-            // backend-agnostic and is covered by the unit suites; only the VALUE is scoped here.
-            bool AtomicCountersAreWired() const { return Gl().BackendName() != "DirectVulkan"; }
-
             unsigned int CompileComputeProgram(const char* source) {
                 const GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
                 glShaderSource(shader, 1, &source, nullptr);

MobileGL/MG_State/GLState/BufferState/BufferObject.cpp

@@ -118,7 +118,13 @@ namespace MobileGL::MG_State::GLState {
         // record that so backends skip uploading the stale shadow bytes.
         m_hasDefinedContent = (data != nullptr) || size == 0;
         m_isImmutableStorage = false;
-        m_storageFlags = 0;
+        // GL 4.6 core 6.2 defines glBufferData as glBufferStorage with
+        // DYNAMIC_STORAGE_BIT | MAP_READ_BIT | MAP_WRITE_BIT, so GL_BUFFER_STORAGE_FLAGS has to
+        // report those three afterwards. Reporting 0 - the value that belongs to a buffer whose
+        // store has never been specified - told an application that a perfectly writable
+        // glBufferData buffer accepted neither glBufferSubData nor a map. Only the IMMUTABLE flag
+        // distinguishes the two cases, and it is cleared just above.
+        m_storageFlags = GL_DYNAMIC_STORAGE_BIT | GL_MAP_READ_BIT | GL_MAP_WRITE_BIT;
         NotifyRespecify();
     }
 

MobileGL/MG_Test/BackendLoader/BackendLoaderTest.cpp

@@ -1087,22 +1087,76 @@ TEST(TextureAnisotropyCapabilities, ExtensionIsAdvertisedOnlyWhenTheHostDriverSu
         return std::find(extensions.begin(), extensions.end(), wanted) != extensions.end();
     };
 
-    const auto without = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false);
+    const auto without = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, false);
     EXPECT_FALSE(contains(without, MobileGL::E_GL_EXT_texture_filter_anisotropic));
     EXPECT_FALSE(contains(without, MobileGL::E_GL_ARB_texture_filter_anisotropic));
 
-    const auto with = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, true, false, false, false);
+    const auto with = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, true, false, false, false, false);
     EXPECT_TRUE(contains(with, MobileGL::E_GL_EXT_texture_filter_anisotropic));
     EXPECT_TRUE(contains(with, MobileGL::E_GL_ARB_texture_filter_anisotropic));
 
     // Same rule on the Vulkan backend, where the gate is the samplerAnisotropy device feature.
-    const auto vkWithout = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false);
+    const auto vkWithout = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, false);
     EXPECT_FALSE(contains(vkWithout, MobileGL::E_GL_EXT_texture_filter_anisotropic));
-    const auto vkWith = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, true, false);
+    const auto vkWith = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, true, false, false);
     EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_EXT_texture_filter_anisotropic));
     EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_ARB_texture_filter_anisotropic));
 }
 
+// Cube map arrays are core at the version MobileGL claims, but there is nothing underneath on a
+// pre-ES-3.2 driver without EXT/OES_texture_cube_map_array, and no VK_IMAGE_VIEW_TYPE_CUBE_ARRAY
+// without the imageCubeArray feature. The string has to follow the capability on both backends -
+// and it has to BE there when the capability is, because KHR-GL4*.texture_gather.*-cube-array
+// gates on the string with no core-version fallback.
+TEST(CubeMapArrayAdvertisement, FollowsTheHostCapabilityOnBothBackends) {
+    const auto contains = [](const MobileGL::Vector<MobileGL::GLExtension>& extensions,
+                             MobileGL::GLExtension wanted) {
+        return std::find(extensions.begin(), extensions.end(), wanted) != extensions.end();
+    };
+
+    const auto esWithout =
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, false);
+    EXPECT_FALSE(contains(esWithout, MobileGL::E_GL_ARB_texture_cube_map_array));
+    const auto esWith =
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, true);
+    EXPECT_TRUE(contains(esWith, MobileGL::E_GL_ARB_texture_cube_map_array));
+
+    const auto vkWithout = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false,
+                                                                                         false);
+    EXPECT_FALSE(contains(vkWithout, MobileGL::E_GL_ARB_texture_cube_map_array));
+    const auto vkWith = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, true);
+    EXPECT_TRUE(contains(vkWith, MobileGL::E_GL_ARB_texture_cube_map_array));
+}
+
+// The core-plumbing strings carry no capability gate: they name entry points that have been real
+// on both backends for as long as the backends have existed, and an application that gates its
+// entry-point resolution on the string (LWJGL does) would otherwise call through null. Pinned
+// together so a future edit cannot quietly drop one, and pinned on BOTH backends so the two
+// cannot disagree about what MobileGL is.
+TEST(CorePlumbingAdvertisement, IsUnconditionalAndIdenticalOnBothBackends) {
+    const auto contains = [](const MobileGL::Vector<MobileGL::GLExtension>& extensions,
+                             MobileGL::GLExtension wanted) {
+        return std::find(extensions.begin(), extensions.end(), wanted) != extensions.end();
+    };
+    const MobileGL::GLExtension expected[] = {
+        MobileGL::E_GL_ARB_sync,          MobileGL::E_GL_ARB_shader_atomic_counters,
+        MobileGL::E_GL_ARB_vertex_array_object, MobileGL::E_GL_ARB_sampler_objects,
+        MobileGL::E_GL_ARB_map_buffer_range,    MobileGL::E_GL_ARB_copy_buffer,
+        MobileGL::E_GL_ARB_copy_image,          MobileGL::E_GL_ARB_texture_swizzle,
+        MobileGL::E_GL_ARB_vertex_type_2_10_10_10_rev, MobileGL::E_GL_ARB_texture_rg,
+        MobileGL::E_GL_ARB_depth_buffer_float,  MobileGL::E_GL_ARB_texture_float,
+        MobileGL::E_GL_ARB_viewport_array};
+
+    // Every gate off: none of these may depend on one.
+    const auto es = MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false,
+                                                                                false);
+    const auto vk = MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, false);
+    for (const auto extension : expected) {
+        EXPECT_TRUE(contains(es, extension)) << "DirectGLES stopped advertising extension " << extension;
+        EXPECT_TRUE(contains(vk, extension)) << "DirectVulkan stopped advertising extension " << extension;
+    }
+}
+
 // Minecraft 26.3 checks ARB_draw_indirect before it considers the already-advertised
 // ARB_multi_draw_indirect, then separately requires ARB_base_instance before enabling its terrain
 // indirect path. Pin both strings and, just as importantly, the non-zero firstInstance gate.
@@ -1113,27 +1167,27 @@ TEST(IndirectDrawAdvertisement, MatchesEachBackendsUsableCommandSemantics) {
     };
 
     const auto esWithoutIndirect =
-        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false);
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, false);
     EXPECT_FALSE(contains(esWithoutIndirect, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_FALSE(contains(esWithoutIndirect, MobileGL::E_GL_ARB_base_instance));
 
     const auto esWithoutBaseInstance =
-        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, false, false);
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, false, false, false);
     EXPECT_TRUE(contains(esWithoutBaseInstance, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_FALSE(contains(esWithoutBaseInstance, MobileGL::E_GL_ARB_base_instance));
 
     const auto esWithBoth =
-        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, true, false);
+        MobileGL::MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, true, true, false, false);
     EXPECT_TRUE(contains(esWithBoth, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_TRUE(contains(esWithBoth, MobileGL::E_GL_ARB_base_instance));
 
     const auto vkWithoutBaseInstance =
-        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false);
+        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, false);
     EXPECT_TRUE(contains(vkWithoutBaseInstance, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_FALSE(contains(vkWithoutBaseInstance, MobileGL::E_GL_ARB_base_instance));
 
     const auto vkWithBoth =
-        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, true);
+        MobileGL::MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, true, false);
     EXPECT_TRUE(contains(vkWithBoth, MobileGL::E_GL_ARB_draw_indirect));
     EXPECT_TRUE(contains(vkWithBoth, MobileGL::E_GL_ARB_base_instance));
 }

MobileGL/MG_Test/Buffer/BufferTest.cpp

@@ -711,6 +711,62 @@ TEST_F(BufferTest, ClearNamedBufferSubDataRepeatsPattern) {
     EXPECT_EQ(actual, (Vector<Uint32>{0, pattern, pattern, pattern, 0}));
     EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
 }
+
+// GL 4.6 core table 8.2 pairs GL_INT with the non-integer base formats as a signed-normalized
+// source, so a GL_R8 clear whose pattern arrives as (GL_RED, GL_INT) is legal. The pair used to be
+// rejected with INVALID_VALUE, which is the first call
+// KHR-GL45.direct_state_access.buffers_functional makes.
+TEST_F(BufferTest, ClearNamedBufferSubDataAcceptsSignedNormalizedIntPattern) {
+    GLuint buffer = 0;
+    MobileGL::MG_Impl::GLImpl::CreateBuffers(1, &buffer);
+
+    const Vector<Uint8> initial(24, 0x7F);
+    MobileGL::MG_Impl::GLImpl::NamedBufferStorage(
+        buffer, initial.size(), initial.data(),
+        GL_MAP_READ_BIT | GL_MAP_WRITE_BIT | GL_DYNAMIC_STORAGE_BIT | GL_MAP_PERSISTENT_BIT);
+    ASSERT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    const GLint zero = 0;
+    MobileGL::MG_Impl::GLImpl::ClearNamedBufferSubData(buffer, GL_R8, 0, sizeof(GLint), GL_RED, GL_INT, &zero);
+    EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    Vector<Uint8> actual(initial.size());
+    auto bufferObject = MobileGL::MG_State::pGLContext->GetBufferObject(buffer);
+    ASSERT_NE(bufferObject, nullptr);
+    Memcpy(actual.data(), bufferObject->AcquireMemory(false, true, false), actual.size());
+    Vector<Uint8> expected(initial);
+    for (SizeT i = 0; i < sizeof(GLint); ++i) expected[i] = 0;
+    EXPECT_EQ(actual, expected);
+
+    MobileGL::MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    DrainPendingGlErrors();
+}
+
+// The same pair on the bound-target entry point: the DSA and the bound call share
+// ClearBufferRange_State, and a regression in either direction has to show up here too.
+TEST_F(BufferTest, ClearBufferSubDataAcceptsSignedNormalizedIntPattern) {
+    GLuint buffer = 0;
+    MobileGL::MG_Impl::GLImpl::GenBuffers(1, &buffer);
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, buffer);
+
+    const Vector<Uint8> initial(8, 0x7F);
+    MobileGL::MG_Impl::GLImpl::BufferData(GL_ARRAY_BUFFER, initial.size(), initial.data(), GL_STATIC_DRAW);
+    // GL_INT is signed-normalized against 2^31-1, so the maximum maps to a saturated GL_R8 texel.
+    const GLint one = 2147483647;
+    MobileGL::MG_Impl::GLImpl::ClearBufferSubData(GL_ARRAY_BUFFER, GL_R8, 0, 4, GL_RED, GL_INT, &one);
+    EXPECT_EQ(MobileGL::MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    Vector<Uint8> actual(initial.size());
+    auto bufferObject = MobileGL::MG_State::pGLContext->GetBufferObject(buffer);
+    ASSERT_NE(bufferObject, nullptr);
+    Memcpy(actual.data(), bufferObject->AcquireMemory(false, true, false), actual.size());
+    EXPECT_EQ(actual, (Vector<Uint8>{0xFF, 0xFF, 0xFF, 0xFF, 0x7F, 0x7F, 0x7F, 0x7F}));
+
+    MobileGL::MG_Impl::GLImpl::BindBuffer(GL_ARRAY_BUFFER, 0);
+    MobileGL::MG_Impl::GLImpl::DeleteBuffers(1, &buffer);
+    DrainPendingGlErrors();
+}
+
 TEST_F(BufferTest, ClearBufferSubDataInitializesIrisStaticSsboRange) {
     GLuint buffer = 0;
     MobileGL::MG_Impl::GLImpl::GenBuffers(1, &buffer);

MobileGL/MG_Test/Program/ParallelShaderCompileTest.cpp

@@ -516,17 +516,17 @@ TEST_F(ParallelShaderCompileTest, MaxShaderCompilerThreadsIgnoresTheCurrentBudge
 TEST_F(ParallelShaderCompileTest, BothBackendsAdvertiseTheExtensionIffAsyncIsEnabled) {
     {
         const AsyncModeScope async(true);
-        EXPECT_TRUE(Advertises(MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false),
+        EXPECT_TRUE(Advertises(MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, false),
                                E_GL_KHR_parallel_shader_compile));
-        EXPECT_TRUE(Advertises(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false),
+        EXPECT_TRUE(Advertises(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, false),
                                E_GL_KHR_parallel_shader_compile));
     }
     {
         const AsyncModeScope async(false);
-        EXPECT_FALSE(Advertises(MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false),
+        EXPECT_FALSE(Advertises(MG_Backend::DirectGLES::BuildAdvertisedExtensions(false, false, false, false, false, false),
                                 E_GL_KHR_parallel_shader_compile))
             << "MOBILEGL_ASYNC_SHADER_COMPILE=0 must withdraw the extension, not only the threading";
-        EXPECT_FALSE(Advertises(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false),
+        EXPECT_FALSE(Advertises(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(false, false, false, false, false),
                                 E_GL_KHR_parallel_shader_compile))
             << "MOBILEGL_ASYNC_SHADER_COMPILE=0 must withdraw the extension, not only the threading";
     }

MobileGL/MG_Test/Texture/TextureTest.cpp

@@ -2227,6 +2227,175 @@ TEST_F(TextureTest, CompressedTextureSubImage2DModifiesTheNamedTextureOnly) {
     EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
 }
 
+namespace {
+    // 8x8x8 RGTC1: 2x2 blocks of 8 bytes per slice, so a slice is 32 bytes and the stack is 256.
+    constexpr GLsizei kRgtc1Size8x8x8 = 256;
+    constexpr GLsizei kRgtc1Slice8x8 = 32;
+
+    GLuint MakeCompressedRgtc1Texture3D() {
+        GLuint texture = 0;
+        MG_Impl::GLImpl::GenTextures(1, &texture);
+        MG_Impl::GLImpl::BindTexture(GL_TEXTURE_3D, texture);
+        MG_Impl::GLImpl::CompressedTexImage3D(GL_TEXTURE_3D, 0, GL_COMPRESSED_RED_RGTC1, 8, 8, 8, 0, kRgtc1Size8x8x8,
+                                              nullptr);
+        return texture;
+    }
+} // namespace
+
+// glCompressedTexImage3D used to answer GL_INVALID_ENUM to every call, which is what threw
+// KHR-GL45.direct_state_access.textures_compressed_subimage out with an InternalError: the CTS
+// asserts no error on it. A 3D compressed image is a stack of per-slice block grids, and the whole
+// stack has to come back byte for byte.
+TEST_F(TextureTest, CompressedTexImage3DShadowsTheWholeStackForReadback) {
+    Uint8 whole[kRgtc1Size8x8x8];
+    for (Int i = 0; i < kRgtc1Size8x8x8; ++i) whole[i] = static_cast<Uint8>(i);
+
+    GLuint texture = 0;
+    MG_Impl::GLImpl::GenTextures(1, &texture);
+    MG_Impl::GLImpl::BindTexture(GL_TEXTURE_3D, texture);
+    MG_Impl::GLImpl::CompressedTexImage3D(GL_TEXTURE_3D, 0, GL_COMPRESSED_RED_RGTC1, 8, 8, 8, 0, kRgtc1Size8x8x8,
+                                          whole);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    Uint8 stored[kRgtc1Size8x8x8] = {};
+    MG_Impl::GLImpl::GetCompressedTexImage(GL_TEXTURE_3D, 0, stored);
+    EXPECT_EQ(std::memcmp(stored, whole, sizeof(whole)), 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // An imageSize that is not the one the format and the three dimensions imply - the depth axis
+    // is the term a 2D-shaped size calculation would drop.
+    MG_Impl::GLImpl::CompressedTexImage3D(GL_TEXTURE_3D, 0, GL_COMPRESSED_RED_RGTC1, 8, 8, 8, 0, kRgtc1Slice8x8,
+                                          whole);
+    ExpectSingleGlError(GL_INVALID_VALUE);
+}
+
+// Where the incoming blocks land. The box below is one block wide, one block high and two slices
+// deep, starting at block (1,1) of slice 3: an implementation that dropped the slice stride, the
+// block-row term or the block-column term puts them somewhere else, and a full-image write would
+// hide all three.
+TEST_F(TextureTest, CompressedTexSubImage3DPlacesBlocksSliceBySlice) {
+    const GLuint texture = MakeCompressedRgtc1Texture3D();
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    Uint8 zeros[kRgtc1Size8x8x8] = {};
+    MG_Impl::GLImpl::CompressedTexSubImage3D(GL_TEXTURE_3D, 0, 0, 0, 0, 8, 8, 8, GL_COMPRESSED_RED_RGTC1,
+                                             kRgtc1Size8x8x8, zeros);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    const Uint8 box[16] = {0xA0, 0xA1, 0xA2, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7,
+                           0xB0, 0xB1, 0xB2, 0xB3, 0xB4, 0xB5, 0xB6, 0xB7};
+    MG_Impl::GLImpl::CompressedTexSubImage3D(GL_TEXTURE_3D, 0, 4, 4, 3, 4, 4, 2, GL_COMPRESSED_RED_RGTC1,
+                                             static_cast<GLsizei>(sizeof(box)), box);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    Uint8 expected[kRgtc1Size8x8x8] = {};
+    // slice 3, block row 1, block column 1 -> 3*32 + 1*16 + 1*8, and the same place one slice on.
+    std::memcpy(expected + 3 * kRgtc1Slice8x8 + 16 + 8, box, 8);
+    std::memcpy(expected + 4 * kRgtc1Slice8x8 + 16 + 8, box + 8, 8);
+
+    Uint8 stored[kRgtc1Size8x8x8] = {};
+    MG_Impl::GLImpl::GetCompressedTexImage(GL_TEXTURE_3D, 0, stored);
+    EXPECT_EQ(std::memcmp(stored, expected, sizeof(expected)), 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+// glCompressedTextureSubImage3D was an exported no-op that raised no error at all. It must reach the
+// NAMED texture and leave the binding it borrowed exactly as it found it.
+TEST_F(TextureTest, CompressedTextureSubImage3DModifiesTheNamedTextureOnly) {
+    const GLuint bound = MakeCompressedRgtc1Texture3D();
+    Uint8 boundImage[kRgtc1Size8x8x8];
+    std::memset(boundImage, 0x11, sizeof(boundImage));
+    MG_Impl::GLImpl::CompressedTexSubImage3D(GL_TEXTURE_3D, 0, 0, 0, 0, 8, 8, 8, GL_COMPRESSED_RED_RGTC1,
+                                             kRgtc1Size8x8x8, boundImage);
+
+    const GLuint named = MakeCompressedRgtc1Texture3D();
+    MG_Impl::GLImpl::BindTexture(GL_TEXTURE_3D, bound); // `named` is NOT the bound texture
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    Uint8 namedImage[kRgtc1Size8x8x8];
+    std::memset(namedImage, 0x22, sizeof(namedImage));
+    MG_Impl::GLImpl::CompressedTextureSubImage3D(named, 0, 0, 0, 0, 8, 8, 8, GL_COMPRESSED_RED_RGTC1,
+                                                 kRgtc1Size8x8x8, namedImage);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    Uint8 stored[kRgtc1Size8x8x8] = {};
+    MG_Impl::GLImpl::GetCompressedTexImage(GL_TEXTURE_3D, 0, stored);
+    EXPECT_EQ(std::memcmp(stored, boundImage, sizeof(stored)), 0);
+
+    MG_Impl::GLImpl::BindTexture(GL_TEXTURE_3D, named);
+    std::memset(stored, 0, sizeof(stored));
+    MG_Impl::GLImpl::GetCompressedTexImage(GL_TEXTURE_3D, 0, stored);
+    EXPECT_EQ(std::memcmp(stored, namedImage, sizeof(stored)), 0);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+}
+
+TEST_F(TextureTest, CompressedTexSubImage3DRejectsTheRegionsGLForbids) {
+    const GLuint texture = MakeCompressedRgtc1Texture3D();
+    Uint8 blocks[kRgtc1Size8x8x8] = {};
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+
+    // A format that is not the one the image is stored in.
+    MG_Impl::GLImpl::CompressedTexSubImage3D(GL_TEXTURE_3D, 0, 0, 0, 0, 8, 8, 8, GL_COMPRESSED_RG_RGTC2, 512, blocks);
+    ExpectSingleGlError(GL_INVALID_OPERATION);
+
+    // A start that is not on a block boundary.
+    MG_Impl::GLImpl::CompressedTexSubImage3D(GL_TEXTURE_3D, 0, 2, 0, 0, 4, 8, 8, GL_COMPRESSED_RED_RGTC1, 128, blocks);
+    ExpectSingleGlError(GL_INVALID_OPERATION);
+
+    // A box that runs past the last slice - the depth bound a 2D-shaped range check never applies.
+    MG_Impl::GLImpl::CompressedTexSubImage3D(GL_TEXTURE_3D, 0, 0, 0, 6, 8, 8, 4, GL_COMPRESSED_RED_RGTC1, 128, blocks);
+    ExpectSingleGlError(GL_INVALID_VALUE);
+
+    (void)texture;
+}
+
+// The DSA name rule the CTS's textures_creation pair does not reach for these two entry points: a
+// name handed out by glGenTextures has no object until it is first bound, so a by-name call on it is
+// INVALID_OPERATION - and, unlike the stub these replaced, it has to SAY so rather than return
+// quietly. A glCreateTextures name is a created object and gets past the name check.
+TEST_F(TextureTest, CompressedTextureSubImage3DRejectsAGeneratedButNeverBoundName) {
+    GLuint generated = 0;
+    MG_Impl::GLImpl::GenTextures(1, &generated);
+    ASSERT_NE(generated, 0u);
+    DrainPendingGlErrors();
+
+    Uint8 blocks[kRgtc1Size8x8x8] = {};
+    MG_Impl::GLImpl::CompressedTextureSubImage3D(generated, 0, 0, 0, 0, 8, 8, 8, GL_COMPRESSED_RED_RGTC1,
+                                                 kRgtc1Size8x8x8, blocks);
+    ExpectSingleGlError(GL_INVALID_OPERATION);
+
+    // A created name is past the name check, so whatever it answers is about the IMAGE (this one
+    // holds none yet), never about the name.
+    GLuint created = 0;
+    MG_Impl::GLImpl::CreateTextures(GL_TEXTURE_3D, 1, &created);
+    ASSERT_EQ(MG_Impl::GLImpl::GetError(), GL_NO_ERROR);
+    MG_Impl::GLImpl::CompressedTextureSubImage3D(created, 0, 0, 0, 0, 8, 8, 8, GL_COMPRESSED_RED_RGTC1,
+                                                 kRgtc1Size8x8x8, blocks);
+    EXPECT_EQ(MG_Impl::GLImpl::GetError(), GL_INVALID_OPERATION)
+        << "a created 3D texture with no compressed image is an image error, not a name error";
+    DrainPendingGlErrors();
+}
+
+// Core GL defines no compressed format for a 1D target, so both the bound and the by-name entry
+// point have to REFUSE the call. The by-name one used to be an exported no-op that raised nothing,
+// which is the one answer an application cannot act on.
+TEST_F(TextureTest, CompressedTextureSubImage1DRefusesLikeTheBoundCall) {
+    GLuint texture = 0;
+    MG_Impl::GLImpl::GenTextures(1, &texture);
+    MG_Impl::GLImpl::BindTexture(GL_TEXTURE_1D, texture);
+    MG_Impl::GLImpl::TexImage1D(GL_TEXTURE_1D, 0, GL_R8, 8, 0, GL_RED, GL_UNSIGNED_BYTE, nullptr);
+    DrainPendingGlErrors();
+
+    Uint8 blocks[16] = {};
+    MG_Impl::GLImpl::CompressedTexSubImage1D(GL_TEXTURE_1D, 0, 0, 8, GL_COMPRESSED_RED_RGTC1,
+                                             static_cast<GLsizei>(sizeof(blocks)), blocks);
+    ExpectSingleGlError(GL_INVALID_ENUM);
+
+    MG_Impl::GLImpl::CompressedTextureSubImage1D(texture, 0, 0, 8, GL_COMPRESSED_RED_RGTC1,
+                                                 static_cast<GLsizei>(sizeof(blocks)), blocks);
+    ExpectSingleGlError(GL_INVALID_ENUM);
+}
+
 TEST_F(TextureTest, CompressedTexSubImage2DRejectsTheRegionsGLForbids) {
     const GLuint texture = MakeCompressedRgtc1Texture8x8();
     Uint8 blocks[kRgtc1Size8x8] = {};

MobileGL/MG_Util/SelfTest/DriverPost.cpp

@@ -1269,7 +1269,7 @@ namespace MobileGL::MG_Util::SelfTest {
                 summary.caps.SupportsDisjointTimerQuery, summary.caps.SupportsTextureFilterAnisotropy,
                 summary.caps.SupportsDrawIndirect,
                 summary.caps.SupportsDrawIndirect && summary.caps.SupportsBaseInstance,
-                summary.caps.SupportsTextureView));
+                summary.caps.SupportsTextureView, summary.caps.SupportsTextureCubeMapArray));
         }
         AppendMobileGLReportedRows(builder, MG_Backend::DirectGLES::GetRendererIdentity(), backendApiVersionString,
                                    advertisedExtensions);
@@ -2009,6 +2009,7 @@ namespace MobileGL::MG_Util::SelfTest {
             Bool samplerAnisotropySupported = false;
             Bool drawIndirectFirstInstanceSupported = false;
             Bool shaderDrawParametersSupported = false;
+            Bool imageCubeArraySupported = false;
         };
     } // namespace
 
@@ -2300,6 +2301,7 @@ namespace MobileGL::MG_Util::SelfTest {
         VkPhysicalDeviceFeatures features{};
         vkGetPhysicalDeviceFeaturesFn(physicalDevice, &features);
         summary.samplerAnisotropySupported = features.samplerAnisotropy == VK_TRUE;
+        summary.imageCubeArraySupported = features.imageCubeArray == VK_TRUE;
         summary.drawIndirectFirstInstanceSupported = features.drawIndirectFirstInstance == VK_TRUE;
         if (features.multiDrawIndirect == VK_TRUE) {
             builder.Pass("multiDrawIndirect", "indirect multi-draw batches run as single native commands");
@@ -2721,7 +2723,8 @@ namespace MobileGL::MG_Util::SelfTest {
                 summary.deviceName, summary.apiVersionString, summary.driverVersionString);
             advertisedExtensions = JoinAdvertisedExtensions(MG_Backend::DirectVulkan::BuildAdvertisedExtensions(
                 summary.shaderSubgroupUsable, summary.timerQueriesSupported, summary.samplerAnisotropySupported,
-                summary.drawIndirectFirstInstanceSupported && summary.shaderDrawParametersSupported));
+                summary.drawIndirectFirstInstanceSupported && summary.shaderDrawParametersSupported,
+                summary.imageCubeArraySupported));
         }
         AppendMobileGLReportedRows(builder, MG_Backend::DirectVulkan::GetRendererIdentity(), backendApiVersionString,
                                    advertisedExtensions);

MobileGL/MG_Util/ShaderTranspiler/Types.h

@@ -20,6 +20,28 @@ namespace MobileGL {
             // buffer, and the trailing number is the only place the GL binding survives.
             inline constexpr const char* ATOMIC_COUNTER_BLOCK_PREFIX = "gl_AtomicCounterBlock";
 
+            // "gl_AtomicCounterBlock_5" -> 5; -1 for any name that is not one of these blocks.
+            // Recovering N from the NAME is not a shortcut, it is the only way: the block reaches
+            // a backend auto-mapped to whatever storage-block slot the IO mapper had free, and
+            // that number has no relation to the GL atomic-counter binding the application asked
+            // for (see TMglGlslIoResolver). A backend that resolves the block from the
+            // shader-storage binding points therefore binds the wrong buffer - or, worse, the
+            // application's own SSBO at the same slot.
+            inline Int AtomicCounterBlockGlBinding(StringView name) {
+                const SizeT prefixLength = StringView(ATOMIC_COUNTER_BLOCK_PREFIX).size();
+                // Needs the prefix, the '_' and at least one digit.
+                if (name.size() <= prefixLength + 1) return -1;
+                if (name.compare(0, prefixLength, ATOMIC_COUNTER_BLOCK_PREFIX) != 0) return -1;
+                if (name[prefixLength] != '_') return -1;
+                Int binding = 0;
+                for (SizeT i = prefixLength + 1; i < name.size(); ++i) {
+                    if (name[i] < '0' || name[i] > '9') return -1;
+                    binding = binding * 10 + (name[i] - '0');
+                    if (binding > 0x0FFFFFFF) return -1; // absurd suffix; treat as not-a-counter
+                }
+                return binding;
+            }
+
             // Atomic-counter limits, in ONE place because GL 4.6 requires glGetIntegerv and the
             // shading language's gl_MaxAtomicCounter* constants to report the same numbers
             // (KHR-GL43.shader_atomic_counters.basic-glsl-built-in compares them directly).

MobileGL/MG_Util/Texture/PixelStoreProcessor.cpp

@@ -442,14 +442,30 @@ namespace MobileGL::MG_Util::PixelStoreProcessor {
                 return packed.fieldCount == mapping.channelCount;
             }
 
+            // GL 4.6 core table 8.2: every unpacked component type pairs with every base format,
+            // with only two exclusions - an integer format takes integer types only, and the two
+            // floating types need a non-integer format. This used to be derived from
+            // GetDirectShadowComponentForType, which answers a different question (is the client
+            // layout byte-identical to some shadow layout) and has no SNorm32 to hand back for
+            // (non-integer format, GL_INT). That legal pair was therefore rejected outright, even
+            // though ConvertUnpackRow decodes it through DecodeComponentToFloat like every other
+            // normalized type - which is what glClearBufferData(GL_R8, GL_RED, GL_INT) needs.
             switch (type) {
             case TexturePixelDataType::UnsignedInt5999Rev:
             case TexturePixelDataType::UnsignedInt101111Rev:
                 return !mapping.isInteger && mapping.channelCount == 3;
-            default: {
+            case TexturePixelDataType::UnsignedByte:
-                ShadowComponent component{};
+            case TexturePixelDataType::Byte:
-                return GetDirectShadowComponentForType(type, mapping.isInteger, component);
+            case TexturePixelDataType::UnsignedShort:
-            }
+            case TexturePixelDataType::Short:
+            case TexturePixelDataType::UnsignedInt:
+            case TexturePixelDataType::Int:
+                return true;
+            case TexturePixelDataType::HalfFloat:
+            case TexturePixelDataType::Float:
+                return !mapping.isInteger;
+            default:
+                return false;
             }
         }