[Fix] (BackendLoader, DirectGLES, DirectVulkan): derive the per-stage shader storage block limits from the backend

MobileGL/MG_Backend/BackendObject.h

@@ -340,6 +340,22 @@ namespace MobileGL {
             Int MaxVertexAttribs = 16;
             Int MaxComputeShaderStorageBlocks = 8;
             Int MaxCombinedShaderStorageBlocks = 32;
+            // Per-stage GL_MAX_*_SHADER_STORAGE_BLOCKS. Zero is a legal answer for the four
+            // non-compute, non-fragment stages and these defaults are the spec minimums, not
+            // placeholders: GL 4.6 table 23.64 and ES 3.2 table 21.44 both set the minimum for
+            // vertex, tessellation control, tessellation evaluation and geometry at 0, and only
+            // fragment (8 in GL, 4 in ES) and compute are guaranteed to have any. Every real ARM
+            // GLES driver takes that allowance - a Mali-G925 reports 0 for all four - so a
+            // backend that cannot honour a graphics-stage storage block MUST report 0 here
+            // rather than a hopeful number. Advertising a non-zero count the driver will refuse
+            // does not make the block work; it only moves the failure from an honest
+            // "unsupported" at query time to a backend link error the frontend never surfaces,
+            // after which every draw with that program silently renders nothing.
+            Int MaxVertexShaderStorageBlocks = 0;
+            Int MaxTessControlShaderStorageBlocks = 0;
+            Int MaxTessEvaluationShaderStorageBlocks = 0;
+            Int MaxGeometryShaderStorageBlocks = 0;
+            Int MaxFragmentShaderStorageBlocks = 8;
             Int MaxComputeUniformBlocks = 12;
             Int MaxComputeWorkGroupInvocations = 128;
             Int MaxShaderStorageBufferBindings = 8;

MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.cpp

@@ -1251,9 +1251,31 @@ namespace MobileGL::MG_Backend::DirectGLES {
                      static_cast<Int>(MG_State::GLState::VertexArrayObject::MAX_VERTEX_ATTRIBS));
         m_dynamicParameters.MaxComputeShaderStorageBlocks = m_GLESCapabilities.MaxComputeShaderStorageBlocks;
         m_dynamicParameters.MaxCombinedShaderStorageBlocks = m_GLESCapabilities.MaxCombinedShaderStorageBlocks;
+        // Per-stage storage-block counts, forwarded from the host driver rather than invented.
+        // A stage the driver cannot serve reports 0, which is a legal answer everywhere these
+        // limits appear (GL 4.6 table 23.64, ES 3.2 table 21.44 - the minimum is 0 for every
+        // graphics stage except fragment) and is the only answer that lets an application take
+        // its own fallback instead of building a program the driver will refuse to link. The
+        // stage limit cannot exceed the combined limit or the number of binding points there
+        // are to bind buffers to, so clamp to both.
+        const auto clampStageStorageBlocks = [this](Int stageLimit) {
+            return std::min({std::max(stageLimit, 0), std::max(m_dynamicParameters.MaxCombinedShaderStorageBlocks, 0),
+                             std::max(m_dynamicParameters.MaxShaderStorageBufferBindings, 0)});
+        };
+        m_dynamicParameters.MaxShaderStorageBufferBindings = m_GLESCapabilities.MaxShaderStorageBufferBindings;
+        m_dynamicParameters.MaxVertexShaderStorageBlocks =
+            clampStageStorageBlocks(m_GLESCapabilities.MaxVertexShaderStorageBlocks);
+        m_dynamicParameters.MaxTessControlShaderStorageBlocks =
+            clampStageStorageBlocks(m_GLESCapabilities.MaxTessControlShaderStorageBlocks);
+        m_dynamicParameters.MaxTessEvaluationShaderStorageBlocks =
+            clampStageStorageBlocks(m_GLESCapabilities.MaxTessEvaluationShaderStorageBlocks);
+        m_dynamicParameters.MaxGeometryShaderStorageBlocks =
+            clampStageStorageBlocks(m_GLESCapabilities.MaxGeometryShaderStorageBlocks);
+        m_dynamicParameters.MaxFragmentShaderStorageBlocks =
+            clampStageStorageBlocks(m_GLESCapabilities.MaxFragmentShaderStorageBlocks);
         m_dynamicParameters.MaxComputeUniformBlocks = m_GLESCapabilities.MaxComputeUniformBlocks;
         m_dynamicParameters.MaxComputeWorkGroupInvocations = m_GLESCapabilities.MaxComputeWorkGroupInvocations;
-        m_dynamicParameters.MaxShaderStorageBufferBindings = m_GLESCapabilities.MaxShaderStorageBufferBindings;
+        // (MaxShaderStorageBufferBindings is assigned above, before the per-stage clamp reads it.)
         // This is the number glGetIntegerv(GL_MAX_TEXTURE_BUFFER_SIZE) hands the application, and
         // on a host without buffer textures it is knowingly a floor MobileGL cannot honour rather
         // than a driver answer (m_GLESCapabilities.MaxTextureBufferSizeIsDriverReported says

MobileGL/MG_Backend/DirectVulkan/BackendObject_DirectVulkan.cpp

@@ -847,6 +847,38 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         m_dynamicParameters.MaxShaderStorageBufferBindings =
             clampLimit("GL_MAX_SHADER_STORAGE_BUFFER_BINDINGS", m_vulkanCaps.MaxShaderStorageBufferBindings,
                        kMaxAdvertisedBufferBlocks);
+        // Per-stage GL_MAX_*_SHADER_STORAGE_BLOCKS. Vulkan has one descriptor limit for every
+        // stage (maxPerStageDescriptorStorageBuffers, which is what MaxComputeShaderStorageBlocks
+        // carries), so the stage limits differ only by whether the stage can have blocks at all.
+        //
+        // Deliberately NOT gated on vertexPipelineStoresAndAtomics, unlike the per-stage image
+        // uniforms below. That gate reads as the obvious one and is wrong here in practice: a
+        // Mali-G925-Immortalis reports vertexPipelineStoresAndAtomics=false (supported AND
+        // enabled) and yet runs all 433 KHR-GL43.constant_expressions.*_tess_* cases correctly
+        // through this backend - those write their result through a storage block declared in a
+        // tessellation stage. Gating would report 0 and turn 433 passing cases into
+        // "unsupported", removing function that demonstrably works.
+        //
+        // The asymmetry with DirectGLES is real and is the point. There, 0 prevents a program
+        // the driver refuses outright at link time; the honest limit converts a silent
+        // wrong-render into a capability an application can route around. Here there is no such
+        // failure to prevent, so the limit stays at what the device can address. If a Vulkan
+        // device is ever found that genuinely rejects such a pipeline, the gate belongs at
+        // pipeline creation where the rejection is observable, not on a feature bit this driver
+        // reports inaccurately.
+        {
+            const Int maxPerStageStorageBlocks =
+                std::min(std::max(m_dynamicParameters.MaxComputeShaderStorageBlocks, 0),
+                         std::min(std::max(m_dynamicParameters.MaxCombinedShaderStorageBlocks, 0),
+                                  std::max(m_dynamicParameters.MaxShaderStorageBufferBindings, 0)));
+            m_dynamicParameters.MaxVertexShaderStorageBlocks = maxPerStageStorageBlocks;
+            m_dynamicParameters.MaxTessControlShaderStorageBlocks = maxPerStageStorageBlocks;
+            m_dynamicParameters.MaxTessEvaluationShaderStorageBlocks = maxPerStageStorageBlocks;
+            // The one hard capability in the set: no geometry stage means no blocks in it.
+            m_dynamicParameters.MaxGeometryShaderStorageBlocks =
+                m_vulkanCaps.SupportsGeometryShader ? maxPerStageStorageBlocks : 0;
+            m_dynamicParameters.MaxFragmentShaderStorageBlocks = maxPerStageStorageBlocks;
+        }
         m_dynamicParameters.MaxTextureBufferSize = clampLimit(
             "GL_MAX_TEXTURE_BUFFER_SIZE", m_vulkanCaps.MaxTextureBufferSize, kMaxAdvertisedTextureBufferSize);
         m_dynamicParameters.TextureBufferOffsetAlignment = m_vulkanCaps.TextureBufferOffsetAlignment;

MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.cpp

@@ -692,8 +692,12 @@ namespace MobileGL::MG_Util::BackendLoader {
             !f.glUnmapBuffer || !f.glMemoryBarrier || !f.glCreateShader || !f.glCreateProgram) {
             return false;
         }
-        GLint maxVertexSsboBlocks = 0;
+        // Read from caps, not re-queried: the per-stage limits are resolved (and their query
-        f.glGetIntegerv(GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, &maxVertexSsboBlocks);
+        // errors drained) before this probe runs, so asking the driver again would be a second
+        // round trip that can disagree with the number MobileGL actually advertises - and, on
+        // the early return below, would leave its own GL_INVALID_ENUM in the queue for the
+        // application's first glGetError to find.
+        const GLint maxVertexSsboBlocks = caps.MaxVertexShaderStorageBlocks;
         if (maxVertexSsboBlocks < 1) {
             // The native indirect machinery cannot read the command buffer from the vertex
             // stage on this driver anyway; assume conforming zero-based gl_InstanceID.
@@ -1036,6 +1040,15 @@ namespace MobileGL::MG_Util::BackendLoader {
         GLint maxVertexAttribs = 16;
         GLint maxComputeShaderStorageBlocks = 8;
         GLint maxCombinedShaderStorageBlocks = 32;
+        // ES 3.2 table 21.44 minimums. Zero for the four graphics stages below fragment is not a
+        // placeholder - it is what the spec permits and what ARM's GLES driver actually reports,
+        // so a probe that never runs (pre-ES 3.2, unsupported pname) leaves behind the truthful
+        // answer rather than an optimistic one.
+        GLint maxVertexShaderStorageBlocks = 0;
+        GLint maxTessControlShaderStorageBlocks = 0;
+        GLint maxTessEvaluationShaderStorageBlocks = 0;
+        GLint maxGeometryShaderStorageBlocks = 0;
+        GLint maxFragmentShaderStorageBlocks = 4;
         GLint maxComputeUniformBlocks = 12;
         GLint maxComputeWorkGroupInvocations = 128;
         GLint maxShaderStorageBufferBindings = 8;
@@ -1127,6 +1140,59 @@ namespace MobileGL::MG_Util::BackendLoader {
             (caps.GLESVersion.Major == 3 && caps.GLESVersion.Minor >= 2)) {
             glesFuncs.glGetIntegerv(GL_MAX_GEOMETRY_IMAGE_UNIFORMS, &maxGeometryImageUniforms);
         }
+        // Per-stage storage-block counts. Deliberately NOT batched with the unconditional probes
+        // above, for the reason GL_MAX_TEXTURE_BUFFER_SIZE is not: the vertex and fragment pnames
+        // are ES 3.1, but the tessellation and geometry ones only exist from ES 3.2 on (or under
+        // EXT_tessellation_shader / EXT_geometry_shader), so on an older context they raise
+        // GL_INVALID_ENUM, leave the local untouched, and - with nothing draining the queue until
+        // some later probe - let that error be misattributed to an unrelated query in between, or
+        // leak into the application's first glGetError.
+        //
+        // A stage whose probe does not run keeps the spec minimum, which for all four graphics
+        // stages is 0. That is the honest answer: DirectGLES emits ESSL 3.10 on an ES 3.1 context,
+        // where those stages do not exist at all.
+        {
+            const auto drainErrors = [&glesFuncs]() {
+                Bool hadError = false;
+                if (glesFuncs.glGetError) {
+                    while (glesFuncs.glGetError() != GL_NO_ERROR) hadError = true;
+                }
+                return hadError;
+            };
+
+            // Isolate from errors raised by the preceding probes so the drain below reports on
+            // these queries only.
+            drainErrors();
+            glesFuncs.glGetIntegerv(GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, &maxVertexShaderStorageBlocks);
+            glesFuncs.glGetIntegerv(GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS, &maxFragmentShaderStorageBlocks);
+            if (drainErrors()) {
+                MGLOG_W("Per-stage shader storage block query failed for the vertex/fragment "
+                        "stages; assuming the ES minimums (vertex 0, fragment 4)");
+                maxVertexShaderStorageBlocks = 0;
+                maxFragmentShaderStorageBlocks = 4;
+            }
+            if (esAtLeast32) {
+                glesFuncs.glGetIntegerv(GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS,
+                                        &maxTessControlShaderStorageBlocks);
+                glesFuncs.glGetIntegerv(GL_MAX_TESS_EVALUATION_SHADER_STORAGE_BLOCKS,
+                                        &maxTessEvaluationShaderStorageBlocks);
+                glesFuncs.glGetIntegerv(GL_MAX_GEOMETRY_SHADER_STORAGE_BLOCKS, &maxGeometryShaderStorageBlocks);
+                if (drainErrors()) {
+                    MGLOG_W("Per-stage shader storage block query failed for the tessellation/"
+                            "geometry stages; assuming the ES minimum of 0");
+                    maxTessControlShaderStorageBlocks = 0;
+                    maxTessEvaluationShaderStorageBlocks = 0;
+                    maxGeometryShaderStorageBlocks = 0;
+                }
+            }
+            // A driver is free to report a negative or nonsensical count into an untouched
+            // out-param; clamp before anything downstream treats it as a capacity.
+            maxVertexShaderStorageBlocks = std::max(maxVertexShaderStorageBlocks, 0);
+            maxTessControlShaderStorageBlocks = std::max(maxTessControlShaderStorageBlocks, 0);
+            maxTessEvaluationShaderStorageBlocks = std::max(maxTessEvaluationShaderStorageBlocks, 0);
+            maxGeometryShaderStorageBlocks = std::max(maxGeometryShaderStorageBlocks, 0);
+            maxFragmentShaderStorageBlocks = std::max(maxFragmentShaderStorageBlocks, 0);
+        }
         glesFuncs.glGetIntegerv(GL_MAX_DRAW_BUFFERS, &maxDrawBuffers);
         glesFuncs.glGetIntegerv(GL_MAX_COLOR_ATTACHMENTS, &maxColorAttachments);
         glesFuncs.glGetIntegerv(GL_MAX_CLIP_DISTANCES, &maxClipDistances);
@@ -1271,6 +1337,11 @@ namespace MobileGL::MG_Util::BackendLoader {
         caps.MaxVertexAttribs = maxVertexAttribs;
         caps.MaxComputeShaderStorageBlocks = maxComputeShaderStorageBlocks;
         caps.MaxCombinedShaderStorageBlocks = maxCombinedShaderStorageBlocks;
+        caps.MaxVertexShaderStorageBlocks = maxVertexShaderStorageBlocks;
+        caps.MaxTessControlShaderStorageBlocks = maxTessControlShaderStorageBlocks;
+        caps.MaxTessEvaluationShaderStorageBlocks = maxTessEvaluationShaderStorageBlocks;
+        caps.MaxGeometryShaderStorageBlocks = maxGeometryShaderStorageBlocks;
+        caps.MaxFragmentShaderStorageBlocks = maxFragmentShaderStorageBlocks;
         caps.MaxComputeUniformBlocks = maxComputeUniformBlocks;
         caps.MaxComputeWorkGroupInvocations = maxComputeWorkGroupInvocations;
         caps.MaxShaderStorageBufferBindings = maxShaderStorageBufferBindings;
@@ -1348,6 +1419,14 @@ namespace MobileGL::MG_Util::BackendLoader {
         MGLOG_I("    GL_MAX_VERTEX_ATTRIBS: %d", caps.MaxVertexAttribs);
         MGLOG_I("    GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS: %d", caps.MaxComputeShaderStorageBlocks);
         MGLOG_I("    GL_MAX_COMBINED_SHADER_STORAGE_BLOCKS: %d", caps.MaxCombinedShaderStorageBlocks);
+        // Worth a line each: a zero here is what stops an application's storage block from ever
+        // working in that stage, and reading it back from an artifact is the difference between
+        // "MobileGL dropped my draw" and "this driver has no SSBOs outside compute".
+        MGLOG_I("    GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS: %d", caps.MaxVertexShaderStorageBlocks);
+        MGLOG_I("    GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS: %d", caps.MaxTessControlShaderStorageBlocks);
+        MGLOG_I("    GL_MAX_TESS_EVALUATION_SHADER_STORAGE_BLOCKS: %d", caps.MaxTessEvaluationShaderStorageBlocks);
+        MGLOG_I("    GL_MAX_GEOMETRY_SHADER_STORAGE_BLOCKS: %d", caps.MaxGeometryShaderStorageBlocks);
+        MGLOG_I("    GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS: %d", caps.MaxFragmentShaderStorageBlocks);
         MGLOG_I("    GL_MAX_COMPUTE_UNIFORM_BLOCKS: %d", caps.MaxComputeUniformBlocks);
         MGLOG_I("    GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS: %d", caps.MaxComputeWorkGroupInvocations);
         MGLOG_I("    GL_MAX_SHADER_STORAGE_BUFFER_BINDINGS: %d", caps.MaxShaderStorageBufferBindings);

MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.h

@@ -1248,6 +1248,17 @@ namespace MobileGL {
             Int MaxVertexAttribs = 16;
             Int MaxComputeShaderStorageBlocks = 8;
             Int MaxCombinedShaderStorageBlocks = 32;
+            // Per-stage GL_MAX_*_SHADER_STORAGE_BLOCKS as the host GLES driver reports them.
+            // The defaults are the ES 3.2 minimums (table 21.44): 0 for every graphics stage
+            // except fragment, which is 4. ES only gained the tessellation and geometry pnames
+            // in 3.2 (or with EXT_tessellation_shader / EXT_geometry_shader), so those two are
+            // queried behind a support check and left at the default otherwise - see
+            // FillInGLESCapabilities.
+            Int MaxVertexShaderStorageBlocks = 0;
+            Int MaxTessControlShaderStorageBlocks = 0;
+            Int MaxTessEvaluationShaderStorageBlocks = 0;
+            Int MaxGeometryShaderStorageBlocks = 0;
+            Int MaxFragmentShaderStorageBlocks = 4;
             Int MaxComputeUniformBlocks = 12;
             Int MaxComputeWorkGroupInvocations = 128;
             Int MaxShaderStorageBufferBindings = 8;