Merge branch 'feat/cts-draw-parameters' into dev

CMakeLists.txt

@@ -279,6 +279,7 @@ set(SOURCE_FILES
     MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/LowerDrawParametersPass.cpp
     MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/PackDoubleVertexInputsPass.cpp
     MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/RebaseInstanceIndexPass.cpp
+    MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/ZeroBaseVertexPass.cpp
     MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/NormalizeRectCoordinatesPass.cpp
     MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/PrivateToEntryLocalPass.cpp
     MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/StripUniformLocationsPass.cpp

MobileGL/MG_Backend/DirectGLES/BackendObject_DirectGLES.cpp

@@ -1002,6 +1002,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
             funcsTable.GL.MultiDrawElementsIndirect = MultiDrawElementsIndirect;
             funcsTable.GL.MultiDrawElementsIndirectCount = MultiDrawElementsIndirectCount;
             funcsTable.GL.MultiDrawArraysIndirect = MultiDrawArraysIndirect;
+            funcsTable.GL.MultiDrawArraysIndirectCount = MultiDrawArraysIndirectCount;
             funcsTable.GL.DrawRangeElementsBaseVertex = DrawRangeElementsBaseVertex;
             funcsTable.GL.DrawRangeElements = DrawRangeElements;
             funcsTable.GL.DrawElementsInstancedBaseVertexBaseInstance = DrawElementsInstancedBaseVertexBaseInstance;

MobileGL/MG_Backend/DirectGLES/DirectGLES.cpp

@@ -2905,11 +2905,40 @@ namespace MobileGL::MG_Backend::DirectGLES {
         }
     }
 
+    void SetCurrentBaseVertex(Int32 baseVertex) {
+        if (const auto program = GetCurrentBackendProgram()) {
+            program->SetBaseVertex(baseVertex);
+        }
+    }
+
     Bool CurrentProgramReadsDrawID() {
         const auto program = GetCurrentBackendProgram();
         return program != nullptr && program->ReadsDrawID();
     }
 
+    Bool CurrentProgramReadsBaseVertex() {
+        const auto program = GetCurrentBackendProgram();
+        return program != nullptr && program->ReadsBaseVertex();
+    }
+
+    // The two questions above, asked from BEFORE PrepareForDraw - where neither can be
+    // answered honestly. GetCurrentBackendProgram only sees a twin that a previous draw
+    // already synced, and a twin from before a relink still carries the previous link's
+    // uniform locations, so "no" there means "not known yet" at least as often as it
+    // means no. The multi-draw compute tier has to decide whether to flatten a batch
+    // before PrepareForDraw runs (its dispatch cannot come after the draw state), and
+    // flattening a batch that turns out to need per-sub-draw values is unrecoverable -
+    // so an unanswerable program counts as needing them.
+    Bool CurrentProgramMayNeedPerSubDrawBuiltins(Bool batchCarriesBaseVertices) {
+        const auto& currentProgram = MG_State::pGLContext->GetProgramForDraw();
+        const auto program = GetCurrentBackendProgram();
+        if (!currentProgram || program == nullptr ||
+            program->GetSyncedLinkVersion() != currentProgram->GetLinkVersion()) {
+            return true;
+        }
+        return program->ReadsDrawID() || (batchCarriesBaseVertices && program->ReadsBaseVertex());
+    }
+
     static Bool SupportsNativeIndirectDraws() {
         const auto& version = g_GLESCapabilities.GLESVersion;
         const Bool esVersionOk = version.Major > 3 || (version.Major == 3 && version.Minor >= 1);
@@ -2947,16 +2976,28 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                                      resource->id);
                 }
             }
+            // gl_BaseVertex has no SSBO view of its own: the command's baseVertex word is read
+            // from the CPU shadow, so a command whose baseVertex a compute shader wrote this
+            // frame is not observable here (baseInstance is, through the view above). Feeding
+            // the stale-but-usually-correct shadow beats leaving the uniform at the previous
+            // draw's value, which is what a program reading gl_BaseVertex saw before.
+            const Bool feedBaseVertex = CurrentProgramReadsBaseVertex();
             for (GLsizei i = 0; i < drawcount; ++i) {
                 const SizeT cmdByteOffset = commandOffset + static_cast<SizeT>(i) * stride;
                 SetCurrentDrawID(static_cast<Uint32>(i));
                 if (paramsBinding >= 0 && backendProgram) {
                     // baseInstance is the 5th word of DrawElementsIndirectCommand.
                     backendProgram->SetBaseInstanceWordIndex(static_cast<Int32>((cmdByteOffset + 16) / 4));
+                    if (feedBaseVertex) {
+                        DrawElementsIndirectCommand cmd{};
+                        std::memcpy(&cmd, commandBytes + static_cast<SizeT>(i) * stride, sizeof(cmd));
+                        SetCurrentBaseVertex(cmd.baseVertex);
+                    }
                 } else {
                     DrawElementsIndirectCommand cmd{};
                     std::memcpy(&cmd, commandBytes + static_cast<SizeT>(i) * stride, sizeof(cmd));
                     SetCurrentBaseInstance(cmd.baseInstance);
+                    SetCurrentBaseVertex(cmd.baseVertex);
                 }
                 g_GLESFuncs.glDrawElementsIndirect(mode, type, reinterpret_cast<const void*>(cmdByteOffset));
             }
@@ -2969,6 +3010,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 }
                 SetCurrentDrawID(static_cast<Uint32>(i));
                 SetCurrentBaseInstance(cmd.baseInstance);
+                SetCurrentBaseVertex(cmd.baseVertex);
                 const auto indexByteOffset = static_cast<SizeT>(cmd.firstIndex) * indexSize;
                 g_GLESFuncs.glDrawElementsInstancedBaseVertex(
                     mode, static_cast<GLsizei>(cmd.count), type, reinterpret_cast<const GLvoid*>(indexByteOffset),
@@ -2977,12 +3019,18 @@ namespace MobileGL::MG_Backend::DirectGLES {
         }
         SetCurrentDrawID(0);
         SetCurrentBaseInstance(0);
+        SetCurrentBaseVertex(0);
     }
 
     static void ExecuteArraysIndirectCommands(GLenum mode, const Uint8* commandBytes, SizeT commandOffset,
                                               const SharedPtr<MG_State::GLState::BufferObject>& drawIndirectBuffer,
                                               GLsizei drawcount, GLsizei stride, const char* label) {
         (void)label;
+        // DrawArraysIndirectCommand has no baseVertex word, so gl_BaseVertex is zero for every
+        // command here. Written BEFORE the draws, not merely restored after them: the previous
+        // draw is what leaves a stale value, and restoring afterwards would only protect the
+        // NEXT draw while these commands ran with the stale one.
+        SetCurrentBaseVertex(0);
         const Bool useNative = drawIndirectBuffer != nullptr && SupportsNativeIndirectDraws();
         if (useNative) {
             const auto backendProgram = GetCurrentBackendProgram();
@@ -3269,7 +3317,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer;
         PrepareForDraw(syncBit);
         CheckPrimitiveRestartSupported(type);
+        SetCurrentBaseVertex(basevertex);
         g_GLESFuncs.glDrawElementsBaseVertex(mode, count, type, indices, basevertex);
+        SetCurrentBaseVertex(0);
     }
 
     void MultiDrawArrays(GLenum mode, const GLint* first, const GLsizei* count, GLsizei drawcount) {
@@ -3279,6 +3329,11 @@ namespace MobileGL::MG_Backend::DirectGLES {
         DrawSyncFlags syncBit = DrawSyncBit::None;
         PrepareForDraw(syncBit);
 
+        // This loop IS the emulation - there is no batched tier for the non-indexed form -
+        // so each sub-draw has to be given its own gl_DrawID here, exactly as the indexed
+        // ladder and the indirect executors do. Without it every sub-draw of a
+        // glMultiDrawArrays read draw index 0.
+        const Bool feedDrawID = CurrentProgramReadsDrawID();
         const auto& currentVAO = MG_State::pGLContext->GetBoundVertexArray();
         for (GLsizei i = 0; i < drawcount; ++i) {
             // Client-side arrays are uploaded per sub-draw range, like the single DrawArrays path.
@@ -3288,8 +3343,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
                     (*backendVAOSlot)->SyncClientSideAttributesForDrawArrays(currentVAO, first[i], count[i]);
                 }
             }
+            if (feedDrawID) SetCurrentDrawID(static_cast<Uint32>(i));
             g_GLESFuncs.glDrawArrays(mode, first[i], count[i]);
         }
+        if (feedDrawID) SetCurrentDrawID(0);
     }
 
     // Both glMultiDrawElements entry points are emulated - ES has neither in core - by the
@@ -3403,6 +3460,15 @@ namespace MobileGL::MG_Backend::DirectGLES {
             return;
         }
 
+        // Both counts are read from the CPU shadow, which a buffer with no shadow does not
+        // have - MappedData() is null there and the reads below would be a null dereference,
+        // not a wrong picture. The DirectVulkan twin declines the same way.
+        if (parameterBuffer->MappedData() == nullptr || drawBuffer->MappedData() == nullptr) {
+            MGLOG_E("MultiDrawElementsIndirectCount skipped: CPU fallback cannot read the parameter or "
+                    "draw-indirect buffer");
+            return;
+        }
+
         Uint32 actualDrawCount = 0;
         std::memcpy(&actualDrawCount, parameterBuffer->MappedData() + drawcount, sizeof(actualDrawCount));
         actualDrawCount = std::min<Uint32>(actualDrawCount, static_cast<Uint32>(maxdrawcount));
@@ -3444,11 +3510,79 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                       drawcount, stride, "MultiDrawArraysIndirect");
     }
 
+    // The non-indexed twin of MultiDrawElementsIndirectCount, and structurally identical to it:
+    // ES has no GL_PARAMETER_BUFFER at all, so the draw count is read from the CPU shadow of the
+    // bound one and the batch degenerates into an ordinary indirect multi-draw of that many
+    // commands. Missing from the backend table until now, which made every
+    // glMultiDrawArraysIndirectCount an INVALID_OPERATION ("backend does not support
+    // indirect-parameter array draws") on DirectGLES while the extension was advertised.
+    void MultiDrawArraysIndirectCount(GLenum mode, const void* indirect, GLintptr drawcount, GLsizei maxdrawcount,
+                                      GLsizei stride) {
+#if MOBILEGL_LOG_ACTIVE_LEVEL <= MOBILEGL_LOG_LEVEL_DEBUG && MOBILEGL_ENABLE_SCOPE_MARKER
+        DebugImpl::OpenGLScopeMarker marker(__func__);
+#endif
+        if (maxdrawcount <= 0) {
+            return;
+        }
+        if (stride == 0) {
+            stride = sizeof(DrawArraysIndirectCommand);
+        }
+        if (stride < static_cast<GLsizei>(sizeof(DrawArraysIndirectCommand))) {
+            MGLOG_E("MultiDrawArraysIndirectCount skipped: stride %d is smaller than command size %zu",
+                    stride, sizeof(DrawArraysIndirectCommand));
+            return;
+        }
+
+        DrawSyncFlags syncBit = DrawSyncBit::IndirectBuffer | DrawSyncBit::Instancing;
+        PrepareForDraw(syncBit);
+
+        auto drawBuffer = MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::DrawIndirect).GetBoundObject();
+        auto parameterBuffer = MG_State::pGLContext->GetBufferBindingSlot(BufferTarget::Parameter).GetBoundObject();
+        if (!drawBuffer) {
+            MGLOG_E("MultiDrawArraysIndirectCount skipped: no GL_DRAW_INDIRECT_BUFFER is bound");
+            return;
+        }
+        if (!parameterBuffer) {
+            MGLOG_E("MultiDrawArraysIndirectCount skipped: no GL_PARAMETER_BUFFER is bound");
+            return;
+        }
+
+        drawBuffer->SyncPersistentMappedRange();
+        parameterBuffer->SyncPersistentMappedRange();
+
+        const SizeT commandOffset = reinterpret_cast<SizeT>(indirect);
+        const SizeT commandBytes = commandOffset + static_cast<SizeT>(stride) * static_cast<SizeT>(maxdrawcount - 1) +
+            sizeof(DrawArraysIndirectCommand);
+        if (commandBytes > drawBuffer->GetSize()) {
+            MGLOG_E("MultiDrawArraysIndirectCount skipped: invalid GL_DRAW_INDIRECT_BUFFER binding or range");
+            return;
+        }
+        if (drawcount < 0 || static_cast<SizeT>(drawcount) + sizeof(Uint32) > parameterBuffer->GetSize()) {
+            MGLOG_E("MultiDrawArraysIndirectCount skipped: invalid GL_PARAMETER_BUFFER binding or range");
+            return;
+        }
+
+        // See the indexed twin: no CPU shadow means no count to read, not a wrong one.
+        if (parameterBuffer->MappedData() == nullptr || drawBuffer->MappedData() == nullptr) {
+            MGLOG_E("MultiDrawArraysIndirectCount skipped: CPU fallback cannot read the parameter or "
+                    "draw-indirect buffer");
+            return;
+        }
+
+        Uint32 actualDrawCount = 0;
+        std::memcpy(&actualDrawCount, parameterBuffer->MappedData() + drawcount, sizeof(actualDrawCount));
+        actualDrawCount = std::min<Uint32>(actualDrawCount, static_cast<Uint32>(maxdrawcount));
+        ExecuteArraysIndirectCommands(mode, drawBuffer->MappedData() + commandOffset, commandOffset, drawBuffer,
+                                      static_cast<GLsizei>(actualDrawCount), stride, "MultiDrawArraysIndirectCount");
+    }
+
     void DrawRangeElementsBaseVertex(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type,
                                      const void* indices, GLint basevertex) {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer;
         PrepareForDraw(syncBit);
+        SetCurrentBaseVertex(basevertex);
         g_GLESFuncs.glDrawRangeElementsBaseVertex(mode, start, end, count, type, indices, basevertex);
+        SetCurrentBaseVertex(0);
     }
 
     void DrawRangeElements(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type, const void* indices) {
@@ -3462,7 +3596,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer | DrawSyncBit::Instancing;
         PrepareForDraw(syncBit);
         SetCurrentBaseInstance(baseinstance);
+        SetCurrentBaseVertex(basevertex);
         g_GLESFuncs.glDrawElementsInstancedBaseVertex(mode, count, type, indices, instancecount, basevertex);
+        SetCurrentBaseVertex(0);
         SetCurrentBaseInstance(0);
     }
 
@@ -3470,7 +3606,9 @@ namespace MobileGL::MG_Backend::DirectGLES {
                                          GLsizei instancecount, GLint basevertex) {
         DrawSyncFlags syncBit = DrawSyncBit::IndexBuffer | DrawSyncBit::Instancing;
         PrepareForDraw(syncBit);
+        SetCurrentBaseVertex(basevertex);
         g_GLESFuncs.glDrawElementsInstancedBaseVertex(mode, count, type, indices, instancecount, basevertex);
+        SetCurrentBaseVertex(0);
     }
 
     void DrawElementsInstancedBaseInstance(GLenum mode, GLsizei count, GLenum type, const void* indices,

MobileGL/MG_Backend/DirectGLES/DirectGLES.h

@@ -40,6 +40,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
     void MultiDrawElementsIndirectCount(GLenum mode, GLenum type, const void* indirect, GLintptr drawcount,
                                         GLsizei maxdrawcount, GLsizei stride);
     void MultiDrawArraysIndirect(GLenum mode, const void* indirect, GLsizei drawcount, GLsizei stride);
+    void MultiDrawArraysIndirectCount(GLenum mode, const void* indirect, GLintptr drawcount, GLsizei maxdrawcount,
+                                      GLsizei stride);
     void DrawRangeElementsBaseVertex(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type,
                                      const void* indices, GLint basevertex);
     void DrawRangeElements(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type, const void* indices);

MobileGL/MG_Backend/DirectGLES/Managers.cpp

@@ -155,6 +155,14 @@ namespace MobileGL::MG_Backend::DirectGLES {
     // (possibly GPU-written) indirect command buffer, so its declaration expands into a
     // std430 SSBO view of that buffer indexed by a CPU-computed word index, with the plain
     // mg_BaseInstance uniform as the fallback for non-indirect draws.
+    //
+    // The word index is stored ONE-BASED, so that zero - the value every GLSL uniform starts
+    // at - is the "not an indirect draw" sentinel. Nothing seeds this uniform before a
+    // program's first draw, and the non-indirect draw entry points never write it at all, so a
+    // zero-based index with a negative sentinel would leave every such draw reading
+    // mg_indirectWords[0] out of a storage buffer no one bound. That is not a silent zero on a
+    // real driver: it returned garbage on Adreno, and a garbage gl_BaseInstance pushed the CTS
+    // shader_draw_parameters geometry clean off screen.
     String PromoteDrawParameterGlobalsToUniforms(String source, GLenum shaderType) {
         if (shaderType != GL_VERTEX_SHADER) {
             return source;
@@ -208,12 +216,12 @@ namespace MobileGL::MG_Backend::DirectGLES {
                          " { highp uint mg_indirectWords[]; };\n";
             if (rebaseInstanceId) {
                 machinery += String("#define ") + ZERO_BASED_INSTANCE_ID_NAME + " (gl_InstanceID - ((" +
-                             BASE_INSTANCE_WORD_INDEX_UNIFORM_NAME + " >= 0) ? int(mg_indirectWords[uint(" +
+                             BASE_INSTANCE_WORD_INDEX_UNIFORM_NAME + " > 0) ? int(mg_indirectWords[uint(" +
-                             BASE_INSTANCE_WORD_INDEX_UNIFORM_NAME + ")]) : 0))\n";
+                             BASE_INSTANCE_WORD_INDEX_UNIFORM_NAME + " - 1)]) : 0))\n";
             }
             machinery += String("#define ") + BASE_INSTANCE_LOWERED_NAME + " ((" +
-                         BASE_INSTANCE_WORD_INDEX_UNIFORM_NAME + " >= 0) ? int(mg_indirectWords[uint(" +
+                         BASE_INSTANCE_WORD_INDEX_UNIFORM_NAME + " > 0) ? int(mg_indirectWords[uint(" +
-                         BASE_INSTANCE_WORD_INDEX_UNIFORM_NAME + ")]) : " + BASE_INSTANCE_UNIFORM_NAME + ")";
+                         BASE_INSTANCE_WORD_INDEX_UNIFORM_NAME + " - 1)]) : " + BASE_INSTANCE_UNIFORM_NAME + ")";
             source.replace(pos, declaration.size(), machinery);
             break;
         }
@@ -4594,6 +4602,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
             m_baseInstanceUniformLocation = g_GLESFuncs.glGetUniformLocation(m_backendProgramId,
                                                                              BASE_INSTANCE_UNIFORM_NAME);
             m_drawIdUniformLocation = g_GLESFuncs.glGetUniformLocation(m_backendProgramId, DRAW_ID_UNIFORM_NAME);
+            m_baseVertexUniformLocation = g_GLESFuncs.glGetUniformLocation(m_backendProgramId,
+                                                                           BASE_VERTEX_UNIFORM_NAME);
             m_baseInstanceWordIndexUniformLocation =
                 g_GLESFuncs.glGetUniformLocation(m_backendProgramId, BASE_INSTANCE_WORD_INDEX_UNIFORM_NAME);
             // The mg_IndirectParams block binding is baked into the ESSL (ES cannot rebind
@@ -4772,14 +4782,21 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 g_GLESFuncs.glUniform1i(m_baseInstanceUniformLocation, static_cast<GLint>(baseInstance));
             }
             // A direct value disables the indirect-command-buffer read.
+            SetBaseInstanceWordIndex(-1);
+        }
+
+        // The uniform is written one-based so that its GLSL initial value, zero, already reads
+        // as "no indirect command" - see PromoteDrawParameterGlobalsToUniforms.
+        void BackendProgramObjectImpl::SetBaseInstanceWordIndex(Int32 wordIndex) const {
             if (m_baseInstanceWordIndexUniformLocation >= 0) {
-                g_GLESFuncs.glUniform1i(m_baseInstanceWordIndexUniformLocation, -1);
+                g_GLESFuncs.glUniform1i(m_baseInstanceWordIndexUniformLocation,
+                                        wordIndex < 0 ? 0 : wordIndex + 1);
             }
         }
 
-        void BackendProgramObjectImpl::SetBaseInstanceWordIndex(Int32 wordIndex) const {
+        void BackendProgramObjectImpl::SetBaseVertex(Int32 baseVertex) const {
-            if (m_baseInstanceWordIndexUniformLocation >= 0) {
+            if (m_baseVertexUniformLocation >= 0) {
-                g_GLESFuncs.glUniform1i(m_baseInstanceWordIndexUniformLocation, wordIndex);
+                g_GLESFuncs.glUniform1i(m_baseVertexUniformLocation, baseVertex);
             }
         }
 

MobileGL/MG_Backend/DirectGLES/Managers.h

@@ -82,14 +82,26 @@ namespace MobileGL::MG_Backend::DirectGLES {
     // GLES core supports only GL_PRIMITIVE_RESTART_FIXED_INDEX. Throws when the app enabled
     // the arbitrary GL_PRIMITIVE_RESTART with a non-fixed index for this index type.
     void CheckPrimitiveRestartSupported(GLenum indexType);
-    // Feed the current program's gl_BaseInstance / gl_DrawID emulation uniforms. Both are
+    // Feed the current program's gl_BaseInstance / gl_DrawID / gl_BaseVertex emulation
-    // no-ops when the program does not read the corresponding builtin.
+    // uniforms. All are no-ops when the program does not read the corresponding builtin.
     void SetCurrentBaseInstance(Uint32 baseInstance);
     void SetCurrentDrawID(Uint32 drawId);
+    // GL's gl_BaseVertex is the base-vertex parameter of an indexed draw and zero for every
+    // command that has none - including all the DrawArrays forms - so every draw path that
+    // does not carry one must leave this at zero rather than inherit the last draw's value.
+    void SetCurrentBaseVertex(Int32 baseVertex);
     // True when the current program actually reads gl_DrawID, i.e. when a batched
     // (single driver call) multi-draw tier would have to feed it one value for the whole
     // batch and would therefore be wrong.
     Bool CurrentProgramReadsDrawID();
+    // Same question for gl_BaseVertex: a batched multi-draw tier cannot give each sub-draw
+    // its own base vertex through a uniform either.
+    Bool CurrentProgramReadsBaseVertex();
+    // Both of the above, conservatively, for a caller that must decide BEFORE PrepareForDraw
+    // has synced the program - where "does not read it" is indistinguishable from "cannot be
+    // asked yet". Answers true whenever the backend twin is missing or predates the current
+    // link.
+    Bool CurrentProgramMayNeedPerSubDrawBuiltins(Bool batchCarriesBaseVertices);
 
     template <typename StateObject, typename BackendObject>
     class StateBackendObjectRegistry {
@@ -1025,9 +1037,13 @@ namespace MobileGL::MG_Backend::DirectGLES {
             void SetBaseInstance(Uint32 baseInstance) const;
             void SetBaseInstanceWordIndex(Int32 wordIndex) const;
             void SetDrawID(Uint32 drawId) const;
+            void SetBaseVertex(Int32 baseVertex) const;
             // True when the transpiled program kept a gl_DrawID uniform, i.e. SetDrawID
             // actually reaches a shader read rather than being discarded.
             Bool ReadsDrawID() const { return m_drawIdUniformLocation >= 0; }
+            // Same for gl_BaseVertex: only a program that reads it pays for the per-draw
+            // uniform write, and only such a program needs the reset after one.
+            Bool ReadsBaseVertex() const { return m_baseVertexUniformLocation >= 0; }
             Int GetIndirectParamsBinding() const { return m_indirectParamsBinding; }
             Uint GetBackendProgramId() const { return m_backendProgramId; }
             // False when the last SyncToBackend could not produce a usable program (a
@@ -1077,6 +1093,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
             Uint m_backendGlobalUBOId = 0;
             Int m_baseInstanceUniformLocation = -1;
             Int m_drawIdUniformLocation = -1;
+            Int m_baseVertexUniformLocation = -1;
             Int m_baseInstanceWordIndexUniformLocation = -1;
             Int m_indirectParamsBinding = -1;
             Uint32 m_snormFallbackClampOutputMask = 0;

MobileGL/MG_Backend/DirectGLES/MultiDraw.cpp

@@ -274,17 +274,22 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
         // the batch's own shape - not the driver - rules it out; the compute tier keeps
         // its remaining feasibility checks inside its implementation, where the data it
         // has to walk is already in hand.
-        GLESMultiDrawMode ResolveTierForBatch(Bool programReadsDrawID, Bool hasIndexBuffer) {
+        GLESMultiDrawMode ResolveTierForBatch(Bool programReadsDrawID, Bool perSubDrawBaseVertex,
+                                              Bool hasIndexBuffer) {
             ResolveTierOnce();
             GLESMultiDrawMode tier = g_resolvedTier;
 
             // Batched tiers issue one driver entry for the whole batch, so the emulated
             // gl_DrawID uniform can only hold one value across every sub-draw. A program
             // that reads gl_DrawID gets an unrolled tier, which feeds each sub-draw its
-            // own index (the spec's value); nothing else observes the difference.
+            // own index (the spec's value); nothing else observes the difference. The
+            // emulated gl_BaseVertex is one uniform for the same reason, so a batch whose
+            // sub-draws carry their own base vertices unrolls too - even the Ext tier,
+            // which hands the driver the whole basevertex array, can only leave ONE value
+            // in the uniform the shader reads.
             const Bool batched = tier == GLESMultiDrawMode::Ext || tier == GLESMultiDrawMode::MultiIndirect ||
                                  tier == GLESMultiDrawMode::Compute;
-            if (batched && programReadsDrawID) {
+            if (batched && (programReadsDrawID || perSubDrawBaseVertex)) {
                 tier = SupportsTier(GLESMultiDrawMode::BaseVertex) ? GLESMultiDrawMode::BaseVertex
                                                                    : GLESMultiDrawMode::DrawElements;
             }
@@ -371,7 +376,8 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
         // ---------------------------------------------------------------------------
 
         Bool RunIndirect(GLenum mode, const GLsizei* count, GLenum type, const GLvoid* const* indices,
-                         GLsizei drawcount, const GLint* basevertex, Bool batched, Bool feedDrawID) {
+                         GLsizei drawcount, const GLint* basevertex, Bool batched, Bool feedDrawID,
+                         Bool feedBaseVertex) {
             if (!SupportsTier(batched ? GLESMultiDrawMode::MultiIndirect : GLESMultiDrawMode::Indirect)) return false;
             const SizeT indexSize = IndexTypeSize(type);
             if (indexSize == 0) return false;
@@ -413,10 +419,12 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
             } else {
                 for (GLsizei i = 0; i < drawcount; ++i) {
                     if (feedDrawID) SetCurrentDrawID(static_cast<Uint32>(i));
+                    if (feedBaseVertex) SetCurrentBaseVertex(basevertex ? basevertex[i] : 0);
                     const SizeT commandOffset = commandBase + static_cast<SizeT>(i) * sizeof(DrawElementsIndirectCommand);
                     g_GLESFuncs.glDrawElementsIndirect(mode, type, reinterpret_cast<const void*>(commandOffset));
                 }
                 if (feedDrawID) SetCurrentDrawID(0);
+                if (feedBaseVertex) SetCurrentBaseVertex(0);
             }
             BufferImpl::BindBufferId(GL_DRAW_INDIRECT_BUFFER, previousIndirectBinding);
             NoteTierExecuted(batched ? GLESMultiDrawMode::MultiIndirect : GLESMultiDrawMode::Indirect);
@@ -428,15 +436,17 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
         // ---------------------------------------------------------------------------
 
         Bool RunBaseVertexLoop(GLenum mode, const GLsizei* count, GLenum type, const GLvoid* const* indices,
-                               GLsizei drawcount, const GLint* basevertex, Bool feedDrawID) {
+                               GLsizei drawcount, const GLint* basevertex, Bool feedDrawID, Bool feedBaseVertex) {
             if (!SupportsTier(GLESMultiDrawMode::BaseVertex)) return false;
             for (GLsizei i = 0; i < drawcount; ++i) {
                 if (count[i] <= 0) continue;
                 if (feedDrawID) SetCurrentDrawID(static_cast<Uint32>(i));
+                if (feedBaseVertex) SetCurrentBaseVertex(basevertex ? basevertex[i] : 0);
                 g_GLESFuncs.glDrawElementsBaseVertex(mode, count[i], type, indices[i],
                                                      basevertex ? basevertex[i] : 0);
             }
             if (feedDrawID) SetCurrentDrawID(0);
+            if (feedBaseVertex) SetCurrentBaseVertex(0);
             NoteTierExecuted(GLESMultiDrawMode::BaseVertex);
             return true;
         }
@@ -446,7 +456,8 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
         // ---------------------------------------------------------------------------
 
         Bool RunRebasedDrawElements(GLenum mode, const GLsizei* count, GLenum type, const GLvoid* const* indices,
-                                    GLsizei drawcount, const GLint* basevertex, Bool feedDrawID) {
+                                    GLsizei drawcount, const GLint* basevertex, Bool feedDrawID,
+                                    Bool feedBaseVertex) {
             const SizeT indexSize = IndexTypeSize(type);
             if (indexSize == 0) return false;
 
@@ -500,11 +511,16 @@ namespace MobileGL::MG_Backend::DirectGLES::MultiDrawImpl {
             for (GLsizei i = 0; i < drawcount; ++i) {
                 if (count[i] <= 0) continue;
                 if (feedDrawID) SetCurrentDrawID(static_cast<Uint32>(i));
+                // The base vertex is folded into the rewritten index stream here, so the
+                // driver sees none - but gl_BaseVertex still has to report the value the
+                // application passed for this sub-draw.
+                if (feedBaseVertex) SetCurrentBaseVertex(basevertex ? basevertex[i] : 0);
                 g_GLESFuncs.glDrawElements(mode, count[i], GL_UNSIGNED_INT,
                                            reinterpret_cast<const void*>(indexBase + cursor * sizeof(Uint32)));
                 cursor += static_cast<SizeT>(count[i]);
             }
             if (feedDrawID) SetCurrentDrawID(0);
+            if (feedBaseVertex) SetCurrentBaseVertex(0);
             BufferImpl::BindBufferId(GL_ELEMENT_ARRAY_BUFFER, previousIndexBinding);
             NoteTierExecuted(GLESMultiDrawMode::DrawElements);
             return true;
@@ -837,8 +853,15 @@ void main() {
         // afterwards would mean unpicking the program, SSBO and index bindings
         // PrepareForDraw just made, and a dispatch inside an open transform feedback
         // span is not legal at all. On success it hands back a flattened index stream.
+        // A batch whose sub-draws carry their own base vertices cannot be flattened either
+        // when the program reads gl_BaseVertex: one draw call leaves one uniform value.
+        // Asked conservatively because this decision precedes PrepareForDraw - see
+        // CurrentProgramMayNeedPerSubDrawBuiltins. Flattening is the irreversible half:
+        // once the batch is one draw the values are gone, whereas declining to flatten only
+        // costs the unrolled tier.
         FlattenedStream flattened;
-        if (ResolvedTier() == GLESMultiDrawMode::Compute && !CurrentProgramReadsDrawID()) {
+        if (ResolvedTier() == GLESMultiDrawMode::Compute &&
+            !CurrentProgramMayNeedPerSubDrawBuiltins(basevertex != nullptr)) {
             FlattenWithCompute(mode, count, type, indices, drawcount, basevertex, flattened);
         }
 
@@ -852,8 +875,11 @@ void main() {
             return;
         }
 
+        // Now that PrepareForDraw has synced the program, both questions have real answers;
+        // the tier choice and the per-sub-draw feeds use those, not the guess above.
         const Bool feedDrawID = CurrentProgramReadsDrawID();
-        const GLESMultiDrawMode tier = ResolveTierForBatch(feedDrawID, hasIndexBuffer);
+        const Bool feedBaseVertex = basevertex != nullptr && CurrentProgramReadsBaseVertex();
+        const GLESMultiDrawMode tier = ResolveTierForBatch(feedDrawID, feedBaseVertex, hasIndexBuffer);
 
         Bool drawn = false;
         switch (tier) {
@@ -861,16 +887,19 @@ void main() {
             drawn = RunExt(mode, count, type, indices, drawcount, basevertex);
             break;
         case GLESMultiDrawMode::MultiIndirect:
-            drawn = RunIndirect(mode, count, type, indices, drawcount, basevertex, /*batched=*/true, feedDrawID);
+            drawn = RunIndirect(mode, count, type, indices, drawcount, basevertex, /*batched=*/true, feedDrawID,
+                                feedBaseVertex);
             break;
         case GLESMultiDrawMode::Indirect:
-            drawn = RunIndirect(mode, count, type, indices, drawcount, basevertex, /*batched=*/false, feedDrawID);
+            drawn = RunIndirect(mode, count, type, indices, drawcount, basevertex, /*batched=*/false, feedDrawID,
+                                feedBaseVertex);
             break;
         case GLESMultiDrawMode::BaseVertex:
-            drawn = RunBaseVertexLoop(mode, count, type, indices, drawcount, basevertex, feedDrawID);
+            drawn = RunBaseVertexLoop(mode, count, type, indices, drawcount, basevertex, feedDrawID, feedBaseVertex);
             break;
         case GLESMultiDrawMode::DrawElements:
-            drawn = RunRebasedDrawElements(mode, count, type, indices, drawcount, basevertex, feedDrawID);
+            drawn = RunRebasedDrawElements(mode, count, type, indices, drawcount, basevertex, feedDrawID,
+                                           feedBaseVertex);
             break;
         case GLESMultiDrawMode::Compute:
             // Its pre-pass ran above; reaching here means it declined this batch's shape.
@@ -883,8 +912,13 @@ void main() {
         // below are the floor: a base-vertex replay where the driver has one, and the
         // rewritten index stream where it does not. Both are safe for any batch these
         // entry points can receive.
-        if (!drawn) drawn = RunBaseVertexLoop(mode, count, type, indices, drawcount, basevertex, feedDrawID);
+        if (!drawn) {
-        if (!drawn) drawn = RunRebasedDrawElements(mode, count, type, indices, drawcount, basevertex, feedDrawID);
+            drawn = RunBaseVertexLoop(mode, count, type, indices, drawcount, basevertex, feedDrawID, feedBaseVertex);
+        }
+        if (!drawn) {
+            drawn = RunRebasedDrawElements(mode, count, type, indices, drawcount, basevertex, feedDrawID,
+                                           feedBaseVertex);
+        }
         if (!drawn) {
             MGLOG_E("DirectGLES multi-draw: no usable tier for a %d sub-draw batch (mode 0x%x, type 0x%x); "
                     "the batch was dropped",

MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.cpp

@@ -1979,13 +1979,27 @@ namespace MobileGL::MG_Backend::DirectVulkan {
     // cannot be corrected and instanced draws with a non-zero baseInstance misrender; this
     // detects the case so the user gets one warning instead of silent corruption.
     Bool ProgramFactory::ReflectedReadsInstanceIndexBuiltin(const SpvReflectShaderModule& reflectModule) {
+        return ReflectedDeclaresInputBuiltin(reflectModule, SpvBuiltInInstanceIndex);
+    }
+
+    // GL's gl_BaseVertex and Vulkan's BaseVertex agree for indexed draws and disagree for every
+    // other command, so a program declaring the builtin needs the ZeroBaseVertex variant when a
+    // non-indexed draw uses it (see CompileOptionBit::ZeroBaseVertex). "Declares" rather than
+    // "reads" is the honest word and the useful one: the zeroing pass keeps the variable, so
+    // both variants of a program answer this question identically.
+    Bool ProgramFactory::ReflectedReadsBaseVertexBuiltin(const SpvReflectShaderModule& reflectModule) {
+        return ReflectedDeclaresInputBuiltin(reflectModule, SpvBuiltInBaseVertex);
+    }
+
+    Bool ProgramFactory::ReflectedDeclaresInputBuiltin(const SpvReflectShaderModule& reflectModule,
+                                                       SpvBuiltIn builtin) {
         for (Uint32 entryIndex = 0; entryIndex < reflectModule.entry_point_count; ++entryIndex) {
             const SpvReflectEntryPoint& entryPoint = reflectModule.entry_points[entryIndex];
             for (Uint32 variableIndex = 0; variableIndex < entryPoint.input_variable_count; ++variableIndex) {
                 const SpvReflectInterfaceVariable* variable = entryPoint.input_variables[variableIndex];
                 if (variable != nullptr &&
                     (variable->decoration_flags & SPV_REFLECT_DECORATION_BUILT_IN) != 0 &&
-                    variable->built_in == SpvBuiltInInstanceIndex) {
+                    variable->built_in == builtin) {
                     return true;
                 }
             }
@@ -2244,6 +2258,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                                              VkProgramObject& entry) const {
         entry.activeVertexInputLocationMask = 0;
         entry.vertexInputTypes.fill(0);
+        entry.readsBaseVertexBuiltin = false;
 
         for (SizeT moduleIndex = 0; moduleIndex < shaders.size() && moduleIndex < spirv.size(); ++moduleIndex) {
             if (!shaders[moduleIndex] || shaders[moduleIndex]->GetShaderStage() != ShaderStage::Vertex) {
@@ -2265,6 +2280,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 continue;
             }
 
+            entry.readsBaseVertexBuiltin = ReflectedReadsBaseVertexBuiltin(reflectModule);
+
             if (!m_shaderDrawParametersEnabled && ReflectedReadsInstanceIndexBuiltin(reflectModule)) {
                 static Bool s_warnedInstanceIndexUnsupported = false;
                 if (!s_warnedInstanceIndexUnsupported) {
@@ -2909,6 +2926,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // a FragCoordYFlip variant also depends on the baked default-framebuffer height, so
         // that height rides in the free high half of the key. Flags occupy the low bits, and a
         // height cannot exceed the 16 bits a swapchain extent fits in.
+        //
+        // "The low bits" is load-bearing and was until now only a comment: a flag that reached
+        // bit 16 would alias the height and two different variants would share one memo slot.
+        static_assert(static_cast<Uint>(CompileOptionBit::ZeroBaseVertex) < (1u << 16),
+                      "CompileOptionBit values must stay below bit 16: GetOrCreateProgram packs the "
+                      "default-framebuffer height into the high half of the same memo key");
         const Uint memoKey = (flags & CompileOptionBit::FragCoordYFlip)
                                  ? (flags.GetRaw() | (m_defaultFramebufferHeight << 16))
                                  : flags.GetRaw();
@@ -3027,6 +3050,26 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 }
             }
 
+            // The non-indexed variant of a vertex stage that reads gl_BaseVertex: GL wants zero
+            // there, Vulkan's builtin would hand it the draw's firstVertex. Requested per draw
+            // through CompileOptionBit::ZeroBaseVertex, so the indexed variant of the same
+            // program keeps the native builtin and stays correct for glDrawElementsBaseVertex
+            // and for the baseVertex word of an indexed indirect command.
+            if (shaders[i] && shaders[i]->GetShaderStage() == ShaderStage::Vertex &&
+                (flags & CompileOptionBit::ZeroBaseVertex)) {
+                Vector<Uint> zeroedSpirv;
+                if (MG_Util::ShaderTranspiler::ShaderCompiler::ZeroBaseVertexForVulkan(moduleSpirvs[i],
+                                                                                       zeroedSpirv)) {
+                    moduleSpirvs[i] = std::move(zeroedSpirv);
+                } else {
+                    // Failing open keeps the native builtin, which is the pre-fix behavior:
+                    // gl_BaseVertex reads firstVertex on a DrawArrays instead of zero.
+                    MGLOG_E("ProgramFactory: failed to zero gl_BaseVertex for program %u; non-indexed "
+                            "draws will read the draw's first vertex from it instead of zero",
+                            program.GetExternalIndex());
+                }
+            }
+
             // A 64-bit vertex input has to arrive as its 32-bit word pair: VK_FORMAT_R64*_SFLOAT is
             // optional and lavapipe advertises none of them at all. The pass is unconditional so it
             // always agrees with the Float64 case in VertexInputStateFactory::ToVkVertexFormat, and

MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.h

@@ -60,6 +60,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // PositionYFlip (the two are the same fact about the same draws) except under a
             // quarter turn, which this renderer does not convert rectangles for either.
             FragCoordYFlip = 1 << 7,
+            // Replaces the vertex stage's gl_BaseVertex reads with zero. GL defines the builtin
+            // as zero for every drawing command that has no baseVertex parameter - all the
+            // DrawArrays forms - while Vulkan's BaseVertex reports firstVertex there. Set only
+            // for a non-indexed draw whose program actually reads the builtin, so nothing else
+            // acquires a second program/pipeline variant. See ZeroBaseVertexPass.
+            ZeroBaseVertex = 1 << 8,
         };
         using CompileOptionFlags = Flags<CompileOptionBit>;
         using HashType = Uint64;
@@ -129,6 +135,11 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             // gl_FragDepth); shader-computed depth is immune to the cross-pipeline
             // position-invariance quirk (see PipelineFactory::ShouldSuppressDepthWrite).
             Bool fragmentReplacesDepth = false;
+            // The vertex module declares the BaseVertex builtin. Selects the ZeroBaseVertex
+            // program variant for non-indexed draws, and is deliberately a property of the
+            // PROGRAM rather than of the variant: the zeroed variant leaves the variable
+            // declared, so both variants answer the same and the draw path can ask either.
+            Bool readsBaseVertexBuiltin = false;
             // Frame-boundary counter value of the last GetOrCreateProgram hit; drives
             // cache eviction (see OnFrameBoundary). Mutable: the draw snapshot's memoised
             // entry pointer re-stamps use through a const reference (StampProgramUse).
@@ -179,6 +190,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 producerOutputComponentCount = other.producerOutputComponentCount;
                 fragmentInputComponentCount = other.fragmentInputComponentCount;
                 fragmentReplacesDepth = other.fragmentReplacesDepth;
+                readsBaseVertexBuiltin = other.readsBaseVertexBuiltin;
                 lastUsedFrame = other.lastUsedFrame;
                 other.hash = 0;
                 other.descriptorSetLayout = VK_NULL_HANDLE;
@@ -192,6 +204,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 other.producerOutputComponentCount = 0;
                 other.fragmentInputComponentCount = 0;
                 other.fragmentReplacesDepth = false;
+                other.readsBaseVertexBuiltin = false;
                 other.lastUsedFrame = 0;
             }
             VkProgramObject& operator=(VkProgramObject&& other) noexcept {
@@ -231,6 +244,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 producerOutputComponentCount = other.producerOutputComponentCount;
                 fragmentInputComponentCount = other.fragmentInputComponentCount;
                 fragmentReplacesDepth = other.fragmentReplacesDepth;
+                readsBaseVertexBuiltin = other.readsBaseVertexBuiltin;
                 lastUsedFrame = other.lastUsedFrame;
                 other.hash = 0;
                 other.descriptorSetLayout = VK_NULL_HANDLE;
@@ -244,6 +258,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 other.producerOutputComponentCount = 0;
                 other.fragmentInputComponentCount = 0;
                 other.fragmentReplacesDepth = false;
+                other.readsBaseVertexBuiltin = false;
                 other.lastUsedFrame = 0;
                 return *this;
             }
@@ -341,6 +356,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         // True when an entry point reads the InstanceIndex builtin. Only gates a diagnostic:
         // without shaderDrawParameters such a shader cannot have gl_InstanceID rebased.
         static Bool ReflectedReadsInstanceIndexBuiltin(const SpvReflectShaderModule& reflectModule);
+        // True when an entry point declares the BaseVertex builtin, i.e. when a non-indexed
+        // draw with this program has to take the ZeroBaseVertex variant.
+        static Bool ReflectedReadsBaseVertexBuiltin(const SpvReflectShaderModule& reflectModule);
+        // Shared by the two above: does any entry point list an input variable decorated with
+        // this builtin?
+        static Bool ReflectedDeclaresInputBuiltin(const SpvReflectShaderModule& reflectModule, SpvBuiltIn builtin);
 
     private:
         struct ProgramLookupCache {

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp

@@ -5821,7 +5821,40 @@ void main() {
                 m_lastLodParamsSum = 0;  // filled below once the sampled set is known
             }
         }
-        const auto& programObj = m_programFactory->GetOrCreateProgram(program, transformFlags);
+        // GL's gl_BaseVertex is zero for every command without a baseVertex parameter, while
+        // Vulkan's builtin reports the draw's firstVertex; a non-indexed draw therefore takes
+        // the zeroed program variant. The question is about the program's SPIR-V, not about
+        // this draw, so it is memoized on (program lifetime, backend-state version): only the
+        // very first draw of a program pays the extra lookup, and a program used exclusively
+        // with non-indexed draws never resolves - never compiles, never re-stamps - the
+        // variant no draw of it would use.
+        const Bool nonIndexedDraw = !(aspects & DrawSetupAspect::IndexBuffer);
+        const Uint64 baseVertexProgramLifetimeId = program.GetLifetimeId();
+        const Uint32 baseVertexProgramVersion = program.GetBackendStateVersion();
+        const Bool baseVertexQueryKnown = m_lastBaseVertexQueryValid &&
+                                          m_lastBaseVertexProgramLifetimeId == baseVertexProgramLifetimeId &&
+                                          m_lastBaseVertexProgramVersion == baseVertexProgramVersion;
+        if (baseVertexQueryKnown && nonIndexedDraw && m_lastBaseVertexReads) {
+            transformFlags |= ProgramFactory::CompileOptionBit::ZeroBaseVertex;
+        }
+        const ProgramFactory::VkProgramObject* resolvedProgramObj =
+            &m_programFactory->GetOrCreateProgram(program, transformFlags);
+        if (!baseVertexQueryKnown) {
+            // Read the answer out of the entry BEFORE any second lookup: that lookup may
+            // insert and move every entry of the open-addressing cache, dangling the
+            // reference. The zeroing pass leaves the variable declared, so the variant just
+            // resolved answers the same as the base one either way.
+            const Bool readsBaseVertex = resolvedProgramObj->readsBaseVertexBuiltin;
+            m_lastBaseVertexQueryValid = true;
+            m_lastBaseVertexProgramLifetimeId = baseVertexProgramLifetimeId;
+            m_lastBaseVertexProgramVersion = baseVertexProgramVersion;
+            m_lastBaseVertexReads = readsBaseVertex;
+            if (nonIndexedDraw && readsBaseVertex) {
+                transformFlags |= ProgramFactory::CompileOptionBit::ZeroBaseVertex;
+                resolvedProgramObj = &m_programFactory->GetOrCreateProgram(program, transformFlags);
+            }
+        }
+        const auto& programObj = *resolvedProgramObj;
         // For the snapshot's memoised entry pointer: if anything below inserts into the
         // program cache (blit/aux program compiles), the epoch moves and the snapshot
         // stores no pointer for this draw - the fast path then re-looks-up once.

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.h

@@ -773,6 +773,19 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         ProgramFactory::CompileOptionFlags m_lastLodBaseFlags = {};
         ProgramFactory::CompileOptionFlags m_lastLodResultFlags = {};
 
+        // Does the current program's vertex stage declare the BaseVertex builtin? A property
+        // of the program's SPIR-V, so (lifetime id, backend-state version) is the whole key.
+        //
+        // Memoized rather than re-asked because asking means resolving the UN-zeroed program
+        // variant, and a program that only ever draws non-indexed would then compile a variant
+        // no draw uses AND re-stamp its use every draw, so the idle sweep could never retire
+        // it. With the memo the answer is known before the first lookup and only the variant
+        // the draw actually needs is resolved.
+        Bool m_lastBaseVertexQueryValid = false;
+        Uint64 m_lastBaseVertexProgramLifetimeId = 0;
+        Uint32 m_lastBaseVertexProgramVersion = 0;
+        Bool m_lastBaseVertexReads = false;
+
         // Snapshot behind TrySetupDrawFastPath. Values only: the program and
         // render-pass caches are open-addressing maps whose entries move on
         // insert, so no pointers into them are cached; the pipeline handle is

MobileGL/MG_IntegrationTest/CMakeLists.txt

@@ -50,6 +50,7 @@ add_executable(MobileGLIntegrationTest
     Scenarios/CrossFrameBufferScenario.cpp
     Scenarios/ResidentIndexScenario.cpp
     Scenarios/MultiDrawScenario.cpp
+    Scenarios/DrawParametersScenario.cpp
     Scenarios/AsyncCompileScenario.cpp
     Scenarios/XfbAfterClipDistanceScenario.cpp
     Scenarios/ThreeChannelAttachmentScenario.cpp

MobileGL/MG_IntegrationTest/Scenarios/DrawParametersScenario.cpp

@@ -0,0 +1,348 @@
+// MobileGL - MobileGL/MG_IntegrationTest/Scenarios/DrawParametersScenario.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+//
+// gl_BaseVertex / gl_BaseInstance / gl_DrawID (GL_ARB_shader_draw_parameters),
+// read straight out of the shader that a draw command produced.
+//
+// Neither backend has these builtins for free, and each is wrong in its own way
+// when nobody watches:
+//
+//   * DirectVulkan HAS a BaseVertex builtin, but Vulkan's carries the draw's
+//     firstVertex on a NON-INDEXED draw where GL's is defined to be zero ("the
+//     value passed to the baseVertex parameter, or zero for a command with no
+//     such parameter"). Only the indexed meaning of the two agrees. Every
+//     DrawArrays form therefore takes the ZeroBaseVertex program variant.
+//   * DirectGLES has no such builtins at all: ESSL knows none of them, so the
+//     transpiler demotes each one to a uniform the draw paths feed. A uniform
+//     nobody writes keeps whatever the previous draw left in it - which is what
+//     made gl_BaseVertex report a stale base vertex, and what made
+//     gl_BaseInstance read an unbound storage buffer on a plain glDrawArrays.
+//
+// The shader paints the three values, so a draw that carries the wrong ones
+// paints the wrong colour rather than merely disagreeing with an expectation
+// somewhere. The framebuffer is cleared to WHITE and no case expects 255 in any
+// channel, so "the draw did not happen" can never be mistaken for a pass.
+
+#include <cstdint>
+#include <string>
+#include <vector>
+
+#include "../Harness/HeadlessGL.h"
+#include "../Harness/ScenarioFixture.h"
+
+#ifdef GLAPI
+#undef GLAPI
+#endif
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+#include <GL/glext.h>
+
+namespace MGITest {
+    namespace {
+
+        // #version 450: glslang only declares the ARB builtins from 440 up.
+        //
+        // Each value is painted at 8 units per count, not 1: the errors these builtins
+        // actually have are OFF BY ONE (a sub-draw that never got its own gl_DrawID reads
+        // the previous one's, a base vertex that arrives one command late), and at one unit
+        // per count no readback tolerance can tell those from rounding.
+        //
+        // And biased by two counts, so that ZERO is not the clamp floor. Five of these cases
+        // expect zero, and an unbiased encoding would let every negative value - the shape a
+        // sign or rebase mistake produces - clamp to the same black and pass.
+        constexpr const char* kVertexSource = R"(#version 450 core
+#extension GL_ARB_shader_draw_parameters : require
+layout(location = 0) in vec2 aPos;
+flat out vec3 vParams;
+void main() {
+    vParams = (vec3(gl_BaseVertexARB, gl_BaseInstanceARB, gl_DrawIDARB) * 8.0 + 16.0) / 255.0;
+    gl_Position = vec4(aPos, 0.0, 1.0);
+}
+)";
+
+        constexpr const char* kFragmentSource = R"(#version 450 core
+flat in vec3 vParams;
+out vec4 oColor;
+void main() {
+    oColor = vec4(vParams, 1.0);
+}
+)";
+
+        struct Vertex {
+            float x, y;
+        };
+
+        // 3 dummy vertices, then the left half of the viewport as two triangles,
+        // then the right half. Nothing here is symmetric by accident:
+        //
+        //   * the padding makes a draw that ignores `first` / baseVertex paint a
+        //     degenerate triangle (i.e. nothing) instead of the right picture;
+        //   * the two halves let one multi-draw show TWO different gl_DrawID
+        //     values in one readback.
+        //
+        // Indices 3..14 together cover the whole viewport, which is what the
+        // single-draw cases use.
+        constexpr int kPad = 3;
+        constexpr int kLeftFirst = kPad;      // 3
+        constexpr int kRightFirst = kPad + 6; // 9
+        constexpr int kHalfCount = 6;
+
+        std::vector<Vertex> SceneVertices() {
+            std::vector<Vertex> vertices(static_cast<std::size_t>(kPad), Vertex{0.0f, 0.0f});
+            const float bounds[2][2] = {{-1.0f, 0.0f}, {0.0f, 1.0f}};
+            for (const auto& half : bounds) {
+                const float x0 = half[0];
+                const float x1 = half[1];
+                vertices.push_back({x0, -1.0f});
+                vertices.push_back({x1, -1.0f});
+                vertices.push_back({x1, 1.0f});
+                vertices.push_back({x0, -1.0f});
+                vertices.push_back({x1, 1.0f});
+                vertices.push_back({x0, 1.0f});
+            }
+            return vertices;
+        }
+
+        // GL's DrawArraysIndirectCommand / DrawElementsIndirectCommand, spelled out
+        // so a test can write one without depending on a GL header's struct.
+        struct ArraysCommand {
+            std::uint32_t count, instanceCount, first, baseInstance;
+        };
+        struct ElementsCommand {
+            std::uint32_t count, instanceCount, firstIndex;
+            std::int32_t baseVertex;
+            std::uint32_t baseInstance;
+        };
+
+        class DrawParametersScenario : public ScenarioTest {
+        protected:
+            void SetUp() override {
+                ScenarioTest::SetUp();
+                if (!Ready()) return;
+                std::string error;
+                m_program = CompileProgram(kVertexSource, kFragmentSource, &error);
+                ASSERT_NE(m_program, 0u) << error;
+
+                const std::vector<Vertex> vertices = SceneVertices();
+                glGenVertexArrays(1, &m_vao);
+                glBindVertexArray(m_vao);
+                glGenBuffers(1, &m_vbo);
+                glBindBuffer(GL_ARRAY_BUFFER, m_vbo);
+                glBufferData(GL_ARRAY_BUFFER, static_cast<GLsizeiptr>(vertices.size() * sizeof(Vertex)),
+                             vertices.data(), GL_STATIC_DRAW);
+                glEnableVertexAttribArray(0);
+                glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, sizeof(Vertex), reinterpret_cast<const void*>(0));
+                ASSERT_EQ(FirstGLError(), GLenum(GL_NO_ERROR)) << "scene setup left a GL error behind";
+            }
+
+            void TearDown() override {
+                if (!Ready()) return;
+                for (GLuint* buffer : {&m_ebo, &m_indirect, &m_parameter, &m_vbo}) {
+                    if (*buffer != 0) glDeleteBuffers(1, buffer);
+                    *buffer = 0;
+                }
+                if (m_vao != 0) glDeleteVertexArrays(1, &m_vao);
+                if (m_program != 0) glDeleteProgram(m_program);
+            }
+
+            template <typename T>
+            void FillBuffer(GLuint& name, GLenum target, const std::vector<T>& data) {
+                if (name == 0) glGenBuffers(1, &name);
+                glBindBuffer(target, name);
+                glBufferData(target, static_cast<GLsizeiptr>(data.size() * sizeof(T)), data.data(), GL_STATIC_DRAW);
+            }
+
+            // Clears to white, runs `draw` and reads the frame back.
+            template <typename DrawFn>
+            Image Render(DrawFn&& draw) {
+                BindDefaultFramebuffer();
+                glViewport(0, 0, HeadlessGL::Get().Width(), HeadlessGL::Get().Height());
+                ClearTo(1.0f, 1.0f, 1.0f, 1.0f);
+                glUseProgram(m_program);
+                glBindVertexArray(m_vao);
+                draw();
+                return ReadPixels(HeadlessGL::Get().Width(), HeadlessGL::Get().Height());
+            }
+
+            // The three builtins as the shader saw them, at a point in one half of
+            // the viewport. `half` is 0 for the left half and 1 for the right.
+            struct DrawParams {
+                int baseVertex = -1, baseInstance = -1, drawId = -1;
+            };
+            // Decodes the biased 8-units-per-count encoding back to the integer the
+            // shader saw. Rounding to the nearest step absorbs any UNORM slop; adjacent
+            // values stay eight units apart, so an off-by-one still reads as one, and a
+            // negative value lands below the bias and decodes negative rather than
+            // clamping into a legitimate zero.
+            static DrawParams ParamsAt(const Image& image, int half) {
+                const int x = image.Width() * (1 + 2 * half) / 4;
+                const Rgba8 pixel = image.At(x, image.Height() / 2);
+                const auto decode = [](std::uint8_t channel) {
+                    return (static_cast<int>(channel) - 16 + 4) / 8;
+                };
+                return {decode(pixel.r), decode(pixel.g), decode(pixel.b)};
+            }
+
+            static void ExpectParams(const Image& image, int half, const DrawParams& expected,
+                                     const std::string& what) {
+                const DrawParams actual = ParamsAt(image, half);
+                EXPECT_EQ(actual.baseVertex, expected.baseVertex)
+                    << what << ": gl_BaseVertex (half " << half << ")";
+                EXPECT_EQ(actual.baseInstance, expected.baseInstance)
+                    << what << ": gl_BaseInstance (half " << half << ")";
+                EXPECT_EQ(actual.drawId, expected.drawId) << what << ": gl_DrawID (half " << half << ")";
+            }
+
+            GLuint m_program = 0;
+            GLuint m_vao = 0;
+            GLuint m_vbo = 0;
+            GLuint m_ebo = 0;
+            GLuint m_indirect = 0;
+            GLuint m_parameter = 0;
+        };
+
+        // ---- the non-indexed forms: gl_BaseVertex is zero, `first` or not ----
+
+        // Vulkan's BaseVertex would answer 3 here (the draw's firstVertex); GL's
+        // must answer 0, because glDrawArrays has no baseVertex parameter at all.
+        TEST_F(DrawParametersScenario, DrawArraysReportsAZeroBaseVertexDespiteItsFirst) {
+            if (!Ready()) return;
+            const Image image = Render([&] { glDrawArrays(GL_TRIANGLES, kLeftFirst, 2 * kHalfCount); });
+            EXPECT_EQ(FirstGLError(), GLenum(GL_NO_ERROR));
+            ExpectParams(image, 0, {0, 0, 0}, "glDrawArrays(first=3)");
+            ExpectParams(image, 1, {0, 0, 0}, "glDrawArrays(first=3)");
+        }
+
+        TEST_F(DrawParametersScenario, DrawArraysInstancedBaseInstanceReportsItsBaseInstance) {
+            if (!Ready()) return;
+            const Image image = Render([&] {
+                glDrawArraysInstancedBaseInstance(GL_TRIANGLES, kLeftFirst, 2 * kHalfCount, 1, 5);
+            });
+            EXPECT_EQ(FirstGLError(), GLenum(GL_NO_ERROR));
+            ExpectParams(image, 0, {0, 5, 0}, "glDrawArraysInstancedBaseInstance(baseInstance=5)");
+        }
+
+        // The base instance of one draw must not survive into the next one. This is
+        // the shape that broke on DirectGLES: the emulation uniform is per-program
+        // state, so a draw that never writes it inherits the last writer's value.
+        TEST_F(DrawParametersScenario, APlainDrawAfterABaseInstancedOneSeesZeroAgain) {
+            if (!Ready()) return;
+            const Image image = Render([&] {
+                glDrawArraysInstancedBaseInstance(GL_TRIANGLES, kLeftFirst, 2 * kHalfCount, 1, 7);
+                glDrawArrays(GL_TRIANGLES, kLeftFirst, 2 * kHalfCount);
+            });
+            EXPECT_EQ(FirstGLError(), GLenum(GL_NO_ERROR));
+            ExpectParams(image, 0, {0, 0, 0}, "plain glDrawArrays after a base-instanced draw");
+        }
+
+        // ---- the indexed forms: gl_BaseVertex IS the base vertex ----
+
+        TEST_F(DrawParametersScenario, DrawElementsBaseVertexReportsItsBaseVertex) {
+            if (!Ready()) return;
+            std::vector<std::uint32_t> indices;
+            for (std::uint32_t i = 0; i < 2 * kHalfCount; ++i) indices.push_back(i);
+            FillBuffer(m_ebo, GL_ELEMENT_ARRAY_BUFFER, indices);
+
+            const Image image = Render([&] {
+                glDrawElementsBaseVertex(GL_TRIANGLES, 2 * kHalfCount, GL_UNSIGNED_INT,
+                                         reinterpret_cast<const void*>(0), kLeftFirst);
+            });
+            EXPECT_EQ(FirstGLError(), GLenum(GL_NO_ERROR));
+            ExpectParams(image, 0, {kLeftFirst, 0, 0}, "glDrawElementsBaseVertex(basevertex=3)");
+            ExpectParams(image, 1, {kLeftFirst, 0, 0}, "glDrawElementsBaseVertex(basevertex=3)");
+        }
+
+        // ... and is zero again for the command that has none, including after one
+        // that did: the same leak the base instance has, on the other builtin. The
+        // preceding draw MUST carry a non-zero base vertex or this case proves nothing -
+        // one index run reaches the geometry through the base vertex, the second through
+        // its own indices, so the two draws paint the same picture with different
+        // gl_BaseVertex and only the second one's value survives in the framebuffer.
+        TEST_F(DrawParametersScenario, DrawElementsAfterABaseVertexDrawReportsZeroAgain) {
+            if (!Ready()) return;
+            std::vector<std::uint32_t> indices;
+            for (std::uint32_t i = 0; i < 2 * kHalfCount; ++i) indices.push_back(i);
+            for (std::uint32_t i = 0; i < 2 * kHalfCount; ++i) indices.push_back(i + kLeftFirst);
+            FillBuffer(m_ebo, GL_ELEMENT_ARRAY_BUFFER, indices);
+            const auto rebasedRun = reinterpret_cast<const void*>(2 * kHalfCount * sizeof(std::uint32_t));
+
+            const Image image = Render([&] {
+                glDrawElementsBaseVertex(GL_TRIANGLES, 2 * kHalfCount, GL_UNSIGNED_INT,
+                                         reinterpret_cast<const void*>(0), kLeftFirst);
+                glDrawElements(GL_TRIANGLES, 2 * kHalfCount, GL_UNSIGNED_INT, rebasedRun);
+            });
+            EXPECT_EQ(FirstGLError(), GLenum(GL_NO_ERROR));
+            ExpectParams(image, 0, {0, 0, 0}, "glDrawElements after a base-vertex draw");
+            ExpectParams(image, 1, {0, 0, 0}, "glDrawElements after a base-vertex draw");
+        }
+
+        // ---- the multi-draw forms: one gl_DrawID per sub-draw ----
+
+        TEST_F(DrawParametersScenario, MultiDrawArraysNumbersItsSubDraws) {
+            if (!Ready()) return;
+            const GLint firsts[2] = {kLeftFirst, kRightFirst};
+            const GLsizei counts[2] = {kHalfCount, kHalfCount};
+
+            const Image image = Render([&] { glMultiDrawArrays(GL_TRIANGLES, firsts, counts, 2); });
+            EXPECT_EQ(FirstGLError(), GLenum(GL_NO_ERROR));
+            ExpectParams(image, 0, {0, 0, 0}, "glMultiDrawArrays sub-draw 0");
+            ExpectParams(image, 1, {0, 0, 1}, "glMultiDrawArrays sub-draw 1");
+        }
+
+        // Every field of an indexed indirect command at once: its own gl_DrawID, the
+        // baseVertex word (which the CPU reads out of the command) and the
+        // baseInstance word (which DirectGLES reads through a storage-buffer view of
+        // the very same buffer).
+        TEST_F(DrawParametersScenario, MultiDrawElementsIndirectCarriesEveryCommandsParameters) {
+            if (!Ready()) return;
+            std::vector<std::uint32_t> indices;
+            for (std::uint32_t i = 0; i < kHalfCount; ++i) indices.push_back(i);
+            FillBuffer(m_ebo, GL_ELEMENT_ARRAY_BUFFER, indices);
+
+            const std::vector<ElementsCommand> commands = {
+                {kHalfCount, 1, 0, kLeftFirst, 0},
+                {kHalfCount, 1, 0, kRightFirst, 4},
+            };
+            FillBuffer(m_indirect, GL_DRAW_INDIRECT_BUFFER, commands);
+
+            const Image image = Render([&] {
+                glMultiDrawElementsIndirect(GL_TRIANGLES, GL_UNSIGNED_INT, reinterpret_cast<const void*>(0), 2,
+                                            sizeof(ElementsCommand));
+            });
+            EXPECT_EQ(FirstGLError(), GLenum(GL_NO_ERROR));
+            ExpectParams(image, 0, {kLeftFirst, 0, 0}, "indirect command 0");
+            ExpectParams(image, 1, {kRightFirst, 4, 1}, "indirect command 1");
+        }
+
+        // glMultiDrawArraysIndirectCount was missing from the DirectGLES backend
+        // table entirely, so the frontend answered INVALID_OPERATION for every call
+        // while GL_ARB_indirect_parameters was advertised. The parameter buffer here
+        // holds a count SMALLER than maxdrawcount, so a path that ignores it draws a
+        // third command over the top of the second and changes the right half.
+        TEST_F(DrawParametersScenario, MultiDrawArraysIndirectCountObeysItsParameterBuffer) {
+            if (!Ready()) return;
+            const std::vector<ArraysCommand> commands = {
+                {kHalfCount, 1, kLeftFirst, 0},
+                {kHalfCount, 1, kRightFirst, 6},
+                {kHalfCount, 1, kRightFirst, 9},
+            };
+            FillBuffer(m_indirect, GL_DRAW_INDIRECT_BUFFER, commands);
+            const std::vector<std::uint32_t> parameters = {2};
+            FillBuffer(m_parameter, GL_PARAMETER_BUFFER, parameters);
+
+            const Image image = Render([&] {
+                glMultiDrawArraysIndirectCount(GL_TRIANGLES, reinterpret_cast<const void*>(0), 0, 3,
+                                               sizeof(ArraysCommand));
+            });
+            EXPECT_EQ(FirstGLError(), GLenum(GL_NO_ERROR));
+            ExpectParams(image, 0, {0, 0, 0}, "counted indirect command 0");
+            ExpectParams(image, 1, {0, 6, 1}, "counted indirect command 1");
+        }
+
+    } // namespace
+} // namespace MGITest

MobileGL/MG_Test/SanityTest.cpp

@@ -346,8 +346,10 @@ void main() {
 
     EXPECT_NE(rewritten.find("int instance = mg_ZeroBasedInstanceID + mg_BaseInstanceLowered;"),
               MobileGL::String::npos);
-    EXPECT_NE(rewritten.find("#define mg_ZeroBasedInstanceID (gl_InstanceID - ((mg_BaseInstanceWordIndex >= 0) ? "
+    // One-based word index: zero is the "not an indirect draw" sentinel because that is
-                             "int(mg_indirectWords[uint(mg_BaseInstanceWordIndex)]) : 0))"),
+    // the value a GLSL uniform starts at and no draw path writes it before the first draw.
+    EXPECT_NE(rewritten.find("#define mg_ZeroBasedInstanceID (gl_InstanceID - ((mg_BaseInstanceWordIndex > 0) ? "
+                             "int(mg_indirectWords[uint(mg_BaseInstanceWordIndex - 1)]) : 0))"),
               MobileGL::String::npos);
     EXPECT_NE(rewritten.find(
                   "layout(std430, binding = 12) readonly buffer mg_IndirectParams { highp uint mg_indirectWords[]; };"),
@@ -356,6 +358,38 @@ void main() {
     EXPECT_EQ(CountOccurrences(rewritten, "gl_InstanceID"), 1u);
 }
 
+// The sentinel itself, on the builtin it exists for. A zero-based index with a
+// negative "off" value made every NON-indirect draw of such a program read
+// mg_indirectWords[0] out of a storage buffer nothing had bound - the uniform starts
+// at zero and no non-indirect draw path writes it - which is where the CTS
+// shader_draw_parameters cases lost their geometry on Adreno. Pinned as text because
+// this contract lives in two places at once: the generated ESSL below and the +1 that
+// BackendProgramObjectImpl::SetBaseInstanceWordIndex applies.
+TEST(DirectGLESSanity, TheIndirectWordIndexIsOneBasedSoItsUnwrittenValueMeansNotIndirect) {
+    const ScopedGLESCapabilitiesOverride capsGuard;
+    auto& caps = MobileGL::MG_Backend::DirectGLES::g_GLESCapabilities;
+    caps.IndirectDrawInstanceIdIncludesBaseInstance = false;
+    caps.MaxShaderStorageBufferBindings = 13;
+
+    const MobileGL::String source = R"(#version 310 es
+highp int mg_BaseInstanceLowered;
+void main() {
+    gl_Position = vec4(float(mg_BaseInstanceLowered));
+}
+)";
+
+    const auto rewritten = MobileGL::MG_Backend::DirectGLES::PromoteDrawParameterGlobalsToUniforms(
+        source, GL_VERTEX_SHADER);
+
+    EXPECT_NE(rewritten.find("#define mg_BaseInstanceLowered ((mg_BaseInstanceWordIndex > 0) ? "
+                             "int(mg_indirectWords[uint(mg_BaseInstanceWordIndex - 1)]) : mg_BaseInstance)"),
+              MobileGL::String::npos)
+        << rewritten;
+    // A zero-based form would spell either of these; neither may survive.
+    EXPECT_EQ(rewritten.find("mg_BaseInstanceWordIndex >= 0"), MobileGL::String::npos);
+    EXPECT_EQ(rewritten.find("uint(mg_BaseInstanceWordIndex)"), MobileGL::String::npos);
+}
+
 TEST(DirectGLESSanity, KeepsInstanceIdWhenIndirectDrawsAreConforming) {
     const ScopedGLESCapabilitiesOverride capsGuard;
     auto& caps = MobileGL::MG_Backend::DirectGLES::g_GLESCapabilities;

MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.cpp

@@ -21,6 +21,7 @@
 #include "SpirvPasses/LowerDrawParametersPass.h"
 #include "SpirvPasses/PackDoubleVertexInputsPass.h"
 #include "SpirvPasses/RebaseInstanceIndexPass.h"
+#include "SpirvPasses/ZeroBaseVertexPass.h"
 #include "SpirvPasses/NormalizeRectCoordinatesPass.h"
 #include "SpirvPasses/PrivateToEntryLocalPass.h"
 #include "SpirvPasses/StripUniformLocationsPass.h"
@@ -758,6 +759,15 @@ namespace MobileGL {
                                            outputBinary);
             }
 
+            bool ShaderCompiler::ZeroBaseVertexForVulkan(const Vector<Uint32>& inputBinary,
+                                                         Vector<uint32_t>& outputBinary) {
+                using namespace spvtools;
+                Optimizer optimizer(SPV_ENV_VULKAN_1_1);
+                optimizer.RegisterPass(ZeroBaseVertexPass::CreateZeroBaseVertexPass());
+
+                return RunOptimizerChecked("ZeroBaseVertexForVulkan", optimizer, inputBinary, outputBinary);
+            }
+
             bool ShaderCompiler::DecoratePositionInvariantForVulkan(const Vector<Uint32>& inputBinary,
                                                                     Vector<uint32_t>& outputBinary) {
                 using namespace spvtools;

MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.h

@@ -65,6 +65,12 @@ namespace MobileGL {
                 static bool LowerRectImages(const Vector<Uint32>& inputBinary, Vector<uint32_t>& outputBinary);
                 static bool RebaseInstanceIndexForVulkan(const Vector<Uint32>& inputBinary,
                                                          Vector<uint32_t>& outputBinary);
+                // Builds the non-indexed-draw variant of a vertex shader: every gl_BaseVertex
+                // read becomes zero, which is what GL defines for a command carrying no
+                // baseVertex parameter while Vulkan's builtin would report firstVertex.
+                // See ZeroBaseVertexPass.
+                static bool ZeroBaseVertexForVulkan(const Vector<Uint32>& inputBinary,
+                                                    Vector<uint32_t>& outputBinary);
                 // Re-declares 64-bit float vertex inputs as their 32-bit unsigned word pair
                 // (double -> uvec2, dvec2 -> uvec4) and bitcasts them back to double at entry, so no
                 // VK_FORMAT_R64*_SFLOAT is needed - lavapipe advertises none of them for vertex

MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/ZeroBaseVertexPass.cpp

@@ -0,0 +1,149 @@
+// MobileGL - MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/ZeroBaseVertexPass.cpp
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+
+#include "ZeroBaseVertexPass.h"
+
+#include "spirv.hpp"
+#include "source/opt/constants.h"
+#include "source/opt/def_use_manager.h"
+#include "source/opt/instruction.h"
+#include "source/opt/ir_context.h"
+#include "source/opt/module.h"
+#include "source/util/make_unique.h"
+
+#include <vector>
+
+namespace MobileGL {
+    namespace MG_Util {
+        namespace ShaderTranspiler {
+            namespace {
+                using spvtools::opt::Instruction;
+                using spvtools::opt::IRContext;
+                using spvtools::opt::Operand;
+
+                // Returns the Input OpVariable decorated with |builtin|, or nullptr if none.
+                Instruction* FindBuiltinInputVariable(IRContext* context, spv::BuiltIn builtin) {
+                    auto* defUseMgr = context->get_def_use_mgr();
+                    for (auto& annotation : context->annotations()) {
+                        if (annotation.opcode() != spv::Op::OpDecorate || annotation.NumInOperands() < 3) {
+                            continue;
+                        }
+                        if (static_cast<spv::Decoration>(annotation.GetSingleWordInOperand(1)) !=
+                            spv::Decoration::BuiltIn) {
+                            continue;
+                        }
+                        if (static_cast<spv::BuiltIn>(annotation.GetSingleWordInOperand(2)) != builtin) {
+                            continue;
+                        }
+
+                        Instruction* variable = defUseMgr->GetDef(annotation.GetSingleWordInOperand(0));
+                        if (variable == nullptr || variable->opcode() != spv::Op::OpVariable ||
+                            static_cast<spv::StorageClass>(variable->GetSingleWordInOperand(0)) !=
+                                spv::StorageClass::Input) {
+                            continue;
+                        }
+                        return variable;
+                    }
+                    return nullptr;
+                }
+            } // namespace
+
+            spvtools::opt::Pass::Status ZeroBaseVertexPass::Process() {
+                auto* irContext = context();
+                auto* defUseMgr = irContext->get_def_use_mgr();
+
+                Instruction* baseVertexVar = FindBuiltinInputVariable(irContext, spv::BuiltIn::BaseVertex);
+                if (baseVertexVar == nullptr) {
+                    return Status::SuccessWithoutChange;
+                }
+                const uint32_t baseVertexVarId = baseVertexVar->result_id();
+
+                // Collect every load before mutating: rewriting invalidates the use list.
+                //
+                // Every OTHER kind of user is enumerated and refused rather than ignored. A read
+                // that reaches the variable through a copied pointer or a pointer function
+                // parameter would keep Vulkan's firstVertex while the pass still reported
+                // success, i.e. a partial rewrite indistinguishable from a complete one. glslang
+                // emits neither shape from GLSL today, so this fails closed on something that
+                // cannot happen yet rather than silently half-doing it when it can.
+                std::vector<Instruction*> baseVertexLoads;
+                Bool sawUnexpectedUser = false;
+                defUseMgr->ForEachUser(baseVertexVar, [&](Instruction* user) {
+                    switch (user->opcode()) {
+                    case spv::Op::OpLoad:
+                        if (user->GetSingleWordInOperand(0) == baseVertexVarId) {
+                            baseVertexLoads.push_back(user);
+                        } else {
+                            sawUnexpectedUser = true;
+                        }
+                        return;
+                    // Declarations of the variable, not reads of it.
+                    case spv::Op::OpDecorate:
+                    case spv::Op::OpDecorateId:
+                    case spv::Op::OpDecorateString:
+                    case spv::Op::OpName:
+                    case spv::Op::OpEntryPoint:
+                        return;
+                    default:
+                        sawUnexpectedUser = true;
+                        return;
+                    }
+                });
+
+                if (sawUnexpectedUser) {
+                    return Status::Failure;
+                }
+                if (baseVertexLoads.empty()) {
+                    // Declared but never read - the variant is already the shader itself.
+                    return Status::SuccessWithoutChange;
+                }
+
+                // Materialize every zero constant BEFORE touching a single instruction, so the
+                // constant/type managers are never consulted against a module this pass has
+                // already half-rewritten - and so the rewrite loop below cannot fail partway
+                // and leave one behind.
+                auto* constantMgr = irContext->get_constant_mgr();
+                auto* typeMgr = irContext->get_type_mgr();
+                std::vector<uint32_t> zeroIds(baseVertexLoads.size(), 0);
+                for (size_t i = 0; i < baseVertexLoads.size(); ++i) {
+                    // The zero is built from the LOAD's own type, because a shader may declare
+                    // the builtin as either int or uint.
+                    const uint32_t typeId = baseVertexLoads[i]->type_id();
+                    const spvtools::opt::analysis::Type* type = typeMgr->GetType(typeId);
+                    if (type == nullptr) {
+                        return Status::Failure;
+                    }
+                    const spvtools::opt::analysis::Constant* zero = constantMgr->GetConstant(type, {0u});
+                    if (zero == nullptr) {
+                        return Status::Failure;
+                    }
+                    const Instruction* zeroInst = constantMgr->GetDefiningInstruction(zero, typeId);
+                    if (zeroInst == nullptr) {
+                        return Status::Failure;
+                    }
+                    zeroIds[i] = zeroInst->result_id();
+                }
+
+                // `OpLoad %ty %res %baseVertex` becomes `OpCopyObject %ty %res %zero`. Keeping
+                // %res makes every downstream use pick the zero up with no further rewriting.
+                for (size_t i = 0; i < baseVertexLoads.size(); ++i) {
+                    baseVertexLoads[i]->SetOpcode(spv::Op::OpCopyObject);
+                    baseVertexLoads[i]->SetInOperands(Instruction::OperandList{
+                        {SPV_OPERAND_TYPE_ID, {zeroIds[i]}}});
+                }
+
+                irContext->InvalidateAnalysesExceptFor(IRContext::kAnalysisNone);
+                return Status::SuccessWithChange;
+            }
+
+            spvtools::Optimizer::PassToken ZeroBaseVertexPass::CreateZeroBaseVertexPass() {
+                return spvtools::Optimizer::PassToken(MakeUnique<ZeroBaseVertexPass>());
+            }
+        } // namespace ShaderTranspiler
+    } // namespace MG_Util
+} // namespace MobileGL

MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/ZeroBaseVertexPass.h

@@ -0,0 +1,40 @@
+// MobileGL - MobileGL/MG_Util/ShaderTranspiler/SpirvPasses/ZeroBaseVertexPass.h
+// Copyright (c) 2025-2026 MobileGL-Dev
+// Licensed under the GNU Lesser General Public License v3.0:
+//   https://www.gnu.org/licenses/gpl-3.0.txt
+//   https://www.gnu.org/licenses/lgpl-3.0.txt
+// SPDX-License-Identifier: LGPL-3.0-only
+// End of Source File Header
+
+#pragma once
+#include "source/opt/pass.h"
+#include "spirv-tools/optimizer.hpp"
+
+#include <Includes.h>
+
+namespace MobileGL {
+    namespace MG_Util {
+        namespace ShaderTranspiler {
+            // GL and Vulkan disagree about gl_BaseVertex on NON-INDEXED draws: GL defines it as
+            // "the value passed to the baseVertex parameter, or zero for a command that has
+            // none", so every DrawArrays form reads zero, while Vulkan's BaseVertex builtin
+            // carries the draw's firstVertex there. (For indexed draws both mean the same thing,
+            // GL's basevertex / Vulkan's vertexOffset, so those must keep the native builtin.)
+            //
+            // This pass produces the non-indexed variant of a vertex shader by replacing every
+            // read of the BaseVertex builtin with a constant zero. The variable itself is left
+            // declared - removing it would also have to reason about the DrawParameters
+            // capability that a BaseInstance read in the same module still needs.
+            //
+            // Vulkan backend only, and only for the ZeroBaseVertex program variant: the
+            // DirectGLES path has no BaseVertex builtin at all (see LowerDrawParametersPass).
+            class ZeroBaseVertexPass : public spvtools::opt::Pass {
+            public:
+                const char* name() const override { return "zero-base-vertex"; }
+                Status Process() override;
+
+                static spvtools::Optimizer::PassToken CreateZeroBaseVertexPass();
+            };
+        } // namespace ShaderTranspiler
+    } // namespace MG_Util
+} // namespace MobileGL